diff --git a/file1.txt b/file1.txt new file mode 100644 index 0000000000000000000000000000000000000000..70e087bb0700f593d69525de8970e90e179b5e58 --- /dev/null +++ b/file1.txt @@ -0,0 +1,342 @@ + + + + +IntroductionThe Center-Terminal Radar Approach Control (TRACON) Automation System (CTAS) is a set of tools developed to help air traffic controllers manage complex air traffic flows to reduce delays and increase safety [1].These tools, known also as CTAS client applications, rely on the Trajectory Synthesizer (TS) [3,4] as the core computational engine for generating accurate predictions of 4Dtrajectories using the detailed aircraft performance characteristics, weather data, and data from the Host Computer System (HCS).TS outputs predicted aircraft position, altitude, and performance parameters, such as drag, thrust, weight, Rate of Climb or Descent (ROCD), and fuel consumption, as a function of time.CTAS currently supports more than four hundred aircraft types, but about 90% of them are supported indirectly by mapping to aircraft types with known drag and engine thrust models.Obtaining the detailed performance data for more aircraft types from manufacturers and validating them may be problematic and time-consuming.Besides that, previous research validating the accuracy of CTAS TS (see [5] and [6]) was limited to certain centers and aircraft types, and the overall accuracy of CTAS predictions remained largely unknown.On the other hand, the Base of Aircraft DATA (BADA) supports more than 300 aircraft types, including about a hundred aircraft types with extensively validated performance data.BADA is an Aircraft Performance Model (APM) developed and maintained by Eurocontrol and available free of charge.This model is documented in the BADA User Manual [7].The complementary "Base of Aircraft DATA (BADA) Aircraft Performance Modelling Report" [8] provides further definition of BADA parameters.Note that BADA documentation [7] does not always make a clear distinction between the physical and operational components.For instance, BADA thrust model blends together the physical parameters, such as maximum engine thrust, and operational considerations like the flight configuration or using reduced climb power.Note also that BADA does not include any official software implementation, but it provides the data that can be used to verify correctness of any software that implements the BADA APM.The main potential advantage of using BADA is that it directly supports more aircraft types in comparison with CTAS and may provide more accurate aircraft parameters for some aircraft types currently supported by CTAS.Besides that, some elements of the BADA operational model, such as default speeds and the use of reduced climb power, may be based on more realistic assumptions.Another important benefit of using BADA is that it is maintained by Eurocontrol and it is expected to be regularly updated and improved over time.To take full advantage of these benefits, BADA APM was integrated in CTAS TS to augment the "native" CTAS APM for aircraft types not supported by CTAS or having the better models in BADA.A set of validation tools was developed and used to judge the accuracy of BADA APM and CTAS APM by comparing trajectory predictions built using both performance models to radar track and other reference data.Software design for BADA-CTAS Integration is presented in Section 2 of this TM.Section 3 describes the tools developed to validate the correctness of software implementation and the accuracy of BADA APM in comparison with CTAS APM.Section 4 briefly describes the results of software validation.Section 5 describes the results of evaluation of CTAS TS with "native" CTAS and integrated BADA model for various aircraft types and operational parameters.The TM is concluded with Section 6, summarizing the main findings.The Appendix includes examples of BADA Operation Performance Files (OPF) and Performance Table Data (PTD) files. +BADA-CTAS IntegrationCTAS TS with support for BADA will be referenced in this TM as "CTAS/BADA TS".The term "CTAS TS" will refer to CTAS/BADA TS using only "native" CTAS APM, which is functionally equivalent to the "original" CTAS TS without support for BADA APM.CTAS TS is a faster-than-real-time trajectory predictor that can be called for every radar track hit, or every 12 seconds, for each of thousands flights managed by CTAS tools.So, the main design requirement was to minimize performance penalty introduced by adding support for both CTAS and BADA models.Also, it was desirable to minimize CTAS TS code changes, to separate clearly the BADA-specific code from the rest of CTAS TS, and to make new code consistent with existing CTAS TS infrastructure.• Easy switch between CTAS "native" and BADA APM by changes in user-defined configuration files; • BADA can be used for specific aircraft types or for all aircraft types; • Each element of BADA APM (drag model, thrust model, speed profile, etc.) can be turned on/off and tested separately; • Some parameters of BADA APM, such as aircraft weight, can be specified as aircraft-specific user-defined values; • The weight of aircraft can be specified as a constant or variable.In the latter case, the weight is calculated as a function of time based on fuel consumption as described in [9]; • BADA thrust model was extended to allow the variable power settings to control the engine thrust in response on client speed requests, such as speed advisories for trial planning.Specifically, the maximum climb power is used for acceleration, and the BADA idle-thrust is used for deceleration in any flight phase; • Fuel consumption is always based on instantaneous thrust value, even if it deviates from nominal BADA thrust due to client requests; • Speed profile can be defined as CTAS, BADA "exact" with nominal BADA speeds defined in [7], and BADA "hybrid" with BADA speeds that can be updated by speed requests from CTAS client applications; • Some BADA options defined as "global" in [7], such as reduced climb power or expedited descent, can be defined as aircraft-specific in CTAS/BADA TS.This rich functionality is controlled by two BADA-CTAS Configuration Files, loaded by CTAS at startup time:• data sources, which specifies the available data sources (currently CTAS and BADA 3.8, but other versions of BADA can be added later); • aircraft data sources, which defines the additional options for each element (sub-model) of APM.If BADA APM is requested, the following BADA Input Files will be used (see [7] for details):• SYNONYM.NEW: the ICAO codes and prefixes for files with aircraft-specific data;• BADA.GPF -Global Parameters Files, containing the values for BADA global parameters;• OPF files -Operation Performance Files (one file for each aircraft type, see Appendix A);• APF files -Airline Procedures Files (one file for each aircraft type).Typical operation of CTAS/BADA TS includes three steps:1. Read BADA-CTAS configuration files; 2. Read CTAS common aircraft data files and BADA data files, and use the data from configuration files to instantiate CTAS or BADA implementations for each APM element; 3. Generate trajectories using CTAS and/or BADA APM as defined in configuration files.The trajectories generated by CTAS/BADA TS may be saved in archive files for further analysis and comparisons with radar track and other benchmark data using validation tools described in next section. +Validation Tools and MethodsThe goals of validation are two-fold.First of all, it should be confirmed that implementation of BADA APM in CTAS/BADA TS is correct by comparison with benchmark data, provided by Eurocontrol in the form of "Performance Table Data" (PTD) files (see [7] and Appendix B).Next, it should be examined how the usage of BADA APM will help improve CTAS.This second goal can be accomplished by running CTAS/BADA TS with CTAS APM and BADA APM side by side and comparing their predictions to each other and to radar track data.For a first goal, the PTD validation tool was developed to validate the implementation of BADA APM by reporting how closely the results produced by the current version of CTAS/BADA TS match the reference data from PTD files.For these comparisons linear interpolation on the TS output is used to get data at flight levels, specified in BADA PTD files.For a second goal, namely, validation of CTAS APM and BADA APM, a number of other tools was created.Two of them are briefly considered in this section:1. TrackComparer tool 2. CmSimTrackComparer tool TrackComparer tool uses methodology similar to what was described in [6].It combines three different analysis techniques:• Systematic analysis -Comparison of every predicted position with actual track data, and of predictions against each other, provided that data can be compared for a specified minimum time or path distance.This analysis generates systematic error statistics for all aircraft types, but it does not reflect the evolution of error along track.• Interval-based sampling technique (IBST) -Samples track data and predictions at regular intervals along the track, as described in [10].• Separation assurance trajectory accuracy algorithm (SA analysis) -Computes metrics at a specific lookahead time starting at (user-defined) climb and descending altitudes.SA analysis is a truncated variant of IBST that computes metrics for only one (first) prediction for each track.This makes it less computationally expensive than IBST, at the cost of a reduced size of data set.The tool requires the following inputs:• Cm sim file, generated by CTAS using HCS data.It contains flight plans, flight plan amendments, and radar tracks.• Trajectory archive files for trajectories generated by CTAS/BADA TS.These files are used for comparison with track data.CmSimTrackComparer tool was originally developed for comparing CTAS/BADA TS predictions against track data for idle-thrust descents with controlled cruise and descent speeds that were recorded for Denver arrivals in 2009, studied in [11].The tool can also be useful for other CTAS data analysis/validation tasks.It combines the following functions:• automatically generate requests to TS from radar track data;• invoke TS for each request;• compare generated predictions against tracks.In contrast to TrackComparer, CmSimTrackComparer has only one required input, a cm sim file, that makes possible to run it without running CTAS.Optional input text files can be used to limit analysis to particular flights and to specify estimated values for descent speeds and aircraft weights.On the other hand, CmSimTrackComparer compares track data to predictions generated using only one APM, either CTAS or BADA, while TrackComparer can simultaneously compare CTAS predictions to track data, BADA predictions to track data, and BADA predictions to CTAS predictions.Both TrackComparer and CmSimTrackComparer tools use the metrics defined in [10]: along-track error (time-correlated):∆AT t = U • V V(1)along-track error (path-correlated):∆AT p = t p -t t(2)cross-track error (time-correlated):∆CT = U × V V(3)altitude error:∆h = h t -h p (4)where t is a time, h is an altitude, and the subcripts p and t designate the path-based and timebased correlations.The vectors U and V are based on the position of aircraft denoted as AC and a trajectory segment between the points T J 1 and T J 2 as shown on Figure 2.All metrics, except for the along-track error for the path distance correlation, have the characteristic that the predicted trajectory is subtracted from the reference (or track).For BADA predictions compared against CTAS, the reference is always the CTAS prediction.Besides these metrics, TrackComparer calculates the metrics for "events," such as the Top of Climb (TOC), the Top of Descent (TOD), and the Estimated Time of Arrival (ETA).In this study the Top of Climb is defined as the end of the longest procedural climb in the flight or prediction, and the Top of Descent is defined as the start of the longest procedural descent in the flight or prediction.A procedural climb is defined as a non-decreasing sequence of altitudes with at least one strictly increasing sequence of altitudes, and procedural descent as a non-increasing sequence of altitudes with at least one strictly decreasing sequence of altitudes.For arriving flights, CTAS makes a clear distinction between Center predictions originating in the ARTCC and ending at the meterfix, and TRACON predictions within the TRACON.Consequently, the metrics for ETA are defined differently for these two classes of predictions.For a Center For each comparison (e.g. for each prediction-reference pair) the maximum, minimum, mean, and absolute mean errors are calculated and stored in SQL-lite database.Then the maximum, mean, minimum, and standard deviation are calculated over each of these metrics separately for CTAS vs. Track, BADA vs. Track, and BADA vs. CTAS comparisons.For example, the "mean of absolute mean errors" is the mean of absolute mean errors for all predictions.Other metrics used in this TM are defined similarly. +CTAS/BADA TS Software ValidationImplementation of BADA APM in CTAS/BADA TS was validated using the PTD validation tool with benchmark PTD files as of the 5th of May 2011.This validation was performed only for BADA "exact" speeds and constant weight, since the PTD files were generated for these conditionsThe following two metrics were considered: a trajectory success rate, and a comparison success rate.The trajectory success rate is defined as a percentage of trajectories successfully generated by CTAS/BADA TS to all trajectory requests.CTAS/BADA TS may fail to generate trajectories for certain requests because of unrealistic constraints or aircraft performance data or due to limitations imposed by implementation of TS computational algorithms.Over all considered aircraft types, e.g.all types for which the BADA provided the PTD files, the PTD validation gives 99.8% trajectory success rate.The comparison success rate is defined as a percentage of successful comparisons to all comparisons.Comparison is considered successful if the mean absolute errors for all trajectory parameters are less than the specified tolerances listed in Table With these tolerances a comparison success rate with the PTD reference was 78.3%.The majority of failed comparisons were due to thrust or ROCD.These failures occurred because "the PTF/PTD files do not take into account the flight envelope limitations", as the BADA support group acknowledged responding on request by the authors of this TM.Therefore, the tables in BADA PTD files include the data up to the maximum altitudes with a negative rate of climb indicating that aircraft cannot climb up to that altitudes.CTAS TS copes with this situation by increasing thrust to achieve a minimum energy rate required for climb, while ROCD is calculated from the Newton's equation of motion and can never be negative in climb.Therefore, the BADA implementation in CTAS/BADA TS disagrees with BADA PTD reference for conditions that are not physically possible.Increasing the thrust tolerance to 2000 Newtons and the ROCD tolerance to 200 feet per minute resulted in an increased comparison success rate of 90.5%.All remaining failures were due to failed comparisons in drag.All data in BADA PTD files are expressed on the coarse grid as functions of the flight level (FL) (see Appendix B).Hence, comparisons of CTAS/BADA TS results against the PTD data require interpolation between this coarse grid and the the grid used by CTAS/BADA TS.The lift coefficient, and hence the drag, appear to be especially sensitive to these interpolation errors.Other parameters, such as altitude and CAS, were not affected by these errors. +Evaluation of CTAS/BADA TSBADA documentation [7] already allows certain variability in some elements of BADA APM.For example, a user can choose to use full or reduced climb power, standard descent or expedited descent with spoilers, and so on.Implementation of BADA APM in CTAS/BADA TS further extends this variability.Hence, any discussion of CTAS/BADA TS evaluation results requires specification of "the variant" of BADA APM used.First, the definitions of CTAS and BADA speed profiles should be considered.CTAS TS uses aircraft-specific default speeds, but they can be updated from client speed requests and the initial speed from track.On another hand, BADA User Manual, [7], defines aircraft-specific speeds for all flight phases, which will be called the "BADA exact" speed profile.In addition to this, CTAS/BADA TS supports the "BADA hybrid" speed profile, defined as BADA exact speeds, replaced when necessary with speeds requested by CTAS clients.A variant of "BADA hybrid" speed profile, which will be referred in this TM as "BADA hybrid hold," holds initial speeds as CTAS TS does.Therefore, the "BADA hybrid hold" speed profile is similar to CTAS speed profile, except the definition of default speeds: the BADA "exact" speeds from [7] are used as default speeds for BADA, while for CTAS the default speeds are explicitly specified for each aircraft type.Next, the "default" initial aircraft weight should be defined, which is based on BADA reference weight, corrected to account for a starting point of aircraft.CTAS TS does not have access to flight history before the first prediction point, so the method of initial weight correction based on estimated fuel burn, suggested in [12] and [13], could not be used.Therefore, a more practical, although less accurate, correction technique was necessary.This technique used heuristics, taking into account a starting point of aircraft.Effectively, these heuristics ensured that the initial weight gradually approached the BADA maximum weight at the beginning of climb, and the "landing weight", defined as BADA empty weight increased by 20%, at the end of descent.The code also ensured that the actual aircraft weight could not fall below the "landing weight".BADA Base case can be defined as follows:1. BADA thrust model; 2. BADA drag model; 3. BADA weight model that implies the weight starting from "default" initial weight and changing as a function of consumed fuel; 4. Reduced climb power, which will give more realistic climb profile than the maximum thrust does (see [7], p.24); 5. BADA "exact" speed profile (requests are ignored); 6. BADA model limits (trajectories assumed invalid if BADA limits are violated).Therefore, the BADA Base case can be seen as the most "authentic" implementation of [7].All other BADA test cases are built upon the BADA Base case as shown in Table 2. Examining these cases allows to study the effects of extensions of BADA APM and, to some degree, the effects of uncertanties in speeds, weight, thrust, and drag.Note that only the BADA Base case can be considered as a "nominal" case, for which the BADA APM was validated by the Eurocontrol.All other cases are "off-nominal" models that require additional validation by comparisons of predicted trajectories against track data as described in this section.All comparisons were made using BADA version 3.8 and RUC weather files with 40km grid for corresponding days and locations. +Fort Worth Center (ZFW), 3 daysThe TrackComparer tool, briefly described in previous section, was used to compare radar track data with predictions obtained using CTAS/BADA TS for different BADA variants for 3 days of traffic from Feb 23 to Feb 25 of 2011 in Fort Worth Center (ZFW).Tables 3 and4 summarize the performance of CTAS and various BADA variants for time-based correlations.Each row in these tables includes the averaged (over all predictions) absolute values of maximal errors, mean errors, absolute values of mean errors, maximal errors, minimal errors, and standard deviations of metrics defined in Section 3.As can be seen from Tables 3 and4, all BADA variants perform as well or slightly better than CTAS in terms of altitude and along-track errors.However, the differences between BADA and CTAS and between different BADA variants in mean of absolute mean errors do not exceed a few percent of error magnitudes, except for BADA Hybrid Hold variant that shows the most noticeable advantage over CTAS in terms of mean of absolute mean along-track errors.The differences between CTAS and BADA become more pronounced when plotted for specific aircraft types and flight phases.Note that the aircraft-type-specific data on all histograms are shown in the order of decreasing aircraft-type frequency from the left to the right, so the data for most frequent aircraft types always appear on the left side.All plots below for along-track and altitude errors are given for time-based correlations.The plots for cross-track errors are not included, since these errors were found to be practically the same for BADA and CTAS.This should be expected since the cross-track errors are mostly related to intent and flight plan amendments, and so they are not sensitive to a particular APM. Figure 3 shows the most frequently observed aircraft types for Fort Worth Center in comparisons between CTAS and BADA Base case.Figures 4 and5 show that overall mean altitude error distributions for all flight phases together are similar for CTAS and BADA.However, Figure 6 indicates that BADA APM may reduce the bias in altitude predictions when compared with CTAS APM because of lower mean of mean altitude errors for 20 of 40 aircraft types and for 6 of 10 most frequent aircraft types.Furthermore, as can be seen from Figure 7, BADA may significantly improve the altitude prediction accuracy of CTAS TS for certain aircraft types, such as CRJ7, CRJ9, BE9L, and E145.Figure 8 shows that using BADA Base case may also help reduce the bias in the along-track error for several aircraft types, including A320, B737, B738, B752, CRJ7, CRJ9, and E145.This may indicate that BADA "exact" speeds can be more realistic than the default CTAS speeds.The event errors are plotted on Figures 9 through 14.These results show an improvement in the TOC time accuracy and generally similar results for TOD time accuracy.The results for ETA indicate that BADA APM may perform slightly better in predicting arrival times for many aircraft types, but for certain aircraft types (E135, BE40, BE36, C182) the BADA ETA errors are much higher when compared with CTAS.Figures 15 through 17 show the IBST analysis results for climb prediction errors expressed as a function of look-ahead time, which is defined as the time interval between the sample time and the future time at which the prediction is made.The plots indicate that BADA may help improve the altitude prediction accuracy in climbs.This can be seen from the fact that the mean altitude error remains essentially flat along the track, and the mean absolute and standard deviation of altitude error are lower than in CTAS for the whole range of look-ahead times.At the same time, in descents BADA underperforms in the interval-based sampling results for mean, mean absolute and standard deviation of altitude error, shown on Figures 18 through 20.Note that the graphs do not start from zero errors because of the logic of the IBST that does not use the track position as the initial position for comparisons between track and predicted points.Instead, the IBST selects a first (most recent) eligible prediction for each track that might be generated some time before a sample point (see [10]).This time difference can be especially large if some predictions after the "first eligible prediction" fail.It is instructive to look at results for the TRACON area only, e.g.excluding en-route predictions.In the TRACON CTAS TS generates predictions only for arriving flights.Modeling of TRACON trajectories is challenging since the arriving aircraft changes its altitude and speed in wide limits, makes frequent turns, uses variable thrust, and deploys flaps, spoilers, and the landing gear affecting the drag.CTAS TS copes with these challenges using the kinematic modeling specifically taylored for arrivals [2].In contrast, BADA APM is kinetic for all flight phases, including the terminal approach and landing.Figure 21 shows most frequent aircraft types with more than 10 flights for subset of flights including TRACON arrivals only.As can be seen from comparison to Figure 3, restricting the data set to the TRACON arrivals results in different most frequent aircraft types, with increased gap between MD82 and other aircraft and significantly reduced fraction of B737.These changes are not unexpected since CTAS is Center-specific, and the flights through particular ARTCC to other Centers can be performed by different airlines using different aircraft.Figures 22 and23 show that BADA may reduce the altitude bias for several most frequent aircraft types in the TRACON, although the mean of absolute mean altitude errors remains roughly the same as in CTAS TS.This finding may not hold for all TRACONs in all Centers, but it is still significant since it proves that BADA kinetic model can outperform the kinematic model, used by CTAS TS in TRACON, [2].All results discussed so far were obtained using the BADA Base case.The BADA Hybrid Hold variant performs similarly, but in the TRACON it reduces the values of mean of mean altitude errors for almost all most frequent aircraft types (except most of Boeing aircraft types), as shown in Figure 24.The improvement over BADA Base case in TRACON area can be attributed to the fact that in BADA Hybrid Hold case the BADA default speeds can be changed by client requests.This advantage becomes even more apparent from comparisons with Denver 2009 descent data (see section 5.4 below).For this reason, for other centers the BADA Hybrid Hold variant was used rather than the BADA Base case.The BADA Expedited Descent variant achieves even more reductions in altitude errors in the TRACON.As can be seen from Figure 25 the mean of mean altitude errors is significantly lower then in CTAS TS for most aircraft types including MD82, MD83, B752, E135 and E145.However, for some aircraft types the mean of mean altitude errors is much larger, as in the case of B733 and B737.The plots for BADA reduced and increased weight variants, shown in Figures 26 through 31 compared with results for BADA Base case with nominal weight in Figures 6 and7, demonstrate the effect of weight uncertainty on prediction accuracy in climb.Again, these plots show the different ordering of aircraft types, because some flights can fly over the center and so the most frequent aircraft types for departing flights are not necessarily the same as for all flights.The reduced weight variant (Figures 26 and29) shows a very good improvement over the BADA Base case in climbs.We can see that in contrast to the BADA Base case, the reduced weight variant performs better than CTAS, rather than worse, for the mean of mean altitude errors for MD82, MD83, E135, B738, and B737 aircraft types.The mean of absolute mean altitude errors have also improved for many aircraft types.The effect of increased weight in climb is less pronounced as can be seen from Figures 28 and 31 It can be noted that CTAS default aircraft weights can also be adjusted to reduce errors of CTAS APM, although this is beyond the scope of this study.One especially interesting BADA variant is the BADA with CTAS speeds since its comparison with CTAS APM allows to study separately the effect of the aircraft performance model for the same speed profile.It is clear from Figures 32 and 33 that the BADA with CTAS speeds reduced the altitude errors for most frequent aircraft types, except CRJ7 and CRJ9.Moreover, for aircraft types with largest altitude errors in CTAS, such as E145, E135, BE20, and BE9L, the BADA with CTAS speeds yields substantially better accuracy.These results demonstrate that the BADA APM is superior for the vast majority of aircraft types, and especially for smaller aircraft types, such as E145 and E135.At the same time, comparison of these plots with Figures 6 and7 does not show significant advantage of using BADA speeds (e.g.Base case variant) over CTAS speeds in terms of altitude accuracy.Aircraft Type +Los Angeles Center (ZLA), 1 dayUsing the same methodology, the radar track data for 1 day of traffic in Los Angeles center (ZLA) were compared with CTAS TS and BADA Hybrid Hold predictions.Table 5 summarizes results obtained from these comparisons: For this center BADA does not have clear advantages over CTAS in terms of altitude and alongtrack errors.This can be partially explained by the different aircraft type decomposition with most frequent aircraft type B737 rather than MD82, as evident from Figure 34.The results plotted on Figures 35 and36 show that for this center BADA clearly outperforms CTAS in terms of altitude errors for A319, CRJ7, and CRJ9.In TRACON area, BADA altitude errors are much lower in comparison with CTAS only for two aircraft types -A319 and CRJ2, and consistently higher for all other aircraft as becomes apparent from Figures 37 and38.Aircraft Type +Denver Center (ZDV), 5 daysThis sub-section discusses the performance of the BADA Hybrid Hold variant against CTAS APM for January 31 st through February 4 th 2011 in Denver Center.This time period was characterized by variable weather, including days with rain and snow, low ceilings, and strong winds.Therefore, it was important to verify whether the comparison results for these conditions are consistent with results for two other centers presented above.Clearly for this center BADA does not have any advantages over CTAS in terms of altitude errors, and it shows only minor improvements in terms of along-track errors.Once again, Figure 39 shows a different type decomposition; the most frequently observed aircraft types were the B737, A320, B752, and A319. +Aircraft TypeFigure 39.Most frequent aircraft types in Denver centerFigures 40 and41 show uneven results for altitude errors, with BADA being substantially more accurate for several aircraft types (CRJ7, E170, B190, BE20, C750, MD82, BE9L), and CTAS much more accurate for other aircraft types (CRJ2, C560, C56X, E120, LJ35).However, the overall BADA performance is affected by the fact that BADA is a little less accurate for the four aircraft types most frequently observed in Denver center: B737, A320, B752, and A319.It is interesting to note that BADA is much more accurate than CTAS in predicting TOC for the vast majority of the most frequently observed aircraft types.This is consistent with results for Los Angeles Center and for several BADA variants for Fort Worth Center as well (compare, for instance, Figure 9 with Figure 42).We could have seen already from comparison between Figures 24 and 22 that the BADA Hybrid Hold variant showed a similar or even better performance than the BADA Base case.This finding could be seen as an indirect evidence of relative insensitivity of BADA accuracy to limited deviations from nominal BADA speeds, allowed by BADA Hybrid Hold speed profile.However, it was desirable to study the effect of off-nominal speeds for better controlled conditions with reduced uncertainty in aircraft weight, speed, and pilot intent.This was done by utilizing the data collected in Denver in 2009, which included descending flights performed according to predefined descent profiles.The data set for Denver-2009 included, in addition to cm sim files with radar tracks, the weather Rapid Update Cycle (RUC) files, the data for advisory descent speeds and, for some flights, the actual descent weights.This is the same data set that was used in study [11].We used for this analysis the CmSimTrackComparer tool, which generated the predictions and calculated their errors for all tracks from cm sim files for the filtered set of flights.Filtering was necessary for two reasons:• The flights with missing/inconsistent data, e.g.all flights not listed in the input files given to us for Denver-2009 data set, had to be excluded, • CmSimTrackComparer does not parse the route from the flight plan, so the flights with cross-track error exceeding 10 nm were filtered out in order to analyze only the flights close to direct-to routes.After this filtering our analysis included more than 40 thousand predictions for 256 flights.Tables 7 and8 summarize the comparison results for path-and time-based correlations: As can be seen from these tables, the altitude errors for time-based correlations are significantly higher than for path-based correlations.This is not surprising because location and altitude of the final prediction point in our analysis were determined from the meterfix position and crossing altitude, while the meterfix crossing time was affected by along-track errors.However, even for path-based correlations the BADA Base case is substantially less accurate than CTAS or BADA Hybrid Hold.This is explained by the fact that ignoring the commanded descent, descent and control speeds results in inaccurate TOD positions and hence in large altitude errors.This can be illustrated by plots for one flight of Boeing 757-200 aircraft, shown on Figures 43 and44.It is interesting to note that the BADA Hybrid Hold model performs almost as well as the CTAS model, with the mean absolute altitude error and standard deviation of altitude error being slightly larger.This can be explained by noting that the advised descent and initial Mach do not differ much from the BADA recommended speeds, hence the altitude profile becomes more accurate.This can be observed from the plots shown on Figures 45 and46 for the same flight.Further analysis is required to determine if BADA accuracy will suffer for off-nominal conditions.These limited results indicate that the performance of BADA APM in idle descents may be similar or slightly worse in comparison with CTAS APM for advised speeds that do not differ significantly from the BADA advised speeds. +ConclusionThe BADA performance and operational model was integrated into the CTAS TS, the core computational engine of CTAS software.The integrated CTAS/BADA TS software incorporated all possible variants of the BADA model, along with the native CTAS model.The CTAS/BADA TS software was thoroughly validated by comparison of predictions for the BADA Base case with the BADA PTD file reference and other available benchmark data.To assess the accuracy of CTAS/BADA TS for different variants of BADA APM, several validation tools for interval-based sampling and systematic analysis were developed and used to perform the analysis for three different centers, Fort Worth, Los Angeles, and Denver, and for several BADA variants over Fort Worth center.The following observations can be made from these comparisons:• In general, some BADA variants perform as well or slightly better than CTAS in terms of altitude and along-track errors.• Cross-track errors are mostly related to intent errors and flight plan amendments, so they remain practically the same for BADA and CTAS.• Overall the differences between BADA and CTAS in terms of mean of absolute mean errors are relatively small (a few percent of error magnitudes) and statistically insignificant.• There is no one BADA variant that would improve performance for all flight phases and all aircraft types.• BADA has more accurate models for many aircraft types, such as AT72, BE20, BE36, BE9L, CRJ7, E145, PC12, and SR22.These are mostly small or regional aircraft.• CTAS has more accurate models for A319, A320, B190, B733, B737, B738, C182, C56X, and E120.Most notably these aircraft types include large commercial aircraft from Airbus and Boeing.Boeing aircraft types that have very good models in CTAS.However, BADA performance for descents can be improved by using the BADA Hybrid Hold variant with speeds responding to client requests, or the BADA Expedited Descent reducing the altitude errors for most aircraft types in the TRACON.• Most BADA variants fail more frequently than CTAS, mainly due to inconsistency in definition of maximum operating altitude in BADA, where it is defined as aircraft-specific, and in CTAS TS, where the value of 60, 000 ft is used for all aircraft types.Our performance metrics did not include fuel consumption.It is known that the BADA fuel consumption model works well in cruise, but does not perform in climb and descent as accurately compared with cruise (see [14]).The analysis of BADA off-nominal performance with controlled cruise and descent speeds confirmed that• the BADA Base Case was substantially less accurate for descents than CTAS;• the BADA Hybrid Hold variant performed almost as well as CTAS TS did.Finally, on the basis of this limited analysis we can conclude that using BADA APM with "hybrid hold" speed profile and expedited descent can improve the accuracy of CTAS for small and regional aircraft types.More substantial improvement in prediction accuracy can be expected using adaptive estimation of aircraft parameters, such as speed and weight, from historical track data for each flight. +Appendix BExample of the BADA PerformanceThese requirements motivated the design of container class APMDef, encapsulating CTAS and BADA APMs in abstract model definition interfaces, such as ADragDef, AThrustDef, etc., as shown in Figure1.Here "[CTAS TS]" denotes any classes of "original" CTAS TS reused in CTAS/BADA TS code.In particular, VSFixed is the base CTAS TS class for all vertical solution classes, and DragModel and ThrustModel are the existing CTAS TS classes modeling drag and engine thrust.Each model definition interface member variable in APMDef class, such as mDragDef and mThrustDef, is instantiated only once to CTAS or BADA implementation in constructors of corresponding Model Definition subclasses for each aircraft type based on BADA configuration files.Hence, extending CTAS TS for BADA APM does not introduce any performance penalty in run time. +Figure 1 .1Figure 1.CTAS/BADA TS Class Diagram +Figure 2 .2Figure 2. Horizontal Trajectory Prediction Metrics +compared with Figures 27 and 30 for BADA Base case with nominal weight.This can be explained by inverse dependency of ROCD, and hence the altitude, on aircraft weight, making the altitude errors less sensitive to weight when the weight increases.Using the BADA reduced climb power further reduces this sensitivity as can be seen from equations (3.8-1) and (3.8-2) in [7]. +Figure 3 .Figure 4 .Figure 5 .Figure 6 .Figure 8 .Figure 9 .Figure 13 .34568913Figure 3.Most frequent aircraft types for Fort Worth center +Figure 36 .36Figure 36.Mean of absolute mean altitude errors in Los Angeles center +Figure 40 .40Figure 40.Mean of mean altitude errors in Denver center +Figure 43 .43Figure 43.Boeing 757-200 BADA Base Case idle descent: ground speed + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +Table 1 .1PTD Comparison TolerancesParameterToleranceTrue Air Speed (TAS)5 knotsCalibrated Air Speed (CAS) 5 knotsMach0.1Thrust1000 NewtonsDrag1000 NewtonsROCD100 feet per minute1: +Table 2 .2BADA APM variantsCTAS speedsBase case, but CTAS speed profileHybrid HoldBase case, but BADA hybrid hold speed profileIncreased WeightBase case, but weight ratio = 110% (10% above "default" weight)Reduced WeightBase case, but weight ratio = 90% (10% below "default" weight)Maximum Climb ThrustBase case, but without reduced climb powerExpedited DescentBase case, but with expedited descentReduced Climb and Descent Maximum climb thrust, but with expedited descent +Table 3 .3Mean altitude errors (ft) for CTAS/BADA TSTest caseComparisonAbs Max Mean Abs Mean MaxMinStd DevBase CaseCTAS vs. Track 25592 BADA vs. Track 25834-267 -176938 89925592 -21308 1932 25834 -21308 1874CTAS SpeedsCTAS vs. Track 22831 BADA vs. Track 24782-281 -223867 84622831 -19415 1835 24782 -19415 1808Hybrid HoldCTAS vs. Track 24506 BADA vs. Track 25109-280 -213898 86624506 -19763 1882 25109 -19763 1837IncreasedCTAS vs. Track 25592-26792725592 -19415 1917WeightBADA vs. Track 25834-17388325834 -21170 1851Reduced WeightCTAS vs. Track 27250 BADA vs. Track 27250-97 -961095 108327250 -19763 2245 27250 -19763 2228MaximumCTAS vs. Track 25592-26392325592 -21308 1916Climb ThrustBADA vs. Track 25834-19689125834 -21308 1867ExpeditedCTAS vs. Track 25592-26793725592 -21308 1932DescentBADA vs. Track 25834-23890325834 -21308 1874Reduced ClimbCTAS vs. Track 25479-26792725479 -19943 1922and DescentBADA vs. Track 25834-26489825834 -21170 1867 +Table 4 .4Mean along-track errors (nmi) for CTAS/BADA TSTest caseComparisonAbs Max Mean Abs Mean Max MinStd DevBase CaseCTAS vs. Track 129.87 BADA vs. Track 129.860.2 0.12.7 2.6145.47 -129.87 4.73 45.47 -129.86 4.52CTAS SpeedsCTAS vs. Track 129.71 BADA vs. Track 129.730.61 0.772.43 2.3545.48 -129.71 4 45.48 -129.73 3.81Hybrid HoldCTAS vs. Track 129.71 BADA vs. Track 129.730.38 0.492.55 2.245.48 -129.71 4.22 45.48 -129.73 3.59IncreasedCTAS vs. Track 129.870.172.745.47 -129.87 4.68WeightBADA vs. Track 131.32-0.03 2.7845.47 -131.32 4.63Reduced WeightCTAS vs. Track 129.87 BADA vs. Track 129.86-0.07 3.09 -0.19 3.0545.47 -129.87 5.62 45.47 -129.86 5.52MaximumCTAS vs. Track 129.870.192.7345.47 -129.87 4.85Climb ThrustBADA vs. Track 129.860.052.6545.47 -129.86 4.65ExpeditedCTAS vs. Track 129.870.182.745.48 -129.87 4.7DescentBADA vs. Track 129.860.042.6645.48 -129.86 4.52Reduced ClimbCTAS vs. Track 129.870.22.745.48 -129.87 4.68and DescentBADA vs. Track 131.32-0.03 2.845.48 -131.32 4.64 +Table 5 .5Mean altitude and along-track errors for BADA Hybrid Hold for Los Angeles centerError typeComparisonAbs Max Mean Abs Mean MaxMinStd DevAltitude (ft)CTAS vs. Track 25960 BADA vs. Track 25598797 8631966 196322475 -25960 3507 19756 -25598 3492Along-trackCTAS vs. Track 122.473.667.3554.26 -122.47 8.9(nm)BADA vs. Track 125.943.836.7351.59 -125.94 8.37 +Table 66summarizes the comparison results for Denver center: +Table 6 .6Mean altitude and along-track errors for BADA Hybrid Hold for Denver centerError typeComparisonAbs Max Mean Abs Mean MaxMinStd DevAltitude (ft)CTAS vs. Track 22156 BADA vs. Track 23969-52 -64524 53422156 -17490 1383 23969 -17688 1414Along-trackCTAS vs. Track 138.231.763.42105.46 -138.23 4.84(nm)BADA vs. Track 138.382.023.22105.46 -138.38 4.37 +Table 7 .7Mean altitude errors (ft), path-based correlationsTest caseAbs Max Mean Abs Mean MaxMinStd DevCTAS TS11667-19250511667 -5113760BADA Hybrid Hold 11528-43356311528 -10608 759BADA Base Case10276-960101410276 -5987815 +Table 8 .8Mean altitude errors (ft), time-based correlationsTest caseAbs Max Mean Abs Mean MaxMinStd DevCTAS TS21426-42784621426 -45271158BADA Hybrid Hold 21515-67393021515 -13799 1136BADA Base Case20981-632109920981 -54041330 +BADA is found to be most beneficial in climb, especially with the BADA Reduced Weight variant.•Typically BADA is much more accurate than CTAS in predicting Top Of Climb (TOC) for the most prevalent aircraft types.• When compared with CTAS, BADA has slightly worse performance in descent, especially for• Both CTAS and BADA have good models for several aircraft types -A306, B752, C750,CRJ2, CRJ9, E135, E170, H25B, MD82, and MD83, including popular McDonnell Douglas,Embraer, and Canadair aircraft.• +Table Data FileDataNot all columns in the following table are shown due to a page width limitation.High mass CLIMBS================FL[-] T[K] p[Pa] rho[kg/m3] a[m/s] TAS[kt] CAS[kt]M[-] mass[kg] Thrust[N] Drag[N]0 288 1013251.22534079.0079.000.12110612285515 287 995081.20734079.5879.000.121106120855110 286 977171.19033980.1679.000.121106118755115 285 959521.17233980.7579.000.1211061167551BADA PERFORMANCE FILE RESULTS 20 284 94213 1.15533881.3579.000.1211061146551============================= 30 282 90812 1.12133782.5779.000.1311061105551============================= 40 280 87511 1.08833683.8179.000.131106106455160 276 812001.02433386.3779.000.131106981551Low mass CLIMBS 80 272 752620.96333189.0579.000.141106899551=============== 100 268 696820.90532891.8679.000.141106817551120 264 644410.84932694.7979.000.151106734551FL[-] T[K] p[Pa] rho[kg/m3] a[m/s] TAS[kt] CAS[kt]M[-] mass[kg] Thrust[N] Drag[N]0 288 101325 Medium mass DESCENTS 1.22534072.1272.120.1173612393675 287 99508 ==================== 1.20734079.5879.000.12736120838010 286 977171.19033980.1679.000.12736118738015 285 959521.17233980.7579.000.127361167380 N]20 284 94213 0 288 1013251.155 1.225338 34081.35 60.9079.00 60.900.12 0.09736 10551146 49380 61430 282 90812 5 287 995081.121 1.207337 34082.57 66.3879.00 65.900.13 0.10736 10551105 197380 57140 280 87511 10 286 977171.088 1.190336 33983.81 77.0279.00 75.900.13 0.12736 10551064 191380 52860 276 81200 15 285 959521.024 1.172333 33986.37 128.7879.00 126.000.13 0.20736 1055981 180380 73280 272 75262 20 284 942130.963 1.155331 33889.05 129.7279.00 126.000.14 0.20736 1055899 177380 732100 268 69682 30 282 908120.905 1.121328 33791.86 131.6579.00 126.000.14 0.20736 1055817 170380 731120 264 64441 40 280 875110.849 1.088326 33694.79 133.6179.00 126.000.15 0.20736 1055734 164380 73160 276 812001.024333137.66126.000.211055073180 272 752620.963331141.90126.000.2210550730Medium mass CLIMBS 100 268 696820.905328146.33126.000.2310550730================== 120 264 644410.849326150.95126.000.2410550729FL[-] T[K] p[Pa] rho[kg/m3] a[m/s] TAS[kt] CAS[kt] TDC stands for (Thrust -Drag) * CredM[-] mass[kg] Thrust[N] Drag[N]0 288 1013251.22534079.0079.000.12105512285235 287 995081.20734079.5879.000.121055120852310 286 977171.19033980.1679.000.121055118752315 285 959521.17233980.7579.000.121055116752320 284 942131.15533881.3579.000.121055114652330 282 908121.12133782.5779.000.131055110552340 280 875111.08833683.8179.000.131055106452360 276 812001.02433386.3779.000.13105598152380 272 752620.96333189.0579.000.141055899523100 268 696820.90532891.8679.000.141055817523120 264 644410.84932694.7979.000.151055734523FL[-] T[K] p[Pa] rho[kg/m3] a[m/s] TAS[kt] CAS[kt] M[-] mass[kg] Thrust[N] Drag[ + + + + +AcknowledgmentsThis work would not have been possible without help and support from many people.We especially want to express our gratitude to Karen Cate for her guidance, to Hassan Eslami for continuous support and encouragement, to Martin Brown for contributions in data analysis and development of validation tools, to Alan Lee and Steven Chan for invaluable help in studying the existing CTAS TS code, to Gano Chatterji and Gilbert Wu for valuable discussions, to Jinn-Hwei Cheng for all her help with check-ins and ClearCase issues, to Pat O'Neal for assistance with data analysis for Denver-2009, and to Laurel Stell for providing data sets for Denver-2009. + + + + + + + + + + + Design of Center-TRACON Automation System + + HErzberger + + + TJDavis + + + SMGreen + + + + Proceedings of the AGARD Guidance and Control Panel 56th Symposium on Machine Intelligence in Air-Traffic Management + the AGARD Guidance and Control Panel 56th Symposium on Machine Intelligence in Air-Traffic ManagementBerlin, GDR + + 1993 + + + + Erzberger, H., Davis, T. J., Green, S. M., Design of Center-TRACON Automation System, Proceedings of the AGARD Guidance and Control Panel 56th Symposium on Machine Intelligence in Air-Traffic Management, Berlin, GDR, 1993, pp. 52-1 -14. + + + + + Terminal Area Trajectory Synthesis for Air Traffic Control Automation, Proceedings of the 1995 American Control Conference + + RASlattery + + + 1995 + + Seattle, WA + + + Slattery, R. A., Terminal Area Trajectory Synthesis for Air Traffic Control Automation, Pro- ceedings of the 1995 American Control Conference, Seattle, WA, 1995, pp. 1206-1210. + + + + + Trajectory Synthesis for Air Traffic Automation + + RASlattery + + + YZhao + + + + Journal of Guidance, Control and Dynamics + + 20 + 2 + + March-April 1997 + + + Slattery, R. A., and Zhao, Y., Trajectory Synthesis for Air Traffic Automation, Journal of Guidance, Control and Dynamics, Vol. 20, No. 2, March-April 1997, pp. 232-238. + + + + + The Trajectory Synthesizer Generalized Profile Interface, 10th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference + + AGLee + + + XBouyssounouse + + + JRMurphy + + + Sep. 2010 + + Fort Worth, TX + + + Lee, A. G., Bouyssounouse, X., Murphy, J. R., The Trajectory Synthesizer Generalized Profile Interface, 10th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference, Fort Worth, TX, 13-15 Sep. 2010. + + + + + Improving and Validating CTAS Performance Models + + WChan + + + RBach + + + JWalton + + AIAA- 2000-4476 + + + AIAA Guidance, Navigation, and Control Conference and Exhibit + Denver, CO + + August 2000 + + + + Chan, W., Bach, R., Walton, J., Improving and Validating CTAS Performance Models, AIAA- 2000-4476, AIAA Guidance, Navigation, and Control Conference and Exhibit, Denver, CO, 14-17 August 2000. + + + + + A Methodology for Automated Trajectory Prediction Analysis + + CGong + + + DMcnally + + AIAA- 2004-4788 + + + AIAA Guidance, Navigation, and Control Conference + Providence, RI + + 19 Aug. 2004 + 16 + + + Gong, C, and McNally, D., A Methodology for Automated Trajectory Prediction Analysis, AIAA- 2004-4788, AIAA Guidance, Navigation, and Control Conference, Providence, RI, 16-19 Aug. 2004. + + + + + No. 2009-009 + Base of Aircraft DATA (BADA) Aircraft Performance Modelling Report + + Eurocontrol Experimental Centre + 2009 + + + EEC + + + Technical/Scientific Report + Eurocontrol Experimental Centre, Base of Aircraft DATA (BADA) Aircraft Performance Mod- elling Report, EEC Technical/Scientific Report No. 2009-009, 2009. + + + + + + RDOaks + + + HFRyan + + Paglione, M. Prototype Implementation and Concept Validation of a 4-D Trajectory Fuel Burn Model Application, AIAA 2010-8164, AIAA Guidance, Navigation, and Control Conference + Toronto, Ontario Canada + + 2010 + + + Oaks, R. D., Ryan, H. F., Paglione, M. Prototype Implementation and Concept Validation of a 4-D Trajectory Fuel Burn Model Application, AIAA 2010-8164, AIAA Guidance, Navigation, and Control Conference, Toronto, Ontario Canada, 2010. + + + + + Implementation and Metrics for a Trajectory Prediction Validation Methodology, AIAA-2007-6517 + + MPaglione + + + ROaks + + + + AIAA Guidance, Navigation, and Control Conference + Hilton Head, SC + + 2007 + + + Paglione, M and Oaks, R. Implementation and Metrics for a Trajectory Prediction Validation Methodology, AIAA-2007-6517, AIAA Guidance, Navigation, and Control Conference, Hilton Head, SC, 2007. + + + + + Predictability of Top of Descent Location for Operational Idle-Thrust Descents, 10th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference + + LStell + + + Sep. 2010 + + Fort Worth, TX + + + Stell, L., Predictability of Top of Descent Location for Operational Idle-Thrust Descents, 10th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference, Fort Worth, TX, 13-15 Sep. 2010. + + + + + Closed Form Takeoff Weight Estimation Model for Air Transportation Simulation + + HLee + + + GBChatterji + + + + AIAA 2010-9156, 10th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference + Fort Worth, TX + + Sep. 2010 + + + + Lee, H and Chatterji, G. B. Closed Form Takeoff Weight Estimation Model for Air Transporta- tion Simulation, AIAA 2010-9156, 10th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference, Fort Worth, TX, 13-15 Sep. 2010. + + + + + Fuel Burn Estimation Using Real Track Data, AIAA-2011-6881, 11th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference + + GBChatterji + + + 20-22 Sep. 2011 + Virginia Beach, VA + + + Chatterji, G. B. Fuel Burn Estimation Using Real Track Data, AIAA-2011-6881, 11th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference, Virginia Beach, VA, 20-22 Sep. 2011. + + + + + Fuel consumption modeling in support of ATM environmental decision-making + + DSenzig + + + GFleming + + + RIovinelli + + + + Proceedings of the Eighth Annual FAA/EUROCONTROL Air Traffic Management Research and Development Seminar + the Eighth Annual FAA/EUROCONTROL Air Traffic Management Research and Development SeminarNapa, CA + + June 29-July 2, 2009 + + + Senzig, D., Fleming, G., and Iovinelli, R. Fuel consumption modeling in support of ATM envi- ronmental decision-making, Proceedings of the Eighth Annual FAA/EUROCONTROL Air Traffic Management Research and Development Seminar, Napa, CA, June 29-July 2, 2009. + + + + + + diff --git a/file10.txt b/file10.txt new file mode 100644 index 0000000000000000000000000000000000000000..29278eb6bbf505e840c9c09ffb166012aaa36b18 --- /dev/null +++ b/file10.txt @@ -0,0 +1,806 @@ + + + + +I. IntroductionSince 2015, the National Aeronautics and Space Administration (NASA) in collaboration with the Federal Aviation Administration (FAA) and flight operators, has been deploying some of its technologies developed for Integrated Arrival Departure Surface (IADS) [1][2][3] to a few chosen airports as part of the Airspace Technology Demonstration-2 (ATD-2) project [4].For its phase 1 & 2, the ATD-2 project focused on operations at Charlotte Douglas International Airport (CLT), Charlotte, North Carolina, due to its complex ground surface, its location in one of the busiest air traffic corridors on the East Coast and its abundant air traffic (# 7 in movements worldwide for 2016, [5]).ATD-2 deployment resulted in an improved operational efficiency that saved an estimated 4.3 million pounds of fuel between October 2017 and January 2020 [6].This improved efficiency is due to a better data exchange, coordination of release time for overhead stream insertion and application of surface metering.After this successful deployment, ATD-2 extended its operation for its phase 3 by deploying to a metroplex environment, the Dallas-Ft.Worth (Texas) Terminal Radar Approach Control (TRACON) where multiple airports operate within common TRACON boundaries.For this phase 3, ATD-2 provided coordination between multiple airports and added the functionality to reroute flights expected to be delayed by restrictions (often due to weather conditions).For fiscal year 2021, the ATD-2 project was extended for an additional year during which its focus is to improve the benefits analysis of its phase 3 and implement a version of its scheduler algorithm that could run for most airports in National Airspace System (NAS).This NAS-ready scheduler is to be enabled by the implementation of Machine Learning (ML) services that will substitute the complex physical model of ground movements and airport configurations designed in the prior implementation of the ATD-2 project.These microservices will lend themselves to a quicker and easier adaptation to new airports and easier maintenance.These microservices will need to be sufficiently accurate to enable the scheduler to provide useful schedules to Air Traffic Control personnel.These microservices will use ML algorithms that will be trained on historical data available for most airports in the NAS.To produce predictions, the trained algorithms will be fed FAA System Wide Information Management (SWIM) data from the ATD-2 Fuser [7], a software that ingests data feeds from various sources, formats and deconflicts them.To enable this new implementation of the ATD-2 scheduler, several microservices would be required such as a service that estimates the unimpeded taxi-out time, one that estimates ON time for arrivals, one that predicts departure runways, etc.This paper describes how the model for predicting unimpeded taxi-out was developed and its performance.In section II, we will present an overview of previous works on taxi-out time predictions.In section III, we will describe the frameworks and architecture used to build and deploy the taxi-out ML prediction service.In section IV, we will present some results produced by the microservice and we will conclude in section V. +II. Taxi-out PredictionsTaxi times of aircraft on the airport surface have been a long standing subject of study.Historically, the community started by developing sophisticated physical models of the airport surface and of aircraft movement [8][9][10][11][12][13][14][15].Recently, models based on historical data and machine learning algorithms have been put forward as more sustainable solutions.The literature of taxi time predictions is fairly large, but the wide range of datasets renders a comparison quite difficult.Prior studies focused on various airports, different measured durations, etc., to calculate their models' performance.The largest portion of the literature concentrates on modeling impeded taxi times, which is the actual duration of time it takes an aircraft to travel on the airport surface, including any time waiting for air traffic control (ATC) instruction and surface traffic congestion along the way.Some models used first-come-first-served queueing theory [16] and others used fast-time simulation calibrated on historical data [17].In recent years, using historical data and machine learning techniques, several studies have tried to estimate the impeded taxi time.Early papers studying the estimation of impeded taxi time with ML techniques used either linear or log-linear regression algorithms [18][19][20][21].Later papers often compared linear regression with more complex algorithms such as Singular Vector Machine (SVM), Random Forest or Fully Connected Neural Network [22][23][24][25][26][27].[28] studied a reinforcement learning approach in which the actions are the taxi out time estimate and the states are defined by parameters such as the number of aircraft taxiing, time, previous taxi times, etc. [29] defined a 2-dimensional embedding to capture the airport state using a fully connected Siamese Deep Neural Network, and used a kNN model to compute the taxi time from the 2D-embedding.In a large number of studies previously listed, the unimpeded taxi time is an input to the impeded taxi time calculation.The unimpeded time represents the duration of time the taxiing aircraft would spend if it had no constraints acting to slow it down.One of the most employed estimate of the unimpeded taxi time is to use the 10 th percentile of the taxi time for a given pair of stand and runway.Another technique developed by the EUROCONTROL PRU [30] uses taxi time as a function of the airport congestion for a given stand-runway pair and assigns unimpeded taxi time as the median value of taxi times that have a congestion below a saturation value.Recently [31] estimated the unimpeded taxi time using a node-link model at the Incheon International Airport (RKSI, 1553 nodes and 2034 links) and Airport Surface Detection Equipment (ASDE) data from April 1 st to 20 th , 2015.The unimpeded taxi time is estimated by the sum of the taxi time of each link along a given path which must be provided (for instance by tools such as departure manager DMAN [32]).The model takes into account the plane weight category, whether the flight is an arrival or a departure, as well as whether aircraft are turning before and after each link.The properties of each link are calculated from historical data.In this paper, we will present a method to evaluate unimpeded taxi out time that uses historical data and a ML algorithm instead of the physical models described previously.This method is similar to the techniques employed to estimate impeded taxi time.We will then use the unique capability of ATD-2 to test its performance. +III. Taxi-out Machine Learning Service +A. Frameworks and LibrariesWe decided to use the Python programming language and several common, open-source packages to develop these ML services.The main packages used for this development are Kedro * , MLFlow † and scikit-learn ‡ (see Fig. 1 to visualize where each framework is used).We endeavored to institute best practices across the Python model development.We decided to implement the Kedro framework as the backbone of each training pipeline used for training and storing the models.The Kedro framework enforces software engineering best practices for developing machine learning code.For instance, Kedro eases collaboration among our team given that the structure of all the pipelines is very similar.It also allows abstract data access (avoid hard-coded local directory path, database access, etc.) and organizes each node of the pipeline into direct-acyclic graphs (DAGs), that enforce input/output readiness.These DAGs can be visualized with an additional Python package called kedro-viz.The ML model, which encompasses feature engineering and the proper data science model, is contained in an extended version of a scikit-learn pipeline and is registered and stored on a MLFlow server.MLFlow allows the developer to store each run of a model (see Fig. 2 for an example) with a wealth of information such as the Git hash of code used in the run, the environment in which the model ran, the parameters of the model, performance metrics, etc.The MLFlow-stored models are then deployed as representational state transfer (REST) application programming interface (API) microservices using Flask § framework.These REST APIs are used by the orchestrator to feed the scheduler.Figure 1 shows a summary of the pipeline steps used to train models and the pieces of the training pipeline that are transferred into the live prediction service through its storage on the MLFlow server. +B. MethodologyThe taxi-out pipelines are separated into three phases: data query, data engineering and data science model.The data query is used in the training model phase to access historical data and primarily accesses data through SQL queries and stores them locally into CSV files.This part of the pipelines is common to all our three pipelines, i.e. the pipeline to estimate the ramp taxi time (from stand to spot), the AMA taxi time (from spot to runway) and the total taxi time (from stand to runway).The data engineering does some basic transformations and filtering to feed the data modeling, such as fusing several sources of information, removing data with incomplete information, and marking data for test or train groups.Some of these processes like getting the proper runway will be re-implemented in the predictive ML service and some, like separating train and test groups, won't be necessary (see Fig. 1).One notable filtering applied to our training and testing data is to limit the data to unimpeded taxis.To characterize a taxi as unimpeded, we used Airport Surface Detection Equipment, Model X (ASDE-X) speed data in the AMA and selected flights whose taxiing speed did not drop below 4 knots for 90% of the taxi time.This filter is applied for our AMA and total taxi estimate.For the ramp taxi, we require the number of arriving and departing aircraft in the ramp to be in the lowest percentile (60% and 30%) as well as the number of departing aircraft in the AMA to be low (40 th percentile).These constraints seem to filter a good amount of impeded taxis while keeping about 15% of the flights in our training set.The data science part of the pipeline processes the data into their final form for modeling (feature engineering), fits the data with a model, and computes some performance metrics.The boundary between feature engineering and data engineering is not always well defined but our main criteria for implementation in feature engineering are complexity and specificity.Transformations applied within the data science part will be saved into our pipeline container that will be served through the REST API.A driving goal in this project was to have a process that can easily be adapted to any airport in the NAS.The feature selection, feature engineering, and model algorithms were all carefully developed with this goal in mind.As such, the data science pipeline of the unimpeded taxi-out time takes in a limited number of accessible information: carrier name, departure runway, and departure gate (the aircraft type has also been used but did not seem to improve the models).These features are then encoded with a standard one-hot-encoder for the carrier and departure runways and some dedicated encoder for the departure gates.We looked at different options for encoding the gate information, two looked the most promising: clustering by terminal name, and clustering by similar historical taxi time.The first of these gate encoders uses the departure gate name to identify the terminal and creates one feature column per terminal.It populates the appropriate column with the gate number.The rationale is that gates at the same terminal have a somewhat similar unimpeded taxi time and that gates with similar numbers within a terminal have an even closer taxi time.The latter assumption is unfortunately dependent on how the gate numbers are distributed in the terminal.The second of the gate encoders uses the data to create a number of clusters to group the gates.This encoder uses the taxi time to three runways for each gate to group them in similar buckets.Gates without a sufficient number of taxis to at least 3 runways are added to the clusters using the available runways. Figure 3 shows an example of the resulting clusters for the unimpeded AMA taxi time at KDFW.The left image shows a top-down map of the KDFW airport, each dot represents a gate location along the 5 main terminals and its color indicates the cluster it belongs to.The right image shows the position of each gate (solid blue circle) in a 3 dimensional taxi-time space, each solid blue circle is highlighted by a larger colored circle indicating its cluster.We found similar accuracy between the two methods, most likely because most of the information resides in the terminal.We opted to use the latter one (clustering by historical taxi time) since it provides us with some good insight into the gate behavior.Once the data has been processed by the feature engineering, the model for the unimpeded taxi-out time is defined and fitted on the calculated features with a gradient boosted decision tree from the XGBoost library ¶ .The loss function is defined as the squared error with respect to the actual taxi time.The model hyperparameters and the number of clusters defined in the gate encoding are tuned to improve the squared error.A baseline model can be executed at the same time as the predictive model.For instance, our current baseline model groups flights by pair of runway and terminal and returns the average taxi time.This baseline model tests the proper behavior and improvement in accuracy of the predictive model.All resulting metrics and plots are stored on the MLFlow server along with information about the run (see Fig. 2).The MLFlow UI allows for quick comparison between models and selection of the best predictive model for each airport for each taxi out pipeline (ramp, AMA, total).These models can be retrieved from the MLFlow server and deployed as a microservice that then can be used by an orchestrator to compute schedules for a specific TRACON or airport. +IV. Results and PerformanceThe unimpeded taxi out models were trained on data from June 1 st 2019 to December 31 st , 2019 at the KDFW and KCLT airports.The total number of departing flights for each airports in this date range was about 220k and 170k respectively.When filtering out incomplete data and impeded taxis, the remaining dataset contains about 47k and 20k flights for the AMA and total taxi at KDFW and KCLT respectively and about 20k flights for the ramp taxi at both airports.We split our data further, using 80% to train our models and 20% to test them.To assert the accuracy of our predictions, we calculated the median absolute deviation (multiplied by 1.4826 to match a Gaussian 1 sigma spread, referred to simply as MAD throughout the rest of this paper) and the median value of the residuals (predicted taxi time minus truth value).The median of the residuals informs us about the amount of bias the model introduces, whereas the MAD indicates the precision of our estimate.The following results show accuracy measurements on the test dataset.Calculating accuracy for unimpeded taxi time presents another challenge compared to impeded taxi time: the true value of the taxi time is not clearly defined, since we do not know if a taxi is really unimpeded.Whereas the model loss function minimizes the residuals with respect to the actual taxi time, we decided to use the STBO (Surface Trajectory Based Operations) system estimates as truth values since they are available at KDFW and KCLT and we believe that STBO unimpeded taxi time estimates are closer to the true unimpeded taxi time than the actual taxi time.Table 1 summarizes the accuracy of our models, and demonstrates that the largest source of uncertainty seems to come from the ramp area.The complexity of the ramp operation explains the difficulty in calculating a good estimate of the unimpeded ramp taxi time and could also imply some larger intrinsic spread of its true value.Another issue is that we do not have reliable ASDE-X data in the ramp, preventing us from filtering our flights with the aircraft speed.The training data for the ramp area might not be as free of unimpeded taxi as the training data used for the AMA taxi time model.Overall the total taxi time has an relative uncertainty of 11 to 14%, the AMA taxi time has an relative uncertainty of 9 to 14 % and the ramp taxi time has an relative uncertainty of about 20%.The models also have an overestimating bias of about 1 to 1.5 minutes compared to the STBO estimates.This difference could be due to some impeded taxis making it into our training data or it could be that the STBO system is underestimating unimpeded taxi time.Figure 4 & 5 show some additional insights into the model accuracy.Figure 4 presents the distribution of the ML model residuals (blue area), defined as the model prediction minus the values coming from the STBO, for the ramp, AMA and total taxi at KDFW.In addition to the ML model predictions, the baseline model residuals (orange area) is shown for comparison.On average, ML models perform about 20% better than baseline models.Figure 5 shows that this improvement might be due to a lower discretization of the taxi time estimate, which can be seen as the orange horizontal lines formed by the baseline orange dots.Another improvement is the ability of the ML model to predict more extreme taxi time (especially longer taxi time).Both these effects reflect the inability of the baseline model to distinguish the different taxi time of gates at the same terminal and the fact that the baseline model does not use the carrier information.While Fig. 4 shows that the residual distributions are fairly tight, AMA taxi time residuals at KDFW have some large wings.After digging through our data, we discovered that most of the largest residuals in this distribution might be due to a few unreliable STBO estimates, that for some reason are either very large or very small.For instance, Fig. 5 shows that some STBO unimpeded AMA taxi time at KCLT are less than 100 seconds.More work needs to be done to improve the truth set and understand why the predictions differ from the true taxi time.In order to test if our model could be served to predict the current unimpeded taxi time estimate, we gathered data from August 1 st , 2020 to September 1 st , 2020 as well as from February 5 th , 2021 to April 21 , 2021.These data come from our PostgreSQL database and not from the ML prediction live system, but they allow us to characterize the performance degradation due to systemic change at the airport.One example of these changes is the median number of departures, which for DFW goes from about 1050 in 2019 to about 750 in 2020 back to about 860 in 2021 for the date range listed previously.Figure 6 shows how much the performance of the models degrades.Most taxi time uncertainties increase by at most 10 to 15%, some biases could get as much as 30% larger (e.g.KCLT total taxi time).To reduce the model degradation, models could be trained again on more recent data, however the fast changing traffic volume might make the new models degrade again.Furthermore, a real true test of the model accuracy will come when the unimpeded taxi out model predictions are computed in the live system. +V. ConclusionIn this paper, we showcased a machine learning service to calculate the unimpeded taxi out time using some open-source frameworks that enable a simplified CI/CD (Continuous Integration/Continuous Deployment) process.Though we developed the process with KCLT and KDFW airports, we limited the features to those which will be readily available across the NAS through SWIM.The trained models have a 1 sigma uncertainty as measured with the median absolute deviation mostly between 1 and 1.5 minutes and an overestimating bias also between 1 and 1.5 minutes.In relative terms, the AMA and total taxi time have an uncertainty of about 10 to 15% whereas the ramp taxi time has an uncertainty of about 20%.These models are currently deployed in a shadow live system for the Dallas -Ft Worth TRACON airports and feed a flight scheduler.The scheduler predictions as well as the service predictions that feed it will be compared to the STBO system.This comparison will allow characterization of the accuracy of each ML service in real-time predictions, as well as to check whether theses service predictions are accurate enough to produce useful schedules for operational planning.A future parallel effort is to train these models for other airports in the NAS, especially for airports in the North East Corridor, where deploying an accurate flight scheduler could provide substantial value.Fig. 11Fig. 1 Diagram of the ML pipeline used to train the taxi-out time estimate (top) and diagram of the ML service implementation (bottom).Kedro constitutes the backbone of the training pipeline, scikit-learn serves as a model container, that will be deployed as a microservice. +Fig. 22Fig. 2 Example of the MLFlow User Interface (UI) on a web-browser, that allows easy access to previous runs/models and to compare several models.A user can access further information about each model run (e.g., model pickle, plots) using the UI hyperlinks. +Fig. 33Fig. 3 Left: KDFW hierarchical clustering on AMA taxi time to three runways.This approach can easily be extended to any airport in the NAS without a complex adaptation.Terminal names are written in black, runways are written in blue.Right: Same color-coded clusters distributed in the 3D AMA taxi time space (spatial 2D and mark size 1D).The blue cluster seems to have the shortest AMA taxi to 17R, green cluster to 18L and the red cluster to 36R. +Fig. 4 Fig. 5 ML45Fig. 4 Histograms (in blue) of the residuals (in seconds, for ramp, AMA, total taxi from left to right) as defined as the STBO estimated unimpeded taxi time (our truth values) minus the predictions from our ML model.As a comparison, the residuals of the baseline model estimate is overplotted as the orange histograms.These distributions include only the KDFW test dataset residuals. +Fig. 66Fig. 6 Evolution with time (datasets from 3 different years) of the median of the residuals with ±1 MAD error bar for KDFW and KCLT using predictions from the 2019 trained model. +Table 1 Metrics of accuracy, median and MAD of residuals as compared to STBO predictions, as well as the median taxi time from STBO for KDFW and KCLT airports. Negative median of the residuals indicate that the estimates are smaller than the STBO predictions.1KDFWKCLTMedian [sec] MAD [sec]Median Taxi [sec]Median [sec] MAD [sec]Median Taxi [sec]ramp-6±280±240162±398±5522AMA61±152±135843±227±2296total95±3113±478693±491±4808 + § https://flask.palletsprojects.com/en/2.0.x/ + ¶ https://xgboost.readthedocs.io + + + + +* https://github.com/quantumblacklabs/kedro† https://mlflow.org‡ https://scikit-learn.org/stable/ + + + + + + + + + + SAEngelland + + + ACapps + + + KDay + + + MSKistler + + + FGaither + + + GJuro + + Precision Departure Release Capability (PDRC) Final Report + + 2013 + + + Engelland, S. A., Capps, A., Day, K., Kistler, M. S., Gaither, F., and Juro, G., "Precision Departure Release Capability (PDRC) Final Report," 2013. + + + + + Performance Evaluation of SARDA: An Individual Aircraft-Based Advisory Concept for Surface Management + + YJung + + + THoang + + + MHayashi + + + WMalik + + + LTobias + + + GGupta + + + + Air traffic control quarterly + + 22 + + 2014 + + + Jung, Y., Hoang, T., Hayashi, M., Malik, W., Tobias, L., and Gupta, G., "Performance Evaluation of SARDA: An Individual Aircraft-Based Advisory Concept for Surface Management," Air traffic control quarterly, Vol. 22, 2014, pp. 195-221. + + + + + Evaluation of Pushback Decision-Support Tool Concept for Charlotte Douglas International Airport Ramp Operations + + MHayashi + + + THoang + + + YJung + + + WMalik + + + HLee + + + VDulchinos + + + 2015 + + + Hayashi, M., Hoang, T., Jung, Y., Malik, W., Lee, H., and Dulchinos, V., "Evaluation of Pushback Decision-Support Tool Concept for Charlotte Douglas International Airport Ramp Operations," 2015. + + + + + + + YCJung + + + SAEngelland + + + RACapps + + + RCoppenbarger + + + BHooey + + + SSharma + + + LStevens + + + SVerma + + + GWLohr + + + EChevalley + + + VDulchinos + + + WMalik + + + LMRuszkowski + + + + Airspace Technology Demonstration + + 2 + 2 + 2018 + Phase 1 Concept of Use + + + Jung, Y. C., Engelland, S. A., Capps, R. A., Coppenbarger, R., Hooey, B., Sharma, S., Stevens, L., Verma, S., Lohr, G. W., Chevalley, E., Dulchinos, V., Malik, W., and Ruszkowski, L. M., "Airspace Technology Demonstration 2 (ATD-2) Phase 1 Concept of Use (ConUse)," 2018. + + + + + ACI releases preliminary 2016 world airport traffic rankings -Robust gains in passenger traffic at hub airports serving trans-Pacific and East Asian routes + + + 2017 + + + Tech. rep., Airports Council International + "ACI releases preliminary 2016 world airport traffic rankings -Robust gains in passenger traffic at hub airports serving trans-Pacific and East Asian routes," Tech. rep., Airports Council International, 2017. Http://www.aci.aero/News/Releases/MostRecent/2017/04/19/ACI-releases-preliminary-2016-world-airport-traffic- rankingsRobustgains-in-passenger-traffic-at-hub-airports-serving-transPacific-and-East-Asian-routes. + + + + + Field Evaluation of the Baseline Integrated Arrival, Departure, Surface Capabilities at Charlotte Douglas International Airport + + YJung + + + WJCoupe + + + ACapps + + + SAEngelland + + + SSharma + + + + 2019. 13th USA/Europe ATM R&D Seminar (ATM2019) + Vienna, Austria + + June 2019 + + + + Jung, Y., Coupe, W. J., Capps, A., Engelland, S. A., and Sharma, S., "Field Evaluation of the Baseline Integrated Arrival, Departure, Surface Capabilities at Charlotte Douglas International Airport," 2019. 13th USA/Europe ATM R&D Seminar (ATM2019), Vienna, Austria, 17-21 June 2019. + + + + + Fuser Deeper Dive (Mediation & Use Cases) + + SMGorman + + + JMBurke + + + IJRobeson + + + BSPhipps + + + 2019 + + + Gorman, S. M., Burke, J. M., Robeson, I. J., and Phipps, B. S., "Fuser Deeper Dive (Mediation & Use Cases)," 2019. + + + + + Dynamic statistical models for the prediction of aircraft take-off times + + RAShumsky + + + 1995 + + + Massachusetts Institute of Technology + + + Ph.D. thesis + Shumsky, R. A., "Dynamic statistical models for the prediction of aircraft take-off times," Ph.D. thesis, Massachusetts Institute of Technology, 1995. + + + + + Modeling and analysis of an airport departure process + + JEHebert + + + DCDietz + + + + Journal of Aircraft + + 34 + 1 + + 1997 + + + Hebert, J. E., and Dietz, D. C., "Modeling and analysis of an airport departure process," Journal of Aircraft, Vol. 34, No. 1, 1997, pp. 43-47. + + + + + Automation tools for enhancing ground-operation situation awareness and flow efficiency + + VCheng + + + DFoyle + + + + AIAA Guidance, Navigation, and Control Conference and Exhibit + + 2002 + 4856 + + + Cheng, V., and Foyle, D., "Automation tools for enhancing ground-operation situation awareness and flow efficiency," AIAA Guidance, Navigation, and Control Conference and Exhibit, 2002, p. 4856. + + + + + Optimisation of airport taxi planning + + JSmeltink + + + MSoomer + + + PDe Waal + + + RVan Der Mei + + + 2003 + + + Smeltink, J., Soomer, M., de Waal, P., and van der Mei, R., "Optimisation of airport taxi planning," 2003. + + + + + Optimal airport surface traffic planning using mixed integer linear programming + + HVisser + + + PRoling + + + 2003 + 6797 + + + AIAA's 3rd Annual Aviation Technology, Integration, and Operations (ATIO) Forum + Visser, H., and Roling, P., "Optimal airport surface traffic planning using mixed integer linear programming," AIAA's 3rd Annual Aviation Technology, Integration, and Operations (ATIO) Forum, 2003, p. 6797. + + + + + Airport surface operation collaborative automation concept + + VCheng + + + + AIAA Guidance, Navigation, and Control Conference and Exhibit + + 2003 + 5773 + + + Cheng, V., "Airport surface operation collaborative automation concept," AIAA Guidance, Navigation, and Control Conference and Exhibit, 2003, p. 5773. + + + + + An optimisation model for airport taxi scheduling + + JSoomera + + + PDe Waalc + + + RVan Der Meib + + + + Oper. Res + + 1 + 2004 + + + Soomera, J., de Waalc, P., and van der Meib, R., "An optimisation model for airport taxi scheduling," Oper. Res, Vol. 1, 2004. + + + + + Development of Surface Management System integrated with CTAS arrival tool + + YJung + + + GMonroe + + AIAA-2005-7334 + + + proceedings of the 5th AIAA Aviation Technology, Integration, and Operations Conference + the 5th AIAA Aviation Technology, Integration, and Operations ConferenceArlington, VA + + September 2005 + + + Tech. rep., Report + Jung, Y., and Monroe, G., "Development of Surface Management System integrated with CTAS arrival tool. In proceedings of the 5th AIAA Aviation Technology, Integration, and Operations Conference, Arlington, VA, September 2005," Tech. rep., Report AIAA-2005-7334, ???? + + + + + Queuing models of airport departure processes for emissions reduction + + ISimaiakis + + + HBalakrishnan + + + + AIAA Guidance, Navigation, and Control Conference + + 2009 + 5650 + + + Simaiakis, I., and Balakrishnan, H., "Queuing models of airport departure processes for emissions reduction," AIAA Guidance, Navigation, and Control Conference, 2009, p. 5650. + + + + + Taxi time prediction at Charlotte Airport using fast-time simulation and machine learning techniques + + HLee + + + WMalik + + + BZhang + + + BNagarajan + + + YCJung + + + + 15th AIAA Aviation Technology, Integration, and Operations Conference + + 2015 + 2272 + + + Lee, H., Malik, W., Zhang, B., Nagarajan, B., and Jung, Y. C., "Taxi time prediction at Charlotte Airport using fast-time simulation and machine learning techniques," 15th AIAA Aviation Technology, Integration, and Operations Conference, 2015, p. 2272. + + + + + A combined statistical approach and ground movement model for improving taxi time estimations at airports + + SRavizza + + + JAAtkin + + + MHMaathuis + + + EKBurke + + + + Journal of the Operational Research Society + + 64 + 9 + + 2013 + + + Ravizza, S., Atkin, J. A., Maathuis, M. H., and Burke, E. K., "A combined statistical approach and ground movement model for improving taxi time estimations at airports," Journal of the Operational Research Society, Vol. 64, No. 9, 2013, pp. 1347-1360. + + + + + A statistical learning approach to the modeling of aircraft taxi time + + RJordan + + + MAIshutkina + + + TGReynolds + + + + 29th Digital Avionics Systems Conference + + IEEE + 2010 + 1 + + + Jordan, R., Ishutkina, M. A., and Reynolds, T. G., "A statistical learning approach to the modeling of aircraft taxi time," 29th Digital Avionics Systems Conference, IEEE, 2010, pp. 1-B. + + + + + Improving departure taxi time predictions using ASDE-X surveillance data + + ASrivastava + + + 2011 + IEEE + + + + 2011 IEEE/AIAA 30th Digital Avionics Systems Conference + Srivastava, A., "Improving departure taxi time predictions using ASDE-X surveillance data," 2011 IEEE/AIAA 30th Digital Avionics Systems Conference, IEEE, 2011, pp. 2B5-1. + + + + + Relationship between airport efficiency and surface traffic + + MKistler + + + GGupta + + + + th AIAA Aviation Technology, Integration, and Operations Conference (ATIO) and Aircraft Noise and Emissions Reduction Symposium + + 2009 + 7078 + + + Kistler, M., and Gupta, G., "Relationship between airport efficiency and surface traffic," 9th AIAA Aviation Technology, Integration, and Operations Conference (ATIO) and Aircraft Noise and Emissions Reduction Symposium (ANERS), 2009, p. 7078. + + + + + Taxi-out time prediction for departures at Charlotte airport using machine learning techniques + + HLee + + + WMalik + + + YCJung + + + + 16th AIAA Aviation Technology, Integration, and Operations Conference + + 2016 + 3910 + + + Lee, H., Malik, W., and Jung, Y. C., "Taxi-out time prediction for departures at Charlotte airport using machine learning techniques," 16th AIAA Aviation Technology, Integration, and Operations Conference, 2016, p. 3910. + + + + + Taxi-out time prediction model at Charles de Gaulle Airport + + FHerrema + + + RCurran + + + HVisser + + + DHuet + + + RLacote + + + + Journal of Aerospace Information Systems + + 15 + 3 + + 2018 + + + Herrema, F., Curran, R., Visser, H., Huet, D., and Lacote, R., "Taxi-out time prediction model at Charles de Gaulle Airport," Journal of Aerospace Information Systems, Vol. 15, No. 3, 2018, pp. 120-130. + + + + + Aircraft taxi time prediction: comparisons and insights + + SRavizza + + + JChen + + + JAAtkin + + + PStewart + + + EKBurke + + + + Applied Soft Computing + + 14 + + 2014 + + + Ravizza, S., Chen, J., Atkin, J. A., Stewart, P., and Burke, E. K., "Aircraft taxi time prediction: comparisons and insights," Applied Soft Computing, Vol. 14, 2014, pp. 397-406. + + + + + Wheels-off time prediction using surface traffic metrics + + GChatterji + + + YZheng + + + + 12th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference and 14th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference + + 2012 + 5699 + + + Chatterji, G., and Zheng, Y., "Wheels-off time prediction using surface traffic metrics," 12th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference and 14th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, 2012, p. 5699. + + + + + Machine learning techniques for taxi-out time prediction with a macroscopic network topology + + JYin + + + YHu + + + YMa + + + YXu + + + KHan + + + DChen + + + + 2018 IEEE/AIAA 37th Digital Avionics Systems Conference (DASC) + + IEEE + 2018 + + + + Yin, J., Hu, Y., Ma, Y., Xu, Y., Han, K., and Chen, D., "Machine learning techniques for taxi-out time prediction with a macroscopic network topology," 2018 IEEE/AIAA 37th Digital Avionics Systems Conference (DASC), IEEE, 2018, pp. 1-8. + + + + + Prediction of Pushback Times and Ramp Taxi Times for Departures at Charlotte Airport + + HLee + + + JCoupe + + + YCJung + + + + AIAA Aviation 2019 Forum + + 2019 + 2933 + + + Lee, H., Coupe, J., and Jung, Y. C., "Prediction of Pushback Times and Ramp Taxi Times for Departures at Charlotte Airport," AIAA Aviation 2019 Forum, 2019, p. 2933. + + + + + Accuracy of reinforcement learning algorithms for predicting aircraft taxi-out times: A case-study of Tampa Bay departures + + PBalakrishna + + + RGanesan + + + LSherry + + + + Transportation Research Part C: Emerging Technologies + + 18 + 6 + + 2010 + + + Balakrishna, P., Ganesan, R., and Sherry, L., "Accuracy of reinforcement learning algorithms for predicting aircraft taxi-out times: A case-study of Tampa Bay departures," Transportation Research Part C: Emerging Technologies, Vol. 18, No. 6, 2010, pp. 950-962. + + + + + Decision Support for Aircraft Taxi Time based on Deep Metric Learning + + JDu + + + MHu + + + WZhang + + + + 2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC) + + IEEE + 2020 + + + + Du, J., Hu, M., and Zhang, W., "Decision Support for Aircraft Taxi Time based on Deep Metric Learning," 2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC), IEEE, 2020, pp. 1-7. + + + + + Additional Taxi-Out Time Performance Indicator Document + + LCapelleras + + + + + EUROCONTROL/PRU + + 2015 + + + Tech. rep. + Capelleras, L., "Additional Taxi-Out Time Performance Indicator Document," Tech. rep., EUROCONTROL/PRU, 2015. Https://ansperformance.eu/library/pru-tx-out-pi.pdf. + + + + + Unimpeded Taxi-Time Prediction Based on the Node-Link Model + + MJeong + + + YEun + + + DJeon + + + HBang + + + + Journal of Aerospace Information Systems + + 17 + 10 + + 2020 + + + Jeong, M., Eun, Y., Jeon, D., and Bang, H., "Unimpeded Taxi-Time Prediction Based on the Node-Link Model," Journal of Aerospace Information Systems, Vol. 17, No. 10, 2020, pp. 591-602. + + + + + The EUROCONTROL DMAN Prototype-Description of DMAN in the A-CDM Context + + Dec. 2010. 0.4 ed + + + Tech. rep + European Org. for the Safety of Air Navigation + "The EUROCONTROL DMAN Prototype-Description of DMAN in the A-CDM Context," Tech. rep., European Org. for the Safety of Air Navigation, Dec. 2010. 0.4 ed. + + + + + + diff --git a/file100.txt b/file100.txt new file mode 100644 index 0000000000000000000000000000000000000000..47ee9a213d0f52c35f01d22007895326b41d2c26 --- /dev/null +++ b/file100.txt @@ -0,0 +1,1178 @@ + + + + +I. IntroductionTerminal airspaces are characterized by narrow portions of airspace in which high traffic volumes must fly through in short time periods.In these constrained environments, most aircraft are climbing or descending at various speeds.In dense terminal operation areas in which busy airports are close to each other, severe bottlenecks can be formed affecting the National Airspace System (NAS) efficiency.Therefore there is a clear need to facilitate air traffic operations and improve their efficiency in the terminal airspace.3][14][15][16] Recent studies 12,[17][18][19][20][21][22] showed that integrated departures and arrivals have the ability to improve the operations efficiency when active metroplex areas are nearby.However, benefits obtained and flight schedules computed under deterministic conditions in terminal operations areas are subject to uncertainty.Uncertainty comes from many sources, such as human factors, aircraft dynamics, wind prediction or weather forecasts that are difficult to model accurately.Uncertainty analyses were conducted to help estimate the robustness of the solutions and benefits obtained from integrated operations.In arrival scheduling problems, Thipphavong et al. used the Stochastic Terminal Arrival Scheduling Software (STASS) to study the relationship between uncertainty and system performance 23 .For the integrated departures and arrivals problem, Xue et al. analysed the consequences of flight time uncertainty on the integrated scheduled operations by investigating the impacts on delays and controller workloads 14 .The integration of uncertainty in algorithm formulations is crucial to better reflect the reality of current air traffic operations.One way to accommodate uncertainty in algorithms is to use buffering or probabilistic sampling techniques.In scheduling problems, the propagation of uncertainty can be represented by Monte Carlo simulations.In previous work, Xue et al. used sequential processors to develop a scheduling algorithm for integrated arrival and departure operations on a model of the Los Angeles terminal airspace 13 .Uncertainty was introduced in the flight times and handled by Monte Carlo simulations.To run Monte Carlo simulations, thousands of sampling points were used.However, solving large scale stochastic optimization problems can become computationally challenging, requiring many hours of computation, which makes testing and application prohibitive.General purpose applications that are computationally expensive are benefiting from the emergence of programmable Graphics Processor Units (GPUs).NVIDIA released a parallel computing development platform called Compute Unified Device Architecture (CUDA) 24 that is implemented by GPUs produced by the company.The programming model uses a environment based on the C/C++ language and existing libraries 25 .CUDA allows developers to use GPUs for general purpose applications that are not exclusively graphics, and are suitable for high performance computing applications 26,27 due to their parallel throughput architecture.Many concurrent threads can be launched at the same time to process individual instructions over multiple data (SIMD).In addition, they represent an investment at low cost when compared to multiple CPU cores.In terms of applications, many areas have already shown significant advantages of using GPUs illustrated with important computing performance increases, for example multitasking 28 , medical applications 29,30 or finance 31,32 .For air traffic management applications, Tandale et al. provided one of the first GPU study contribution and accelerated by 30 times a CPU implementation of a large-scale Traffic Flow Management (TFM) problem with 17, 000 aircraft 33 .In terms of algorithms, evolutionary computing techniques present an inherent parallel nature whose performance can be significantly improved when implemented on GPUs.Recent studies [34][35][36][37][38][39] show important computation time savings and large speedup when genetic algorithms are run on GPUs.The GPU's current maturity and its great capability allow a large community of researchers to successfully solve complex problems by exploiting such technology.This paper contributes to the air traffic management field by providing a fast decision support algorithm enabled by the use of GPU.It provides a GPU-based parallelization of an existent scheduling optimization 13 with uncertainty.The Monte Carlo simulations used for the uncertainty cost computation are transferred to the GPU in order to save computation time.Parallelization strategies are investigated and applied to data transfer, random number generation and GPU-occupancy optimization.Overall, the performance of the algorithm was significantly increased.The resulting acceleration enables a large range of experiments to be explored and hundreds of scenarios to be rapidly tested.An analysis is carried out on two types of experiments which explore different values of traffic densities and arrival-to-departure ratios.The characterization of the algorithm and solutions obtained are also provided as well as the extraction of delay and controller intervention results and trends.In the paper, the sections are organized as follows.Section II provides some background on the problem studied and revisits briefly the existing algorithm implementation.The GPU implementation of the scheduler is described in Section III.Different accelerating strategies and associated performances are also provided in Section III.An application of the GPU-based scheduler in the Los Angeles terminal airspace is presented in Section IV.Section V presents multiple scenario evaluations and corresponding results characterization.Finally, conclusions and future research are provided in Section VI. +II. BackgroundThis section provides background on the existing model of optimization with uncertainty, its original implementation and its application. +A. Los Angeles case studyThis effort is based on previous work 13 studying optimal integrated operations with uncertainty for partial flows in the Los Angeles terminal airspace.The interactions between arrivals and departures in this area are worth a special case study because of their complex natures and layouts.Figure 1 shows the Standard Terminal Arrival Routes (STARs) and the Standard Instrumental Departures (SIDs) +B. Optimization with uncertaintyThe optimization model is formulated as a multiple objective optimization with both spatial and temporal separations as aircraft separation methods.The objective is to minimize the total amount of delay for all departure and arrival flights and the number of controller interventions.In the formulation, the delay is equivalent to the sum of exit times at waypoint SUTIE for arrivals and at waypoint WPT1 for departures.A controller intervention is defined as the need to intervene to avoid a loss of separation between aircraft.Deterministic solutions are first computed using a nondominated sorting genetic algorithm (NSGA).Because flight times are subject to uncertainty, unexpected separation loss or flight delay might occur.Therefore, error sources are added to the flight times in the model to generate uncertainties.Work performed on trajectory and flight time predictions show that predicted flight delays can be represented by shifted normal distributions. 41 42Therefore as illustrated in Figure 2, samples drawn from a normal distribution are used for simplicity to perturb flight times at FIM and SUTIE for arrival flows and at the runway for departure flights 14 .When separation losses occur, Monte Carlo simulations are used to simulate controller interventions.The design variables allow planes to fly two route options (direct or indirect), fly at different speeds and allow both air and ground delays to be computed to ensure conflict-free operations.The route assignments are determined by the solutions generated under deterministic scenarios and are not modified by the Monte Carlo simulations.The controllers are assumed to follow a First-Come-First-Serve (FCFS) rule to process the aircraft.The separation requirements at the fixes are hard constraints.If these are not respected, extra delay will be imposed which will involve extra controller interventions.More details about the optimization formulation and handling of uncertainty are covered in previous work 13 .The original implementation of the stochastic scheduler reported in previous work 13 was carried out on traditional sequential processors.Because genetic algorithms (GAs) are sensitive to the initial population, different seeds corresponding to different starting populations were explored.To compute the various seeds at the same time, multithreading was coded using Tcl scripting to parallelize the computation on the CPU and distribute the jobs evenly among the CPU cores.The whole procedure is illustrated in Figure 3. Figure 3 shows that each thread is assigned to run on different seeds.The structure of the code allows each thread to run smaller operation chunks.The code uses 10 seeds which are distributed among 4 Tcl threads.In the GA, the population size is set to 832 individuals.The cost evaluation function consists of two different steps, the deterministic cost evaluation and the uncertainty cost computation.The cost function evaluation is the most time consuming process because it computes the uncertainty cost for each individual, which is itself evaluated by thousands of Monte Carlo simulations. +III. GPU ImplementationThis section presents and describes the implementation of the stochastic scheduler on a mixed CPU-GPU (hostdevice) platform development environment.Accelerating strategies are investigated to optimize the overall performance.The results that are presented in the following sub-sections are obtained using the author's implementation.The implementation is novel and was not used in previous work.In this work, the NVIDIA GeForce GTX690 was used on a Linux platform with 2.30 GHz 12-Core Intel Xeon and 16 GB RAM.The graphics card has a CUDA Capability of 3.0 and combines two Kepler GPUs, consisting each of 1, 536 CUDA Cores.The NVIDIA GeForce GTX690 has a total global memory of 2, 048 MB and a total shared memory of 49,152 bytes per block.The maximum number of threads per streaming processor is 2,048 and the number of threads per block cannot exceed 1,024.A maximum of 16 blocks per streaming processor can be launched in parallel. +A. Implementation of the Monte Carlo simulations on the GPUThe robustness of the uncertainty cost computation depends directly on the number of Monte Carlo simulations evaluated.When a large number of Monte Carlo simulations is computed on the CPU, it becomes a computational challenge and compromises the optimization feasibility.Monte Carlo simulations rely on a large number of independent simulations which make them perfectly suitable for parallelization.A GPU is therefore introduced to offload thousands of uncertainty evaluations from the CPU and reduce the associated computation time.The following pseudo-algorithm provides the processing model of the parallel implementation of the Monte Carlo simulations on the GPU.Each Monte Carlo simulation requires a large amount of intermediate variables to evaluate the two objectives of this optimization.Therefore the global memory of the GPU is used to store them.Average delay and workloadCounter over maxNumberMonteCarloSim 10:end for 11: end procedure Thousands of Monte Carlo simulations are computed in parallel for each individual.In CUDA terms, this is performed by launching from the host at the same time 32 thread blocks, each containing 128 threads.The GPU-based implementation of the Monte Carlo simulations consists of a kernel that is called from the host.For each individual of the GA population, the kernel computes thousands of uncertainty costs.Because the Monte Carlo simulations of each seed are grouped together, millions of operations are performed simultaneously on the GPU.In order to enable the computation of these Monte Carlo simulations on the GPU, a combination of block number and thread number per block is assigned.Because the number of blocks that can be called and run simultaneously is limited by the physical properties of the graphics card used, the entire population cannot be processed at the same time on the GPU.Therefore the population is split into smaller group chunks, and a loop is created to handle all the chunks.The degree of parallelism is not explicitly expressed on the GPU.The kernels look like serial programs, and functions are written as if they would be run by a single thread.To address this, different thread granularity options were investigated.By launching a large number of threads and a large number of thread blocks, the GPU runs the program in parallel on many threads (128) and many blocks (32).A high-level flow chart is provided in Figure 4 and represents the task organization and mechanism of the scheduler.In this current implementation, 10 seeds are distributed among 10 CPU-cores.The threads synchronize before the GPU call so that the whole population set is sent to the GPU. Figure 4 shows that the inner loop on each individual originally represented in Figure 3 has been removed for efficiency.Accelerating strategies are investigated for the purpose of increasing speed.The parallelization strategy chosen induces the distribution of problem instances to multiple GPU-processors.A code performance evaluation study is conducted in order to show different tuning strategies that can be used to improve the performance. +Data transferBecause the host-device communication between CPU and GPU is time consuming, the operations of memory allocation and data transfer have to be minimized.For the GPU-based implementation of the Monte Carlo simulations, data needed in the GPU-kernel are first copied from the host memory to the device memory.Then, necessary memory is allocated on the device.This data-related operation requires two CUDA function calls: CUDA malloc and CUDA memcpy.This test compares two alternative strategies to perform the memory allocation and data transfer operations.On one hand, the first method allocates and copies the input data separately for each individual.Therefore this first method requires 832 CUDA malloc and 832 CUDA memcpy to address the whole population.On the other hand, in the second method, the input data is allocated and copied only once for the whole population.Only one CUDA malloc and one CUDA memcpy are then required.Table 1 presents the resulting computation times for both strategies when one generation is considered in the simulation.The computation times obtained show that calling CUDA memory functions multiple times increases the computation time of the Monte Carlo simulations by 81.8x.This test shows that host-device communication through the use of CUDA function calls can be very expensive when used intensively.Therefore, the code performance can be improved when memory operations are executed in a small number.The flow chart of the original code is modified to address this and is presented in Figure 5.The uncertainty cost evaluation loop over each individual has been replaced by a single large memory operation on the GPU.Once the deterministic cost is evaluated for a whole population on the CPU, the results for the population are transferred to the GPU and the uncertainty cost is computed. +Random numbers generationIn the Monte Carlo evaluation, random numbers that will be added as source errors in the flight times have to be generated for each simulation.Random numbers are required to follow a normal distribution.Therefore, this section explores different ways of generating random numbers following a normal distribution on the CPU and on the GPU.Five generators are implemented, two on the GPU and three on the CPU.In this algorithm, 1,024 Monte Carlo simulations are evaluated per population.In the reference scenario 14 flights are considered, therefore 14*1,024=14,336 random variables are needed per population.The aim of this test is to compare running times of the different random number generators.Each random number generation is performed in a separate function or kernel.Resulting random numbers are gathered in a main CPU function which also computes the different running times.To make sure that overheads and warmup times are minimized in the computation times for all benchmarks, 100 generation trials are performed and resulting running times are averaged.The different types of random number generators and associated averaged running times can be viewed in Table 2.This test shows that GPU might not be very efficient at generating large sets of random numbers.Moreover, this test demonstrates that using the Boost library on the CPU is more time consuming than implementing a C-function.Generator 5 was implemented by hand and uses the Box-Muller transformation 43 to transform uniformly distributed random variables to a new set of random variables with a normal distribution.Based on the average running times, this test demonstrates that Generator 5 is more capable for this task and is therefore chosen for this implementation. +GPU-occupancy studyThe Monte Carlo simulations require thousands of evaluations to cover each individual.The hardware being used has physical limitations and cannot compute every single evaluation simultaneously.Thus the Monte Carlo kernel has to be launched multiple times to ensure the completion of the computation.In CUDA terms, a grid size defined by a fixed number of blocks and a fixed number of threads per block has to be chosen.To ensure that the grid configuration chosen is optimal, an occupancy study is performed using the CUDA Occupancy calculator tool provided by NVIDIA.The GPU-occupancy is defined as the number of active thread blocks per processor divided by the maximum number of thread blocks per processor.It is a number in the range of 0 to 100%.At any time, some threads may be processing data and some others may be accessing memory.When this occurs, the number of threads available for computation is limited, degrading the occupancy.Occupancy of 100% means that the GPU-cores are fully utilized.The internal GPU thread block scheduler controls when thread blocks are executing data.Therefore, understanding how the scheduler works is crucial.According to Luitjens and Rennich 44 , the occupancy calculator can help efficiently use the GPU.However, the resulting configurations of the occupancy study might not provide optimal computation times.To reach an optimal trade-off between occupancy and computation time, a few experiments are conducted to investigate different grid configurations for the Monte Carlo kernel.The inputs needed by the calculator are GPUrelated (compute capability and size of total shared memory) and kernel-specific (grid size, number of registers and size of shared memory used).The amount of registers and shared memory provided to the tool are defined by how the kernel is coded.The Monte Carlo kernel implemented requires 32 registers and 167 bytes of shared memory.The outputs of the calculator are the resulting occupancy and occupancy limiters.Occupancy limiters are given so that the user gets an idea of which parameter (grid, register or shared memory) constrain the number of threads per block that can be launched.Three tests are conducted to study how the grid size affects the GPU-occupancy.Table 3 gathers occupancy results when the grid size and the number of threads per block are varied.Because the total number of threads is capped at 2, 048 for the GPU used in this work, the number of active blocks will decrease when the number of threads per block increases.If higher numbers of blocks are defined than the one defined by the GPU physical limits, they will be active sequentially.The block sequencing is arranged internally by the GPU scheduler.The outcome of the occupancy study shows that high GPU-occupancy is obtained when the number of threads launched at the same time is large, or in other words, occupancy is high when the number of Monte Carlo simulations computed in parallel is large.Maximum occupancy is reached for the last three rows corresponding to three different numbers of threads per block.To find the optimal trade-off between GPU-occupancy and computation time, the code is run for the last three rows.Table 4 presents the GPU-kernel and overall algorithm computation times associated when one generation is simulated.Despite 100% occupancy obtained for block sizes of 256 or 512 threads, the corresponding running times are slower.From these results, the solution that compromises occupancy and computation time is obtained when each block contains 128 threads.The following set of plots in Figure 6 illustrates how the inputs of the CUDA occupancy calculator can influence and impact the occupancy of the GPU.The curves illustrate how the different input impact the occupancy when the inputs values are varied.The triangle shows the occupancy of the Monte Carlo kernel when the number of threads per block is set to 128.Because the registers and shared memory usages are fixed by the kernel, the top graph shows that for this kernel a block size of 128 threads is the optimal setting in terms of computation time.In addition, the curves show that maximum occupancy can be obtained for different numbers of threads per block, such as 256 or 512 as previously tested.The middle and bottom graphs provide a visual representation of how occupancy can be degraded and demonstrate that memory coding choices might affect the GPU-occupancy. +C. Performance resultsThe different experiments provide some interesting results, some of which are unexpected.To illustrate the performance achieved by the final version of the code, the algorithm runs 400 generations.The number of Monte Carlo simulations run for each individual is 1,024 and 10 seeds are considered.Table 5 presents the comparison of the computation times between the original and the current code.The GPU computes 1, 024 Monte Carlo simulations for popSize * numSeeds = 8, 320 individuals.Thanks to the GPU, the computation time of the Monte Carlo simulations has been decreased from 18.5 seconds to 0.029 seconds.The GPU performs well at 8.5 million operations and significantly outperforms the original inner loop computing only 1, 024 operations at a time.The computation time is now 152.19 seconds, which is less than 2.5 minutes.The code +IV. Application to the Los Angeles terminal airspaceThis section provides a direct application of the scheduling tool supported by GPU technology on a representative schedule of 14 flights flying on partial flows in the Los Angeles terminal airspace.The schedule covers 30 minutes of historical traffic operations on December 4, 2012 from 9:00 AM to 9:30 AM.It includes two flows: 8 arrivals from FIM fix and 6 departures to the North from Runway 24L.According to statistical data mentioned previously, the numbers of flights considered in the arrival flow represent typical operational traffic on SADDE6.However the number of flights considered in the departure flow is twice the average number of departures using CASTA2.Therefore, because the aircraft will fly in a mixed environment with shared resources, 14 flights represent a typical scenario case that requires scheduling.Table 6 provides a representative schedule of the 14 flights considered.The initial times are relative to simulation start time and the flights order is determined through a sorted ordering of the initial times.In previous work 13 , Xue et al. applied the scheduling tool developed on sequential processors to this representative schedule and presented the solutions obtained that were optimized under uncertainty.The results showed that under uncertain environment, the integrated arrivals and departures in terminal airspace can provide great savings in flight delays.Moreover, the proposed method demonstrated trade-offs between delay savings and controller intervention counts.However, the optimization process took about 6.5 hours which would be a problem for real-time application and was prohibitive for further experiments and extension of the algorithm.In this present work, the GPU-based scheduling optimization was applied to the same scenario.The solutions obtained and results trends are the same as with the sequential version of the scheduling tool.However, the GPU technique enabled solving the optimization process in about 2.5 minutes resulting in a speedup of 154x.With this significant speedup, the GPU-based scheduling tool now presents characteristics to be a candidate for real-time/fasttime tool. +V. Experiments and characterization of the algorithmPrevious section showed that the application of the GPU-based scheduling tool to a concrete traffic flow scenario presents great benefits in terms of computing time.The important speedup obtained allows hundreds of scenarios to be run quickly.The main aim of this section is to use the speedup gain brought by the GPU technique to test diverse experiments.This section describes the experiments that were designed along with the results obtained.The goal of these experiments is to characterize the algorithm and understand its performance on a large range of different scenarios.Scenarios are generated randomly but the computed schedules are restrained to respect the authorized aircraft separation times.Some scenarios might not necessarily represent realistic conditions as they might stress the algorithm outside its solution space search.The representative scenario tested in previous section establishes the baseline.Two different types of experiments are designed and tested.Both types investigate the variation of a single parameter that can take different values on a defined range.Details are provided in Table 7.For each specific value of the defined range of the total number of aircraft, 400 schedule-scenarios are randomly generated then tested in simulations.In the first experiment, the range chosen for the total number of aircraft starts with 14 aircraft which represents average traffic operations for the partial flows considered in the Los Angeles terminal airspace.The total number of aircraft to be scheduled is then increased up to 24 aircraft.One of the goals of the experiment is to test the algorithm For each schedule-scenario, new initial flight times are generated as pseudo-random numbers that follow a uniform distribution.It is assumed that the probability of having a flight time in a 1-minute interval is the same as having another flight time in any 1-minute interval but the time values are constrained to stay within a 30-minute time range.A uniform distribution was used to allow more varied schedule permutations.Because the minimum separation distance is four nautical miles, and the speed range considered in the model implementation is between 180 kts and 350 kts, aircraft separation times are maintained between 41 and 80 seconds.In this work, the aircraft fleet considered is homogeneous flights are assumed to be the same aircraft type.For analysis purposes, a set of Pareto fronts are computed for each schedule-scenario.Because ten different seeds are considered, the algorithm computes ten different fronts per scenario.Previous Pareto front studies focusing on multi-objective optimization 45,46 showed that the leading edge is a good representation of the optimal front.Therefore in this application, the leading edge of the 400 Pareto fronts is retrieved for each simulation to provide a visual representation of the set of minimum pairs (controller intervention and delay).A. Experiment 1: Variation of the total number of aircraft +SetupIn the first experiment, ten different values of the total number of aircraft are tested; the range of values considered is provided in Table 7.For each value of the total number of aircraft, a simulation is run to test 400 schedule-scenarios.Although the total number of aircraft is kept as a constant in the scenario generation procedure, the number of arrivals and the number of departures are randomly computed while constraining their sum.Therefore the arrival-to-departure ratio is also a random variable. +ResultsFigures 7-8-9 present the superposed results of each simulation.As the number of total aircraft was increased, the computation time of each simulation increased.To run 400 scenarios, the computation time varied between 5 hours when the number of aircraft was fixed to 24 and a maximum of 20 hours when the number of aircraft was fixed to 14.This large computation time variation illustrates that the program cannot compute solutions when testing scenarios with 40% greater than typical number of aircraft.Because the algorithm has a fixed delay range that can be assigned, it cannot assign delays outside this range.Therefore if the program encounters this situation, it cannot compute a solution.Define an infeasible scenario as a scenario for which the algorithm cannot compute a solution.After each simulation, data were analyzed to determine the number of scenarios for which the algorithm could not find a solution (i.e.infeasible).Once infeasible scenarios were discarded from the solution set, a time discretization was performed on the range of delay obtained.Then, the number of scenarios having a solution in each time delay bin was computed.Figure 7 shows the scenario distributions of each simulation.After each simulation, the leading edge of the 400 Pareto fronts was retrieved and saved.When every simulation was done, the global leading edge was computed from the ten different leading edges.The global delay range was sampled with a time delay bin size of five seconds and the averaged number of controller interventions was computed for each time delay bin.The standard deviation of the number of controller interventions was also computed for each time delay bin. Figure 8 and Figure 9 show the resulting curves. +AnalysisThe set of graphs generated for this first experiment allows characterization of the solutions of the algorithm for different traffic densities.The scenario distribution in Figure 7 shows that the number of scenarios that have a solution in each simulation follows a distribution with a shape similar to the standard normal curve.However each simulation provides different distribution characteristics.As the total number of aircraft is increased, the number of scenarios having a solution decreases.The data analysis performed after each simulation demonstrates that the number of scenarios to be discarded from the solution set increases when the total number of aircraft is increased.A visual representation is provided in Figure 10.The experiment results show that when the traffic density becomes greater than a total of 20 scheduled aircraft in a 30 minute time range, the algorithm is not able to compute solutions for each of the scenarios generated.For current day operations using the partial flows modeled in the Los Angeles terminal airspace, this does not present a problem because scenarios with more than 20 scheduled aircraft represent conditions 40% busier than typical.For example, when the total number of aircraft is set to 24, only 29.5% of the schedules generated have a solution.Therefore, for these larger values, the algorithm is stressed beyond its capacity to produce optimal solutions.Based on the data analysis of the results, one of the reasons for such behavior comes from the random computations of the number of arrivals and departures.In the scenario generation procedure, because the only requirement is on the total number of aircraft, extreme values of arrivals and departures are allowed to be generated.For examples, extreme values generated could be 1 arrival for 19 departures and vice versa.When running schedules with such values, the algorithm cannot delay aircraft more than the delay range defined in the code.In consequence, the algorithm cannot compute a solution for these schedules.Infeasible scenarios which distributions are represented in Figure 10 are thus characterized by schedules with extreme values of arrivals and departures.According to Figure 8 and Figure 9, one can observe that there exists a clear trade-off between delays and number of controller interventions.To obtain low delays, higher numbers of controller interventions are needed.This trend is even more noticeable as the total number of aircraft becomes larger.Figure 7 shows that each simulation has a range of delay for which the algorithm computes the most solutions.Therefore, highest confidence should be given to data corresponding to these ranges in Figure 8. Data points corresponding to average time delay of 200 seconds when the number of aircraft is equal to 16 are not representative of the main trend because the number of scenarios corresponding is low.For each time delay bin, the number of controller interventions increases as the total number of aircraft is increased.Figure 9 provides the statistical dispersion of the data from the average values presented in Figure 8. Observations of the standard deviation graph demonstrates that most of the data points obtained are concentrated and the number of outliers is low.For each simulation, when focusing on the most represented time delay range, the variability of the averaged number of controller interventions is low and confined.In this specific time delay range and for each simulation, the deviation of the average is no more than two controller interventions.Therefore, the computed solution is not deteriorated too much when testing different schedule-scenarios and this is indicative of algorithm robustness.B. Experiment 2: Variation of the arrival-to-departure ratio +SetupThe second experiment focuses on varying the arrival-to-departure ratio.Because results from experiment one demonstrate that when the number of aircraft is greater than 20, the algorithm cannot compute a solution for each scenario tested, the second experiment only considers aircraft number between 14 and 20.Four different cases are investigated where each case corresponds to a fixed number of aircraft.In each case, the arrival-to-departure ratio is varied such that the only constraint is the number of aircraft.The number of arrivals is computed randomly between 1 and the fixed total number of aircraft and the number of departures is determined by the difference between the total number of aircraft and the number of arrivals.For each value of the number of aircraft, 400 schedule-scenarios are generated then tested in different simulations.The arrival-to-departure ratios ranged between 0.05 and 8.00.Because in real operations and for the partial flows considered, the arrival-to-departure ratio is typically 3, it is covered in the range of values tested. +ResultsFigures 11-12-13 present the results of the four different cases.The same set of data from experiment 1 for a given fixed number of aircraft was segregated by arrival-to-departure ratio.Then a time discretization was performed on the range of delay obtained.Finally, the number of scenarios having a solution in each time delay bin was computed.Figure 11 shows the scenario distributions of each case study when varying the arrival-to-departure ratio.For each fixed number of aircraft case, the leading edge of the Pareto fronts with the same arrival-to-departure ratio was retrieved and saved.When every simulation was done, the global leading edge was computed from the ten different leading edges.The global delay range was sampled with a time delay bin size of five seconds, and the average number of controller interventions was computed for each time delay bin.The standard deviation of the number of controller interventions was also computed for each time delay bin. Figure 12 and Figure 13 show the resulting curves after identifying the different ratio values tested. +AnalysisFigure 11 provides the four different scenario distributions corresponding to different numbers of aircraft.The figures show that the number of scenarios having a solution in each of the 4 cases follow a distribution with similar properties to a standard normal curve.However, as the number of aircraft is increased the distributions are shifted to the right in time and they present fewer scenarios having a solution.Two trends can be extracted from Figure 12.On one hand the ratio variation does not influence the results as much as the total number of aircraft.As the number of aircraft is increased, the Pareto fronts are shifted to the upperright corner of the figures and the computed delay range is vertically translated to the right.Moreover, the number of controller interventions increases with the aircraft number considered in the scenarios.However when observing individually the different cases, it can be seen that larger ratios produce less number of controller interventions than smaller ratios for the whole range of time delay.Smaller ratios corresponds to scenarios with larger number of departures whereas larger ratios corresponds to scenarios with larger number of arrivals.Increasing the number of departures increases the number of controller interventions because departing aircraft are slower than arriving aircraft and because they require larger separation distances than arrivals.Wake vortex separations are imposed at the runway and larger amounts of uncertainty are modeled for departures than arrivals in this work.According to Figure 13, a similar result trend can be observed for all case studies.As the number of aircraft considered in the scenarios is increased, the standard deviation of the number of controller interventions is shifted to the right in time.Moreover from this figure, high confidence can be given to these results.Most of the data points obtained have a low standard deviation and the number of outliers is low.Figure 13 shows that lower amount of data points can be observed for large ratios when the number of aircraft is 18 or 20.This confirms previous trends.Finally when focusing on the most represented time delay range, the variability of the averaged number of controller interventions is low and the deviation of the average does not exceed one.Therefore, this experiment is indicative of algorithm robustness. +C. DiscussionsBecause of the GPU introduction and accelerating strategies investigated, the significant computation time speedup allows the stochastic scheduler to be run hundreds of scenarios under reasonable time for fast-time simulations.The experiments that were designed enable a diverse range of scenarios to be tested and help to identify trade-offs of integrated departures and arrivals between reduced number of controller interventions and low delays.It is important to keep in mind that the schedules were generated randomly and therefore do not necessarily represent realistic conditions.However they covered parameter values corresponding to typical operations for the partial flows considered in the Los Angeles terminal airspace.The goal of the proposed methodology was to characterize the algorithm and its solution when computing very different sets of schedules.In reality, searching spaces of decision variables are often smaller in operations.Testing only real schedules obtained from historical data was not the focus of this study.In this work, it was shown that when dense traffic conditions (number of aircraft greater than 20) are considered within a 30-minute time range, the algorithm is not often able to compute a solution because the algorithm cannot assign delays greater than the delay range defined in the code.However when applied to the Los Angeles terminal airspace, this trend corresponds to traffic conditions 40% greater than typical operations.Therefore this is not a problem for the Los Angeles application.The program was able to compute solutions for busy operations and show compromise solutions to help decision maker reduce controller interventions and achieve low delays.Moreover, it was found that varying the traffic density affects more significantly the results than varying the arrival-to-departure ratio.The total number of aircraft considered and the number of departures have the biggest impacts on the results obtained with this algorithm.The random nature of the scenario generation allowed the production of extreme valued schedules in both flight times and total number of aircraft.In the experiments, the algorithm was pushed to its limits when computing such scenarios.Within the thousands of scenarios tested, a range of schedules can be identified for which the algorithm computes realistic solutions.It would require additional work to further analyze their properties.Although the acceleration methods proposed in this work were applied to stochastic scheduling, speedup gain and insights from this study can be used for other applications requiring real time application.Example of applications are search algorithms (e.g.genetic/evolutionary algorithms) and machine learning (e.g.neural networks). +VI. ConclusionA GPU implementation of Monte Carlo simulations was developed using a mixed CPU-GPU development environment to obtain a massive computation parallelization.The GPU implementation of the stochastic scheduler is a success.Transferring the Monte Carlo computations to the GPU enabled a speedup of 637x.When the entire algorithm is parallelized on the mixed CPU-GPU development environment, the computation time is about 2.5 minutes to run a representative traffic flow scenario for partial flows in the Los Angeles terminal airspace.This computation time represents an acceleration of 154x the original run time.Despite relatively important CPU overhead times, the GPU is still able to accelerate the entire process and the speedup performance enabled to solve fast an air traffic management problem.To support a traffic variation analysis, several scheduling experiments were generated and tested on a model of the Los Angeles terminal airspace covering partial flows.The new computation time enabled testing different traffic configurations, characterize the computed solutions and elucidate the behavior of the algorithm itself.Although the scenarios were generated randomly and some of them did not represent realistic conditions, the scenarios that were tested produced realistic amount of delays and number of controller interventions.When comparing the generated scenarios with typical operations on the partial flows modeled in the Los Angeles terminal airspace, the program was able to compute solutions for traffic conditions up to 40% more than typical operations.There are two main takeaways from the results.First, there exists a clear compromise between achieving low time delays and low number of controller interventions.Moreover, the exploration of a variety of schedule-scenarios and the low dispersion of the results proves great algorithm robustness.Second, it was found that the variation of traffic density and the number of departures have greater impacts on the results than the variation of arrival-to-departure ratios.Moreover, it was found that for scenarios having extreme valued flight times or extreme number of aircraft, the algorithm could not compute a solution because of the fixed authorized delay range in the code.To complete the traffic variation analysis, future work should be conducted to test scenarios generated from historical data.This paper proves that massive parallel implementation on Graphics Processing Units enables fast/real time solution to air traffic management applications.The integrated arrival/departure scheduling problem with uncertainty can be solved in realistic times when using GPUs.This approach outperforms the traditional serial implementation on CPUs developed in previous work despite the use of multi-threading on serial processors.This paper encourages the use of such techniques to remove research struggles associated with air traffic management numerical optimization techniques.Figure 1 :1Figure 1: Route interactions between arrivals and departures in the LA terminal airspace +Figure 2 :2Figure 2: Flight time perturbations setup +Figure 3 :3Figure 3: Integrated arrivals and departures scheduler with uncertainty -Original Implementation 13 +Algorithm 1 : 2 : 3 :: end for 5 : 8 :12358A single Monte Carlo simulation evaluating the uncertainty cost for a set of schedule-scenarios 1: procedure MONTE CARLO(mean, standardDeviation, delay, workloadCounter) for all schedules ∈ setof (departure, arrival)f lights do computation on CPU Generate random number from given normal distribution create uncertainties on departure/arrival times 4for M onteCarloSimu = 0 → maxN umberM onteCarloSim do computation on GPU 6: Delay and intervention function evaluation 1 MC sim.per thread return Delay and interventionCounter 7: end for for all individual ∈ (GApopulation) do computation on CPU 9: +Figure 4 :4Figure 4: Integrated arrivals and departures scheduler with uncertainty -Final implementation in current study +Figure 5 :5Figure 5: Modification of the original flow chart (see Figure 3), resulting in less host-device communication +Figure 6 :6Figure 6: GPU-occupancy study for 128 threads, 32 registers and 167 bytes of shared memory +Figure 7 :7Figure 7: Distribution of scenarios having a solution in the delay range [0,2000] seconds +Figure 8 :Figure 9 :89Figure 8: Average distribution of the number of controller interventions in the delay range [0,2000] seconds +Figure 10 :10Figure 10: Number of scenarios to discard of the solution set +Figure 12 : 20 Figure 13 :122013Figure 12: Average distribution of the number of controller interventions in the delay range [0,2000] seconds + + + + + + + + +Table 1 :1Effect of CUDA function call grouping on Monte Carlo computation timesNumber of CUDA malloc and CUDA memcpySize unitMonte Carlo computation time using GPU for one generation832 CUDA malloc and 832 CUDA memcpy1 individual26.75 secs1 CUDA malloc and 1 CUDA memcpy832 individuals0.327 secs +Table 2 :2Comparison of different random number generatorsGeneratorTypeLocation Average running time1cuRAND library, MRG32k3aGPU1.97 ms2cuRAND library, XORWOWGPU2.08 ms3Boost library, Mersenne 19937CPU2.65 ms4Boost library, Ecuyer 1988CPU2.60 ms5C-function GaussianCPU1.2 ms +Table 3 :3GPU-occupancy study# threads Active thread blocks Occupancy641650%12816100%2568100%5124100% +Table 4 :4GPU and CPU Computation times for one generation using different number of threads# threads GPU-kernel Computation Time Algorithm Computation Time1280.031 secs0.382 sec2560.032 secs0.388 sec5120.034 secs0.391 sec +Table 5 :5Comparison of computation times between original and current version of the codeCode compared# seeds # generationsOriginal codeCurrent code using 2 GPUs SpeedupMonte Carlo simulations1118.5 secs0.029 secs637xTotal1040023,400 secs (6.5h)152.19 secs154x +Table 6 :6Scheduled initial timesFlight Order FIM Arrivals (sec) RWY Departures (sec)13968244616537283634110652951332161361475183071613NA81770NA +Table 7 :7Experiments setup 70% greater than typical) traffic operations.For each number of aircraft, a random number of arrivals is computed.For example, if the number of aircraft is fixed to 18, the number of arrivals can take any integer value between one and 18.Because 400 scenarios are generated per value of the number of aircraft, up to 400 different arrival-to-departure ratios can be generated.The lower fixed ratio is 0.05 and represents almost only departures (e.g. one arrival and 20 departures) whereas the maximum fixed ratio is 8 and represents mainly arrivals (e.g.16 arrivals and two departures).Based on average traffic data for the partial flows considered in the Los Angeles terminal airspace, any ratio equal to 3 ± 20% represents typical operations.ExperimentVaried parameterRange of values tested1total number of aircraft14 to 242arrival-to-departure ratio0.05 to 8.00with higher than normal ( + of 22 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 20, 2014 | http://arc.aiaa.org| DOI: 10.2514/6.2014-2024 + of 22 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 20, 2014 | http://arc.aiaa.org| DOI: 10.2514/6.2014-2024 + of 22 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 20, 2014 | http://arc.aiaa.org| DOI: 10.2514/6.2014-2024 + of 22 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 20, 2014 | http://arc.aiaa.org| DOI: 10.2514/6.2014-2024 + of 22 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 20, 2014 | http://arc.aiaa.org| DOI: 10.2514/6.2014-2024 + Downloaded by NASA AMES RESEARCH CENTER on June 20, 2014 | http://arc.aiaa.org| DOI: 10.2514/6.2014-2024 + of 22 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 20, 2014 | http://arc.aiaa.org| DOI: 10.2514/6.2014-2024 + of 22 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 20, 2014 | http://arc.aiaa.org| DOI: + 10.2514/6.2014-2024 + of 22 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 20, 2014 | http://arc.aiaa.org| DOI: 10.2514/6.2014-2024 + of 22 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 20, 2014 | http://arc.aiaa.org| DOI: 10.2514/6.2014-2024 + of 22 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 20, 2014 | http://arc.aiaa.org| DOI: 10.2514/6.2014-2024 + of 22 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 20, 2014 | http://arc.aiaa.org| DOI: 10.2514/6.2014-2024 + of 22 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 20, 2014 | http://arc.aiaa.org| DOI: 10.2514/6.2014-2024 + + + + +American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 20, 2014 | http://arc.aiaa.org| DOI: 10.2514/6.2014-2024 + + + + +Mail Stop 210-8.AIAA senior member ‡ Computer Engineer, Aerospace High Density Operations Branch.Mail Stop 210-6.AIAA senior member + + + + + + + + + + + The dynamic scheduling of aircraft in the near terminal area + + RGDear + + 10.1016/0041-1647(78)90133-8 + + + Transportation Research + Transportation Research + 0041-1647 + + 12 + 3 + + 1976 + Elsevier BV + Cambridge, Mass + + + Flight Transportation Laboratory, Massachusetts Institute of Technology + + + Tech. rep. + Dear, R. G., "The Dynamic Scheduling of Aircraft in the Near Terminal Area," Tech. rep., Cambridge, Mass.: Flight Transportation Labo- ratory, Massachusetts Institute of Technology, 1976. + + + + + An algorithm for computer assisted sequencing and scheduling of terminal area operations + + RogerGDear + + + YosefSSherif + + 10.1016/0191-2607(91)90132-a + + + Transportation Research Part A: General + Transportation Research Part A: General + 0191-2607 + + 25 + 2-3 + + 1991 + Elsevier BV + + + Dear, R. G. and Sherif, Y. S., "An algorithm for Computer Assisted Sequencing and Scheduling of Terminal Area Operations," Transporta- tion Research Part A: General, Vol. 25, No. 2, 1991, pp. 129-139. + + + + + Analysis of Delay Reducing and Fuel Saving Sequencing and Spacing Algorithms for Arrival Traffic + + FNeuman + + + HErzberger + + + 1991 + Citeseer + + + Neuman, F. and Erzberger, H., Analysis of Delay Reducing and Fuel Saving Sequencing and Spacing Algorithms for Arrival Traffic, Citeseer, 1991. + + + + + Scheduling Aircraft Landings Under Constrained Position Shifting + + HamsaBalakrishnan + + + BalaChandran + + 10.2514/6.2006-6320 + + + AIAA Guidance, Navigation, and Control Conference and Exhibit + Keystone, CO + + American Institute of Aeronautics and Astronautics + 2006 + + + Balakrishnan, H. and Chandran, B., "Scheduling Aircraft Landings under Constrained Position Shifting," AIAA Guidance, Navigation, and Control Conference, Keystone, CO, 2006. + + + + + Scheduling Aircraft Landings—The Static Case + + JEBeasley + + + MKrishnamoorthy + + + YMSharaiha + + + DAbramson + + 10.1287/trsc.34.2.180.12302 + + + Transportation Science + Transportation Science + 0041-1655 + 1526-5447 + + 34 + 2 + + 2000 + Institute for Operations Research and the Management Sciences (INFORMS) + + + Beasley, J. E., Krishnamoorthy, M., Sharaiha, Y. M., and Abramson, D., "Scheduling Aircraft Landings The Static Case," Transportation science, Vol. 34, No. 2, 2000, pp. 180-197. + + + + + Scheduling Aircraft Landings to Closely Spaced Parallel Runways + + MKupfer + + + + Eighth USA/Europe Air Traffic Management Research and Development Seminar (ATM2009) + Napa, CA + + June 2009 + + + + Kupfer, M., "Scheduling Aircraft Landings to Closely Spaced Parallel Runways," Eighth USA/Europe Air Traffic Management Research and Development Seminar (ATM2009), Napa, CA, June 2009, pp. 1-10. + + + + + A Mixed Integer Linear Program for Airport Departure Scheduling + + GautamGupta + + + WaqarMalik + + + YoonJung + + 10.2514/6.2009-6933 + + + 9th AIAA Aviation Technology, Integration, and Operations Conference (ATIO) + Hilton Head, South Carolina + + American Institute of Aeronautics and Astronautics + 2009 + + + Gupta, G., Malik, W., and Jung, Y. C., "A Mixed Integer Linear Program for Airport Departure Scheduling," 9th AIAA Aviation Technology, Integration, and Operations Conference (ATIO), Hilton Head, South Carolina, 2009. + + + + + A Metaheuristic Approach to Aircraft Departure Scheduling at London Heathrow Airport + + JasonA DAtkin + + + EdmundKBurke + + + JohnSGreenwood + + + DaleReeson + + 10.1007/978-3-540-73312-6_12 + + + Lecture Notes in Economics and Mathematical Systems + Berlin Heidelberg, Berlin, Heidelberg + + Springer Berlin Heidelberg + 2008 + + + + Atkin, J. A. D., Burke, E. K., Greenwood, J. S., and Reeson, D., "A Metaheuristic Approach to Aircraft Departure Scheduling at London Heathrow Airport," Springer Berlin Heidelberg, Berlin, Heidelberg, 2008, pp. 235-252. + + + + + A Generalized Dynamic Programming Approach for a Departure Scheduling Problem + + SivakumarRathinam + + + ZacharyWood + + + BanavarSridhar + + + YoonJung + + 10.2514/6.2009-6250 + + + AIAA Guidance, Navigation, and Control Conference + Chicago, IL + + American Institute of Aeronautics and Astronautics + 2009 + + + Rathinam, S., Wood, Z., Sridhar, B., and Jung, Y. C., "A generalized dynamic programming approach for a departure scheduling problem," AIAA Guidance, Navigation, and Control Conference, Chicago, IL, 2009. + + + + + A Dynamic Programming Algorithm for Robust Runway Scheduling + + BalaChandran + + + HamsaBalakrishnan + + 10.1109/acc.2007.4282922 + + + 2007 American Control Conference + New York, NY + + IEEE + 2007 + + + + Chandran, B. and Balakrishnan, H., "A Dynamic Programming Algorithm for Robust Runway Scheduling," American Control Conference, 2007. ACC '07, New York, NY, 2007, pp. 1161-1166. + + + + + Scheduling of airport runway operations using stochastic branch and bound methods + + GustafSölveling + + + John-PaulClarke + + 10.1016/j.trc.2014.02.021 + + + Transportation Research Part C: Emerging Technologies + Transportation Research Part C: Emerging Technologies + 0968-090X + + 45 + + Aug. 2012 + Elsevier BV + + + Georgia Institute of Technology + + + Ph.D. thesis + Sölveling, G., Stochastic Programming Methods for Scheduling of Airport Runway Operations under Uncertainty, Ph.D. thesis, Georgia Institute of Technology, Aug. 2012. + + + + + Optimal Integration of Departures and Arrivals in Terminal Airspace + + MinXue + + + ShannonZelinski + + 10.2514/6.2012-4977 + + + AIAA Guidance, Navigation, and Control Conference + Mineapolis, MN + + American Institute of Aeronautics and Astronautics + August 2012 + + + Xue, M. and Zelinski, S., "Optimal Integration of Departures and Arrivals in Terminal Airspace," AIAA Guidance, Navigation, and Control Conference, Mineapolis, MN, August 2012. + + + + + Optimization of Integrated Departures and Arrivals Under Uncertainty + + MinXue + + + ShannonZelinski + + 10.2514/6.2013-4322 + + + 2013 Aviation Technology, Integration, and Operations Conference + Los Angeles, CA + + American Institute of Aeronautics and Astronautics + August 2013 + + + + Xue, M. and Zelinski, S., "Optimization of Integrated Departures and Arrivals Under Uncertainty," AIAA 13th Aviation Technology, Integra- tion, and Operation Conference (ATIO), Los Angeles, CA, August 2013, pp. 1-9. + + + + + Uncertainty Analysis of Integrated Departures and Arrivals: A Los Angeles Case Study + + MinXue + + + ShannonZelinski + + + DanielGMulfinger + + 10.2514/6.2013-4347 + + + 2013 Aviation Technology, Integration, and Operations Conference + Los Angeles, CA + + American Institute of Aeronautics and Astronautics + August 2013 + + + + Xue, M., Zelinski, S., and Mulfinger, D., "Uncertainty Study of Integrated Departures and Arrivals: A Los Angeles Case Study," AIAA 13th Aviation Technology, Integration, and Operation Conference (ATIO), Los Angeles, CA, August 2013, pp. 1-10. + + + + + Multiple-Point Integrated Scheduling of Terminal Area Traffic + + HemingChen + + + YiyuanJZhao + + + ChrisProvan + + 10.2514/1.c031332 + + + Journal of Aircraft + Journal of Aircraft + 0021-8669 + 1533-3868 + + 48 + 5 + + 2011 + American Institute of Aeronautics and Astronautics (AIAA) + + + Chen, H., Zhao, Y. J., and Provan, C., "Multiple-Point Integrated Scheduling of Terminal Area Traffic," Journal of Aircraft, Vol. 48, No. 5, 2011, pp. 1646-1657. + + + + + Dynamic Real-time Scheduling of Terminal Traffic + + HemingChen + + + YiyuanZhao + + + ChrisProvan + + 10.2514/6.2011-6364 + + + AIAA Guidance, Navigation, and Control Conference + Portland, Oregon + + American Institute of Aeronautics and Astronautics + August 2011 + + + Chen, H., Zhao, Y. J., and Provan, C., "Dynamic Real-time Scheduling of Terminal Traffic," AIAA Guidance, Navigation, and Control Conference, Portland, Oregon, August 2011. + + + + + A Hybrid Optimization Approach to Air Traffic Management for Metroplex Operations + + BrianCapozzi + + + StephenAtkins + + 10.2514/6.2010-9062 + + + 10th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference + Forth Worth, TX + + American Institute of Aeronautics and Astronautics + 2010 + + + Capozzi, B. and Atkins, S., "A Hybrid Optimization Approach to Air Traffic Management for Metroplex Operations," 10th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference, Forth Worth, TX, 2010. + + + + + Towards Optimal Routing and Scheduling of Metroplex Operations + + BrianCapozzi + + + StephenAtkins + + + SeongimChoi + + 10.2514/6.2009-7037 + + + 9th AIAA Aviation Technology, Integration, and Operations Conference (ATIO) + Hilton Head, South Carolina + + American Institute of Aeronautics and Astronautics + 2009 + + + Capozzi, B., Atkins, S., and Choi, S., "Towards Optimal Routing and Scheduling of Metroplex Operations," 9th AIAA Aviation Technology, Integration, and Operations Conference (ATIO), Hilton Head, South Carolina, 2009. + + + + + Analysis of Dependencies and Impacts of Metroplex Operations + + DADelaurentis + + + SAyyalasomayajula + + + + Tech. rep + + 2010 + + + DeLaurentis, D. A. and Ayyalasomayajula, S., "Analysis of Dependencies and Impacts of Metroplex Operations," Tech. rep., 2010. + + + + + Discussion and Comparison of Metroplex-Wide Arrival Scheduling Algorithms + + AdityaSaraf + + + John-PaulClarke + + + EvanMcclain + + 10.2514/6.2010-9258 + + + 10th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference + Forth Worth, TX + + American Institute of Aeronautics and Astronautics + 2010 + + + Saraf, A., Clarke, J.-P., and McClain, E., "Discussion and Comparison of Metroplex-Wide Arrival Scheduling Algorithms," 10th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference, Forth Worth, TX, 2010. + + + + + Observation and measurement of metroplex phenomena + + StephenAtkins + + 10.1109/dasc.2008.4702816 + + + 2008 IEEE/AIAA 27th Digital Avionics Systems Conference + + IEEE + 2008. 2008 + 3 + + + DASC 2008. IEEE/AIAA 27th + Atkins, S., "Observation and Measurement of Metroplex Phenomena," Digital Avionics Systems Conference, 2008. DASC 2008. IEEE/AIAA 27th, 2008, p. 3. + + + + + A simulation-based method for estimating metroplex efficiency + + LeihongLi + + + JongWookPark + + + John-PaulClarke + + 10.1109/dasc.2011.6096199 + IEEE/AIAA 30th + + + 2011 IEEE/AIAA 30th Digital Avionics Systems Conference + + IEEE + 2011. 2011 + 2 + + + Li, L., Park, J. W., and Clarke, J.-P., "A Simulation-based Method for Estimating Metroplex Efficiency," Digital Avionics Systems Conference (DASC), 2011 IEEE/AIAA 30th, 2011, p. 2. + + + + + Design Considerations for a New Terminal Area Arrival Scheduler + + JaneThipphavong + + + DanielMulfinger + + + AlexanderSadovsky + + 10.2514/6.2010-9290 + + + 10th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference + Forth Worth, TX; Zone, N. D. + + American Institute of Aeronautics and Astronautics + September 2010 + 24 + + + CUDA Toolkit Documentation + Thipphavong, J. and Mulfinger, D., "Design Considerations for a New Terminal Area Arrival Scheduler," 10th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference, Forth Worth, TX, September 2010. 24 Zone, N. D., "CUDA Toolkit Documentation," . + + + + + CURAND :: CUDA Toolkit Documentation + + NDZone + + + + Zone, N. D., "CURAND :: CUDA Toolkit Documentation," . + + + + + Accelerating Compute-Intensive Applications with GPUs and FPGAs + + ShuaiChe + + + JieLi + + + JeremyWSheaffer + + + KevinSkadron + + + JohnLach + + 10.1109/sasp.2008.4570793 + + + 2008 Symposium on Application Specific Processors + + IEEE + 2008 + + + + Che, S., Li, J., Sheaffer, J. W., Skadron, K., and Lach, J., "Accelerating Compute-Intensive Applications with GPUs and FPGAs," 2008 Symposium on Application Specific Processors (SASP), IEEE, 2008, pp. 101-107. + + + + + GPU accelerated computing–from hype to mainstream, the rebirth of vector computing + + SatoshiMatsuoka + + + TakayukiAoki + + + ToshioEndo + + + AkiraNukada + + + ToshihiroKato + + + AtushiHasegawa + + 10.1088/1742-6596/180/1/012043 + + + Journal of Physics: Conference Series + J. Phys.: Conf. Ser. + 1742-6596 + + 180 + 012043 + August 2009 + IOP Publishing + + + Matsuoka, S., Aoki, T., Endo, T., Nukada, A., Kato, T., and Hasegawa, A., "GPU accelerated computing-from hype to mainstream, the rebirth of vector computing," Journal of Physics: Conference Series, Vol. 180, August 2009, pp. 012043. + + + + + The case for GPGPU spatial multitasking + + JacobTAdriaens + + + KatherineCompton + + + NamSungKim + + + MichaelJSchulte + + 10.1109/hpca.2012.6168946 + + + IEEE International Symposium on High-Performance Comp Architecture + + IEEE + 2012 + + + + Adriaens, J. T., Compton, K., Kim, N. S., and Schulte, M. J., "The case for GPGPU spatial multitasking," 2012 IEEE 18th International Symposium on High Performance Computer Architecture (HPCA), 2012, pp. 1-12. + + + + + Real-time nonlinear finite element computations on GPU – Application to neurosurgical simulation + + GrandRomanJoldes + + + AdamWittek + + + KarolMiller + + 10.1016/j.cma.2010.06.037 + + + Computer Methods in Applied Mechanics and Engineering + Computer Methods in Applied Mechanics and Engineering + 0045-7825 + + 199 + 49-52 + + 2010 + Elsevier BV + + + Joldes, G. R., Wittek, A., and Miller, K., "Real-time nonlinear finite element computations on GPU-Application to neurosurgical simula- tion," Computer methods in applied mechanics and engineering, 2010, pp. 3305-3314. + + + + + Accelerating advanced mri reconstructions on gpus + + SamuelSStone + + + JustinPHaldar + + + StephanieCTsao + + + Wen-Mei W.Hwu + + + Zhi-PeiLiang + + + BradleyPSutton + + 10.1145/1366230.1366276 + + + Proceedings of the 5th conference on Computing frontiers + the 5th conference on Computing frontiersIschia, Italy + + ACM + 2008 + 251 + + + Stone, S., Haldar, J., Tsao, S., Hwu, W.-m., Liang, Z.-P., and Sutton, B., "Accelerating Advanced MRI Reconstructions on GPUs ," CF'08, ACM Press, Ischia, Italy, 2008, p. 251. + + + + + GPU based sparse grid technique for solving multidimensional options pricing PDEs + + AbhijeetGaikwad + + + IoaneMuniToke + + 10.1145/1645413.1645419 + + + Proceedings of the 2nd Workshop on High Performance Computational Finance + the 2nd Workshop on High Performance Computational FinancePortland, Oregon + + ACM + 2009 + + + + Gaikwad, A. and Toke, I. M., "GPU based Sparse Grid Technique for Solving Multidimensional Options Pricing PDEs," the 2nd Workshop, ACM Press, Portland, Oregon, 2009, pp. 1-9. + + + + + Parallel Iterative Linear Solvers on GPU: A Financial Engineering Case + + AbhijeetGaikwad + + + IoaneMuniToke + + 10.1109/pdp.2010.55 + + + 2010 18th Euromicro Conference on Parallel, Distributed and Network-based Processing + + IEEE + 2010 + + + Gaikwad, A. and Toke, I. M., "Parallel Iterative Linear Solvers on GPU: A Financial Engineering Case," 2010 18th Euromicro Conference on Parallel, Distributed and Network-based Processing, 2010. + + + + + Massively Parallel Optimal Solution to the Nationwide Traffic Flow Management Problem + + MonishDTandale + + + SandyWiraatmadja + + + VeeraVVaddi + + + JosephLRios + + 10.2514/6.2013-4349 + + + 2013 Aviation Technology, Integration, and Operations Conference + Los Angeles, CA + + American Institute of Aeronautics and Astronautics + August 2013 + + + Tandale, M. D., Wiraatmadja, S., Vaddi, V., and Rios, J. L., "Massively Parallel Optimal Solution to the Nationwide Traffic Flow Manage- ment Problem," AIAA 13th Aviation Technology, Integration, and Operation Conference (ATIO), Los Angeles, CA, August 2013. + + + + + Implementation Method of Genetic Algorithms to the CUDA Environment using Data Parallelization + + MasashiOiso + + + YoshiyukiMatsumura + + + ToshiyukiYasuda + + + KazuhiroOhkura + + 10.3156/jsoft.23.18 + + + Journal of Japan Society for Fuzzy Theory and Intelligent Informatics + J. SOFT + 1347-7986 + 1881-7203 + + 23 + 1 + + 2011. July 2013 + Japan Society for Fuzzy Theory and Intelligent Informatics + + + Oiso, M., Matsumura, Y., Yasuda, T., and Ohkura, K., "Implementic Genetic Algorithms to CUDA Environment Using Data Parallelization," Tehnički vjesnik 18, 4(2011), 511-517, July 2013, pp. 1-7. + + + + + Gpu-based acceleration of the genetic algorithm + + RShah + + + PNarayanan + + + KKothapalli + + + PJaros + + + J + + + + Proceedings of The 3rd Workshop on Parallel Architectures for Bio-inspired Algorithms(WPABA) in conjunction with Parallel Architectures for Compilation Techniques (PACT'10) + The 3rd Workshop on Parallel Architectures for Bio-inspired Algorithms(WPABA) in conjunction with Parallel Architectures for Compilation Techniques (PACT'10)Vienna, Austria + + 2010. 36 Pospíchal,. 2009 + + + GPU-Accelerated Genetic Algorithms. GECCO'09 competition + Shah, R., Narayanan, P., and Kothapalli, K., "GPU-Accelerated Genetic Algorithms," Proceedings of The 3rd Workshop on Parallel Archi- tectures for Bio-inspired Algorithms(WPABA) in conjunction with Parallel Architectures for Compilation Techniques (PACT'10), Vienna, Austria, 2010. 36 Pospíchal, P. and Jaros, J., "Gpu-based acceleration of the genetic algorithm," GECCO'09 competition, 2009. + + + + + A Parallel Genetic Algorithm with GPU Accelerated for Large-scale MDVRP in Emergency Logistics + + JianmingLi + + + XimengLv + + + LinlinLiu + + 10.1109/cse.2011.106 + + + 2011 14th IEEE International Conference on Computational Science and Engineering + + IEEE + 2011. 2011 + + + + IEEE + Li, J., Lv, X., and Liu, L., "A Parallel Genetic Algorithm with GPU Accelerated for Large-scale MDVRP in Emergency Logistics," 2011 IEEE 14th International Conference on Computational Science and Engineering (CSE), 2011, pp. 602-605. + + + + + Optimization of parallel Genetic Algorithms for nVidia GPUs + + MohamedWahib + + + AsimMunawar + + + MasaharuMunetomo + + + KiyoshiAkama + + 10.1109/cec.2011.5949701 + + + 2011 IEEE Congress of Evolutionary Computation (CEC) + + IEEE + 2011. 2011 + + + + Wahib, M., Munawar, A., Munetomo, M., and Akama, K., "Optimization of Parallel Genetic Algorithms for NVIDIA GPUs," 2011 IEEE Congress on Evolutionary Computation (CEC), 2011, pp. 803-811. + + + + + Fast Genetic Programming on GPUs + + SimonHarding + + + WolfgangBanzhaf + + 10.1007/978-3-540-71605-1_9 + + + Lecture Notes in Computer Science + Berlin Heidelberg, Berlin, Heidelberg + + Springer Berlin Heidelberg + 2007 + + + + Genetic Programming + Harding, S. and Banzhaf, W., "Fast Genetic Programming on GPUs," Genetic Programming, Springer Berlin Heidelberg, Berlin, Heidelberg, 2007, pp. 90-101. + + + + + Super Density Operations Airspace Modeling for the Southern California Metroplex + + SebastianTimar + + + GauravNagle + + + AdityaSaraf + + + PeterYu + + + PeterHunt + + + AndrewTrapani + + + NickJohnson + + 10.2514/6.2011-6535 + + + AIAA Modeling and Simulation Technologies Conference + + American Institute of Aeronautics and Astronautics + 2011 + + + Timar, S. D., Nagle, G., Saraf, A., Yu, P., Hunt, P., Trapani, A., and Johnson, N., "Super Density Operations Airspace Modeling for the Southern California Metroplex," AIAA Modeling and Simulation Technology Conference, 2011. + + + + + Short-Term National Airspace System Delay Prediction Using Weather Impacted Traffic Index + + BanavarSridhar + + + NeilYChen + + 10.2514/1.38798 + + + Journal of Guidance, Control, and Dynamics + Journal of Guidance, Control, and Dynamics + 0731-5090 + 1533-3884 + + 32 + 2 + + 2009 + American Institute of Aeronautics and Astronautics (AIAA) + + + Sridhar, B. and Chen, N., "Short-Term National Airspace System Delay Prediction Using Weather Impacted Traffic Index," Journal of guidance, control, and dynamics, Vol. 32, No. 2, 2009, pp. 657-662. + + + + + Trajectory-Based Takeoff Time Predictions Applied to Tactical Departure Scheduling: Concept Description, System Design, and Initial Observations + + ShawnEngelland + + + RichardCapps + + 10.2514/6.2011-6875 + + + 11th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference + Virginia Beach, VA + + American Institute of Aeronautics and Astronautics + 2011. 2014 + 43 + + + Box-Muller transform -Wikipedia, The Free Encyclopedia + Engelland, S. A. and Capps, A., "Trajectory-Based Takeoff Time Predictions Applied to Tactical Departure Scheduling: Concept Descrip- tion, System Design, and Initial Observations," AIAA Aviation Technology, Integration, and Operations Conference (ATIO), Virginia Beach, VA, 2011. 43 "Box-Muller transform -Wikipedia, The Free Encyclopedia," 2014. + + + + + Approximating the Pareto Front of Multi-criteria Optimization Problems + + JulienLegriel + + + ColasLe Guernic + + + ScottCotton + + + OdedMaler + + 10.1007/978-3-642-12002-2_6 + + + Tools and Algorithms for the Construction and Analysis of Systems + + Springer Berlin Heidelberg + January 2010 + + + + Legriel, J., Le Guernic, C., Cotton, S., and Maler, O., "Approximating the Pareto Front of Multi-Criteria Optimization Problems," January 2010, pp. 1-15. + + + + + Pareto Multi Objective Optimization + + PNgatchou + + + AZarei + + + AEl-Sharkawi + + 10.1109/isap.2005.1599245 + + + Proceedings of the 13th International Conference on, Intelligent Systems Application to Power Systems + the 13th International Conference on, Intelligent Systems Application to Power Systems + + IEEE + 2005. 2005 + + + + Ngatchou, P., Zarei, A., and El-Sharkawi, M., "Pareto multi objective optimization," Proceedings of the 13th International Conference on Intelligent Systems Application to Power Systems, 2005., IEEE, 2005, pp. 84-91. + + + + + + diff --git a/file12.txt b/file12.txt new file mode 100644 index 0000000000000000000000000000000000000000..b4b574cd91b4e574e7d7e84425cc3bbb9152ce29 --- /dev/null +++ b/file12.txt @@ -0,0 +1,157 @@ + + + + +INTRODUCTIONA number of analytic and simulation studies have attempted to assess the potential benefits resulting from deployment of the Final Approach Spacing Tool (FAST) that is a part of the Center TRACON Automation System (CTAS). 1,2,3,4 On of the primary sources of FAST benefits is the increased precision of control, which is presumed to reduce the average landing time intervals (LTIs) at each runway.In general, it is assumed that achieved separations contain some amount of excess spacing (not required by separation standards) and that by allowing more precise control, this excess is reduced in a uniform way for all arrivals to which FAST advisories are applied.By saving a few seconds of runway time for each arriving pair, this mechanism provides an increase in runway ________________ capacity.The delay savings that accrue over an extended period of operation are found by integrating the delay reductions achieved over a variety of traffic and weather conditions.In this paper, data from actual airport operations is analyzed and applied to the problem of validating, calibrating, and extending the model for the key FAST benefits mechanism -landing time interval reduction.The analysis of actual operations data is also helpful in prioritizing research activities to focus upon areas where the greatest opportunity lies.This work extends the capabilities used in earlier data analysis conducted by Boswell and Ballin and Erzberger. 1,2 he emphasis is upon robust statistical measures that can be produced through automated analysis routines, thus enabling large amounts of data to be analyzed. +PACKAGE FOR ANALYSIS OF RUNWAY OPERATIONS (PARO)All major airports acquire and archive radar data on traffic in the terminal area using the Automated Radar Terminal System (ARTS).When combined with basic flight plan information and knowledge of the runway layout, this data provides insight into the flow rates into the terminal and the manner in which particular runways were being utilized.A software package called Package for Analysis of Runway Operations (PARO) was written to automatically process such data and produce analyses relevant to the efficiency of operations.PARO is written in the C++ programming language.Data analyses presented in this paper will focus primarily upon analysis of four DFW data sets that were available during the software development period.These data sets were used to develop the analysis techniques and give some preliminary insight into AFAST benefits questions.Analysis of additional sets of data are being analyzed currently.PARO processing takes place in three phases designated G0, G1 and G2.Phase G0 involves the reading of raw data files, correcting certain errors and anomalies, and producing new radar data files.In the original data files, tracks appear in order of the time of the first radar report in the track.Phase G1 involves reading the G0 radar data files, correcting and validating the input data, estimating velocities, and conducting certain analyses that require complete track data.Phase G2 involves reading and processing the summary data files produced from G1 processing.Among the variables that may be analyzed are the path length flown, the time of crossing the outer marker, the interarrival interval relative to the preceding arrival, etc.By operating only upon the summary files, G2 analyses can run more rapidly without having to process the more voluminous track data. +Bayesian Runway Assignment Procedure (BRAP)The ability to properly assign each observed operation to a particular runway is essential for reliable analysis of multi-runway operations.If radar data were complete and of sufficient accuracy, such assignment might require a simple comparison of the surface intercept projection of tracks with the known runway locations.However, several imperfections in the radar data (particularly altitude coverage limitations) lead to the need for a somewhat more sophisticated approach to runway assignment.A Bayesian approach to runway assignment has been developed as part of PARO.Under the Bayesian approach, the runway assignment is viewed both as a parameter that determines the likelihood of any given set of radar observations and as a random variable that has its own probability distribution.The Bayesian approach allows an optimum utilization of all available information about how runways are being used and what was observed with radar.The result is a runway assignment algorithm that is more accurate than any algorithm based solely upon radar data for a single track. +Data CompletenessThe completeness of the data is of great concern when evaluating the efficiency of airport operations.Missing tracks create gaps in the arrival stream that can be mistakenly attributed to system inefficiencies.As a general rule, data should be approximately 99% complete to perform all the PARO analyses of interest.(That is, not more than 1 aircraft in 100 should be missing from the set of radar tracks).The DFW data is judged to be adequate in this respect. +GENERAL INSPECTION OF AIRPORT OPERATIONS AT DFWIn this section we will discuss some general attributes of the traffic flow that are relevant to the analysis of interarrival spacing.Figure 1 shows the runway layout at DFW.There are seven runways and thus 14 possible landing directions.When traffic is flowing to the north, the airport is said to be in a "north flow".When traffic is flowing to the south, the airport is said to be in a "south flow".Figure 2 shows a selection of tracks plotted during a period when the airport was in a north flow.It is difficult to determine exactly how efficiently the runways were being used by casual inspection of the actual tracks.However, the plots and analyses that will now be described are designed to provide insight into this question.31L 36L 36R 35L 35C 35R 18R 18L 17R 17C 13L 17L 31R3 +American Institute of Aeronautics and AstronauticsThe four DFW data sets analyzed are listed in Table 1.These data sets contain over 3500 tracks of which about half are arrivals.The weather for all data sets was VMC with reported visibility's exceeding 10 nmi.However for data set DFW.03, a period of IMC weather ended only four hours before the data set began.Figure 3 depicts the time history of operations for one of the DFW data sets.In this figure, the time of each individual arrival and departure is shown in association with the runway of operation.Several features of the traffic flow can be seen.Note that at approximately 10:10 there is a change in runway configuration -from "north flow" to "south flow".The rate of operations varies greatly with time.Periods of intense activity lasting for 45-60 minutes are followed by lulls in which only modest numbers of operations occur.Such irregularities are attributable primarily to airline scheduling, but they can also be produced by the impact of convective weather upon traffic flow and approach routes. +ANALYSIS OF LANDING TIME INTERVALS (LTIS) DEFINITIONS AND BASIC RELATIONSHIPSWhile the plots shown in the previous section provide some insight into how the airport was operating, they do not allow us to assess with any quantitative precision the efficiency of the spacings being achieved for any particular runway.In part, this is because the spacing achievable under radar separation standards varies with aircraft weight class, approach speed, approach geometry, and other factors.Techniques for such analyses will now be described.The key feature of the analysis is a focus upon the landing time intervals (LTIs) achieved.The LTI at the runway is defined as the time separation between two successive landings.It is the difference between the time one aircraft crosses the runway threshold and the time the previous landing aircraft crossed the same threshold.(Note: It is also possible to measure LTIs at the outer marker (OM), but such time intervals will not be employed in this paper.).The throughput of a runway over any arbitrary time period is simply the inverse of the average LTI during that period.For example, if the average LTI is 120 seconds, then the throughput must be 1/120 aircraft/sec or 30 aircraft/hour.We will define the capacity of a runway as the sustained throughput achieved under saturated traffic conditions.Then the mean LTI under saturated conditions is the inverse of the runway capacity.Figure 4 shows the LTIs for three hours of operations at Dallas/Ft.Worth International Airport (data set DFW.04).Six arrival rushes are clearly seen during the 10 hours of data.This plot provides insight into the extent to which the loading upon the arrival runways was balanced.It can be seen that during the 8:00AM push, runway 35C was not as heavily loaded as the other three runways.But during the 9:30AM push, all four runways appear to have been loaded equally.There were brief periods in which landing intervals of 60 seconds or so were achieved for several successive aircraft.While we have chosen to measure LTIs at the runway, interarrival times can also be measured at the outer marker.A comparison of the LTI measured at the outer marker and the runway is provided in Figure 5 (using data from data set DFW.01).It can be seen that there does not appear to be any clear tendency for the LTI's to become either greater or smaller between the outer marker and the runway.This implies that control actions taken within the outer marker are not significantly changing the interarrival times.A histogram of the observed landing time intervals for the four combined data sets is provided in Figure 6.The most common separation was in the 90-100 second range.Figure 7 provides a similar histogram for the minimum in-trail separations observed for 801 arrival pairs in which both aircraft were in the "large" weight class.Separations below 2 nmi appear to be mostly due to the occasional use of visual procedures in which altitude separation was maintained visually.In both figures a line showing a theoretical fit to the histogram is shown (An explanation of the theory follows.) +American Institute of Aeronautics and Astronautics +American Institute of Aeronautics and Astronautics +VANDEVENNE MODEL FOR INTERARRIVAL SEPARATIONSAny analysis of actual runway operations must recognize that at some times the traffic flow will be less than the runway capacity and that gaps will occur between aircraft that are not due to any inefficiency in the spacing process.A statistical model that takes this into account helps avoid confusing these gaps with excess spacing inserted by the final spacing process.Vandevenne 5 developed such a model for the distribution of observed interarrival separations.PARO employs the Vandevenne model to provide a more robust analysis of final spacing performance.This section describes that model.The Vandevenne model is motivated as follows: Let us assume that controllers attempt to achieve an interarrival time separation D that represents the closest comfortable target spacing for specified separation standards and operational conditions.The actual time separation achieved, S, will differ from D for two reasons.First, there is imprecision in spacing.Second, there may be gaps in the arrival stream that are too large to be closed by the level of control available.The Vandevenne model assumes that the errors and gaps are additive so thatS = D + ε + g (1)where ε is the imprecision error and g is the time gap that cannot be closed.The model assumes that ε is normally distributed according to Ν[0, σ 2 ].Vandevenne noted that if the arrival stream is random at a given average arrival rate λ, the time gaps between arrivals prior to application of any control actions will have a Poisson distribution such thatf g (x) = λ exp(-λx) , x ≥ 0 (2)It should be noted that although time separations in a single arrival stream will not be random because of intrail separation standards, the merging of multiple independent streams results in an initial set of interarrival times that is approximately Poisson.Vandevenne assumed that all interarrival spacings will include a time gap component, and that this time gap will have a Poisson distribution.The components of S are summarized in Table 2.Vandevenne showed that the resulting probability density function for S isf S (y) = λexp -λ(s -D - λσ 2 2 )       F SN s -D -λ σ 2 σ       (3)where F SN is the standard normal distribution.In many analyses of actual data, the value of λ changes during the period of observation.This violates the assumptions in the Vandevenne model.For that reason, λ should not be viewed as providing a good indication of the actual arrival flow rate in the data.It is better to view it as merely a parameter of the distribution that is used to correct for the existence of time gaps in the interarrival time observations.f ε (x) = 1 σ 2π exp - x 2 2σ 2      g Time gaps in arrival stream that cannot be closed by control in terminal area.Poissonf g (x) = λexp -λx ( ) +American Institute of Aeronautics and AstronauticsThe targeted time spacing, D, is a key parameter since the inverse of D is the inherent capacity of the runway.Note that when unsaturated flow exists, the mean observed spacing can be significantly greater than D.A positive bias results if the mean spacing is assumed to be equal to the targeted spacing.The Vandevenne model can produce a nearly unbiased estimate of D under such conditions.This results in a more robust analysis that is better suited to automated processing.Experience has shown that the form of the Vandevenne model provides a good fit to actual data.It has the essential characteristic of a major peak reflecting the predominant normally distributed errors and a long tail reflecting gaps arising from other processes. +INTERARRIVAL SEPARATION AND CAPACITYWe will now discuss how the parameters of the Vandevenne distribution relate to runway capacity and to potential FAST benefits.The capacity of a runway (defined as the sustainable throughput when saturated with traffic) is approximately 1/D.FAST capacity benefits are assumed to be derived from reductions in the value of D.At first glance, it appears that the parameter σ has no effect on capacity since the spacing error it produces tends to average to zero.However, it is commonly assumed that in actual operations the value of D is affected by σ because of a need to insert a safety buffer between each pair of aircraft.This buffer guarantees that imprecision will not cause frequent violations of separation standards.The size of the buffer is selected to keep the rate of separation violations below some level, α.If σ is decreased, the safety buffer can be decreased.For the Vandevenne model, we can model the target separation asD = D 0 + σ F SN -1 (1-α) (4)where D 0 is the required minimum time separation, F SN -1 is the inverse of the standard normal distribution (with zero mean), and α is the allowed rate of violating this separation.Figure 8 shows how the capacity of a runway is affected by the value of σ when the uncertainty buffer corresponds to either 2σ or 3σ. +American Institute of Aeronautics and AstronauticsIn many cases, the major determinant of D 0 is the intrail wake vortex separation standard.This standard depends upon the aircraft weight class combination for a pair of successive arrivals.In translating the distance standard to an equivalent time standard, we must also consider the speed profiles of the aircraft on final approach.To provide a more relevant comparison of aircraft with different weight classes and speeds, we will usually subtract the computed separation standard from the observed separation to yield the excess separation S* defined as S* = S -D 0 .When the value of S* is negative it means that the actual separation achieved was less than that indicated by the applicable radar separation standard.This does not necessarily mean that any standards were violated since under visual meteorological conditions the radar separation standards do not have to be applied to aircraft that have their traffic in sight.The advantage of using S* is that it allows combining pair separations values for all aircraft types under the assumption that the applicable values of σ and λ are independent of an aircraft's weight class and final approach speed.This assumption appears justified by data analysis completed to date for Dallas/Ft.Worth, but should be verified again when different airports are analyzed. +MAXIMUM LIKELIHOOD ESTIMATION OF MODEL PARAMETERSGiven a set of interarrival time separations, how do we go about finding the model parameters for fitting the Vandevenne model to the data?Vandevenne suggested using a maximum likelihood estimation technique.Suppose that we observe N arrival pairs.Let the i th pair have separation y i .Then the log likelihood function isL = ln[f y (y i i =1 N ∑ )](5)where f y is the probability density function for the separation.The maximum likelihood set of parameters is the set that maximizes this function.Vandevenne found the maximum likelihood values by generating contours of likelihood and using search techniques on these contours.While this method is theoretically sound, the estimation of the likelihood function for each point on a contour involves N separate evaluations of the density function f y .If large databases containing tens of thousands of arrivals are to be analyzed, the computational load could be a hindrance to the analysis.For this reason, an alternative technique was developed that computes an approximate likelihood value directly from the histogram.For this technique, the N data points are compiled into a histogram with H bins.The likelihood factor is calculated for a separation at the midpoint of the histogram bin.The same factor is then assumed to apply to each point in the bin.For example, let the count of separations falling into bin i be n j .Let the mid-point of the separation interval for bin i be y i .Then the approximate likelihood function can be writtenL = n i ln[f y ( y i i =1 H ∑ )](6)With this approach, the number of times the f y function must be computed is equal to the number of histogram bins instead of the number of points within those bins.This greatly expedites the search for the maximum likelihood values.Inspection of several cases indicates that as long as the histogram bin width is less than approximately one-half σ, the maximum likelihood parameters derived in this way are almost indistinguishable from those derived by using all N original data points.What is the accuracy with which PARO is able to estimate the three parameters of the Vandevenne distribution?Clearly, the accuracy will depend on the number of points that are available for forming the estimate.It will also depend upon the bin size used in the histogram.Table 3 presents simulation results for a case in which the true parameter values are D = 72.0seconds, σ = 18.0 seconds, and λ = 90/HR.For each entry, Monte Carlo simulation was used to generate 100 histograms, each with a bin size of 10 seconds.The standard deviations of the parameter estimation error decreases roughly as the inverse of the square root of the number of points used to construct the histograman expected result.For σ, a standard deviation of error that is less than 10 percent of the true value is achieved with 400 data points. +Analysis of LTIs from DFWThe LTI distributions that exist in the four DFW data sets were analyzed by first generating LTI histograms for each set separately and then for the combined data.The maximum likelihood fit of the Vandevenne distribution to each histogram was computed.Figure 9 depicts the histogram of excess LTIs for all four data sets combined.The results of the analysis are summarized in Table 4. +American Institute of Aeronautics and AstronauticsIt can be seen from the histograms that the shape of the distribution closely resembles the Vandevenne distribution.The maximum likelihood parameters are similar for all sets.The combined value of ρ was 17.7 seconds.This is only slightly less than the 19-20 second values reported by Ballin and Erzberger. 2e fact that D tends to be slightly less than zero means that the aircraft were often achieving intervals smaller than would be possible under radar separation standards.This implies for this predominantly VMC data, it is unlikely that an AFAST calibrated to preserve radar separation standards could have increased throughput by reducing the "imprecision buffers" incorporated in D. Nevertheless, AFAST might have been able to provide benefits by anticipating and removing the larger interarrival gaps that are related to the overall flow to the runways.Additional data analysis is being pursued to confirm this and to address the same question for IMC conditions. +ANALYSIS OF FACTORS AFFECTING SEPARATIONSThis section presents the results of several types of analysis conducted to investigate the reasons for the differences between the LTIs obtained under different conditions. +American Institute of Aeronautics and Astronautics +Correlation Analysis for LTIsOne way of searching for factors that affect final spacing efficiency is to examine the linear correlation coefficient between various variables and the excess separation.This type of analysis can fail to detect certain types of nonlinear dependencies, but will nevertheless identify a number of relationships that deserve further scrutiny.Figure 10 shows the parameters used to characterize the final approach geometry.The parameters are defined in the runway coordinate system for which the origin is the runway threshold.The parameter YBASE is the y value at which the base leg was established.The parameter yCL2 is the y value at which the aircraft achieved flight along the centerline of the runway.The criteria for "centerline" status is that the aircraft has to be within 2 nmi of the centerline and have a heading within 15 degrees of the runway heading.Table 5 defines the "W variables" that were used to describe the pair of aircraft generating a single LTI value.Table 6 provides the serial correlation coefficient ρ, for these variables when correlated against the full value of the LTI.Only large/large weight class pairs were used to avoid variations due to differing wake vortex separation requirements.If the initial gap between arrivals was too large to be closed by typical control actions, the pair was excluded from this analysis.In this table, p is the probability that the observed correlation coefficient would be as far from zero as observed if the actual value were in fact zero.The following observations apply to Table 6:• Separation increases when the total path length flown increases (ρ = 0.147).This may reflect greater constraints and traffic interactions encountered by aircraft that fly longer paths.It could also reflect the fact that having to maneuver within terminal airspace introduces imprecision into the spacing.• There is high correlation (ρ = 0.880) between LTI measured at the runway and LTI measured at outer marker.This suggests that if efficient spacing isn't achieved at the outer marker, then it is unlikely to improve much due to actions taken within the marker.Table 7 provides a correlation analysis for W variables correlated against the excess landing time interval, S*.Here all weight classes can be combined.Note that• Excess separation is negatively correlated (ρ = -0.203)with weight class difference (lead minus follower).This indicates that when the lead aircraft is heavier, the separation relative to wake separation standards is less.• Excess separation increases when centerline intercept is closer to the runway (ρ=0.135 for yCL2).Excess separation decreases with more time spent on the centerline (ρ=-0.122for t_on_CL).The reason for this is not clear, but may have something to do with the ability to tighten separation through speed control as compared to trying to achieve tight separation by a precise turn from a short base leg.• Excess separation increases when pathlength increases.(See earlier comment for Table 6).• There is negative correlation (ρ = -0.254)with absolute LTI allowed by separation standards.This suggests that there is a tendency to space closer than the standard for the larger standards, perhaps through use of VMC procedures.There is high correlation (ρ = 0.905) with excess LTI measured at outer marker.Again, this indicates that actions taken after the outer marker have little impact on the final time separations. +ADDITIONAL ANALYSIS OF FACTORS AFFECTING LTIS +Differences Between RunwaysAn obvious question to ask is whether LTI distributions are the same for all runways.Figure 11 provides histograms of LTIs for each runway at DFW using the combined data sets.The LTI distribution for each runway appears to be similar except possibly for runway 18L for which very few arrivals were observed.If it is more difficult for controllers to anticipate the effect of such speed differences, then differences might be expected to appear in the histograms.For the data in Figure 12, the values of the targeted time separation, D, are quite similar except possibly for lower D value for the case of a following aircraft more than 20% faster than the lead aircraft.In general, it appears that controllers at DFW are quite skilled at taking the differing landings speeds of aircraft into account when spacing them. +Approach PatternsIt seems possible that the precision of interarrival spacing can be affected by the geometry available for making the final spacing adjustment.For aircraft that fly a downwind segment, the controller is able to choose the location of the base segment to achieve proper spacing.But the turn required doing so can be a source of imprecision.Is there a difference in spacing performance that should be addressed by tools such as FAST?An analysis of approach patterns was conducted by dividing all pairs of successive aircraft into four groups depending upon whether or not the approach involved a downwind-base trajectory.Approaches without a downwind phase were called "straight-in", although it should be noted that some of these aircraft approached the runway centerline at a fairly large angle.The set labeled "downwind->straight" includes all pairs for which an aircraft on straight-in approach was followed by an aircraft flying a downwind segment.The four histograms that result are shown in Figure 13.At DFW, aircraft are generally first directed to the cornerpost fixes that are most consistent with the north/south direction of flow, and hence straight-in patterns predominate for final approach.While there are no statistically significant differences in the S* distribution for this data, it does appear that aircraft following a downwind leader tend to have higher median values.One additional question concerns the relationship between time intervals and actual in-trail spacing.In this report, we have expressed separations in terms of time, but actual separation standards are expressed in distance.Figure 14 shows the relationship between excess landing time intervals and excess in-trail spacings for arrival pairs in data set DFW.01.It can be seen that there is high correlation (ρ = 0.843) between the two measures.The dotted line shows the space-totime conversion that would if the excess spacing is traversed by the trailing aircraft at a speed of 150 knots.This analysis suggests that conclusions about performance deduced from inspection of time separations are likely to be the same as those of an analysis that used only spatial separations. +American Institute of Aeronautics and Astronautics +CONCLUSIONS AND RECOMMENDATIONSThis work has developed robust, computationally practical data analysis routines that can be used to provide insights into runway operations through analysis of radar data.The questions that can be addressed are relevant to the benefits mechanisms of AFAST and to determining the total benefits that can be achieved with AFAST implementation.The following observations apply to the four data sets analyzed for DFW.It should be noted that all these data sets involved VMC weather and the generality of the conclusions drawn from this limited data set has not been proven.• In short rush periods (of 10 minutes or so), very high landing rates of 60 AC/HR or more are obtained on single runways.It is not clear that the high peak rates observed can be obtained in IMC.Nor is it clear that they can be sustained in periods of prolonged saturation.• In general, the LTI achieved at the outer marker is preserved at the runway.Variations appear to be random with no discernible tendency to change in a given manner.Thus, there is little evidence that visual separation practices applied within the outer marker are having a significant impact upon spacings.• The targeted separation, D, appears to be about 72 seconds, which corresponds to a throughput of about 50 aircraft/hour.The fact that this rate is almost never sustained in practice suggests that there is an opportunity to increase throughput if the consistency of flow to the final vector position is improved. +American Institute of Aeronautics and Astronautics• Targeted separation (D) tends to be slightly less (by about 7 seconds or 0.3 nmi) than the value that would be expected if radar separation standards were the sole determinant of target separation.This suggests that visual procedures near the runway may have allowed separations to be tightened enough to overcome the effects of any "safety buffers" that were applied during the earlier radar separation process.• The occasional presence of larger LTIs during periods of saturated flow is a further indication that irregularities in the flow may be contributing to a loss of throughput in VMC.This phenomenon deserves further study since the ability of AFAST to reduce such irregularities provide capacity benefits under VMC conditions.Figure 1 .1Figure 1.Runway layout at Dallas/Ft.Worth International Airport. +Data set: DFW1, 8:00-9:00 +Figure 2 .2Figure 2. Traffic flow sample at DFW for data set DFW.01. +Figure 3 .3Figure 3. Runway utilization showing a change in runway configuration for data set DFW.01 at 10:30 local time. +Figure 4 .Figure 5 .45Figure 4. Landing time intervals for data set DFW.04 +Figure 6 .Figure 7 .67Figure 6.Landing time interval histogram for combined DFW arrivals (1758 pairs). +Figure 8 .8Figure 8.Effect of imprecision ( upon runway capacity. +ExcessFigure 9 .9Figure 9. Distribution of excess interarrival spacings (S*) for 1758 DFW arrivals. +Figure 10 .10Figure 10.Definition of approach geometry using YBASE and yCL2. +Figure 11 .11Figure 11.LTI differences between runwaysFigure12examines the effect on the LTI distribution of the speed ratio between the lead and following aircraft.If it is more difficult for controllers to anticipate the effect of such speed differences, then differences might be expected to appear in the histograms.For the data in Figure12, the values of the targeted time separation, D, are quite similar except possibly for lower D value for the case of a following aircraft more than 20% faster than the lead aircraft.In general, it appears that controllers at DFW are quite skilled at taking the differing landings speeds of aircraft into account when spacing them. +Figure 12 .Figure 13 .1213Figure 12.Effect of follower/leader speed ratio on landing time intervals +Figure 14 .14Figure 14.Comparison of excess spacing expressed in time and distance +Table 1 . DFW data sets analyzed.1Data SetDateNo.Start TimeNameTracks(local 24 hr)DFW.016 DEC 998267:00DFW.0210 JAN 0081012:35DFW.037 FEB 0050617:59DFW.0410 FEB 0014318:19 +American Institute of Aeronautics and Astronautics Runway 14 13 12 11 10 9 8 7 13L 13R 17L 17C 17R 18L 18R 31L 31R 35L 35R 35C 36L 36RData Set: DFW.01SOUTH FLOWNORTH FLOWArrivals Departures +Table 2 The Vandevenne Model2VariableDefinitionDistributionDTime separation thatFixed for a given aircraft paircontroller attempts toεachieve. Error in achieving targetednormal, zero meantime separation +Table 3 . Landing Time Interval (LTI) Analysis3No. Points inMean Error in DStd. Dev.of DMean Error in σStd. Dev. of σΜean Error in λStd dev of λHistogram200-0.2543.249-0.4512.7301.95410.2394000.3072.4830.1071.7812.5627.150800-0.2561.774-0.2161.3000.4714.7231600-0.0751.009-0.0080.8100.5363.303 +Table 4 . Errors in Estimation of Vandevenne Parameters4Data SetNo. of AircraftDσλPairs(sec)(sec)(AC/hr)DFW.01418-7.914.362.8DFW.02391-0.317.269.0DFW.03242-0.719.968.1DFW.04707-12.317.664.4All sets1758-7.117.765.4combinedyRUNWAYxOUTER MARKERyCL2YBASE +Table 5 . Descriptive "W Variables" for an Arrival Pair5Variable, WDefinitionabsolute LTI Absolute value of landing time interval (sec).AP_PATTERN Approach pattern type of follower (1000=straight-in, 2000=downwind/base)AZ_FIRST Azimuth at which aircraft first appeared (degrees).LTI_OM Landing time interval at outer marker (sec)LTImin Minimum LTI permitted by separation standards (sec).pathlength Total path flown in terminal airspace by follower (meters)S* Excess landing time interval at runway (sec)t_on_CL Time spent "on centerline" (CL) state before landing (sec)V2/V1 Speed ratio (final) of follower to lead aircraft.vel_op2 Final speed (at landing) of follower (KT).vOM2 Velocity of follower at outer marker (KT).WEIGHT_CL_DIF Weight class code of lead minus that of follower,wtclass_lead Weight class code of lead (1=light, 2=large, 3=B757, 4=heavy)YBASE y coordinate of base segment for follower (meters).yCL2 Centerline intercept coordinate of follower (meters)yCL2-yCL1 Difference in centerline intercept coordinate of follower and lead (meters). +Table 6 . Linear Correlation Analysis of Full LTI: S vs. Variable W, Large/Large Weight Class Pairs6Variable, Wnmean SStd.mean WStd.ρpsigni-DeviationDeviationficanceof Sof WV2/V11211 105.8642.8901.0140.148-0.1490.00000 •••vOM21211 105.8642.89086.314.2-0.1200.00003 •••vel_op21211 105.8642.89067.2216.287-0.0820.00459 •••AZ_FIRST21211 105.8642.890299.6455.8-0.0450.11802yCL2-yCL11211 105.8642.890245.89371.40.0110.70289AP_PATTERN1211 105.8642.8901067.5263.30.0200.49171WEIGHT_CL_DIF1211 105.8642.8900.0150.7680.0500.08173 •wtclass11211 105.8642.8902.0450.6090.0920.00143 •••LTImin1211 105.8642.8900.418.00.1310.00001 •••yCL21211 105.8642.890-17056.99231.80.1360.00000 •••pathlength21211 105.8642.890 150929.823251.40.1490.00000 •••YBASE21043 107.4244.733-17325.35428.30.1570.00000 •••S*1211 105.8642.89029.043.60.9050.00000 •••LTI_OM1211 105.8642.890105.945.00.9110.00000 •••Significance code: • = significant at 0.10 level, •• = at 0.05 level, ••• at 0.01 levelAmerican Institute of Aeronautics and Astronautics +Table 7 . Linear Correlation Analysis of Excess LTI: S* vs. Variable W, all aircraft pairs7Variable, Wn mean SStd.mean WStd.ρpsigni-Deviation ofDeviationficanceSof WV2/V1787101.3636.3491.0170.154 -0.138 0.00011 •••vel_op2787101.3636.34967.6726.245 -0.123 0.00056 •••vOM2787101.3636.34986.713.6 -0.035 0.32483AZ_FIRST2787101.3636.349284.5448.8 -0.025 0.48288yCL2-yCL1787101.3636.349167.19450.60.035 0.32807AP_PATTERN787101.3636.3491079.9281.50.048 0.17857LTImin787101.3636.349-0.217.60.116 0.00121 •••yCL2787101.3636.349-16967.19141.70.144 0.00006 •••pathlength2787101.3636.349152726.222616.60.147 0.00004 •••YBASE2688102.9837.131-17549.55447.40.179 0.00000 •••LTI_OM787101.3636.349101.737.90.880 0.00000 •••S*787101.3635.34931.835.20.955 0.00000 ••• + + + + +*This work was performed for the National Aeronautics and Space Administration under Air Force Contract No. F19628-00-C-0002. + + + + + + + + + Evaluation of the Capacity and Delay Benefits of Terminal Air Traffic Control Automation + + StevenBBoswell + + No. DOT/FAA/RD-92/28 + + 14 April 1993 + + + MIT Lincoln Laboratory + + + Report + Boswell, Steven B., "Evaluation of the Capacity and Delay Benefits of Terminal Air Traffic Control Automation," Project Report ATC-192, Report No. DOT/FAA/RD-92/28, MIT Lincoln Laboratory, 14 April 1993. + + + + + Benefits Analysis of Terminal-Area Air Traffic Automation at the Dallas/Fort Worth International Airport + + MarkGBallin + + + HeinzErzberger + + + + AIAA 96-3723, AIAA Guidance, Navigation and Control conference + San Diego + + July 29-31, 1996 + + + Ballin, Mark G., and Erzberger, Heinz, "Benefits Analysis of Terminal-Area Air Traffic Automation at the Dallas/Fort Worth International Airport," AIAA 96-3723, AIAA Guidance, Navigation and Control conference, San Diego, July 29-31, 1996. + + + + + CTAS Benefits Extrapolation Preliminary Analysis + + TWeidner + + + GJCouluris + + + CGHunter + + + February 1998 + Seagull Technology Inc + + + Weidner, T., Couluris, G.J., and Hunter, C.G., "CTAS Benefits Extrapolation Preliminary Analysis," 98156-01, Seagull Technology Inc., February 1998. + + + + + Terminal Area separation Standards: Historical Development, Current Standards, and Processes for Change + + SDThompson + + ATC-258 + + 16 January 1997 + + + MIT Lincoln Laboratory + + + Project Report + Thompson, S.D., "Terminal Area separation Standards: Historical Development, Current Standards, and Processes for Change," Project Report ATC-258, MIT Lincoln Laboratory, 16 January 1997. + + + + + Using Maximum Likelihood Estimation to Determine Statistical Model Parameters for Landing Time Separations + + HFVandevenne + + + MALippert + + PM-AATT-0006 + + 17 March 2000. 27 April 1992 + 92 + + + Vandevenne, H.F., and Lippert, M.A., "Using Maximum Likelihood Estimation to Determine Statistical Model Parameters for Landing Time Separations", 92PM-AATT-0006, 17 March 2000 [Originally published as internal memorandum 27 April 1992]. + + + + + + diff --git a/file13.txt b/file13.txt new file mode 100644 index 0000000000000000000000000000000000000000..ce8b70c97f3a518252a8029617504e6e80b925f2 --- /dev/null +++ b/file13.txt @@ -0,0 +1,836 @@ + + + + +Introduction:The 1960s and 70s witnessed an expansion in the capabilities and mission requirements of rotorcraft and the introduction of jet-powered vertical and short take-off and landing (VSTOL) aircraft.The ability to hover and maneuver precisely at low-speed and then transition to high-speed flight made them useful for a variety of missions that could not be fulfilled by conventional fixed-wing aircraft.These aircraft, however, presented an unique set of pilot-vehicle interface challenges not seen in conventional fixed-wing aircraft: 1) unfavorable stability and control characteristics at low-speed and hover; and 2) static and dynamic behavior that changed significantly in a relatively small speed range during transitions to high speed flight.Advancing the capabilities of VSTOL aircraft demanded that these challenges be properly understood and overcome, and this required a simulation environment that could recreate the pilot feedback cueing available in an actual aircraft.The Vertical Motion Simulator (VMS) at NASA Ames Research Center was engineered to provide the realistic pilot cueing environment necessary to be a viable surrogate to flight test research for evaluating VSTOL aircraft and other future aircraft concepts.The VMS became operational in 1979 and incorporated the largest and most realistic motion system in the world, a distinction it holds to this day.The high-fidelity independent rotational and translational motion displacements are combined with an interchangeable cab system housing realistic computergenerated visual displays of the outside world and adaptable cockpit interfaces with accurate control-feel systems, flight instruments and displays.These features make the VMS an extremely adaptable and efficient platform for studying all types of aircraft with widely varying and challenging flight dynamics and performance issues, as well as pilot-vehicle interface concepts.Its large motion capability made it ideally suited to investigating pilot-vehicle interactions and interfaces, because the high-fidelity motion cueing increased the likelihood that the research findings would translate to a flight environment.Over the past three decades, the VMS has made significant contributions to aeronautical research in the broad area of pilot-vehicle interaction for a range of existing and conceptual aircraft, both fixed and rotarywing.With its unparalleled level of fidelity, the VMS was also ideally suited to fundamental research on pilot cueing and simulation fidelity.Concurrent and interacting studies, therefore, used the VMS's unmatched six degree-of-freedom motion capability to examine human pilot cueing and the level of simulation fidelity required to accurately recreate the pilot-vehicle interaction observed in flight.Other studies on the VMS over the years included handling qualities evaluations of aircraft concepts with unconventional and challenging flight characteristics (e.g., Space Shuttle Orbiter, high-speed civil transports, large fixed-wing transports, tilt-wing, tilt-rotor, etc), flight control design and handling qualities assessment of production aircraft (e.g., C-17A, CH-47F, and UH-60), pilot guidance and cueing display design and development, and accident investigations (e.g., USAir 427, and AA 587).Recent studies on the VMS are evaluating the handling qualities and flight control system requirements for the next generation of US spacecraft -the Orion Crew Exploration Vehicle and the Altair Lunar Lander vehicle.This paper describes the VMS and its motion characteristics and summarizes its contributions and impact on aeronautics research.A detailed facility description emphasizing the motion system is provided followed by a description of the prior simulation experience that influenced the design of the VMS motion system.This is followed by a summary of the research conducted on the VMS through its history and its impact on aircraft programs and projects.As complete coverage of all vehicles simulated in the history of the VMS is prohibitive in scope, this paper focuses on simulation fidelity and cueing, VSTOL research, and Space Shuttle development and training that comprise the majority of the simulations conducted on the VMS. +VMS Facility DescriptionThe VMS combines a high-fidelity simulation capability with an adaptable simulation environment that enables customization for numerous human-in-the-loop research applications.The distinctive feature of the VMS is its unparalleled large amplitude, high-fidelity motion capability.An overall high level of simulation fidelity is achieved by combining this motion fidelity with realistic visual and cockpit interfaces.Interchangeable cabs allow different crew vehicle interfaces and vehicle types to be evaluated with rapid turnaround times between simulation projects.The Interchangeable Cab (ICAB) capability allows for tailoring the cockpit to the research application.The VMS has five portable ICABs with different out-the-window visual fields-of-view.For each simulation, an ICAB is selected and equipped to meet the study's requirements and then tested with the complete simulation environment without motion.Configuring the cab includes installation of flight controls, flight instruments and displays, and seats (Fig. 1).Following cab configuration and checkout, the ICAB is transported and installed on the motion system.The ICAB capability, simulation architecture and resources enable the VMS facility to conduct fixed-base and moving-base simulation studies simultaneously. +Fixed-baseCab Interior Moving-base A custom digital-control interface allows for comprehensive adjustment of the controller's static and dynamic characteristics.Force-loader characteristics may be varied during simulated flight as necessary for studying pilot cueing concepts using inceptors.A variety of aircraft manipulators, ranging from the regular column-and-wheel type to conventional rotorcraft controls and side sticks, are available and may be combined with the force-loader systems.A Rockwell-Collins EPX-5000 image generation system creates the out-the-window visual scene and provides a high-resolution and complex visual environment at update rates exceeding 60Hz.Fifteen channels are available, allowing the simultaneous conduct of up to three different simulation experiments in the facility.In-house graphics expertise customizes the visual databases to meet widely varying simulation requirements.Separate graphics processors generate the content for the primary flight displays, head-up displays, sensor imagery, etc, which can be fully customized.All the essential elements of the simulation are linked with the host environment through a dedicated network, and the simulation is managed from a fully equipped control room.The flexible simulation architecture makes it convenient to interface and evaluate custom software and hardware modules.This capability may be used to evaluate sensors, vehicle dynamic models, flight control systems, etc. +Motion System DescriptionThe VMS motion system, shown in Fig. 2, is an uncoupled, six-degree-of-freedom, combined electromechanical/electro-hydraulic servo system (Ref.2).It is located in, and partially supported by, a specially constructed 120-ft tower.The motion system includes a beam structure, called the vertical platform, which spans the width of the tower.The vertical platform is mounted on two columns, called equilibrators which extend down into 75-ft deep shafts under the tower floor.Wheel assemblies, which ride along vertical guide rails attached to the tower walls, restrain the vertical platform at both ends and the center of its span. +Figure 2. VMS Motion SystemThe two equilibrators act as pneumatic counterweights.The hollow equilibrator columns slide over inner columns so that the two, along with a gas-tight seal between them, form a cylinder/piston arrangement.Nitrogen gas, supplied by a special storage system, pressurizes the equilibrators such that the pressure forces balance the weight of the 140,000-lb cab and platform structure.This counterbalancing force reduces the power requirement of the vertical drive motors and results in a linear motion response in both directions of vertical travel.An added benefit is that if vertical drive power is lost, the motion system will float to an equilibrium position towards the center of the tower.Eight mechanically-coupled, 150-hp direct-current servomotors power the vertical motion through reduction gearboxes and a rack-and-pinion drive system with the racks mounted on the equilibrator columns (Fig. 2).Four 40-hp direct-current servomotors power the lateral carriage along the vertical platform using reduction gearboxes and a rack-and-pinion drive system with the rack mounted on the top of the vertical platform.A linear hydraulic actuator powers the longitudinal carriage, located atop the lateral carriage.A 48-inch diameter roller bearing provides the yaw motion, which is mounted on top of longitudinal carriage that is driven by another linear hydraulic actuator.The yaw bearing supports a conical center structure, which has a two-axis gimbal on top that provides roll and pitch motion using two linear hydraulic actuators.A unique feature of the VMS is that the yaw actuator may be attached at two different locations, 90° apart, allowing for the large ±20 ft. of translational displacement in either the aircraft's longitudinal or lateral axis, as desired by the particular simulation. +Motion System PerformanceTable 1 summarizes the VMS motion capability.Included are two sets of limits: system limits, that represent the absolute maximum levels attainable under controlled conditions; and operational limits, that represent attainable levels for normal piloted operations.The operational limits listed in Table 1 include the effects of all the system limiters, both hardware and software.The motion system also incorporates a parabolic limiter, which is not reflected explicitly in Table 1.When triggered, the parabolic limiter commands a maximum acceleration opposite to the direction of travel so that the simulator stops just short of a displacement limit.The motion drive dynamics may be modeled as equivalent time delays ranging from 90 msec in pitch and roll to approximately 130 msec in the yaw and translational axes (Ref.3).The VMS motion system includes digital feed-forward compensators (motion lead compensators) in each degree of freedom that may be used to alter and improve the overall motion system dynamics within limits.The motion lags are typically larger in the translational axes than documented (when they are documented) for small hexapod simulators, which one would expect when considering the relative scale difference between a small hexapod and the VMS.These inherent lags can be effectively eliminated, if a particular task deems it necessary, by modifying the math model so that lags due to actuators and digital effects are removed in exchange for the motion system lag so that the overall equivalent delay in the simulated vehicle is maintained for the evaluation (Ref.4). +Motion Washout FiltersThe cockpit motion cueing algorithm uses a high-pass (washout) filter and a rotational/translational crossfeed arrangement shown schematically in Fig. 3.The computed pilot station accelerations of the modeled aircraft are high-pass filtered, and attenuated, before commanding the motion drive system.Turn coordination and induced acceleration compensation account for the cross-coupled motion commands and provide the correct cues at the pilot's station.A low-pass filter tilts the simulator to provide steady-state longitudinal and lateral acceleration cueing at low frequency (Fig. 3). +Residual tilt for low frequency acceleration +VMS Motion System DesignThe design of the VMS is based on the experience gained from a series of ground-based flight simulators at NASA Ames beginning in the late 1940's (Ref.5).Some key aspects of the VMS motion system design directly trace to the research experience from three past flight simulators at NASA Ames -the sizing of the translational envelope was based on insights gained from the Height Control Test Apparatus (HCTA) and the Flight Simulator for Advanced Aircraft (FSAA), while the equilibrator design was based on experience with the Six-Degree-of-Freedom Motion Simulator (Fig. 4).The HCTA was a single degree-of-freedom flight simulator with 80 ft. of vertical travel that became operational in 1961.In the early 1970s a study on the HCTA determined the significance of vertical acceleration cues when simulating the visual approach and landing maneuver (Ref.6).The results indicated that vertical motion cues are important for the landing task, particularly for aircraft with marginal longitudinal handling qualities.The study concluded that a simulator must have a vertical excursion capability of at least ±20 ft. to provide realistic pilot cueing for the approach and landing task.The sizing of the VMS vertical travel envelope was based on these findings.The FSAA was a six degree-of-freedom flight simulator with ±40 ft. of lateral travel but limited vertical (±4 ft.) and longitudinal (±3.5 ft.) travel.Since its inception in 1969, the FSAA contributed to several important fixed-wing aircraft research programs, but its small vertical travel limited the ability to simulate VTOL aircraft accurately.Similarly, the longitudinal travel was adequate for conventional aircraft, but more travel was needed for simulating the low-speed maneuvers of VTOL aircraft.A study on the FSAA to determine motion simulation requirements for helicopter flight research determined that at least ±16 ft. of lateral travel is required (Ref.7).The sizing of the VMS lateral-travel envelope and the ability to orient the cockpit with either the lateral or longitudinal axis were based on these findings.The Six DOF Motion Simulator was the first flight simulator to use equilibrators, instead of counterweights to help offset gravitational effects and thereby improve vertical dynamic performance (Ref.5).It became operational in 1964 and had ±9 ft. of travel in all translational axes and ±45 degrees in all rotational axes.Experience with the equilibrators led to the improved design used in the VMS. +Figure 4. Primary Influences on the VMS Motion System Design Simulation Fidelity Studies in the VMSA host of experiments aimed at understanding simulation fidelity requirements for aircraft have been performed on the VMS since its inception.Given its one-of-a-kind capabilities, the first natural question to ask was "how good is it compared to flight?"Initial answers to that question used a model of a UH-60 and identified deficiencies in all of the components that supply the cues: the math model, the inceptors, the visuals, and the motion.While the gap in these components continued to taper relative to flight, more general studies investigated cuing requirements.Given the motion envelope of the VMS, predominant emphasis was placed on motion cues.These studies led to the validation and use of motion fidelity criteria.As technology improved, the cueing investigations became more detailed, especially as the interactions between the motion fidelity and visual requirements were considered.For instance, rudimentary orthogonal grids composed of lines were previously used to artificially convey visual feedback of positions, orientation, and rates.Even as texture and increasing levels of detail were introduced, the visual cueing was still deemed inadequate, so questions were raised on visual requirements.Another byproduct of the technology improvements was the hopefulness that simulators could be used to effectively simulate difficult tasks such as autorotation and shipboard landing.Since these tasks place stringent cueing demands on simulation, the cueing studies that had been performed to date were valuable.The VMS studies described next encompass various questions that have been posed on motion fidelity, visual cueing, and the ever-hopeful attempts of using simulation for the most-challenging in-flight tasks.These studies provided valuable data for improving the validity of other studies conducted on the VMS and elsewhere. +UH-60A Simulation Validation (1984-1993)Several organizations cooperatively conducted the first attempt to compare how well the VMS could simulate helicopter flight (Ref.8).Pilot handling qualities ratings (HQRs) were Level 1 (satisfactory without improvement) in flight but Level 2 (not satisfactory without improvement) in simulation.These differences led to investigations to improve the rotor model, servo dynamics, and engine.Improvements to the math model, visual systems, motion configurations, and experimental tasks and protocols led to experiments documented in Ref. 9.This experiment compared performance and pilot opinion using a UH-60 flown at NASA Ames.Extensive frequency-domain identification compared the simulation performance with flight, with the conclusion that the model was a reasonable representation of the flight vehicle.Improvements were self-evident since the 1984 simulation, as the HQRs overlapped between simulation and flight, but the pilots noted deficiencies in the visual and heave motion cueing.Field-ofview in the simulator was inadequate for some tasks, and the lack of texture and detail in the simulated visual scene made the simulated tasks more difficult than the in-flight tasks.The heave motion cue in simulation was noted to be marginal, and it made the anticipation of finding the proper stopping point during the bob-up (an altitude repositioning task) difficult. +Initial Cueing Fidelity Studies (1985-1999)Early research by Bray (Ref.10) on the VMS showed how far removed the simulated cues in a typical helicopter are from the real world and exposed the need for systematic investigations with objective measures to determine fidelity requirements.Bray observed that while pilot-vehicle performance and opinion is sensitive to degradations in motion and visual cues, pilot opinion has not been particularly helpful in identifying the source of these deficiencies.Thus, objective measures of fidelity were required.Early studies emphasized the relative timing between the visual and motion cues that was a suspected cause of simulator sickness.This was a concern with all training simulators and a particular concern in the VMS.Given the salient nature of the increased motion cues in the VMS, it was a natural facility to investigate simulator sickness.Using four different motion conditions, McCauley (Ref.11) found that simulator sickness increased with both time and the level of maneuvering.Using a UH-60 model, Sharkey (Ref.12) subsequently found that false motion cues had an adverse effect similar to having no motion cues.The effects of asynchrony between the motion and visual cues were also investigated for handling qualities effects.Mitchell and Hart (Ref.3) examined variations in visual time delays and motion washout filters.They suggested tailoring the motion system to the task and minimizing the mismatch between motion and visual delays.Chung and Schroeder (Ref.13) studied the motion and visual synchrony among the roll and lateral axes using a predominantly lateral axis task.They recommended that the equivalent time delay mismatch between the roll and lateral motion cues not exceed 40 msec.Their work also suggested that the equivalent delays in the motion cuing could exceed the equivalent visual delay without a resulting degradation in handling qualities ratings.This went against the prevailing conventional wisdom that believed the motion cue should not lag the visual cue.In a ground-based simulator where motion displacement is limited, there is a necessary trade-off between the desired initial, or short-term acceleration and the desired sustaining, or long-term, acceleration.The only instance when this trade-off maybe avoided is when the task requires displacements that are within the physical envelope of the simulator used.The Ref. 13 study considered such an instance for a sidestep task in the VMS for which 1:1 motion was possible.It is reasonable to conclude that such studies add to the validity of motion-and-visual fidelity investigations, as the motion system is providing the full physical motion as calculated by the mathematical model.Considering the effects of motion cues on handling qualities, Mitchell (Ref.14) showed that the addition of motion improved pilot opinion ranging from ½ to 2 HQR points.For precision tasks, sustained acceleration cues were preferred (reduced washout natural frequencies of the motion filter), while, for aggressive tasks, short-term acceleration cues were preferred.A reason for this preference is that, if the immediate acceleration feedback cue provided by the motion gain is more representative of the acceleration the model is actually producing, it allows the pilot to adjust his input to achieve the desired level of aggressiveness.Schroeder (Ref.15) used pilot describing-function measurements to examine a variety of motion gains and motion washout filter variations on a classical single-axis compensatory tracking task.The results showed that motion cues allowed the pilot to generate lead compensation and improve target tracking phase margins with increasing filter gain or decreased natural frequency.Tracking errors increased significantly when all motion was removed.The study also showed no effects for any of the pure yaw motion configurations, which led to subsequent investigations.To help answer the question on what characteristics a motion filter should have so that simulation is a reasonable representation of flight, Schroeder (Ref.16) evaluated the proposed Sinacori motion fidelity criteria.Objective and subjective results showed that the original criteria could be relaxed, and these criteria (shown in Fig. 5) are still used today as a guide when configuring the VMS motion system as well as other simulators.When compared against these criteria, the gain and phase mismatch for the VMS motion system spans the "like flight" and "different from flight" regions, depending on how the motion filter settings are optimized (Fig. 5). +Figure 5. Motion Fidelity CriteriaA question that often arises when configuring the motion system is the level of fidelity required in each axis.That is, should the reduction in cue be equivalent in all axes or should it be weighted in favor of one axis over the other?It appears that all of the motion cues are not of equal importance when combined with other motion cues, visual cues, or both.The earlier study that showed little effect of yaw cueing led to a more detailed study (Ref.17), which evaluated a helicopter in a single degree-of-freedom hovering yaw task.Four variations in the motion cueing were studied: full motion, only lateral translational motion, only yaw rotational motion, and no motion.The study found that the lateral acceleration cue was of predominant importance in both performance and opinion.This suggested that if you had strong lateral translational cues combined with compelling visual yaw rotational cues, then the yaw motion system rotational cues might be redundant and unnecessary.In a study that considered visual cueing aspects as well as motion, Johnson (Ref.18) investigated variations in how the displayed level-of-detail might change in the image generator as one gets closer or further away from an object for a height control task.Platform motion was also a variable.The results showed that changing the level-of-detail to maintain constant optical density as the altitude changed, like that of the real world (yet different than how visual systems provide it), improved altitude awareness.Separately, adding platform motion improved speed regulation and altitude perception.Further systematic changes in visual scene, via changing spatial frequency with alternating black-andwhite stripes, and motion cues were evaluated in the vertical axis (Ref.19).The variations in visual scene evaluated had no effect, while the motion configurations did have an effect.These configurations were subsequently analyzed using a structural pilot model (Ref.20).The intent was to develop and calibrate a model that would predict pilot opinion for a given rotorcraft and task, and the model's predictions correlated well, in a ranking sense, with the subjective ratings. +Cueing Requirements for Autorotation and Shipboard Landing (1982-2001)The VMS was used to evaluate cueing effects on autorotation in two studies separated by more than a decade (Refs.21 and 22).In the first study, autorotation task performance decreased with degraded motion cues, yet acceptable performance could be attained as long as adequate visual cues were present.In the second study with improved visual cueing technology, the effect of visual texture and motion variations on autorotation task performance were evaluated.Visual texture affected all measures, but the finest texture did not perform the best, debunking a myth that more texture is better, as one has to be mindful of the aliasing that can occur due to both the image generator and human visual dynamics.This result was also supported by a fixed-base psychophysics study (Ref.23).Visual detail affected only pilot subjective opinion.Pilot performance, as well as opinion of motion fidelity, improved with increased motion cueing.Two simulation studies were conducted under the Dynamic Interface Modeling and Simulation System as part of the Joint Shipboard Helicopter Integration Process (JSHIP), to investigate the simulation fidelity required to recreate the shipboard-landing task accurately (Ref.24).Considerable effort was placed on achieving the visual and dynamic fidelity required to ensure that pilot workload was consistent with that experienced in the actual task.This study was preceded almost two decades earlier by a similar one that evaluated whether high-fidelity simulation could be used for assessing the shipboard landing environment using the VMS and a model of a SH-2F helicopter (Ref.25).The more recent simulation showed that the significant improvement in vehicle modeling and visual display fidelity in the time period between these studies had made the overall cueing environment more realistic, but that challenges remained in accurately modeling and simulating the ship airwake and its interaction with the rotorcraft. +Adverse pilot-vehicle interactions:A study by Schroeder in 1997 (Ref.26) evaluated the effect of simulator platform motion on predicting Pilot Induced Oscillations (PIOs).The goal of the study was to understand the level of motion fidelity necessary to reproduce PIOs observed in a previous in-flight experiment (Ref.27).Three platform motion characteristics were examined: large, small, and no motion, where small motion simulated the motion envelope of a conventional hexapod with 60-in stroke actuators, and large motion used the full range of motion available on the VMS.Pilot opinion ratings and performance data from the study indicated that large motion was necessary to match the results of the in-flight experiment.The results indicated that the pitch-rate cues from small and large motion were both adequate, but the more realistic vertical acceleration cues available in the large motion made the difference in pilot performance and opinion.A study by Hoh in 2006 investigated rudder flight control system requirements for passenger aircraft (Ref.28).Specifically, the study investigated the rudder control system characteristics that could contribute to pilot overcontrol or PIO in the directional axis.A primary objective of the study was to assess the role of motion cueing fidelity on pilot perception and tendency to PIO.To this end, experimental configurations were evaluated using the full motion envelope of the VMS and a simulated hexapod motion envelope using a directional control task.Here again, pilot opinion data and comments indicated that full motion provided more compelling and accurate cues for this task than the simulated hexapod.These studies indicate that a range of motion similar to that available on the VMS is required for providing the level of realism necessary to ensure that pilot cueing modalities and control technique does not differ significantly from flight for the evaluated tasks.They serve as a point of departure for further research on the minimum motion fidelity required for various flight tasks. +VSTOL Aircraft ResearchVSTOL aircraft research from two broad programs by NASA and other collaborating agencies accounts for approximately 60% of the simulations conducted at the VMS.The two programs developed technologies for rotorcraft and short take-off vertical landing (STOVL) aircraft that are presently incorporated in modern rotorcraft and the F-35B Joint Strike Fighter (JSF), respectively.The rotorcraft research began in earnest in the 1970s as the US Army recognized that understanding of rotorcraft performance and flying qualities issues had to be expanded and improved to provide a solid base of knowledge for designing and developing the next generation of military rotorcraft.This eventually led to the development of a military standard for rotorcraft handling qualities, and modern flight control and pilotvehicle interface technologies.A similar recognition of the deficits in understanding jet-lift aircraft and the need to develop new flight control concepts for these aircraft resulted in another multi-agency program that led to the flight control technologies incorporated in the JSF.In both cases, data and insights from VMS simulations reduced technical risk and enhanced safety by maturing concepts and technologies in a safe but realistic environment. +STOVL Aircraft:The flight control requirements for STOVL operations were examined using simulations on the VMS and flight tests under the NASA/UK MoD Joint Aeronautical Program between 1980 and 1996.In parallel, the advanced short take off and vertical landing program initiated by the Defense Advanced Research Projects Agency investigated flying qualities requirements for integrated flight and propulsion concepts.The JSF program and the manufacturers leveraged this body of knowledge to design the X-32B and X-35B prototypes leading to the selection of the F-35B as the next generation of fighter/attack aircraft for the US and UK (Ref.29).Research in support of these programs included flying qualities for advanced control modes, maximum control power used, control system dynamic response associated with thrust transfer rates for attitude control, thrust margin in the presence of ground effect and hot gas ingestion, dynamic thrust response for the engine core, and flight path control during transition (Ref.30).Determining flight control and aircraft performance requirements for shipboard landing were the focus of many of these studies and the highfidelity motion cueing available on the VMS provided confidence that the simulation results would translate to flight.In one study to evaluate minimum thrust-to-weight ratio required for precise vertical control prior to shipboard landing, the VMS was configured to provide 1:1 motion in the vertical axis to provide the most realistic vertical motion cueing to the pilots.The study established the thrust-to-weight ratio required for satisfactory performance in controlling sink rate during shipboard landings (Ref.29). +Rotorcraft Handling Qualities:In the 1970s, the Army needed a handling qualities specification that could guide the development of rotorcraft to meet the more demanding missions and tactics envisioned in the future.The strategy for developing a helicopter handling qualities database of sufficient validity for use in a military specification was to combine high-fidelity simulation with a limited amount of flight test activity.Almost all the simulation data incorporated into ADS-33 came from VMS studies.The process of developing the database is ongoing, and simulation studies on the VMS continue to fill known gaps in the database and refine others.Early studies investigated the effect of design variations on rotorcraft dynamic characteristics and handling qualities.Later studies investigated some of the fundamental precepts under development for the specification -required response type and response bandwidth (Ref.32).An innovative concept introduced in ADS-33 is the trade-off between augmentation (which defines response type) and the visual cueing environment.An important outcome of these studies was the progressive development of evaluation tasks that were designed to be representative of the mission tasks expected of rotorcraft but also constrained to allow repeatability and promote consistency in handling qualities ratings.These evaluation tasks were refined over the course of many experiments on the VMS and included in ADS-33 as demonstration maneuvers for evaluating the overall handling qualities of a rotorcraft.Since its release, ADS-33 has guided the procurement of the CH-47F and CH-53K helicopters, and the UH-60Mu and AH-64D fly-by-wire upgrades by the Army and Navy.It was designed for, and used as, the guiding specification for the RAH-66 Comanche flight control system that is the basis for the full-authority fly-by-wire flight control systems implemented on the UH-60Mu, CH-53K, H-92 Superhawk, and the digital automatic flight control system on the CH-47F which was evaluated on the VMS (Ref.33).The RAH-66 control system design was, in turn, based on extensive research on the Army's advanced digital optical control system program that also used the VMS for initial design and development. +Spacecraft Research:Spacecraft research on the VMS began soon after it was built, with engineering development studies of the Space Shuttle Orbiter.Simulations on the VMS made important contributions, starting with initial engineering studies in 1979 and 1980 to refine the Orbiter flight control system prior to first flight (Ref.34).Subsequently, NASA used the VMS for engineering studies on longitudinal and lateral handling qualities evaluations, and landing rollout systems and procedures development.As the US considers Shuttle replacement, studies on the VMS are evaluating the handling qualities and flight control system requirements for the next generation of US spacecraft -the Orion Crew Exploration Vehicle (CEV) and the Altair Lunar Lander vehicle.The ability of pilots to successfully carry out their missions will be determined in part by the handling qualities of these new spacecraft.Some operational tasks may be fully automated, while others will be manually controlled.Even for the nominally automated tasks, NASA requires a backup manual control capability when an automated system or critical subcomponent of the spacecraft fails.In these cases of emergency reversion to manual control, when the pilot switches abruptly from monitoring to active control, it is even more important that the vehicle have good handling qualities.At this time, no reference standards exist for handling qualities of piloted spacecraft.Handling qualities data do exist for some space vehicles; however, the focus of these studies was on evaluating or addressing deficiencies in the handling qualities of an existing design for a specific vehicle.A more systematic approach is needed to map out handling qualities variations for a range of design variables and identify regions of satisfactory handling qualities in the design space for a class of vehicles.A project to develop design guidelines for spacecraft handling qualities was initiated by NASA in 2007 and four simulations were conducted on the VMS with more to follow.Two simulations investigated the effect of flight control system and guidance display design on CEV handling qualities during docking with a simulated International Space Station (Ref.37).Two simulations investigated the effect of control power and guidance display design on the handling qualities of the Lunar Lander during precision approach and landing (Ref.38).With its realistic cues and flexible simulation architecture, the VMS could play a similar role in the engineering development and training with these new space vehicles as it did with the Space Shuttle.Figure 1 .1Figure 1.VMS Transport Cab on and off the motion systemThe high-fidelity flight controls are heavily modified and optimized McFadden hydraulic force-loader systems (Ref.1).A custom digital-control interface allows for comprehensive adjustment of the controller's static and dynamic characteristics.Force-loader characteristics may be varied during simulated flight as necessary for studying pilot cueing concepts using inceptors.A variety of aircraft manipulators, ranging from the regular column-and-wheel type to conventional rotorcraft controls and side sticks, are available and may be combined with the force-loader systems. +Figure 3 .3Figure 3. VMS Motion Algorithm Schematic +Over 20 Space Shuttle flight rules changes (changes in mission operation procedures) have resulted from engineering development studies conducted on the VMS (Ref.35).Initial experience and success of the VMS as an engineering development simulator led to its use as an astronaut training simulator for the landing and rollout phase of flight.Astronaut training sessions in the VMS are held semi-annually with each session containing unique objectives related to specific mission profiles and maintaining pilot training currency under nominal and off-nominal conditions.A Space Shuttle Mission Simulator team lead stated that the realistic pilot cueing environment in the VMS "by far affords the most realistic shuttle rollout simulation" (Ref.36).Every Shuttle pilot has trained on the VMS and over 65,000 landing and rollout training and engineering runs have been completed. + + + +Table 1 . VMS motion system performance limits (from Ref. 2)1DegreeDisplacementVelocityAccelerationofSystemOperationalSystemOperationalSystemOperationalFreedomLimitsLimitsLimitsLimitsLimitsLimitsLongitudinal ± 4 ft± 4 ft± 5 ft/sec± 4 ft/sec± 16 ft/sec 2± 10 ft/sec 2Lateral± 20 ft± 15 ft± 8 ft/sec± 8 ft/sec± 13 ft/sec 2± 13 ft/sec 2Vertical± 30 ft± 22 ft± 16 ft/sec± 15 ft/sec± 22 ft/sec 2± 22 ft/sec 2Roll± 0.31 rad± 0.24 rad± 0.9 rad/sec ± 0.7 rad/sec ± 4 rad/sec 2± 2 rad/sec 2Pitch± 0.31 rad± 0.24 rad± 0.9 rad/sec ± 0.7 rad/sec ± 4 rad/sec 2± 2 rad/sec 2Yaw± 0.42 rad± 0.24 rad±0.9 rad/sec ± 0.8 rad/sec ± 4 rad/sec 2± 2 rad/sec 2 +32) military specification for rotorcraft handling qualities, MIL-H-8501A, was written in 1952 and was inadequate (Ref.31).Several attempts to update MIL-H-8501A met with little success and were not adopted, primarily due to a lack of background data of adequate quality.To remedy this, the U. S. Army Aeroflightdynamics Directorate (AFDD), in collaboration with NASA Ames, began an effort to build a database of handling qualities data and design criteria that could be incorporated into a new handling qualities specification.The VMS was central to this effort from the outset.In 1982, the AFDD began the development of a new rotorcraft handling qualities specification to supersede MIL-H-8501A (Ref.31).The specification, US Army Aeronautical Design Standard -33 or ADS-33 (Ref.32), was completed and published in 1987.Initial development of the specification was for the procurement of the modern light/attack/scout helicopter by the Army leading to the development of a prototype helicopter, the RAH-66 Comanche.Sikorsky Aircraft built and flew two prototypes; however, the Army cancelled the production program in 2004 to provide renovation funds for its existing helicopter fleet. + Chief, Aerospace Simulation Research and Development Branch + Simulation Engineer, Aerospace Simulation Research and Development Branch + Chief, Aviation Systems Division + + + +Concluding RemarksThe NASA Ames Vertical Motion Simulator's motion system design leverages past experience with flight simulation at NASA Ames to achieve the best compromise between size and cueing fidelity necessary for flight research.The result is a ground-based flight simulator with an unmatched level of realism in pilot feedback cueing.This realism increases the likelihood that information and conclusions gathered from simulations on the VMS will translate to flight with few changes, thus enabling the VMS to be the best available ground-based alternative to flight testing.This realism has, over the past three decades, enabled researchers using the VMS to generate a wealth of data and knowledge that has enhanced the understanding of pilot cueing and the art of simulating these cues with minimum loss of fidelity.It has also provided critical design data on aircraft handling qualities and for flight control development resulting in reduced risk for major programs such as current rotorcraft flight control system upgrades and the Joint Strike Fighter.The high fidelity cueing environment also enabled the VMS to play an important role in the Space Shuttle program as an engineering development tool and later as a pilot training tool.Its ability to recreate pilot cueing in unconventional vehicles is presently exemplified by its use in evaluating the handling qualities and flight control requirements for the next generation of US spacecraft.Engineered for realism, the VMS is a safe and cost-effective solution for research and development of vehicles and human-vehicle interaction concepts. + + + + + + + Optimizing the Performance of the Pilot Control Loaders at the NASA Vertical Motion Simulator + + RAMueller + + + + AIAA Paper 2008-6349, AIAA Modeling and Simulation Technologies Conference + Honolulu, HI + + Aug. 2008 + + + Mueller, R. A., "Optimizing the Performance of the Pilot Control Loaders at the NASA Vertical Motion Simulator," AIAA Paper 2008-6349, AIAA Modeling and Simulation Technologies Conference, Honolulu, HI, Aug. 2008. + + + + + Vertical Motion Simulator Familiarization Guide + + GeorgeLDanek + + + May 1993 + + + NASA TM 103923 + Danek, George L., "Vertical Motion Simulator Familiarization Guide," NASA TM 103923, May 1993. + + + + + Effects of Simulator Motion and Visual Characteristics on Rotorcraft Handling Qualities + + DGMitchell + + + DCHart + + + + American Helicopter Society Conference on Piloting Vertical Flight Aircraft + San Francisco, CA + + Jan. 1993 + + + Mitchell, D.G. and Hart, D.C., "Effects of Simulator Motion and Visual Characteristics on Rotorcraft Handling Qualities," American Helicopter Society Conference on Piloting Vertical Flight Aircraft, San Francisco, CA, Jan. 1993. + + + + + Identification and Simulation Evaluation of an AH-64 Helicopter Hover Math Model + + JASchroeder + + + MBTischler + + + DCWatson + + + MMEshow + + + + AIAA Paper 91-2877, AIAA Atmospheric Flight Mechanics Conference + New Orleans, LA + + Aug. 1991 + + + Schroeder, J.A., Tischler, M.B., Watson, D.C., and Eshow, M.M., "Identification and Simulation Evaluation of an AH-64 Helicopter Hover Math Model," AIAA Paper 91-2877, AIAA Atmospheric Flight Mechanics Conference, New Orleans, LA, Aug. 1991. + + + + + Historical Review of Piloted Simulation at NASA Ames + + SethBAnderson + + + + Proceedings of the AGARD FVP Symposium on "Flight Simulation -Where are the Challenges? + the AGARD FVP Symposium on "Flight Simulation -Where are the Challenges?Braunschweig, Germany + + + Anderson, Seth B., "Historical Review of Piloted Simulation at NASA Ames," Proceedings of the AGARD FVP Symposium on "Flight Simulation -Where are the Challenges?" Braunschweig, Germany. + + + + + Vertical Motion Requirements for Landing Simulation + + RichardSBray + + + Feb 1973 + 236 + + + NASA TM X-62 + Bray, Richard S., "Vertical Motion Requirements for Landing Simulation," NASA TM X-62,236, Feb 1973. + + + + + Initial Operating Experience with an Aircraft Simulator Having Extensive Lateral Motion + + RichardSBray + + + May 1972 + 155 + + + NASA TM X-62 + Bray, Richard S., "Initial Operating Experience with an Aircraft Simulator Having Extensive Lateral Motion," NASA TM X-62,155, May 1972. + + + + + Assessment of Simulation Fidelity Using Measurements of Piloting Technique in Flight + + WFClement + + + WBCleveland + + + DLKey + + + + AGARD Conference Proceedings No. 359 + Monterey, CA + + May 1984 + + + Clement, W.F., Cleveland, W.B., and Key, D.L., "Assessment of Simulation Fidelity Using Measurements of Piloting Technique in Flight," AGARD Conference Proceedings No. 359, Monterey, CA, May 1984. + + + + + Fidelity Assessment of a UH-60A Simulation on the NASA Ames Vertical Motion Simulator + + AAtencioJr + + 93-A-005 + + + NASA TM 104016, USAATC + + Sept. 1993 + + + Tech. Report + Atencio, Jr., A. "Fidelity Assessment of a UH-60A Simulation on the NASA Ames Vertical Motion Simulator," NASA TM 104016, USAATC Tech. Report 93-A-005, Sept. 1993. + + + + + Visual and Motion Cuing in Helicopter Simulation + + RSBray + + + Sept. 1985 + + + NASA TM-86818 + Bray, R. S., "Visual and Motion Cuing in Helicopter Simulation," NASA TM-86818, Sept. 1985. + + + + + The Effects of Simulator Visual-Motion Asynchrony on Simulator Induced Sickness + + MEMccauley + + + LJHettinger + + + TJSharkey + + + JBSinacori + + + + AIAA Flight Simulation Technologies Conference and Exhibit + Dayton, OH + + Sept. 1990 + + + McCauley, M.E., Hettinger, L.J., Sharkey, T.J., and Sinacori, J.B., "The Effects of Simulator Visual- Motion Asynchrony on Simulator Induced Sickness," AIAA Flight Simulation Technologies Conference and Exhibit, Dayton, OH, Sept. 1990. + + + + + Does a Motion Base Prevent Simulator Sickness? + + TJSharkey + + + MEMccauley + + + + AIAA/AHS Flight Simulation Technologies Conference + Hilton Head, SC + + Aug. 1992 + + + Sharkey, T.J. and McCauley, M.E., "Does a Motion Base Prevent Simulator Sickness?" AIAA/AHS Flight Simulation Technologies Conference, Hilton Head, SC, Aug. 1992. + + + + + Visual and Roll-Lateral Motion Cueing Synchronization Requirements for Motion-Based Flight Simulations + + WWChung + + + JASchroeder + + + + 53 rd Annual Forum Proceedings + Virginia Beach, VA + + American Helicopter Society + Apr. 1997 + + + Chung, W.W. and Schroeder, J.A., "Visual and Roll-Lateral Motion Cueing Synchronization Requirements for Motion-Based Flight Simulations," American Helicopter Society 53 rd Annual Forum Proceedings, Virginia Beach, VA, Apr. 1997. + + + + + Ground Based Simulation Evaluation of the Effects of Time Delays and Motion on Rotorcraft Handling Qualities + + DGMitchell + + + RHHoh + + + AAtencioJr + + + DLKey + + AVSCOM-TR-91-A-010 + + Aug. 1990 + + + Mitchell, D.G., Hoh, R.H., Atencio Jr., A., and Key, D.L., "Ground Based Simulation Evaluation of the Effects of Time Delays and Motion on Rotorcraft Handling Qualities," AVSCOM-TR-91-A-010, Aug. 1990. + + + + + Simulation Motion Effect on Single Axis Compensatory Tracking + + JASchroeder + + + + AIAA Flight Simulation Technologies Conference + Monterey, CA + + 1993 + + + Schroeder, J.A. "Simulation Motion Effect on Single Axis Compensatory Tracking," AIAA Flight Simulation Technologies Conference, Monterey, CA, 1993. + + + + + Evaluation of Simulation Motion Fidelity Criteria in the Vertical and Directional Axes + + JASchroeder + + + + Journal of the American Helicopter Society + + 41 + 2 + + Apr. 1996 + + + Schroeder, J.A., "Evaluation of Simulation Motion Fidelity Criteria in the Vertical and Directional Axes," Journal of the American Helicopter Society, Vol. 41, No. 2, Apr. 1996, pp. 44-57. + + + + + Flight Simulation -Where are the Challenges? + + JASchroeder + + + WWJohnson + + AGARD CP-577 + + Apr. 1996 + Braunschweig, Germany + + + Yaw Motion Cues in Helicopter Simulation + Schroeder, J.A. and Johnson, W.W., "Yaw Motion Cues in Helicopter Simulation," Paper No. 5, AGARD CP-577, "Flight Simulation -Where are the Challenges?" Braunschweig, Germany, Apr. 1996. + + + + + Visual-Motion Cueing in Altitude and Yaw Control + + WJohnson + + + JASchroeder + + + + 38 th Annual Human Factors and Ergonomics Society Meeting + Nashville, TN + + Oct. 1994 + + + Johnson, W. and Schroeder, J.A., "Visual-Motion Cueing in Altitude and Yaw Control," 38 th Annual Human Factors and Ergonomics Society Meeting, Nashville, TN, Oct. 1994. + + + + + Evaluation of a Motion Fidelity Criterion with Visual Scene Changes + + JASchroeder + + + WW YChung + + + RAHess + + + + Journal of Aircraft + + 37 + 4 + + July-August, 2000 + + + Schroeder, J.A., Chung, W.W.Y., and Hess, R.A., "Evaluation of a Motion Fidelity Criterion with Visual Scene Changes," Journal of Aircraft, Vol. 37, No. 4, July-August, 2000, pp. 580-587. + + + + + Modeling Human Pilot Cue Utilization with Applications to Simulator Fidelity Assessment + + YZeyada + + + RAHess + + + + Journal of Aircraft + + 37 + 4 + + July-Aug., 2000 + + + Zeyada, Y. and Hess, R.A., "Modeling Human Pilot Cue Utilization with Applications to Simulator Fidelity Assessment," Journal of Aircraft, Vol. 37, No. 4, July-Aug., 2000, pp. 588-597. + + + + + Pilot Use of Simulator cues for Autorotation Landings + + WADecker + + + CFAdam + + + RMGerdes + + + June 1986 + American Helicopter Society's 42 nd Annual Forum + Washington, DC + + + Decker, W.A., Adam, C.F., and Gerdes, R.M., "Pilot Use of Simulator cues for Autorotation Landings," American Helicopter Society's 42 nd Annual Forum, Washington, DC, June 1986. + + + + + Effects of Visual Texture, Grids, and Platform Motion on Unpowered Helicopter Landings + + MGDearing + + + JASchroeder + + + BTSweet + + + MKKaiser + + + + AIAA Modeling and Simulation Conference + Montreal, Canada + + 2001 + + + Paper 2001-4251 + Dearing, M.G., Schroeder, J.A., Sweet, B.T., and Kaiser, M.K., "Effects of Visual Texture, Grids, and Platform Motion on Unpowered Helicopter Landings," Paper 2001-4251, AIAA Modeling and Simulation Conference, Montreal, Canada, 2001. + + + + + Runway Texture and Grid Pattern Effects on Rate-of-Descent Perception + + JASchroeder + + + MGDearing + + + BTSweet + + + MKKaiser + + + AAAtencioJr + + + + Paper 2001-4307, AIAA Modeling and Simulation Conference + Montreal, Canada + + 2001 + + + Schroeder, J.A., Dearing, M.G., Sweet, B.T., Kaiser, M.K., and Atencio, Jr., A.A., "Runway Texture and Grid Pattern Effects on Rate-of-Descent Perception," Paper 2001-4307, AIAA Modeling and Simulation Conference, Montreal, Canada, 2001. + + + + + DIMSS -JSHIPs Modeling and Simulation Process for Ship/Helicopter Testing and Training + + MFRoscoe + + + CHWilkinson + + + + AIAA Modeling and Simulation Technologies Conference and Exhibit + Monterey, CA + + Aug. 2002 + + + Roscoe, M. F., and Wilkinson, C. H., "DIMSS -JSHIPs Modeling and Simulation Process for Ship/Helicopter Testing and Training," AIAA Modeling and Simulation Technologies Conference and Exhibit, Monterey, CA, Aug. 2002. + + + + + Simulation and Evaluation of the SH-2F Helicopter in a Shipboard Environment Using the Interchangeable Cab System + + CHPaulk + + + Jr + + + DLAstill + + + STDonley + + NASA-TM-84387 + + Aug. 1983 + + + Paulk, C. H., Jr., Astill, D. L., and Donley, S. T., "Simulation and Evaluation of the SH-2F Helicopter in a Shipboard Environment Using the Interchangeable Cab System," NASA-TM-84387, Aug. 1983. + + + + + Pilot-Induced Oscillation Prediction with Three Levels of Simulation Motion Displacement + + JefferyASchroeder + + + Chung + + + WYWilliam + + + Tran + + + TDuc + + + Laforce + + + Soren + + + NormanJBengford + + AIAA- 98-4333 + + + AIAA Atmospheric Flight Mechanics Conference + + Aug. 1998 + Boston, MA + + + Schroeder, Jeffery A, Chung, William, W. Y., Tran, Duc T., Laforce, Soren, and Bengford, Norman J., "Pilot-Induced Oscillation Prediction with Three Levels of Simulation Motion Displacement," AIAA- 98-4333, AIAA Atmospheric Flight Mechanics Conference, Boston, MA, Aug. 1998. + + + + + Flight Test Evaluation of Techniques to Predict Pilot Induced Oscillations + + EABjorkman + + + + Dec 1986 + + + Air Force Institute of Technology + + + Masters Thesis + Bjorkman, E. A., "Flight Test Evaluation of Techniques to Predict Pilot Induced Oscillations," Masters Thesis, AFIT/GAE/AA/86J-1, Air Force Institute of Technology, Dec 1986. + + + + + Piloted Simulation Study to Develop Transport Aircraft Rudder Control System Requirements: Phase I -Simulation Motion System Requirements and Initial Results + + RogerHHoh + + + PaulDesrochers + + + ThomasKNicoll + + 1137-4 + + April 2007 + Hoh Aeronautics, Inc + + + Technical Report + Hoh, Roger H., Desrochers, Paul, and Nicoll, Thomas K., "Piloted Simulation Study to Develop Transport Aircraft Rudder Control System Requirements: Phase I -Simulation Motion System Requirements and Initial Results," Hoh Aeronautics, Inc. Technical Report 1137-4, April 2007. + + + + + STOVL Integrated Flight and Propulsion Control: Current Successes and Remaining Challenges + + JamesWDenham + + + + AIAA 2002-6021, Presented at the 2002 Biennial International Powered Lift Conference and Exhibit + Williamsburg, Virginia + + Nov. 2002 + + + Denham, James W., "STOVL Integrated Flight and Propulsion Control: Current Successes and Remaining Challenges," AIAA 2002-6021, Presented at the 2002 Biennial International Powered Lift Conference and Exhibit, Williamsburg, Virginia, Nov. 2002. + + + + + Experience with Integrated Flight/Propulsion Controls from Simulation of STOVL Fighter Concepts + + JamesAFranklin + + + + AIAA International Powered Lift Conference + Santa Clara, CA + + Dec. 1993 + + + AIAA-93-4874-CP + Franklin, James A., "Experience with Integrated Flight/Propulsion Controls from Simulation of STOVL Fighter Concepts," AIAA-93-4874-CP, Presented at the AIAA International Powered Lift Conference, Santa Clara, CA, Dec. 1993. + + + + + Handling Qualities Specifications for U.S. Military Helicopters + + DLKey + + + + AIAA Journal of Aircraft + + 19 + 2 + Feb. 1982 + + + Key, D. L., "Handling Qualities Specifications for U.S. Military Helicopters," AIAA Journal of Aircraft, Vol. 19, no. 2, Feb. 1982. + + + + + Handling Qualities Requirements for Military Rotorcraft + + Anon + + + + Aeronautical Design Standard-33 (ADS-33E-PRF), US Army Aviation and Missile Command + + Mar. 2000 + + + Anon., "Handling Qualities Requirements for Military Rotorcraft," Aeronautical Design Standard-33 (ADS-33E-PRF), US Army Aviation and Missile Command, Mar. 2000. + + + + + ADS-33E Predicted and Assigned Low-Speed Handling Qualities of the CH-47F with Digital AFCS + + JGIrwin + + + PGEinthoven + + + DGMiller + + + CLBlanken + + + + AHS 63 rd Annual Forum + Virginia Beach, VA + + May 2007 + + + Irwin, J. G., Einthoven, P. G., Miller, D. G., and Blanken, C. L., "ADS-33E Predicted and Assigned Low-Speed Handling Qualities of the CH-47F with Digital AFCS," AHS 63 rd Annual Forum, Virginia Beach, VA, May 2007. + + + + + Space Shuttle Longitudinal Landing Flying Qualities + + BruceGPowers + + + + AIAA Journal of Guidance and Control + + 9 + 5 + + Sep-Oct 1986 + + + Powers, Bruce G., "Space Shuttle Longitudinal Landing Flying Qualities," AIAA Journal of Guidance and Control, Vol. 9, No. 5, Sep-Oct 1986, p. 566-572. + + + + + Space Shuttle Landing and Rollout Training at the Vertical Motion Simulator + + StevenDBeard + + + LeslieARingo + + + BrianMader + + + EstelaHBuchmann + + + ThomasTanita + + + + AIAA Modeling and Simulation Technologies Conference + Honolulu, HI + + Aug. 2008 + + + Beard, Steven D., Ringo, Leslie A., Mader, Brian, Buchmann, Estela H., and Tanita, Thomas, "Space Shuttle Landing and Rollout Training at the Vertical Motion Simulator," presented at the AIAA Modeling and Simulation Technologies Conference, Honolulu, HI, Aug. 2008. + + + + + The Flights Before the Flight: An Overview of Shuttle Astronaut Training + + JohnTSims + + + MichaelRSterling + + + + AIAA 89-3265-CP, AIAA Flight Simulation Technologies Conference + Boston, MA + + Aug. 1989 + + + Sims, John T., and Sterling, Michael R., "The Flights Before the Flight: An Overview of Shuttle Astronaut Training," AIAA 89-3265-CP, AIAA Flight Simulation Technologies Conference, Boston, MA, Aug. 1989. + + + + + Handling Qualities Evaluation for Spacecraft Docking in Low Earth Orbit + + EricMueller + + + KarlDBilimoria + + + ChadFrost + + + + AIAA Guidance, Navigation, and Control Conference + Honolulu, HI + + Aug. 2008 + + + Mueller, Eric, Bilimoria, Karl D., and Frost, Chad, "Handling Qualities Evaluation for Spacecraft Docking in Low Earth Orbit," presented at the AIAA Guidance, Navigation, and Control Conference, Honolulu, HI, Aug. 2008. + + + + + Effects of Control Power and Guidance Cues on Lunar Lander Handling Qualities + + KarlDBilimoria + + + + AIAA SPACE 2008 Conference + San Diego, CA + + Sep. 2008 + + + Bilimoria, Karl D., "Effects of Control Power and Guidance Cues on Lunar Lander Handling Qualities," presented at the AIAA SPACE 2008 Conference, San Diego, CA, Sep. 2008. + + + + + + diff --git a/file14.txt b/file14.txt new file mode 100644 index 0000000000000000000000000000000000000000..de9e75b8d08644ae552f33cfedef10e7208048db --- /dev/null +++ b/file14.txt @@ -0,0 +1,446 @@ + + + + +I. IntroductionThe "Next Generation Air Transportation System," known as NextGen, is a comprehensive transformation of the National Airspace System (NAS) to a system that will be safer, more reliable, more efficient, and which will reduce the impact of aviation on the environment.The transition to NextGen is vital to improving system performance, meeting continued growth in air traffic, and increasing the Nation's mobility to support economic progress.The Federal Aviation Administration (FAA) estimates that by 2020, NextGen improvements will reduce delays by 41 percent, reduce cumulative carbon dioxide emissions by 16 million metric tons, and reduce fuel use by 1.6 billion gallons in cumulative.Taken together, these and other efficiency improvements will provide an estimated $38 billion in cumulative benefits to aircraft operators, the traveling public, and the FAA by 2020 (Ref.1).The Aeronautics Research Mission Directorate (ARMD) of the National Aeronautics and Space Administration (NASA) is collaborating with the FAA and other industry partners to enhance the transition to NextGen by developing advanced automation concepts and tools that provide air traffic controllers, pilots, and other airspace users with more accurate real-time information and advisories about the nation's traffic flow, weather, and routing.Improving the precision of these information sources and intelligently combining them to enable better decision-making in the NAS are key enablers of NextGen.NASA's NextGen related programs research and develop new concepts and technologies, evaluates them in an operationally relevant environment, demonstrates their feasibility and benefits, and transitions them to an external organization (primarily the FAA) for implementation in the NAS.Examples of NASA technologies that have completed this cycle include Traffic Management Advisor (TMA) (Ref.2) and the Efficient Descent Advisor (EDA)/3D Path Arrival Management (3D-PAM) concept (Ref.3).To streamline this process, ARMD created the Airspace Technology Demonstration (ATD) Project that builds upon fundamental research conducted by NASA and its partners and demonstrates integrated, domain focused solutions for NAS operations.The goal of this project is to accelerate the process of maturing new concepts and technologies and transition them to stakeholders while demonstrating their feasibility and quantifying their benefit to the NAS.The current and planned ATD sub-projects are: ATD-1: Interval Management -Terminal Area Precision Scheduling and Spacing. ATD-2: Integrated Arrival/Departure/Surface Operations (IADS) Metroplex Traffic Management. ATD-3: Applied Traffic Flow Management. + TASA: Technologies for Airplane State AwarenessThe TASA sub-project seeks to mitigate flight crew loss of airplane state awareness as a causal factor in commercial aviation accidents and incidents by developing cockpit systems with predictive algorithms, models for aircraft stall performance, and flight crew training methods that may be transitioned to the aviation community together with the Commercial Aviation Safety Team (CAST).The ATD-1 sub-project enhances the initial TMA concept and improves the precision and throughput of terminal arrivals by developing and operationally demonstrating an integrated set of NASA arrival management software technologies for planning and executing efficient arrival operations in the terminal environment of a highdensity airport.The key enabler is NASA's Terminal Sequencing and Spacing (TSS) technology (Ref.4) that was transferred to the FAA and will be implemented in the NAS in the near future.The ATD-2 and ATD-3 sub-projects are in their advanced and preliminary planning stages, respectively.ATD-2 will investigate improving the efficiency of arrival, departure, and surface operations while ATD-3 will investigate efficient routing concepts for en-route weather avoidance and oceanic operations.This paper describes the initial planning process for ATD-2.NASA's research portfolio includes several promising technologies for improving surface and departure operations that are in relatively advanced stages of development with active stakeholder partnership.As a result, NASA decided to explore the potential for improving IADS operations using these technologies in the ATD-2 subproject.The first step in this process, however, was to clearly understand the IADS technology needs from the stakeholder's point-of-view.As part of planning the project and setting its goals, therefore, NASA needed to identify areas in the current arrival, departure, surface operational environment where an integrated solution could offer the greatest benefit to the stakeholders.With this understanding, the ATD-2 sub-project could be structured to develop and demonstrate an integrated concept that would be valued and accepted by these stakeholders.To understand the critical issues related to IADS operations, NASA reached out to relevant stakeholders and gathered information and insights that enhanced NASA's understanding of the challenges faced by these stakeholders and identified the benefit (value) of potential solutions.The results of the stakeholder survey were combined with a literature survey American Institute of Aeronautics and Astronautics on IADS issues and previous research, sorted, categorized, and converted to project goals and technical objectives through the process illustrated in Fig. 1.This paper presents the stakeholder survey process, results, and an initial discussion of project goals that were distilled from this information.NASA is presently formally planning and formulating the ATD-2 sub-project to address the goals derived from the information provided by the stakeholders. +II. BackgroundArrival, departure, surface operations involve a complex interplay between stakeholders.For an integrated solution to be successful in this environment, it must address the economic and operational issues that influence this interplay.Before approaching the stakeholders to understand their viewpoint on these issues, a preliminary literature survey was conducted to get a broad understanding of this environment.This survey identified two basic factors that need to be considered by an IADS solution.These are: the economic factors that influence airline arrival, departure, surface operations, and the technical landscape that a potential IADS solution needs to integrate with.These are introduced and discussed briefly here to set a context for the stakeholder survey that is discussed later. +A. Relevant Economic Factors -Airline Block Times and On-Time MetricsEfficiencies in NAS operations, and IADS operations in particular, will be measured primarily in economic terms.It was, therefore, important to understand the basic economic factors driving one of the major stakeholder groupsthe commercial airlines.A key measure of airline cost performance is block time (Ref.5).The block time is the elapsed time for a flight when the airline's fixed asset (the airplane) is earning revenue by transporting passengers.To earn this revenue, an airline incurs direct operating costs including primarily crew and fuel costs that scale with block time.Typically, block time is determined as the time between the cabin door closing at departure and opening on arrival (Ref.5).On any flight, there are four distinct event-time signals that are transmitted from the aircraft to the airline (Ref.6).These are depicted in Fig. 2 and defined as: OUT (cabin door close/parking brake release), OFF (weight off landing gear), ON (weight on landing gear), and IN (cabin door open/parking brake set).Block time is defined as the elapsed time between the OUT and IN signals (Ref. 5).A key airline cost factor that is linked to scheduled or actual block time is the flight crew compensation rate (Ref.5).Airlines maintain comprehensive databases of actual historical block times for each city-pair within their route network.These databases are extremely sophisticated and include all the possible variables that influence block time, including type of aircraft, day-of-year, time-of-day, seasons, typical air-traffic constraints, airport issues, etc (Ref.5).This historic data is used to forecast future block times (known as scheduled block times) for planning route networks, flight schedules, revenues/costs, and business strategy.A report by an industry/air-navigation service provider focus group explains that air-traffic management (ATM) performance is critical to how airlines plan and operate their schedules and proposes several metrics that assesses the effect of the ATM system on airline economic performance (Ref.7).Flight schedule delay and predictability are two of the key metrics that are defined.The discussion surrounding these metrics in Ref. 7 indicates the mechanism by which increased delay and reduced predictability affect costs and increases scheduled block times.Ref. 7 also points out that scheduled block time for flights among the 29 hub airports in the U.S. increased by an average of 1.25 minutes from 1993 to 1997, and discusses the adverse effect of poor predictability on the integrity of an airline's network, schedule, and productivity.The MIT International Center for Air Transportation (ICAT) delved into actual block times and investigated the causes of delays on US domestic routes by assessing delays incurred at each "segment" of block timetaxi out, departure, en-route, arrival, and taxi-in (Ref.8).This analysis showed how airlines tended to be conservative in determining their scheduled block times -lengthening the scheduled flight time (scheduled block time) by adding "buffer" times to increase the likelihood that the aircraft will arrive at, or before, the scheduled arrival time and, thereby, improve the reliability of their schedules (Ref.8).Scheduled block times are also indirectly influenced by the "on-time" data that are voluntarily reported by the airlines to the US Department of Transportation's (USDOT) Bureau of Transportation Statistics (BTS).These data assesses airline performance against their schedules -a flight is counted as "on time" if its actual arrival or departure time (from the gate) is less than 15 minutes after the published scheduled time.Flights arriving early are counted as on time.Airline on-time performance statistics based on these data are compiled and published monthly by the BTS and are used by non-governmental organizations to rank airline schedule performance.As a result, performing well on these metrics is a priority for airlines. +B. Relevant Technical Factors -Existing Concepts and TechnologiesThe FAA has several programs that are designed to introduce new technologies and improve the efficiency of the air-traffic management system.These technology programs typically serve areas of the NAS where an aircraft is managed by air-traffic control (ATC)from the time the aircraft enters an active taxiway to the time it leaves it (known as the "movement area").Airport gate and ramp areas are not usually controlled by ATC and are known as "non-movement areas."They are usually managed by the airline or airport authority.Many of the upgrades to the ATM system are consolidated under three large FAA Decision Support System programs that introduce new technologies in discrete "builds" as technologies and funding becomes available (Ref. 1).These three programs are: Time Based Flow Management (TBFM)is an existing system that facilitates time-based metering for managing arrivals that is based on NASA's TMA technology (Ref.2).The TSS technology developed under NASA's ATD-1 project (Ref.4) will be introduced into future builds of this system.Future builds are also expected to include NASA's EDA (Ref.3) and Precision Departure Release Capability (PDRC -Ref.9) technologies. Terminal Flight Data Manager (TFDM)is a proposed system that will improve airport surface and terminal area operations by integrating airport surveillance, flight data, and automation systems available to ATC in the surface area.A Concept of Operations (ConOps) for TFDM is available (Ref.10) and procurement of an initial capability is being planned.Future builds of TFDM are expected to include time-based management of departures. Traffic Flow Management System (TFMS) -this system is the FAA's primary tool for traffic flow management.It gathers and processes flight plan and schedule information as well as NAS capacity constraints to enable the demand on the NAS to be assessed and compared with the available constraints.Future builds are expected to include automation and decision-support tools to alleviate NAS constraints.In addition, the FAA's System Wide Information Management (SWIM) initiative supports these and other NextGen programs by improving architecture and infrastructure to enable interoperability and data sharing between NAS stakeholders (Ref.11).In parallel with these programs, the FAA is studying concepts and technologies to include in future builds of these systems.One such concept promotes collaborative planning for airport departure operations using mechanisms and procedures similar to those used in Collaborative Decision Making (CDM) for strategic traffic flow management.The basic precepts of collaborative planning are that better information will lead to better decisionmaking, and that tools and procedures need to be in place to enable air navigation service providers and the flight operators to more easily respond to changing conditions.The Surface CDM ConOps (Ref.10), developed by the FAA and industry stakeholders, is founded on the premise that access to real-time aircraft position data on the airport surface coupled with timely sharing of accurate operational data among stakeholders affords the opportunity to better predict and manage demand on the airport surface.When implemented, the Surface CDM ConOps is expected to change the ATC's current "first-come, first-served" operating paradigm for managing surface traffic by incorporating strategic planning through information sharing and collaborative planning between airlines and service providers.The TFDM program encompasses the concepts in the Surface CDM ConOps (Ref.10).The FAA also sponsored a broader investigation on improving IADS operations.This investigation, summarized in Ref. 12, found that developing and implementing an integrated schedule for arrivals and departures at an airport could improve IADS operations.This integrated schedule would be updated using data sharing between the key stakeholders.The concept called for incorporating airspace and surface constraints in the integrated schedule.An integrated scheduler would allow airlines to absorb delay in the non-movement area (preferably at the gate) as opposed to the movement area where parked aircraft might contribute to congestion.Airlines and airports also use industry provided customized solutions to assess arrival/departure demand at airports and use this information to manage flights at the gate and maximize schedule integrity at specific airports.At a few airports in the US, these customized solutions use data sharing and collaborative planning to improve overall throughput while allowing airlines to maintain schedule integrity (Ref.13).In the US, these systems manage flights in the non-movement area and their information is not used by the ATC system, which operates on the "firstcome, first-served" operating paradigm. +III. Stakeholder Survey ProcessA preliminary examination of IADS operations and technologies identified four broad stakeholder categories.These are:  Airlines;  Airports;  Air Navigation Service Providers (ANSPs); and  Industry service/product providers.American Institute of Aeronautics and Astronautics The primary goal of the survey was to uncover all the key issues faced by stakeholders in IADS operations thus, emphasis was placed on the breadth of understanding of the issues as opposed to depth on specific operational issues.Also, an understanding of the economic and operational impact of any issue to a stakeholder was critical, therefore, particular emphasis was placed on identifying sources with knowledge of operational and business/financial aspects of their companies or organizations.Figure 3 shows the companies and organizations that provided information for the stakeholder survey.They included major US commercial passenger and cargo carriers; airport authorities of major US airports; groups within the FAA with responsibility for surface and terminal operations, operational infrastructure planning, and NextGen implementation; and industry providers of aircraft, avionics, and software/hardware tools that support airport surface and terminal area traffic management.The survey was conducted in two phases.The first-round of the survey was designed to uncover broad issues in arrival, departure, surface operations and their impact on the stakeholders.The second-round of the survey assessed stakeholder opinion on a proposed high-level concept by NASA for IADS operations that addressed the major shortfalls identified by stakeholders in the first round.In both phases, stakeholders were interviewed in-person or by telephone.A statement on the purpose of the survey and several questions on arrival, departure, surface operations were provided prior to the interview to indicate the scope of the interview.All interviews were conducted in the February to September 2014 time frame.To ensure consistency in the survey, a standardized set of questions was used as a basis for discussion with all stakeholders.Stakeholders were always encouraged, however, to expand the conversation as necessary to explain issues relevant to them.The standard set of questions were:  What are the top three challenges to airport operations?(Airport Authorities) +IV. First-Round Stakeholder Survey ResultsIn general, the stakeholders interviewed overwhelmingly supported NASA's initiative to improve IADS operations and appreciated being engaged in the process to develop a project.Several mentioned that NASA was uniquely positioned to "balance" the frequently competing interests of the stakeholders and formulate a solution that could benefit all NAS stakeholders.As a result, they provided a wealth of information on issues, experiences, and opinions.These responses were sorted, interpreted, and combined with other sources of information to arrive at the following broad issues from the point-of-view of each stakeholder category. +A. AirlinesThe inability to accurately and consistently predict flight times is an important issue for airlines.This unpredictability in flight schedules adds to costs and hampers an airline's ability to plan and execute business strategy.Schedule predictability is critical to maintaining the integrity of an airline's route network with its interconnecting and interdependent flights.This predictability is also needed to efficiently utilize their primary assets, the crews and aircraft, while conforming to crew duty time and aircraft maintenance requirements and regulations.Block time is a key metric in airline schedule and cost planning (Ref.5).As mentioned previously, airlines have comprehensive databases of block times for routes on their network that are used to determine scheduled block times when developing future flight schedules.Although airlines probably have different methods for estimating block times, it is reasonable to assume that they all rely on historical actual block time data, incorporating "offnominal" effects such as seasonal weather effects and holiday congestion, as a basis for estimating future block times (Ref.5).These scheduled block times are used to set flight schedules and estimate costs.In addition to scheduled block times, an airline's published schedule specifying departure and arrival times for flights throughout its network is also based on other marketing and economic factors.Two of the factors that were identified in the survey were:  Passenger ticket sales (revenues) are sensitive to the order in which a flight appears on a reservation screen or website when compared with the competition, as well as the preferred travel times on a particular day for a particular city pair; and  Customer opinion (and, therefore, ticket revenue) and the airline's brand image are sensitive to an airline's on-time arrival/departure performance that is reported to the DOT and is publicly available.As late departures and arrivals inconvenience passengers, results in negative publicity, and disrupts the integrity of the route network, airlines are conservative when estimating block times and building their schedulesthe more unpredictable the actual historical block time for a given city pair, the more conservative the scheduled block time that is used to set the scheduled departure and arrival times.This conservatism improves the likelihood that the flight will meet the on-time arrival performance metrics despite unanticipated delays.This conservatism is manifested as block time buffersthe more unpredictable the actual block time from experience, the greater the buffer that is applied to the scheduled block time when developing the schedule (Ref.8).As a result, airlines have performed very well when assessed against the on-time statistics but block times have grown steadily for a majority of US city pairs over the last several decades (Refs.14 and 15).For example, Richard Anderson, the CEO of Delta Airlines, remarked that "Delta's block time in 1956 flying between Atlanta and Washington National in a pistonengine [Douglas] DC-6 wasand is the same as it is today on a Boeing 757-200 jet" (Ref.16).As mentioned previously, airline operating costs scale directly with block time.The improvement in ontime performance using increased block times, therefore, results in increased cost to the airline.An FAA sponsored study estimated that block time buffers cost airlines $3.7 billion in 2007 (Ref.14).To manage these costs, stakeholders indicated that airline management expected operations to achieve actual block times that are a fraction (typically 65 to 75%) of the scheduled block times.Block time buffers also indirectly increase airline cost.The focus on being conservative by padding flight schedules increases the likelihood that flights will arrive early.Ref. 14 reports that, in 2012, 20% of U.S. domestic American Institute of Aeronautics and Astronautics flights by major airlines arrived 15-minutes or more ahead of their scheduled arrival time and notes that early arrivals result in increased costs due to inefficient crew and fleet utilization.Customer satisfaction can also be adversely affected when flights arrive early and cannot find an open gate, partially or completely negating the benefit of arriving early.The Ref. 8 study dissected the cause of schedule delay and variability by examining component segments of block time.The study showed that the elapsed time for the departure phase of a flight (from OUT to OFFsee Fig. 2) is the most variable in US domestic operations and is, therefore, the most difficult to estimate.As a result, it is probably the primary cause for adding block time buffers when developing flight schedules.Survey respondents indicated that ATC's first-come, first-served operating principle for departures is a primary factor in this uncertainty.The airline has no strategic or tactical information on the estimated OFF time prior to entering the movement area controlled by ATC.Under normal operations (i.e., when not subject to an ATCmandated traffic management program), the first time the airline is made aware of the departure queue and the estimated OFF time is when the pilot calls for taxi clearance.The need for better long-term (strategic) and short-term (tactical) information on the status of the NAS was echoed by all the survey participants across all stakeholder categories.The airline stakeholders mentioned that there is, presently, no warning of the timing and extent of FAA Traffic Management Initiatives (TMIs) and no preview of impending NAS capacity limitations (both surface and airspace) and that improved information sharing would improve their ability to better estimate actual block times and strategically manage their schedules.The final key issue for the airline stakeholders was the limited use of Performance Based Navigation (PBN) capability in the NAS.Many of the major domestic airlines have equipped their fleets with PBN capable avionics and the FAA has begun implementing these procedures for arrivals and departures at most airports.In practice, however, PBN procedures are not followed continuously in many airports due to several factors including airspace congestion and the presence of non-PBN equipped aircraft in the airspace.There seemed to be a consensus among the airline stakeholders that the promise of "better equippedbetter served" in the NAS was not being met. +B. AirportsIn an IADS context, the major operational priority for airports is maintaining throughput.These stakeholders indicated that the primary cause of loss of throughput is due to inaccurate aircraft estimated arrival times (ETAs).As the arrival stream of traffic to an airport is planned based on ETAs, late arrivals relative to these ETAs used for planning result in gaps in the arrival stream (or missed arrival "slots") that results in reduced arrival throughput.Most often, these missed slots are caused by the conservatism built into FAA mandated Ground Delay Programs (GDPs) that hold aircraft prior to departure due to a downstream constraint in the airspace or arrival airport.The decision to adopt a GDP is based on an estimation of demand at a constraint point in the airspace or destination airport using ETAs.When this demand estimation has a large degree of uncertainty (for example, if it includes a significant number of aircraft that have not yet departed and, therefore, have uncertain ETAs), the GDP will account for this uncertainty by being conservative and that can result in a loss of throughput at the destination airport.Another major cause of loss of throughput at airports is a loss of airspace capacity due to adverse weather.Airports would like to maintain a consistent throughput regardless of weather, with the exception of extreme weather situations, through improved use of available airspace capacity.Airport stakeholders indicated that improved data quality and decision-support tools would help alleviate potential loss of throughput.They mentioned that airport operational data was available through data sources such as the FAA's Aviation System Performance Metrics (ASPM) database but that this data was mostly unprocessed and not in a form that could be used by airports to make operational decisions.They identified a need for management tools that could help them interpret and understand the data and support decision-making.Airports were also interested in solutions that were recognized by the FAA and widely applied across U.S. airportsthey were reluctant to expend resources on solutions that were customized to a single airport or airspace that, in their experience, would not be supported by industry over the long-term. +C. Air Navigation Service ProviderSeveral groups within the FAA were interviewed including management and those involved in ATC operations.Following are the arrival, departure, surface operations issues that were identified together with ideas on implementing an eventual IADS solution: Reduce inter-ATC communication delays ("transactional friction") caused by manual transmission of information and approval requests by introducing intelligent data sharing (for example, coordinating American Institute of Aeronautics and Astronautics OFF times at the departure airport to meet an available arrival slot at a destination airport via data exchange rather than voice communication). Use collaborative decision making capability to facilitate smooth traffic flows from smaller airports into a busy large airport in the metroplex or congested en-route airway.This will allow the ATM system to plan and schedule flows to an airport or airway over a longer time horizon, as opposed to reacting to a flight after it departs. Integrate existing and planned FAA capabilities to improve arrival, departure, surface operationsthis refers primarily to integrating current and planned capabilities in TBFM, TFMS, and TFDM. Leverage on-going SWIM and TBFM/TFMS/TFDM developments to enhance the potential for implementing an IADS solution. Consider IADS operational improvements to alleviate congestion in busy metroplex areas (the New York terminal area, for example). Develop and demonstrate operational metrics that will incentivize improved operational efficiency by all stakeholders +D. IndustryResponses from the industry stakeholders largely reflected their individual areas of business focus and expertise.The key issues related to arrival, departure, surface operations were: +Operations management and control processes are largely manual, high workload, and responds slowly to constraints. +Need integrated solutions as opposed to the single-instance, customized, solutions at various airports that are used presently. +Need improved information sharing between stakeholders, and more intelligent automation. +Need improved metrics to measure NAS performance and provide incentives to the stakeholders to encourage more efficient operations. +Improve airline schedule predictability. +Improve airport throughput through better use of available runway capacity.V. Literature search on arrival, departure, surface issues and solutions A thorough literature search was conducted for research related to arrival, departure, and surface operations that might identify arrival, departure, surface issues as well as potential solutions.This search unearthed a broad set of issues in this area.The search spanned a multitude of references too numerous to include in this paper, therefore, an example reference is provided with each broad issue.The key issues identified in the literature search, and their respective consequences, are: Lack of information sharing and inaccurate system (NAS) predictability (Ref.17);  Status and planning information is not shared between users and ANSP's. Unclear system status awareness leads to an inability to accurately predict its future state.The issues exposed by the literature search correlated broadly with those identified in the stakeholder survey. +VI. Proposed Initial IADS SolutionAnalysis of issues identified by the stakeholder and literature surveys indicated that arresting the steady increase in block times and the related economic inefficiencies this causes would be an appropriate long-term goal for a potential IADS solution.Properly developed, implemented, and executed, such a solution would create an environment that would allow scheduled block times to trend towards their ideal unimpeded block times over the long-term.The survey information also indicated that improving the predictability and efficiency of departure operations by adopting time-based metering was the most likely mechanism to affect an improvement in scheduled block times.Three over-arching issues, outlined by the stakeholders, formed the basis for this conclusion.These were: Minimize variability and improve predictability of flight schedules. Enable more efficient routing using existing advanced navigation capabilities. Reduce lost airport throughput due to lack of status information and inadequate planning.To validate this assumption, NASA developed an initial sketch of a concept that addressed this goal and discussed it with stakeholders in a second round of interviews.In addition to assessing whether the initial stakeholder feedback was understood and interpreted correctly, this second-round stakeholder survey also provided the opportunity to obtain information on the perceived benefits of the proposed solution and the characteristics and features that stakeholders perceived as being useful or essential.The proposed solution was an efficient departure scheduler concept that will account for taxi through the top-of-climb phases of a flight.This scheduler would concurrently develop both tactical and strategic solutions with appropriate levels of fidelity and confidence.The strategic scheduler would develop an optimized schedule that balances airport and airspace demand and capacity over a long time horizon (roughly 20-min or more prior to pushback) while the tactical scheduler would adjust the schedule closer to departure (roughly 20-min or less prior to pushback) to account for actual variations in the schedule (delayed flights, unplanned airport constraints, etc.).The goal of the strategic scheduler would be to provide an overall awareness of airport operations by developing the expected departure sequence at the runway and other relevant points on the surface over a long time horizon.Inputs to the strategic scheduler would include flight ready times provided by airlines, arrival information, airport capacity, taxi routes and speeds, surface surveillance, traffic management initiative restrictions, overhead stream traffic and departure fix restrictions for merging and de-confliction, wake vortex separation restrictions, and airport/runway configuration.Outputs would include de-conflicted target take-off times and associated target airport movement area entry times and intermediate times at other appropriate points on the airport surface.The strategic scheduler would facilitate collaborative planning with airline operators to allow schedules to be adjusted when needed.The goal of the tactical scheduler component would be to enable uninterrupted taxi/climb trajectories while maintaining surface/terminal area throughput in all traffic conditions.This scheduler's primary goal would be to reduce flight time, fuel burn, noise, and emissions by using proven optimization methods to facilitate largely unimpeded departure operations from the gate to top-of-climb.The overall scheduling concept would build on relevant existing scheduling technologies and concepts of operation. +VII. Second-Round Stakeholder Survey QuestionsA subset of the stakeholders from the first-round survey, composed primarily of airlines and organizations that provide arrival, departure, surface operations-related services and products to the aviation industry, was approached for the second-round survey.A description of the proposed departure scheduler concept was provided to the stakeholders (with the level of detail provided in the description above) together with the following set of standard questions (as before, the stakeholders were encouraged to expound on any question or expand the discussion beyond these questions, if desired): Would this departure scheduler concept be useful to your organization -why or why not?American Institute of Aeronautics and Astronautics +VIII. Results of the Second-Round SurveyThe key results of second-round stakeholder survey were:  All the stakeholders surveyed unanimously agreed that a departure scheduler would benefit the NAS, for the following reasons:  It would improve schedule predictability that is of paramount importance to airlines. It would reduce taxi-times and fuel use. NASA's "holistic" (integrated) approach was better than the many custom solutions currently available. Moving to a time-based paradigm on the surface is critical to improving IADS (change from the current first-come, first-served paradigm). It promises to improve schedule predictability. The stakeholders indicated that the scheduler should exhibit the following characteristics: Timeliness -provide necessary information reasonably in advance (to facilitate strategic and tactical decision-making). Logical -users can have confidence in the information the system provides and will accept its recommendations. Accuracy -information provided is repeatable and reliable;  Flexibilityshould incorporate user/stakeholder preferences and input;  Robustnessit should adapt reliably (with repeatability) to unexpected events; and  Transparency/equitabilityits operating principles and performance are available to all stakeholders and are clearly understood. The scheduler must work in metroplex environments with multiple airports in relatively close proximity.It should:  Incorporate airspace constraints and allow alternate/efficient departure routes. Be transportable; i.e., it must work at different airports with different airport systems. The stakeholders maintained that both strategic and tactical components of the scheduler were equally importantthe strategic would facilitate long-term planning with less precision and the tactical would give them a preview of near-term adjustments to the schedule. It should provide decision support to ATC on departure scheduling while allowing flexibility when necessary. This concept would be a valuable enhancement to the FAA's terminal and surface management programs.American Institute of Aeronautics and Astronautics The concept's potential to synergistically integrate the FAA's surface and terminal area traffic management system upgrade programs was of particular interest.IX. Conclusions NASA surveyed NAS stakeholders to understand the key issues and potential solutions in IADS operations.The intent was to use this information to formulate a project that would develop and demonstrate a solution that maximized benefit to all the stakeholders.Four stakeholder groups were surveyed -they were: airlines, airports, airnavigation service providers, and industry service/product providers.An initial survey indicated that improving the predictability and efficiency of departure operations by adopting time-based metering would improve the predictability of flight schedules, and decrease costs and fuel use.In a follow-on survey, all the stakeholders supported a departure scheduling concept that could enable unimpeded transit from gate-to-cruise in a busy metroplex airspace environment, and would allow stakeholders to share information and express schedule preferences through a collaborative decision-making process.The stakeholders agreed that this proposed solution would improve flight schedule predictability, facilitate better strategic planning for airlines and airports, and benefit all stakeholders.The stakeholders also outlined the characteristics that the departure scheduler should exhibit in order to benefit IADS operations.The results of this stakeholder survey are being used to formulate the goals for NASA's ATD-2 sub-project that is presently in an advanced planning stage.Figure 1 .1Figure 1.Process for developing project objectives from stakeholder feedback. +Figure 2 .2Figure 2. Event-time communications from a commercial flight. +Figure 3 .3Figure 3. Stakeholders surveyed. +Interfering terminal area trajectories (Ref.18);  Usually in dense metroplex environments with common fixes and shared routes. Poor predictability of arrival/departure demand exacerbates the problem.Some TMIs are local to a particular ATC area and are not reported outside that local area. Manual surface operations (Ref.20);  ATC's first-come, first-served operating paradigm leads to unpredictable behavior.American Institute of Aeronautics and Astronautics  Lack of integration between arrival, surface, and departure operations (Ref.21);  These operations are planned and managed independently. Lack of user/ANSP collaboration (Ref.10);  User priorities and preferences not taken into account in arrival/departure queuing. Lack of coordination of FAA Traffic Management Initiatives (TMIs) (Ref. 19); Many TMIs are not coordinated or shared with users (some flights are affected by more thanone TMI, for example). +What do you see as the most beneficial features of such a scheduler and how would these benefit your organization? What features are essential to a departure scheduler? What is the downside to a scheduler (i.e., what must it not do)? What metrics would your organization use to assess the value of such a scheduler? NASA has a relatively short time-frame (five years) to develop and demonstrate the feasibility of this concept -what are the minimum features should be included in a demonstration that would give you confidence that the new concept will be useful? How would you like to participate in developing and demonstrating this concept? Additional Questions for Vendors of existing schedulers:  What are the inputs to your scheduler? What do you optimize on in your scheduler? What do you output from your scheduler? What would you like to add to improve the optimization of your scheduler? + Downloaded by NASA AMES RESEARCH CENTER on July 2, 2015 | http://arc.aiaa.org| DOI: 10.2514/6.2015-2590 + + + + +AcknowledgementsThe authors greatly appreciate the contributions of the following: Robert Rosen, Matthew Blake, and others at Crown Consulting, Inc., for their support of the stakeholder and literature surveys; Capt.Robert Kotesky of San Jose State University for explaining the intricacies of airline crew operations; and most importantly, the stakeholders who enthusiastically responded to the survey and devoted their time to educate the NASA team on IADS operations and economics. + + + + + + + + + Federal Aviation Administration + + June 2013 + + + NextGen Implementation Plan + Federal Aviation Administration, "NextGen Implementation Plan," June 2013. + + + + + Design and Operational Evaluation of the Traffic Management Advisor at the Fort Worth Air Route Traffic Control Center + + HNSwenson + + + THoang + + + SEngelland + + + + Europe Air Traffic Management Research & Development Seminar + + June, 1997 + Saclay, France + + + Presented at the 1st USA + Swenson, H. N., Hoang, T., Engelland, S., et al, "Design and Operational Evaluation of the Traffic Management Advisor at the Fort Worth Air Route Traffic Control Center," Presented at the 1st USA/Europe Air Traffic Management Research & Development Seminar, Saclay, France, June, 1997. + + + + + AIAA 2012-5611, Presented at the 12th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference and 14th AIAA/ISSM + + RCoppenbarger + + + MHayashi + + + September 2012 + Indianapolis, Indiana + + + The Efficient Descent Advisor: Technology Validation and Transition + Coppenbarger, R., Hayashi, M., et al, "The Efficient Descent Advisor: Technology Validation and Transition," AIAA 2012-5611, Presented at the 12th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference and 14th AIAA/ISSM, September 2012, Indianapolis, Indiana. + + + + + NASA's ATM Technology Demonstration 1 -Moving NextGen Arrival Concepts from the Laboratory to the Operational NAS + + HSwenson + + + JERobinson + + + Iii + + + SWinter + + + + Journal of Air Traffic Control + + 55 + 2 + + 2013 + Summer + + + Swenson, H., Robinson, J. E. III, and Winter, S., "NASA's ATM Technology Demonstration 1 -Moving NextGen Arrival Concepts from the Laboratory to the Operational NAS," Journal of Air Traffic Control, Summer 2013, Volume 55, No. 2, pp. 27-37. + + + + + Chip Off The Old Block + + MCenteno + + + UVitt + + + + Ascend Magazine + + 2 + 2011 + + + Centeno, M., and Vitt, U., "Chip Off The Old Block," Ascend Magazine, 2011 Issue No. 2. + + + + + Air Carrier Flight Operations + + AHMidkiff + + + JRHansman + + + TGReynolds + + + + MIT International Center for Air Transportation + + 2004-3, July 2004 + + + Report + Midkiff, A. H., Hansman, J. R., and Reynolds, T. G., "Air Carrier Flight Operations," MIT International Center for Air Transportation, Report No. 2004-3, July 2004. + + + + + Airline Metric Concepts for Evaluating Air Traffic Service Performance + + + Report of the CNS/ATM Focused Team, Air Traffic Services Performance Focus Group + + Feb. 1999 + + + "Airline Metric Concepts for Evaluating Air Traffic Service Performance," Report of the CNS/ATM Focused Team, Air Traffic Services Performance Focus Group, Feb. 1999. + + + + + + GSkaltsas + + + PBelobaba + + + ACohn + + Presented at the MIT Global Airline Industry Program -IAB/Industry Consortium Joint Meeting + + Nov. 2010 + + + Flight Time Components and Their Delays on Domestic Routes + Skaltsas, G., Belobaba, P., and Cohn, A., "Flight Time Components and Their Delays on Domestic Routes," Presented at the MIT Global Airline Industry Program -IAB/Industry Consortium Joint Meeting, Nov. 2010. + + + + + + SAEngelland + + + ACapps + + + KDay + + + MKistler + + + FGaither + + + GJuro + + NASA/TM-2013-216533 + Precision Departure Release Capability (PDRC) Final Report + + June 2013 + + + Engelland, S.A., Capps, A., Day, K., Kistler, M., Gaither, F., and Juro, G., "Precision Departure Release Capability (PDRC) Final Report," NASA/TM-2013-216533, June 2013. + + + + + FAA Terminal Flight Data Manager (TFDM) Program Office + PMO-02-TFDM-13-001 + + + Federal Aviation Administration + + Nov. 2013 + + + TFDM Core for ATCTs -Concept of Operations + Federal Aviation Administration, "TFDM Core for ATCTs -Concept of Operations," FAA Terminal Flight Data Manager (TFDM) Program Office, ConOps PMO-02-TFDM-13-001, Rev. 2.1, Nov. 2013. + + + + + + System Wide Information Management (SWIM) + + American Institute of Aeronautics and Astronautics + Dec 2014 + + + Federal Aviation Administration + Federal Aviation Administration, "System Wide Information Management (SWIM)," www.faa.gov/nextgen, Dec 2014. American Institute of Aeronautics and Astronautics + + + + + A Functional Analysis of Integrated Arrival, Departure, Surface Operations in NextGen + + MSimons + + + + Presented at the 31st Digital Avionics Systems Conference + Williamsburg, VA + + October, 2012 + + + Simons, M., "A Functional Analysis of Integrated Arrival, Departure, Surface Operations in NextGen," Presented at the 31st Digital Avionics Systems Conference, Williamsburg, VA, October, 2012. + + + + + Estimating the Achievable Benefits of Airport Surface Metering + + DHowell + + + TMcinerney + + AIAA 2012-5653 + + + Presented at the 12th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference and 14th AIAA/ISSM Conference + Indianapolis, Indiana + + September 2012 + + + Howell, D., and McInerney, T., "Estimating the Achievable Benefits of Airport Surface Metering," AIAA 2012-5653, Presented at the 12th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference and 14th AIAA/ISSM Conference, September 2012, Indianapolis, Indiana. + + + + + Airlines pad flight schedules to boost on-time records + + TFrank + + + + USA TODAY + + Feb. 15, 2013 + + + Frank, T., "Airlines pad flight schedules to boost on-time records," USA TODAY, Feb. 15, 2013. + + + + + Why a Six-Hour Flight Now Takes Seven + + SMccartney + + + + Wall Street Journal + + Feb. 4, 2010 + + + McCartney, S., "Why a Six-Hour Flight Now Takes Seven," Wall Street Journal, Feb. 4, 2010. + + + + + Future of the Aviation Industry + + RAnderson + + + + Remarks to the U.S. Chamber of Commerce + Washington, D.C. + + April 12, 2012 + + + Anderson, R., "Future of the Aviation Industry," Remarks to the U.S. Chamber of Commerce, Washington, D.C., April 12, 2012. + + + + + Managing Departure Aircraft Release for Efficient Airport Surface Operations + + WMalik + + + GGupta + + + YCJung + + + + Presented at the AIAA Modeling and Simulation Technologies Conference + Toronto, Canada + + Aug. 2010 + + + Malik, W., Gupta, G., and Jung, Y.C., "Managing Departure Aircraft Release for Efficient Airport Surface Operations," Presented at the AIAA Modeling and Simulation Technologies Conference, Toronto, Canada, Aug. 2010. + + + + + Evaluation of the Integrated Departure Route Planning (IDRP) Tool 2011 Prototype + + RADelaura + + + NKUnderhill + + + LMHall + + + RGRodriguez + + ATC-388 + + May 2012 + + + MIT Lincoln Laboratories + + + Project Report + DeLaura, R. A., Underhill, N. K., Hall, L. M., and Rodriguez, R. G., "Evaluation of the Integrated Departure Route Planning (IDRP) Tool 2011 Prototype, MIT Lincoln Laboratories, Project Report ATC-388, May 2012. + + + + + Airport Collaborative Decision Making -Enhancing Airport Efficiency + + 2012 + + + Airports Council International + Airports Council International, "Airport Collaborative Decision Making -Enhancing Airport Efficiency," 2012. + + + + + Improving Departure Taxi Time Predictions Using ASDE-X Surveillance Data + + ASrivastava + + + + Proceedings of the 30 th IEEE/AIAA Digital Avionics Systems Conference + the 30 th IEEE/AIAA Digital Avionics Systems Conference + + 2011 + + + Srivastava, A., "Improving Departure Taxi Time Predictions Using ASDE-X Surveillance Data," Proceedings of the 30 th IEEE/AIAA Digital Avionics Systems Conference, 2011. + + + + + Automated Integration of Arrival/Departure Schedules + + PDiffenderfer + + + ZTao + + + GPayton + + + + Tenth USA/Europe Air Traffic Management Research and Development Seminar (ATM2013) + + June 2013 + + + Diffenderfer, P., Tao, Z., and Payton, G., "Automated Integration of Arrival/Departure Schedules," Tenth USA/Europe Air Traffic Management Research and Development Seminar (ATM2013), June 2013. + + + + + + diff --git a/file15.txt b/file15.txt new file mode 100644 index 0000000000000000000000000000000000000000..f493af99fb94ea67ea4dc426695bc4cbb525d1b9 --- /dev/null +++ b/file15.txt @@ -0,0 +1,1537 @@ + + + + +Following a description of the present capabilities of the VMS, this paper will summarize the research conducted in each of these streams of research. +Description of the VMSThe Vertical Motion Simulator (VMS) combines a highfidelity simulation capability with an adaptable simulation environment that allows the simulator to be customized to a wide variety of human-in-the-loop research applications.The distinctive feature of the VMS is its unparalleled large amplitude, high-fidelity motion capability.The high level of simulation fidelity is achieved by combining this motion fidelity with excellent visual and cockpit interface fidelities.An interchangeable cab arrangement allows different crew vehicle interfaces and vehicle types to be evaluated allowing fast turnaround times between simulation projects.The VMS motion system is a six degree-of-freedom combined electro-mechanical/electro-hydraulic servo system, shown in Figure 1.It is located in and partially supported by a specially constructed 120 ft tower.The motion platform consists of a 40-ft long beam that travels ±30 ft vertically.On top of the beam is a carriage that traverses the 40-ft length of the beam.A sled sits atop the carriage providing the ±4-ft of travel in a third translational degree of freedom.A conically shaped structure is mounted on the sled and rotates about the vertical axis providing yaw motion.A two-axis gimbal allows pitch and roll motion.The ICAB is attached to the top gimbal ring.The motion capability of the VMS is summarized in the Table 1. +Figure 1. Cutaway Diagram of the VMS Facility Table 1. VMS motion capabilityThe Interchangeable Cab (ICAB) capability in the VMS allows the cockpit to be tailored to the research application.The VMS has five portable ICABs with different out-thewindow (OTW) visual fields-of-view.For each simulation study, the cab is equipped based on the requirements that simulation and tested fixed-base with the complete simulation environment except for motion.Equipping the cab includes installation of flight controls, flight instruments and displays, and seats (Fig. 2).Following equippage and checkout, the ICAB is transported and installed on the motion system.The ICAB capability allows the VMS facility to conduct fixed-base and moving base simulation studies simultaneously. +Figure 2. Exterior and interior of Transport Cab (TCAB) with eight out-the-window visual displaysThe high-fidelity flight controls are heavily modified and optimized McFadden hydraulic force loader systems (Ref.2).A digital control interface allows all aspects of the controller's static and dynamic characteristics to be adjusted.It also allows characteristics to be varied in realtime as necessary for researching pilot cueing concepts using controllers.The force loader is available on a variety of aircraft control manipulators from the regular column and wheel type to conventional rotorcraft controls and side sticks.The out-the-window visual scene is generated by a Rockwell-Collins EPX-5000 image generation (IG) system providing a high-resolution and complex visual environment at fast frame rates (Fig. 3).Presently, capability is for nine channels of IG that will be expanded to 15 this year.This IG capability allows the VMS to operate up to three different simulation experiments at the same time.In-house graphics expertise at the VMS customizes the visual databases to particular simulation requirements.Separate graphics processors generate the displays required for primary flight displays, head-up displays, sensor imagery, etc.These displays may also be customized to the researcher's needs. +Figure 3. Typical outside visual scenesAll the essential elements of the simulation are linked with the host environment through a dedicated network and controlled through a fully equipped control room.The flexible simulation architecture makes it convenient to interface and to evaluate custom software and hardware modules on the VMS.This capability may be used to evaluate sensors, vehicle dynamic models, flight control systems, etc. +Handling Qualities StudiesIn the 1970's, the Army needed a handling qualities specification that could guide the development of rotorcraft that would meet the more demanding missions and tactics envisioned in the future.The military specification for rotorcraft handling qualities, MIL-H-8501A, was written in 1952 and was inadequate (Ref.3).Several attempts to update MIL-H-8501A met with little success and were not adopted, primarily due to a lack of background data of adequate quality.To overcome the deficiencies in MIL-H-8501A, the Army used Prime Item Development Specifications (PIDS) for procuring new helicopters (Ref.The strategy for developing a helicopter handling qualities database of sufficient quality and validity for use in a military specification was to combine high-fidelity simulation with a limited amount of flight test activity.Almost all the simulation data incorporated into ADS-33 came from VMS studies.The process of developing the database is ongoing, and several simulation studies in 2008 will fill known gaps in the database and refine others.Table 2 lists the VMS simulation studies that have supported this process.Reference 1 contains a comprehensive summary of the rotorcraft research conducted in the VMS in its first decade of operation.Initial studies focused on two issues of primary concern: 1) rotorcraft design requirements for lowaltitude or nap-of-the-earth (NOE) flight, and 2) flight in adverse visual conditions.Not all these studies were documented or published, but their data were used to develop criteria in ADS-33 and are available in the Background Information and Users Guide to ADS-33 (Ref. +4).Initial studies investigated the effect of design variations on rotorcraft dynamic characteristics and handling qualities.These included the effect of vertical damping and thrust available (Ref.5), pitch/roll and collective/yaw cross coupling (Refs.6 and 7), yaw control (Refs.8 and9), and roll control (Ref.10).Other studies investigated some of the fundamental precepts under development for the specification -required response type and bandwidth.An innovative concept introduced in ADS-33 is the trade-off between augmentation (which defines response type) and the visual cueing environment.Ground-based and in-flight simulation studies established that when the visual cueing environment is degraded, increased compensation is required to maintain handling qualities at a satisfactory level.Bandwidth is a measure that defines the quickness in the response to control inputs and represented a shift, at the time, from time-domain based performance criteria to a more accurate and discriminating frequency-domain measure.Studies on the VMS that developed these concepts include those reported in Refs.11, 12, and 13.These studies indicated that the mission management and demanding NOE flight environment placed an unacceptable workload on a single pilot that may be alleviated by increasing augmentation through the flight control system.An important outcome of these studies was the progressive development of evaluation tasks that were designed to be representative of the mission tasks expected of rotorcraft but also constrained to allow repeatability and promote consistency in handling qualities ratings.These evaluation tasks, or Mission Task Elements (MTEs), were refined over the course of many experiments on the VMS and included in ADS-33 as demonstration maneuvers for evaluating the overall handling qualities of a rotorcraft.Task performance displays for the MTEs were also developed over the course of many experiments on the VMS and these displays have transitioned to flight testing.Finally, a series of eight simulation studies investigated the effect of externally slung loads on helicopter handling qualities.The results of the first five simulation studies by the AFDD are reported in Ref. 19.These simulation studies established a database of handling qualities data for future inclusion in ADS-33 as guidelines for the design of rotorcraft that carry external loads.The database on handling qualities is being further expanded by follow-on simulation studies on the VMS that will explore heavy cargo rotorcraft operations including the effect of pitch/roll response-type on handling qualities with slung loads.Since its release, ADS-33 has guided the procurement of the CH-47F and CH-53K helicopters, and the UH-60Mu and AH-64D fly-by-wire upgrades by the Army and Navy.It was designed for and used as the guiding specification for the RAH-66 Comanche flight control system that is the basis for the full-authority fly-by-wire flight control systems implemented on the UH-60Mu, CH-53K, H-92 Superhawk, and the digital automatic flight control system (DAFCS) on the CH-47F.The CH-47F flight control system was evaluated on the VMS.The RAH-66 control system design was, in turn, based on extensive research on the Army's advanced digital optical control system (ADOCS) program that also used the VMS for initial design and development.The difficulty of landing rotorcraft in blowing sand, known as brownout, became evident in the first Gulf War. +Flight Control Design StudiesConducting operations under cover of darkness using night vision goggles (NVG) was also becoming more prevalent as the quality of the devices improved.At that time, the knowledgebase on rotorcraft handling qualities included significant evidence that control augmentation could alleviate handling qualities deficiencies in degraded visual environments such as brownout and NVGs.A series of simulation studies on the VMS investigated the possibility of modifying the existing limited authority flight control system on the UH-60 Black Hawk to provide added augmentation to improve handling qualities when flying in degraded visual environments (Refs.22 -24).The simulation studies explored different methods for adding this increased augmentation using the existing flight control system mechanization and evaluated the effect of saturating the stability augmentation system for larger pilot inputs.Based on the favorable results of these studies, increased augmentation was proposed for the Black Hawk and tested on the VMS (Ref.25).The CH-47F with its digital automatic flight control system (DAFCS) was the first production Army helicopter to be acquired using ADS-33 requirements.The DAFCS included advanced response types to improve handing qualities in degraded visual environments.Following development and initial testing by the manufacturer, they were refined and evaluated by Army pilots in an AFDD conducted simulation study on the VMS (Ref.26).The simulation investigations showed that the DAFCS significantly improved handling qualities in day and night (using NVGs) visual conditions when compared with a CH-47D.Transitions between response type modes in the DAFCS were also improved based on the findings of VMS simulation study. +Studies on Specific Rotorcraft ProgramsThe high-fidelity simulation capability of the VMS has been used to evaluate and test new rotorcraft concepts and configurations such as tilt-wing and tilt-rotor as well as to support existing programs such as Comanche and Apache by implementing and evaluating simulation models (Ref.27).Table 4 lists the simulation studies that fall in this category.Army's Aeroflightdynamics Directorate.A particular concern during development was controllability of the helicopter if there was a failure in the research flight control system that operated in parallel with the production flight control system.Several studies investigated and quantified the effect of failure transients and assessed the handling qualities following a failure (Ref.34) and evaluated RASCAL safety systems.A pilot failure rating scale was developed to assess the safety and handling qualities requirements associated with dynamic failures that could occur on a full-authority flight control system as envisioned for RASCAL (Ref.35).Recent studies on the VMS have focused on modifying the RASCAL safety systems to allow flight testing at lower altitudes. +Simulation Fidelity StudiesA host of experiments aimed at understanding simulation fidelity requirements for helicopters have been performed on the VMS since its inception including those listed in Table 5.Since the VMS represented the most capable simulator in the world, the first natural question to ask was how good is it compared to flight.While answering critical questions necessary for advancing simulation-based flight training, the results of these studies were also fundamental to the validity of the studies conducted in the other streams of research outlined in this paper.The first attempt to compare how well the VMS could simulate helicopter flight was conducted cooperatively among several organizations (Ref.36).Pilot opinion (HQRs) were Level 1 in flight but Level 2 in simulation.These differences led to investigations to improve the rotor model, servo dynamics, and engine.Improvements to the math model, visual systems, motion configurations, and experimental tasks and protocols led to experiments documented in Ref. 37.This experiment compared performance and pilot opinion using a UH-60 operated out of NASA Ames.Extensive frequency-domain identification compared the simulation performance with flight, with the conclusion that the model was a reasonable representation of the flight vehicle.This was an improvement from the 1984 simulation study but deficiencies in visual and heave motion fidelity were noted.When simulating motion in a ground-based simulator where motion displacement is limited, there is a necessary tradeoff between the desired initial, or short-term acceleration and the desired sustaining, or long-term, acceleration.Mitchell (Ref. 43) showed that the addition of motion improved pilot opinion ranging from ½ to 2 HQR points.For precision tasks, sustained acceleration cues were preferred (reduced washout natural frequencies of the motion filter), while, for aggressive tasks, short-term acceleration cues were preferred.Schroeder (Ref.44) used pilot describing function measurements to examine a variety of motion gains and motion washout filter variations on a classical single-axis compensatory tracking task.The results showed that motion cues allowed the pilot to generate lead compensation and improve target tracking phase margins with increasing filter gain or decreased natural frequency.The study also showed no effects for any of the pure yaw motion configurations, which led to subsequent investigations.To help answer the question on what characteristics a motion filter should have so that simulation is a reasonable representation of flight, Schroeder (Ref.45) evaluated the proposed Sinacori motion fidelity criteria.Objective and subjective results showed that the original criteria could be relaxed, and these criteria are still used today as a guide when configuring the VMS motion system.A question that often arises when configuring the motion system is the level of fidelity required in each axis.Specifically, the earlier study that showed little effect of yaw cueing led to a more detailed study (Ref. 46).This study evaluated a helicopter in a single degree of freedom hovering yaw task.Four variations in the motion cueing were studied: full motion, only lateral translational motion, only yaw rotational motion, and no motion.The study found that the lateral acceleration cue was of predominant importance in both performance and opinion.This suggested that if you had strong lateral translational cues, then the yaw rotational cues might be redundant and unnecessary.In a study that considered visual cueing aspects as well as motion, Johnson (Ref. 47) investigated how the displayed level-of-detail changes as one gets closer or further away from an object for a height control task.The results showed that changing the level-of-detail to maintain constant optical density as the altitude changed, like that of the real world, improved altitude awareness.Texture at this time was only beginning to be used in the VMS.Separately, adding platform motion improved speed regulation and altitude perception.Further systematic changes in visual scene, via changing spatial frequency with alternating black-and-white stripes, and motion cues were evaluated in the vertical axis (Ref.48).The variations in visual scene evaluated had no effect, while the motion configurations did have an effect.These configurations were subsequently analyzed using a structural pilot model (Ref.49).The intent was to develop and calibrate a model that would predict pilot opinion for a given rotorcraft and task, and the model's predictions correlated well, in a ranking sense, with the subjective ratings.The VMS was used to evaluate cueing effects on autorotation in two studies separated by more than a decade (Refs.50 and 51).In the first study, autorotation task performance decreased with degraded motion cues, yet acceptable performance could be attained as long as there were adequate visual cues.In the second study with improved visual cueing technology, the effect of visual texture and motion variations on autorotation task performance was evaluated.Visual texture affected all measures, but the finest texture did not perform the best, debunking a myth that more texture is better.This result was also supported by a fixed-base psychophysics study (Ref.52).Visual detail affected only pilot subjective opinion.Pilot performance as well as opinion of motion fidelity improved with increased motion cueing. +Guidance and Display StudiesThe high-level of motion and cockpit fidelity offered by the VMS made it an ideal platform to assess and prototype rotorcraft guidance and display concepts (Table 6).Early studies on the VMS evaluated the rotorcraft handling characteristics necessary for low-level nap-of-the-earth missions envisioned for the next generation of military rotorcraft.These studies showed that the workload for such missions was unacceptable for a single pilot.A series of studies over several years beginning in 1988 conducted research to develop an automated helicopter flight capability for potential application in the U.S. Army light, attack, scout helicopter (LHX) program (Fig. 4).Several simulation studies on the VMS developed and refined the critical components of a guidance system for terrain and obstacle detection, path prediction, guidance displays and symbology, and automatic control concepts for terrain following/terrain avoidance (TF/TA) (Refs.53, 54, and 55).The resulting guidance system was implemented on the Army's UH-60 System Testbed for Avionics Research (STAR) and tested in flight.The outcome of the project included several fundamental concepts for displaying flight information superimposed on sensor imagery.Another series of studies by the AFDD investigated methods for predicting helicopter maneuver limits and communicating this information to the pilot via tactile cueing through the pilot inceptor (Ref.56).The results showed that tactile cueing through a conventional inceptor or a sidestick enabled the pilots to easily track helicopter structural and power-train limits and avoid exceedances while maneuvering aggressively without referring to flight instruments.The cueing allowed the pilots to extract maximum maneuvering performance from the helicopter without risk of damage.This technology was developed for evaluation on an AH-64 Apache helicopter under the Army's Helicopter Active Control Technology program (Ref.57).The increased operation of helicopters in urban areas brought with it the risk of an accident, particularly in an emergency situation such as an engine failure.Updated procedures for recovering from an engine failure when taking off or landing in confined areas were needed.Two simulation studies on the VMS by NASA and the FAA investigated methods to reduce pilot workload and increase safety for rotorcraft terminal area operations (Ref.58).The studies examined the benefits of optimal procedures, cockpit displays, and alternate cueing methods for safe terminal area procedures with one-engine inoperative. +Table 6. Research into guidance and display concepts +SimulationResults showed that an integrated display reduced pilot workload and improved safety when compared with conventional instruments.Pilot opinion showed a preference for the MIL-STD-1295 symbology.Specific improvements to the Contact Analog symbology were recommended.Another AFDD study, reported in Ref. 61, examined ways to optimize the alerting effectiveness of helmet display symbology.The research investigated two approaches to increasing the effectiveness of alerts -using the entire display surface and including information about the required response in the alert itself.Helmet display symbology was based on the AH-64's pilot night vision system (PNVS), cruise mode symbology.The data showed a small benefit from both the full-screen alert and the partial information alert.The continued development and flight-testing of the military V-22 Osprey prepared the way for the introduction of a civil tilt-rotor transport.The potential of a civil tilt-rotor transport into the National Airspace System presented challenges and opportunities for vertical flight solutions to airspace congestion.The Federal Aviation Administration (FAA) had developed a Vertiport design guide based on helicopter capabilities and projected civil tilt-rotor transport performance.As a complement to the FAA guide, NASA undertook an effort to develop controls and display technology to fully utilize the capability of this rotorcraft.This effort took the form of a series of ten piloted simulation studies conducted at the VMS (from 1988 to 2001) to investigate tilt-rotor terminal operations and certification issues (Fig. 5).The general objectives of these simulation studies were: 1) to develop design guidance for safe, all-weather, low-noise flight operations, 2) to develop controls and cockpit displays to support tiltrotor transport operations, and 3) to develop tilt-rotor transport terminal area procedures.All these objectives were met.Initial studies evaluated steep instrument (IFR) approaches to confined spaces.These steep approaches would reduce the required obstruction-free approach zone and could significantly reduce the noise footprint of terminal operations.Two display concepts were investigated to provide guidance for steep IFR approaches (Refs.62 and 63).Glide slopes ranging from the nominal 3-degrees up to 25-degrees were investigated. +Tilt-Rotor StudiesThe next series of experiments further evaluated these steep IFR approaches under One-Engine-Inoperative (OEI) conditions.This was followed by handling qualities evaluations of noise abatement landing approaches and comparing them with acoustic measurements from flight tests using the XV-15 aircraft flying similar trajectories (Ref.64).A potential two-segment approach with initial deceleration at a threedegree glide slope converting to final approach along a nine-degree glide slope was also investigated.The final experiments were full mission simulation studies that evaluated operation in congested airspace and led to the development and use of pursuit displays, particularly for the transition from the airplane type cruise configuration to the helicopter configuration for final approach and landing (Refs.65 and 66).In contrast, a study for the Marine Corps, NASA evaluated the effectiveness of the V-22 Osprey tilt-rotor in one-onone air combat maneuvering on the VMS (Ref.67).The study showed that the unique speed and maneuvering characteristics of the V-22 enhanced its survivability against both fixed-wing and rotorcraft aggressors. +Concluding RemarksOver the past three decades, the NASA Vertical Motion Simulator has provided a wealth of data and knowledge to further rotorcraft technology and safety.The collaboration between NASA and the Army Aeroflightdynamics Directorate at the Ames Research Center has been, and continues to be, a primary reason for the prolific output of valuable research from this facility.This collaboration has developed a database of knowledge on a variety of interacting disciplines on rotorcraft including handling qualities, simulation fidelity, guidance and displays, and design.This database is now used in military procurement and in civil applications on training simulation and guidance displays.NASA programs as well as those in collaboration with the FAA have also contributed significantly to the development of the tilt-rotor aircraft and its civil derivative.Specific contributions include:• Data on rotorcraft handling qualities that formed a basis for the current military specification on rotorcraft handling qualities, ADS-33E;• Understanding on human motion and visual cueing, and developing guidelines for configuring simulation cueing environments;• Designing and evaluating novel rotorcraft configurations including the tilt-wing and tilt-rotor• Designing and evaluating production flight control systems including the CH-47F;• Designing and evaluating rotorcraft guidance and display concepts for low-level terrain following and helmet-mounted displays; and• Civil tilt rotor operation and certification.3).The U. S. Army Aeroflightdynamics Directorate (AFDD), in collaboration with NASA Ames Research Center, began an effort to build a database of handling qualities data and design criteria that could be incorporated into a new handling qualities specification.In 1982, the AFDD began the development of a new rotorcraft handling qualities specification to supersede MIL-H-8501A (Ref.3).The specification, US Army Aeronautical Design Standard -33 or ADS-33 (Ref.4), was completed and published in 1987.Initial development of the specification was for the procurement of the modern light/attack/scout helicopter (LHX) the Army intended to acquire.This program led to the development of a prototype helicopter, the RAH-66 Comanche. +Figure 4 .4Figure 4. Automated NOE cockpit and test courseIn the area of head-mounted displays, a study on the VMS investigated the handling qualities benefits that could be attained using new display law design methods for hover displays (Ref.59).The display law design was applied to the Apache helmet-mounted display format, using the Apache vehicle dynamics to tailor the dynamics of the velocity predictor symbol.The new symbol dynamics improved the pilots' ability to maneuver about hover in poor visual cuing environments and improved pilot opinion.More recently, the AFDD conducted a simulation study to examine the performance of the Comanche Contact Analog world-referenced symbology displayed on the Comanche's helmet-mounted display when compared with a compressed symbology design similar to that specified by the former MIL-STD-1295 (Ref.60).Pilot opinion showed a preference for the MIL-STD-1295 symbology.Specific improvements to the Contact Analog symbology were recommended.Another AFDD study, reported in Ref.61, examined ways to optimize the alerting effectiveness of helmet display symbology.The research investigated two approaches to increasing the effectiveness of alerts -using the entire display surface and including information about the required response in the alert itself.Helmet display symbology was based on the AH-64's pilot night vision system (PNVS), cruise mode symbology.The data showed a small benefit from both the full-screen alert and the partial information alert. +In 1981 the Deputy Secretary of Defense directed that the services reviewed V/STOL technology with the intent of establishing a joint rotary wing aircraft development program to satisfy service lift requirements for medium lift V/STOL aircraft in the 1990s and beyond.This would take advantage of the advanced, but mature, tilt-rotor technology already in place by the early 1970s.The joint NASA/Army XV-15 Tilt-Rotor Research Aircraft program had already begun in the early 1970's and became the foundation for the full-scale development of the JVX, later designated as the V-22 Osprey.A number of simulation studies conducted at the VMS from 1980 to 1985 validated the JVX math model and evaluated the flight control system characteristics.Previous studies on the FSAA had led to the evaluation and selection of Bell to build the XV-15 prototype. +Figure 5 .5Figure 5. Civil tilt-rotor cockpit (with HUD) and typical visual scene +Table 2 . Rotorcraft handling qualities simulation studies Simulation Description Year(s)2Heavy-Lift Rotorcraft (HLR) stability margin/handling qualities2008Handling qualities with external slung loads (8 simulations)1994 -2008Civil handling qualities specification2001LHX/RAH 66 handling qualities (4 simulations)1982 -1992Helicopter cross-coupling studies (3 simulations)1986 -1994Handling qualities for helicopter air combat (6 simulations)1984 -1992Handling qualities for yaw control1984Handling qualities for roll control (2 simulations)1984, 1985Single/dual pilot advanced cockpit and handling qualities (2 simulations)1986Handling qualities for shipboard landing1984Handling qualities for vertical control1983-1984Effect of rotorcraft design on handling qualities (3 simulations)1980-1982Several studies by the AFDD also investigated helicopterand flight control system design requirements foraggressive air-to-air combat (Refs. 14, 15, and 16) at low-altitude. These studies provided data on the effect ofresponse type, pitch/roll/yaw bandwidth, and turncoordination on handling qualities when the rotorcraft itselfis used as a pointing platform. Another study investigatedthe yaw dynamic requirements for a rotorcraft with aturreted gun engaged in low-level air combat operations(Ref. 17). The study investigated the potential trade-offbetween gun slewing angular movement capability andrequired rotorcraft yaw axis response. A separate series ofenvisioned, at that time, for the LHX program (Ref.18). +Table 3321)luding multi-axis side sticks and advanced pilot displays(Ref.20).Outcomes of these studies were recommendations for the design of multiaxis pilot controllers, flight control system designs with advanced response types; control law mode switching logic; automatic stick force trimming; and helmet mounted display.These concepts were implemented and flighttested on the ADOCS UH-60 helicopter.The designs translated well from simulation to flight and only minor parametric changes were required in flight to optimize handling qualities (Ref.21).The program provided invaluable data on advanced flight control system design.The fundamental flight control concepts and architectures developed in the program have influenced all the rotorcraft fly-by-wire flight control systems designed since that time.lists the VMS simulation studies that evaluated flight control system concepts and designs.All these development efforts were closely linked with the development of a handling qualities knowledgebase.To evaluate new flight control concepts in preparation for the procurement of a new attack helicopter (LHX), the Army initiated the ADOCS program to investigate modern flight control system concepts and pilot-vehicle interfaces.A series of simulation studies on the VMS evaluated modern control laws, flight control system architectures, and cockpit interface concepts +Table 3 . Flight control systems research simulations3Simulation DescriptionYear(s)CH-47F digital automatic flight control system (2 simulations)2004 -2005UH-60 MCLAWS evaluation2002Partial authority flight control systems (4 simulations)1991 -1998Advanced Digital Optical Control System(ADOCS) program1981 -1985(4 simulations) +Table 4 . Simulation studies supporting specific rotorcraft programs433)er programs include several unique rotorcraft configurations including a tilt-wing, X-wing, and the Piasecki Vectored Thrust Combat Agility Design (VTCAD).The tilt-wing studies ranged over a period of 10 years and investigated novel control methods such as the geared flap, that were necessary to control the aircraft during wing tilt (Ref.30), and the effect of augmentation on handling qualities during approach and landing (Ref.31).The VTCAD concept involved the addition of a ducted fan with thrust vectoring capability in lieu of a tail rotor on an AH-64 helicopter to increase its speed and agility.The AFDD evaluated the efficacy of this design in two VMS investigations.The X-wing program was a joint Army/NASA project to investigate high-speed rotorcraft that culminated in the development of the prototype Rotor Systems Research Aircraft (RSRA) to test rotor and propulsion concepts.The complex fly-by-wire flight control system for the X-wing required control transitions as the aircraft transitioned from a rotorcraft to a fixed-wing aircraft (Ref.32).These control laws were refined and on the VMS.A joint NASA/Sikorsky study on the VMS compared Sikorsky's Variable Diameter Tilt Rotor (VDTR) concept with a conventional fixed diameter tilt-rotor (Ref.33).Pilots from government and industry evaluated the VDTR in regular and emergency (engine failure) operations.The study demonstrated the enhanced performance potential of the VDTR and identified areas for further study.Simulation DescriptionYear(s)RASCAL safety systems (3 simulations)1989 -2008Joint Shipboard Helicopter Integration Process (JSHIP) (2 simulations)2000 -2001RAH-66 Comanche1997Tilt-Wing/Advanced Theater Transport (ATT) (4 simulations)1991 -2002Variable Diameter Tilt Rotor (VDTR)1996AH-64 Apache (4 simulations)1988 -1993Piasecki VTCAD (2 simulations)1991 -1993X-Wing/RSRA (4 simulations)1984 -1987SH-2F1982Two simulation studies were conducted under the DynamicInterface Modeling and Simulation System as part of theJoint Shipboard Helicopter Integration Process (JSHIP),sponsored by the Office of the Secretary of Defense, toinvestigate the simulation fidelity required to accuratelyrecreate the shipboard-landing task (Ref. 28). Considerableeffort was placed on achieving the visual and dynamicfidelity required to ensure that the pilot workload wasconsistent with that experienced in the actual task. Themodeling and simulation of the ship airwake and itsinteraction with the rotorcraft was a particular technical29allenge.This study was preceded almost two decades earlier by a similar one that evaluated whether high-fidelity simulation could be used for assessing the shipboard landing environment using the VMS and a model of a SH-2F helicopter (Ref.29). +Table 5 . Simulation fidelity research5Simulation DescriptionYear(s)Visual scene height perception -PsychoPath (2 simulations)2001 -2006Autorotation Cues -AutoCue (2 simulations)2000 -2001Computation situational awareness model --SAMSIM2000Simulation fidelity requirements (7 simulations)1996 -1999Motion and visual evaluation -MOTIVE1993Simulation validation -SIMVAL (3 simulations)1990 -1993Visual/motion synchronization -VISMOSYNC (2 simulations)1990 -1992Visual and motion delay -SIMVAC1992Blackhawk validation1989Simulator sickness study1989Helicopter autorotation (2 simulations)1984 -1985Early research on the VMS showed how far away thesimulated cues in a typical helicopter are from the realworld (Ref. 38) and exposed the need for systematicinvestigations with objective measures to determine fidelityrequirements. Early emphasis was on the relative timingbetween the visual and motion cues that was a suspectedcause of simulator sickness. Using four different motionconditions, McCauley (Ref. 39) found that simulatorsickness increased with both time and the level ofmaneuvering. Using a UH-60 model, Sharkey (Ref. 40) + + + + +AcknowledgementsThe authors would like to thank the engineers at the U.S. Army Aeroflightdynamics and NASA Ames Research Center who reviewed this paper and provided valuable comments and insight. + + + + + + + + + Rotorcraft Handling-Qualities Design Criteria Development + + EWAiken + + + VJLebacqz + + + RTChen + + + DLKey + + SEE N88-16632 09- 01 + + + Materials and Structures, Propulsion and Drive Systems, Flight Dynamics and Control, and Acoustics + + 2 + + + NASA/Army Rotorcraft Technology + Aiken, E. W., Lebacqz, V. J., Chen, R. T., and Key, D. L., "Rotorcraft Handling-Qualities Design Criteria Development," NASA/Army Rotorcraft Technology. Volume 2: Materials and Structures, Propulsion and Drive Systems, Flight Dynamics and Control, and Acoustics, SEE N88-16632 09- 01; + + + + + + + WashingtonNasa + + + 1988 + + + + NASA, Washington, 1988, p 948-998. + + + + + Pilot Force Measurement with Inertia and Gravity Compensation + + RAMueller + + + GHHardy + + + + AIAA Paper 2007-6563, AIAA Modeling and Simulation Technologies + + + Mueller, R. A., and Hardy, G. H., "Pilot Force Measurement with Inertia and Gravity Compensation," AIAA Paper 2007-6563, AIAA Modeling and Simulation Technologies + + + + + + + Conference + + + Aug. 2007 + Hilton Head, SC + + + Conference, Hilton Head, SC, Aug. 2007. + + + + + Handling Qualities Specifications for U.S. Military Helicopters + + DLKey + + + + AIAA Journal of Aircraft + + 19 + 2 + Feb. 1982 + + + Key, D. L., "Handling Qualities Specifications for U.S. Military Helicopters," AIAA Journal of Aircraft, Vol. 19, no. 2, Feb. 1982. + + + + + Handling Qualities Requirements for Military Rotorcraft + + Anon + + + + Aeronautical Design Standard-33 (ADS-33E-PRF), US Army Aviation and Missile Command + + Mar. 2000 + + + Anon., "Handling Qualities Requirements for Military Rotorcraft," Aeronautical Design Standard-33 (ADS-33E-PRF), US Army Aviation and Missile Command, Mar. 2000. + + + + + Effects of Rotor Inertia and RPM Control on Helicopter Handling Qualities + + LDCorliss + + + CLBlanken + + + KNelson + + + + AIAA Paper 1983-2070, AIAA Atmospheric Flight Mechanics Conference + Gatlinburg, TN + + Aug. 1983 + + + Corliss, L. D., Blanken, C. L., and Nelson, K., "Effects of Rotor Inertia and RPM Control on Helicopter Handling Qualities," AIAA Paper 1983-2070, AIAA Atmospheric Flight Mechanics Conference, Gatlinburg, TN, Aug. 1983. + + + + + An Investigation of the Effects of Pitch-Roll Cross Coupling on Helicopter Handling Qualities for Terrain Flight + + DCWatson + + + EWAiken + + + + AIAA Conference on Guidance, Navigation, and Control + Monterey, CA + + Aug. 1987 + + + Watson, D. C., and Aiken, E. W., "An Investigation of the Effects of Pitch-Roll Cross Coupling on Helicopter Handling Qualities for Terrain Flight," AIAA Conference on Guidance, Navigation, and Control, Monterey, CA., Aug. 1987. + + + + + An Investigation of the Effects of Pitch-Roll (De)coupling on Helicopter Handling Qualities + + CLBlanken + + + HJPausder + + + CJOckier + + NASA-TM-110349 + + May 1995 + + + Blanken, C. L., Pausder, H. J., and Ockier, C. J., "An Investigation of the Effects of Pitch-Roll (De)coupling on Helicopter Handling Qualities," NASA-TM-110349, May 1995. + + + + + A Simulation Investigation of Scout/Attack Helicopter Directional Control Requirements for Hover and Low-Speed Tasks + + CCBivens + + + JGGuercio + + + Mar. 1987 + + + NASA-TM-86755 + Bivens, C. C., and Guercio, J. G., "A Simulation Investigation of Scout/Attack Helicopter Directional Control Requirements for Hover and Low-Speed Tasks," NASA-TM-86755, Mar. 1987. + + + + + A Piloted Simulation Investigation of Yaw Attitude Quickness in Hover and Yaw Bandwidth in Forward Flight + + MSWhalley + + + + NASA TM-103948 + + Sep. 1992 + + + Whalley, M. S., "A Piloted Simulation Investigation of Yaw Attitude Quickness in Hover and Yaw Bandwidth in Forward Flight," NASA TM-103948, Sep. 1992. + + + + + + RKHeffley + + + SMBourne + + + HCCurtiss + + + Jr + + + WSHindson + + + RAHess + + NASA-CR-177404 + Study of Helicopter Roll Control Effectiveness Criteria; Final Report + + Apr. 1986 + + + Heffley, R. K., Bourne, S. M., Curtiss, H. C., Jr., Hindson, W. S., and Hess, R. A., "Study of Helicopter Roll Control Effectiveness Criteria; Final Report," NASA-CR-177404, Apr. 1986. + + + + + Helicopter Control Response Types for Hover and Low-Speed Near-Earth Tasks in Degraded Visual Conditions + + CLBlanken + + + DCHart + + + RHHoh + + + + AHS 47th Annual Forum + Phoenix, AZ + + May 1991 + + + Blanken, C. L., Hart, D. C., and Hoh, R. H., "Helicopter Control Response Types for Hover and Low-Speed Near-Earth Tasks in Degraded Visual Conditions," AHS 47th Annual Forum, Phoenix, AZ, May 1991. + + + + + An Investigation of the Use of Bandwidth Criteria for Rotorcraft Handling Qualities Specifications + + CLBlanken + + + CCBivens + + + MSWhalley + + + + AHS International Conference on Rotorcraft Basic Research + Research Triangle Park, NC + + Feb. 1985 + + + Blanken, C. L., Bivens, C. C., and Whalley, M. S., " An Investigation of the Use of Bandwidth Criteria for Rotorcraft Handling Qualities Specifications," AHS International Conference on Rotorcraft Basic Research, Research Triangle Park, NC, Feb. 1985. + + + + + Investigation of the Effects of Bandwidth and Time Delay on Helicopter Roll-Axis Handling Qualities + + Heinz-JuergenPausder + + + CLBlanken + + + + Piloting Vertical Flight Aircraft: A Conference on Flying Qualities and Human Factors (SEE N94-13294 02-08) + + Jul.1993 + + + + Pausder, Heinz-Juergen, and Blanken, C. L., "Investigation of the Effects of Bandwidth and Time Delay on Helicopter Roll-Axis Handling Qualities," NASA Ames Research Center, Piloting Vertical Flight Aircraft: A Conference on Flying Qualities and Human Factors (SEE N94-13294 02- 08), Jul.1993, p 91-110. + + + + + Piloted Simulation of One-on-One Helicopter Air Combat at NOE Flight Levels + + MSLewis + + + EWAiken + + NASA-TM-86686 + + Apr. 1985 + + + Lewis, M. S., and Aiken, E. W., "Piloted Simulation of One-on-One Helicopter Air Combat at NOE Flight Levels," NASA-TM-86686, Apr. 1985. + + + + + A Piloted Simulation of Helicopter Air Combat to Investigate Effects of Variations in Selected Performance and Control Response Characteristics + + MSLewis + + + MHMansur + + + RTChen + + USAAVSCOM TM-87-A-3 + + Aug. 1987 + + + Lewis, M. S., Mansur, M. H., and Chen, R. T., "A Piloted Simulation of Helicopter Air Combat to Investigate Effects of Variations in Selected Performance and Control Response Characteristics," USAAVSCOM TM-87-A-3, Aug. 1987. + + + + + A Piloted Simulation Investigation of Forward Flight Handling Qualities Requirements for Helicopter Air-to-Air Combat + + MSWhalley + + + WRCarpenter + + + + NASA TM-103919 + + May 1992 + + + Whalley, M. S., and Carpenter, W. R., "A Piloted Simulation Investigation of Forward Flight Handling Qualities Requirements for Helicopter Air-to-Air Combat," NASA TM-103919, May 1992. + + + + + A Piloted Simulation Investigation of Yaw Dynamics Requirements for Turreted Gun Use in Low-Level Helicopter Air Combat + + WMDecker + + + PMMorris + + + JNWilliams + + + + 44th AHS Annual Forum + + June 1988 + Washington, DC + + + Decker, W. M., Morris, P. M., and Williams, J. N., "A Piloted Simulation Investigation of Yaw Dynamics Requirements for Turreted Gun Use in Low-Level Helicopter Air Combat," 44th AHS Annual Forum, Washington, DC, June 1988. + + + + + Advanced Helicopter Cockpit And Control Configurations for Helicopter Combat Missions + + LAHaworth + + + AAtencio + + + Jr + + + CBivens + + + RShively + + + DDelgado + + + Dec. 1987 + + + NASA-TM-100017 + Haworth, L. A., Atencio, A., Jr., Bivens, C., Shively, R., and Delgado, D., "Advanced Helicopter Cockpit And Control Configurations for Helicopter Combat Missions," NASA-TM- 100017, Dec. 1987. + + + + + Development of Handling Qualities Criteria for Rotorcraft with Externally Slung Loads + + RogerHHoh + + + RobertKHeffley + + + DavidGMitchell + + + + NASA CR + + 2006-213488, Oct. 2006 + + + Hoh, Roger H.; Heffley, Robert K.; and Mitchell, David G., "Development of Handling Qualities Criteria for Rotorcraft with Externally Slung Loads," NASA CR-2006-213488, Oct. 2006. + + + + + Development of ADOCS Controllers and Control laws + + KHLandis + + + SIGlusman + + + + NASA-CR-177339 + + Mar. 1985 + 1 + + + Executive Summary + Landis, K. H., and Glusman, S. I., "Development of ADOCS Controllers and Control laws, Volume 1: Executive Summary," NASA-CR-177339, Mar. 1985. + + + + + Evaluation of ADOCS Demonstrator Handling Qualities; Advanced Digital Optical Control System + + SIGlusman + + + CDabundo + + + KHLandis + + + + AHS, 43rd Annual Forum + Saint Louis, MO + + May 1987 + + + Glusman, S. I., Dabundo, C., and Landis, K. H., "Evaluation of ADOCS Demonstrator Handling Qualities; Advanced Digital Optical Control System," AHS, 43rd Annual Forum, Saint Louis, MO, May 1987. + + + + + Optimizations of Partial Authority Automatic Flight Control Systems for Hover/Low-Speed Maneuvering in Degraded Visual Environments + + MWhalley + + + JHowitt + + + + Journal of the American Helicopter Society + + 47 + 2 + Apr. 2002 + + + Whalley, M., and Howitt, J., "Optimizations of Partial Authority Automatic Flight Control Systems for Hover/Low-Speed Maneuvering in Degraded Visual Environments," Journal of the American Helicopter Society, Vol. 47, No. 2, Apr. 2002. + + + + + + DLKey + + + RKHeffley + + NASA-CR-2002-211391 USAAMCOM- TR-02-A-003 + Piloted Simulator Investigation of Techniques to Achieve Attitude Command Response with Limited Authority Servos + + Jan. 2002 + + + Key, D. L., and Heffley, R. K., Piloted Simulator Investigation of Techniques to Achieve Attitude Command Response with Limited Authority Servos, NASA-CR-2002-211391 USAAMCOM- TR-02-A-003; Jan. 2002. + + + + + Increased Stabilization for UH-60A Blackhawk Night Operations + + DGMitchell + + + BLAponso + + + AAtencio + + + DLKey + + + RHHoh + + USAAVSCOM TR-92-A-007 + + Nov. 1992 + + + Mitchell, D. G., Aponso, B. L., Atencio, A., Key, D. L., and Hoh, R. H., "Increased Stabilization for UH-60A Blackhawk Night Operations," USAAVSCOM TR-92-A-007, Nov. 1992. + + + + + Piloted Evaluation of Modernized Limited Authority Control Laws in the NASA-Ames Vertical Motion Simulator (VMS) + + VSahasrabudhe + + + EMelkers + + + AFaynberg + + + CLBlanken + + + + AHS 59 th Annual Forum + Phoenix, AZ + + May 2003 + + + Sahasrabudhe, V., Melkers, E., Faynberg, A., and Blanken, C. L., "Piloted Evaluation of Modernized Limited Authority Control Laws in the NASA- Ames Vertical Motion Simulator (VMS)," AHS 59 th Annual Forum, Phoenix, AZ, May 2003. + + + + + ADS-33E Predicted and Assigned Low-Speed Handling Qualities of the CH-47F with Digital AFCS + + JGIrwin + + + PGEinthoven + + + DGMiller + + + CLBlanken + + + + AHS 63 rd Annual Forum + Virginia Beach, VA + + May 2007 + + + Irwin, J. G., Einthoven, P. G., Miller, D. G., and Blanken, C. L., "ADS-33E Predicted and Assigned Low-Speed Handling Qualities of the CH-47F with Digital AFCS," AHS 63 rd Annual Forum, Virginia Beach, VA, May 2007. + + + + + Identification and Simulation Evaluation of an AH-64 Helicopter Hover Math Model + + JASchroeder + + + DCWatson + + + MBTischler + + + MMEshow + + + + AIAA Paper 91-2877, AIAA Atmospheric Flight Mechanics Conference + New Orleans, LA + + Aug. 1991 + + + Schroeder, J. A., Watson, D. C., Tischler, M. B., and Eshow, M. M., "Identification and Simulation Evaluation of an AH-64 Helicopter Hover Math Model," AIAA Paper 91-2877, AIAA Atmospheric Flight Mechanics Conference, New Orleans, LA, Aug. 1991. + + + + + DIMSS -JSHIPs Modeling and Simulation Process for Ship/Helicopter Testing and Training + + MFRoscoe + + + CHWilkinson + + + + AIAA Modeling and Simulation Technologies Conference and Exhibit + Monterey, CA + + Aug. 2002 + + + Roscoe, M. F., and Wilkinson, C. H., "DIMSS - JSHIPs Modeling and Simulation Process for Ship/Helicopter Testing and Training," AIAA Modeling and Simulation Technologies Conference and Exhibit, Monterey, CA, Aug. 2002. + + + + + Simulation and Evaluation of the SH-2F Helicopter in a Shipboard Environment Using the Interchangeable Cab System + + CHPaulk + + + Jr + + + DLAstill + + + STDonley + + NASA-TM-84387 + + Aug. 1983 + + + Paulk, C. H., Jr., Astill, D. L., and Donley, S. T., "Simulation and Evaluation of the SH-2F Helicopter in a Shipboard Environment Using the Interchangeable Cab System," NASA-TM-84387, Aug. 1983. + + + + + Initial Piloted Simulation Study of Geared Flap Control for Tilt-Wing V/STOL Aircraft + + LMGuerrero + + + Corliss + + + LD + + + + NASA TM-103872 + + Oct. 1991 + + + Guerrero, L. M., and Corliss, L. D., "Initial Piloted Simulation Study of Geared Flap Control for Tilt- Wing V/STOL Aircraft," NASA TM-103872, Oct. 1991. + + + + + Evaluation of Flying Qualities and Guidance Displays for an Advanced Tilt-Wing STOL Transport Aircraft in Final Approach and Landing + + CRFrost + + + JAFranklin + + + GHHardy + + + + Biennial International Powered Lift Conference and Exhibit + Williamsburg, Virginia + + 2002. Nov. 2002 + + + AIAA Paper 2002-6016 + Frost, C. R., Franklin, J. A., and Hardy, G. H., Evaluation of Flying Qualities and Guidance Displays for an Advanced Tilt-Wing STOL Transport Aircraft in Final Approach and Landing," AIAA Paper 2002-6016, 2002 Biennial International Powered Lift Conference and Exhibit, Williamsburg, Virginia, Nov. 2002. + + + + + RSRA/X-Wing Flight Control System Development -Lessons Learned + + LDCorliss + + + WRDunn + + + MAMorrison + + + + AHS 45th Annual Forum + + May 1989 + Boston, MA + + + Corliss, L. D., Dunn, W. R.; and Morrison, M. A., "RSRA/X-Wing Flight Control System Development -Lessons Learned," AHS 45th Annual Forum, Boston, MA, May 1989. + + + + + Simulation of a Variable Diameter Tiltrotor + + KStudebaker Fletcher + + + WADecker + + + DCMatuska + + + PMMorris + + + MTSmith + + + Vms + + + + AHS 53 rd Annual Forum + Virginia Beach, VA + + May 1992 + + + Studebaker Fletcher, K., Decker, W. A., Matuska, D. C., Morris, P.M., and Smith, M. T, VMS Simulation of a Variable Diameter Tiltrotor," AHS 53 rd Annual Forum, Virginia Beach, VA, May 1992. + + + + + + MHMansur + + + JASchroeder + + An Investigation of the Ability to Recover from Transients Following Failures for Single-Pilot Rotorcraft + + May1988 + + + NASA-TM-100078 + Mansur, M. H., and Schroeder, J. A., "An Investigation of the Ability to Recover from Transients Following Failures for Single-Pilot Rotorcraft," NASA-TM-100078, May1988. + + + + + A Pilot Rating Scale for Evaluating Failure Transients in Electronic Flight Control Systems + + WSHindson + + + MMEshow + + + JASchroeder + + + + AIAA Paper 90-2827, AIAA Atmospheric Flight Mechanics Conference + Portland, OR + + Aug. 1990 + + + Hindson, W. S., Eshow, M. M., and Schroeder, J. A., "A Pilot Rating Scale for Evaluating Failure Transients in Electronic Flight Control Systems," AIAA Paper 90-2827, AIAA Atmospheric Flight Mechanics Conference, Portland, OR, Aug. 1990. + + + + + Assessment of Simulation Fidelity Using Measurements of Piloting Technique in Flight + + WFClement + + + WBCleveland + + + DLKey + + + + AGARD Conference Proceedings No + + 359 + May 1984 + Monterey, CA + + + Clement, W.F., Cleveland, W.B., and Key, D.L., "Assessment of Simulation Fidelity Using Measurements of Piloting Technique in Flight, AGARD Conference Proceedings No 359, Monterey, CA, May 1984. + + + + + Fidelity Assessment of a UH-60A Simulation on the NASA Ames Vertical Motion Simulator + + AAtencioJr + + 93-A-005 + + + NASA TM 104016, USAATC + + Sept. 1993 + + + Tech. Report + Atencio, Jr., A. "Fidelity Assessment of a UH-60A Simulation on the NASA Ames Vertical Motion Simulator," NASA TM 104016, USAATC Tech. Report 93-A-005, Sept. 1993. + + + + + Visual and Motion Cuing in Helicopter Simulation + + RSBray + + + Sept. 1985 + + + NASA TM-86818 + Bray, R. S., "Visual and Motion Cuing in Helicopter Simulation," NASA TM-86818, Sept. 1985. + + + + + The Effects of Simulator Visual-Motion Asynchrony on Simulator Induced Sickness + + MEMccauley + + + LJHettinger + + + TJSharkey + + + JBSinacori + + + + AIAA Flight Simulation Technologies Conference and Exhibit + Dayton, OH + + Sept. 1990 + + + McCauley, M.E., Hettinger, L.J., Sharkey, T.J., and Sinacori, J.B., "The Effects of Simulator Visual-Motion Asynchrony on Simulator Induced Sickness," AIAA Flight Simulation Technologies Conference and Exhibit, Dayton, OH, Sept. 1990. + + + + + Does a Motion Base Prevent Simulator Sickness? + + TJSharkey + + + MEMccauley + + + + AIAA/AHS Flight Simulation Technologies Conference + Hilton Head, SC + + Aug. 1992 + + + Sharkey, T.J. and McCauley, M.E., "Does a Motion Base Prevent Simulator Sickness?" AIAA/AHS Flight Simulation Technologies Conference, Hilton Head, SC, Aug. 1992. + + + + + Effects of Simulator Motion and Visual Characteristics on Rotorcraft Handling Qualities + + DGMitchell + + + DCHart + + + + American Helicopter Society Conference on Piloting Vertical Flight Aircraft + San Francisco, CA + + Jan. 1993 + + + Mitchell, D.G. and Hart, D.C., "Effects of Simulator Motion and Visual Characteristics on Rotorcraft Handling Qualities," American Helicopter Society Conference on Piloting Vertical Flight Aircraft, San Francisco, CA, Jan. 1993. + + + + + Visual and Roll-Lateral Motion Cueing Synchronization Requirements for Motion-Based Flight + + WWChung + + + JASchroeder + + + + Chung, W.W. and Schroeder, J.A., "Visual and Roll-Lateral Motion Cueing Synchronization Requirements for Motion-Based Flight + + + + + American Helicopter Society 53 rd Annual Forum Proceedings + + Apr. 1997 + Virginia Beach, VA + + + Simulations" American Helicopter Society 53 rd Annual Forum Proceedings, Virginia Beach, VA, Apr. 1997. + + + + + Ground Based Simulation Evaluation of the Effects of Time Delays and Motion on Rotorcraft Handling Qualities + + DGMitchell + + + RHHoh + + + AAtencioJr + + + DLKey + + AVSCOM-TR-91- A-010 + + Aug. 1990 + + + Mitchell, D.G., Hoh, R.H., Atencio Jr., A., and Key, D.L., "Ground Based Simulation Evaluation of the Effects of Time Delays and Motion on Rotorcraft Handling Qualities," AVSCOM-TR-91- A-010, Aug. 1990. + + + + + Simulation Motion Effect on Single Axis Compensatory Tracking + + JASchroeder + + + + AIAA Flight Simulation Technologies Conference + Monterey, CA + + 1993 + + + Schroeder, J.A. "Simulation Motion Effect on Single Axis Compensatory Tracking," AIAA Flight Simulation Technologies Conference, Monterey, CA, 1993. + + + + + Evaluation of Simulation Motion Fidelity Criteria in the Vertical and Directional Axes + + JASchroeder + + + + Journal of the American Helicopter Society + + 41 + 2 + + Apr. 1996 + + + Schroeder, J.A., "Evaluation of Simulation Motion Fidelity Criteria in the Vertical and Directional Axes," Journal of the American Helicopter Society, Vol. 41, No. 2, Apr. 1996, pp. 44-57. + + + + + AGARD CP-577, Flight Simulation -Where are the Challenges? + + JASchroeder + + + WWJohnson + + + Apr. 1996 + Braunschweig, Germany + + + Yaw Motion Cues in Helicopter Simulation + Schroeder, J.A. and Johnson, W.W., "Yaw Motion Cues in Helicopter Simulation," Paper No. 5, AGARD CP-577, Flight Simulation -Where are the Challenges?," Braunschweig, Germany, Apr. 1996. + + + + + Visual-Motion Cueing in Altitude and Yaw Control + + WJohnson + + + JASchroeder + + + + 38 th Annual Human Factors and Ergonomics Society Meeting + Nashville, TN + + Oct. 1994 + + + Johnson, W. and Schroeder, J.A., "Visual-Motion Cueing in Altitude and Yaw Control," 38 th Annual Human Factors and Ergonomics Society Meeting, Nashville, TN, Oct. 1994. + + + + + Evaluation of a Motion Fidelity Criterion with Visual Scene Changes + + JASchroeder + + + WW YChung + + + RAHess + + + + Journal of Aircraft + + 37 + 4 + + July-August, 2000 + + + Schroeder, J.A., Chung, W.W.Y., and Hess, R.A., "Evaluation of a Motion Fidelity Criterion with Visual Scene Changes," Journal of Aircraft, Vol. 37, No. 4, July-August, 2000, pp. 580-587. + + + + + Modeling Human Pilot Cue Utilization with Applications to Simulator Fidelity Assessment + + YZeyada + + + RAHess + + + + Journal of Aircraft + + 37 + 4 + + July-Aug., 2000 + + + Zeyada, Y. and Hess, R.A., "Modeling Human Pilot Cue Utilization with Applications to Simulator Fidelity Assessment," Journal of Aircraft, Vol. 37, No. 4, July-Aug., 2000, pp. 588- 597. + + + + + Pilot Use of Simulator cues for Autorotation Landings + + WADecker + + + CFAdam + + + RMGerdes + + + June 1986 + American Helicopter Society's 42 nd Annual Forum + Washington, DC + + + Decker, W.A., Adam, C.F., and Gerdes, R.M., "Pilot Use of Simulator cues for Autorotation Landings," American Helicopter Society's 42 nd Annual Forum, Washington, DC, June 1986. + + + + + Effects of Visual Texture, Grids, and Platform Motion on Unpowered Helicopter Landings + + MGDearing + + + JASchroeder + + + BTSweet + + + MKKaiser + + + + AIAA Modeling and Simulation Conference + Montreal, Canada + + 2001 + + + Paper 2001-4251 + Dearing, M.G., Schroeder, J.A., Sweet, B.T., and Kaiser, M.K., "Effects of Visual Texture, Grids, and Platform Motion on Unpowered Helicopter Landings," Paper 2001-4251, AIAA Modeling and Simulation Conference, Montreal, Canada, 2001. + + + + + Runway Texture and Grid Pattern Effects on Rate-of-Descent Perception + + JASchroeder + + + MGDearing + + + BTSweet + + + MKKaiser + + + AAAtencioJr + + + + Paper 2001-4307, AIAA Modeling and Simulation Conference + Montreal, Canada + + 2001 + + + Schroeder, J.A., Dearing, M.G., Sweet, B.T., Kaiser, M.K., and Atencio, Jr., A.A., "Runway Texture and Grid Pattern Effects on Rate-of- Descent Perception," Paper 2001-4307, AIAA Modeling and Simulation Conference, Montreal, Canada, 2001. + + + + + Simulation Evaluation of Display/FLIR Concepts for Low-Altitude Terrain-Following Helicopter Operations + + HNSwenson + + + CHPaulk + + + RLKilmer + + + FGKilmer + + NASA TM- 86779 + + 1985 + + + Swenson, H. N., Paulk, C. H., Kilmer, R. L., and Kilmer, F. G., "Simulation Evaluation of Display/FLIR Concepts for Low-Altitude Terrain- Following Helicopter Operations," NASA TM- 86779, 1985. + + + + + Rotary Wing Aircraft Terrain Following/Terrain Avoidance System Development + + DWDorr + + NASA TM-88322 + + 1986 + + + Dorr, D. W., "Rotary Wing Aircraft Terrain Following/Terrain Avoidance System Development," NASA TM-88322, 1986. + + + + + Simulation Evaluation of a Low-Altitude Helicopter Flight Guidance System Adapted for a Helmet-Mounted Display + + HNSwenson + + + REZelenka + + + GHHardy + + + MGDearing + + NASA TM-103883 + + Feb. 1992 + + + Swenson, H. N., Zelenka, R. E., Hardy, G. H., and Dearing, M. G., "Simulation Evaluation of a Low- Altitude Helicopter Flight Guidance System Adapted for a Helmet-Mounted Display," NASA TM-103883, Feb. 1992. + + + + + A Comparison of Active Sidestick and Conventional Inceptors for Helicopter Flight Envelope Tactical Cueing + + MSWhalley + + + + AHS 56 th Annual Forum + Virginia Beach, VA + + May 2000 + + + Whalley, M. S., "A Comparison of Active Sidestick and Conventional Inceptors for Helicopter Flight Envelope Tactical Cueing," AHS 56 th Annual Forum, Virginia Beach, VA, May 2000. + + + + + Helicopter Active Control Technology Demonstrator Program + + MSWhalley + + + JKeller + + + RBuckanin + + + JRoos + + + + AHS 57 th Annual Forum + + May 2001 + Washington, DC + + + Whalley, M. S., Keller, J., Buckanin, R., and Roos, J., "Helicopter Active Control Technology Demonstrator Program," AHS 57 th Annual Forum, Washington, DC, May 2001. + + + + + Simulator Investigation of Pilot Aids for Helicopter Terminal Area Operations with One Engine Inoperative + + LIseler + + + RChen + + + MDearing + + + WDecker + + + EWAiken + + + + AGARD Flight Vehicle Integration Panel Spring + + 1996 + + + Iseler, L., Chen, R., Dearing, M., Decker, W., and Aiken, E. W.; "Simulator Investigation of Pilot Aids for Helicopter Terminal Area Operations with One Engine Inoperative," AGARD Flight Vehicle Integration Panel Spring 1996 + + + + + + + Symposium + + + CanadaOttawa + + + May 1996 + + + Symposium, Ottawa, Canada, May 1996. + + + + + Improvements in Hover Display Dynamics for a Combat Helicopter + + MMEshow + + + JASchroeder + + + + Piloting Vertical Flight Aircraft: A Conference on Flying Qualities and Human Factors + San Francisco, CA + + 1993 + + + Eshow, M. M., and Schroeder, J. A., "Improvements in Hover Display Dynamics for a Combat Helicopter," Piloting Vertical Flight Aircraft: A Conference on Flying Qualities and Human Factors, San Francisco, CA, 1993. + + + + + + TSTurpin + + + SRDowell + + + ZSzoboszlay + + + + Comanche Helmet-Mounted Display Symbology Simulation; Final Report + + 2006-212834, Oct. 2006 + + + Turpin, T. S., Dowell, S. R., and Szoboszlay, Z., "Comanche Helmet-Mounted Display Symbology Simulation; Final Report," NASA CR 2006- 212834, Oct. 2006. + + + + + Evaluation of Helmet Mounted Display Alerting Symbology + + JDemaio + + + MRutkowski + + + + NASA TM + + 2000-209603, Sep. 2000 + + + DeMaio, J., and Rutkowski, M., "Evaluation of Helmet Mounted Display Alerting Symbology," NASA TM 2000-209603, Sep. 2000. + + + + + Piloted Simulator Investigations of a Civil Tilt-Rotor Aircraft on Steep Instrument Approaches + + WADecker + + + + American Helicopter Society 48 th Annual Forum + Washington, D.C. + + Jun. 1992 + + + Decker, W.A., "Piloted Simulator Investigations of a Civil Tilt-Rotor Aircraft on Steep Instrument Approaches," American Helicopter Society 48 th Annual Forum, Washington, D.C., Jun. 1992. + + + + + Evaluation of Two Cockpit Display Concepts for Civil Tiltrotor Instrument Operations on Steep Approaches + + WADecker + + + RBray + + + RCSimmons + + + GETucker + + + + Piloting Vertical Flight Aircraft: A Conference on Flying Qualities and Human Factors + San Francisco, CA + + American Helicopter Society + Jan. 1993 + + + Decker, W.A., Bray, R., Simmons, R.C., and Tucker, G.E., "Evaluation of Two Cockpit Display Concepts for Civil Tiltrotor Instrument Operations on Steep Approaches," American Helicopter Society Piloting Vertical Flight Aircraft: A Conference on Flying Qualities and Human Factors, San Francisco, CA, Jan. 1993. + + + + + Handling Qualities Evaluation of XV-15 Noise Abatement Landing Approaches Using a Flight Simulator + + WADecker + + + + 57 th Annual Forum + Washington, D.C. + + American Helicopter Society + May 2001 + + + Decker, W.A., "Handling Qualities Evaluation of XV-15 Noise Abatement Landing Approaches Using a Flight Simulator," American Helicopter Society 57 th Annual Forum, Washington, D.C., May 2001. + + + + + Pursuit Display Review and Extension to a Civil Tilt Rotor Flight Director + + GHHardy + + + + AIAA Paper 2002-4925, AIAA Guidance, Navigation, and Control Conference and Exhibit + Monterey, CA + + Aug. 2002 + + + Hardy, G.H., "Pursuit Display Review and Extension to a Civil Tilt Rotor Flight Director," AIAA Paper 2002-4925, AIAA Guidance, Navigation, and Control Conference and Exhibit, Monterey, CA, Aug. 2002. + + + + + Civil Tiltrotor Transport Procedure and Requirement Development Using a Flight Simulator + + WADecker + + + GHHardy + + + + AIAA Paper 2002-4530, AIAA Guidance, Navigation, and Control Conference and Exhibit + Monterey, CA + + Aug. 2002 + + + Decker, W.A., and Hardy, G.H., "Civil Tiltrotor Transport Procedure and Requirement Development Using a Flight Simulator," AIAA Paper 2002-4530, AIAA Guidance, Navigation, and Control Conference and Exhibit, Monterey, CA, Aug. 2002. + + + + + + WADecker + + + DIsleib + + + Major + + + JJohn + + A Simulator Investigation of Air-to-Air Combat Maneuvering for Tilt-Rotor Aircraft + New Bern, NC + + Sep. 1989 + + + AHS National Technical Specialists' Meeting on Tactical V/STOL + Decker, W. A., Isleib, D., Major, and John, J., "A Simulator Investigation of Air-to-Air Combat Maneuvering for Tilt-Rotor Aircraft," AHS National Technical Specialists' Meeting on Tactical V/STOL, New Bern, NC, Sep. 1989. + + + + + + diff --git a/file16.txt b/file16.txt new file mode 100644 index 0000000000000000000000000000000000000000..ef2112feafa5d4a6c5e3c627333bdd5de8bd260e --- /dev/null +++ b/file16.txt @@ -0,0 +1,1557 @@ + + + + +I. IntroductionN the 1960's, the role of the military helicopter evolved from primarily a utility role to encompass more demanding roles including ground assault transport and ground attack.Recent combat experience in the 1970's indicated that low-level nap-of-the-earth (NOE) flight, air-to-air combat, and night operations were necessary future tactical capabilities for rotorcraft.Designing a new generation of rotorcraft to meet these evolving and increasingly demanding missions required a significant expansion and improvement of the understanding of rotorcraft performance and flying characteristics.Acquiring this understanding and applying it to develop new technologies and rotorcraft designs that could accomplish these more demanding missions with improved safety required fundamental research.The single-main-rotor helicopter was the most common rotorcraft configuration at the time.With unstable dynamics at low-speed and significant cross-coupling between axes of control, the single-rotor helicopter is demanding to fly and requires closed-loop pilot control at all times.Expanding the understanding of rotorcraft flying characteristics, therefore, required a high-fidelity flight simulation environment.This requirement was met at NASA Ames Research Center using a combination of ground-based simulation, and in-flight simulation using variable stability research rotorcraft.For almost four decades, the Vertical Motion Simulator (VMS) and its predecessor, the Flight Simulator for Advanced Aircraft (FSAA) at NASA Ames Research Center, have served as cornerstones for rotorcraft research.Built in 1969, the FSAA was originally designed for fixed-wing aircraft research and, as it was increasingly used for rotorcraft research, the need for improved motion fidelity, particularly in the vertical axis, became evident (Ref.1).The VMS was designed to meet this need and became operational in 1979.The VMS included the highest fidelity motion system in the world, a distinction it holds to this day.In addition to high-fidelity vertical and lateral motion cueing, the VMS included an interchangeable cab system with high-fidelity visual displays of the outside world and adaptable cockpit interfaces with accurate control feel systems, flight instruments and displays.The need to expand the understanding of rotorcraft performance and flying qualities was recognized by the Army who, in collaboration with NASA, began a program to meet this need.Together with research on rotorcraft, a parallel and interacting stream of studies examined human pilot cueing and the level of simulation fidelity required to accurately recreate the pilot-rotorcraft interaction in flight.In almost three decades of operation, simulation studies on the VMS by a team of NASA and U.S. Army Aeroflightdynamics Directorate (AFDD) researchers at the Ames Research Center generated a majority of this data.Jet powered V/STOL aircraft and technologies were also evaluated on the VMS through collaboration between NASA and industry.This paper focuses on rotorcraft and provides an overview of rotorcraft research at the VMS -summarizing the impact of this work on current and future rotorcraft design, development, procurement, and operations.The research is grouped into six streams: handling qualities; guidance and displays; simulation fidelity and cueing; flight control design and evaluation, specific programs; and tilt-rotor.Following a description of the present capabilities of the VMS and the genesis of its design, this paper summarizes the research conducted in each of these streams. +II. Description of the VMSThe VMS combines a high-fidelity simulation capability with an adaptable simulation environment that enables customization for numerous human-in-the-loop research applications.The distinctive feature of the VMS is its unparalleled large amplitude, high-fidelity motion capability.An overall high level of simulation fidelity is achieved by combining this motion fidelity with realistic visual and cockpit interfaces.Interchangeable cabs allow different crew vehicle interfaces and vehicle types to be evaluated with rapid turnaround times between simulation projects.The Interchangeable Cab (ICAB) capability allows the cockpit to be tailored to the research application.The VMS has five portable ICABs with different out-the-window visual fields-of-view.For each simulation, an ICAB is selected and equipped to meet the study's requirements and then tested with the complete simulation environment without motion.Configuring the cab includes installation of flight controls, flight instruments and displays, and seats (Fig. 1).Following cab configuration and checkout, the ICAB is transported and installed on the motion system.The ICAB capability, simulation architecture, and resources enable the VMS facility to conduct fixed-base and moving-base simulation studies simultaneously. +Fixed-baseCab Interior Moving-base +Figure 1. VMS Transport Cab on and off the motion systemThe high-fidelity flight controls are heavily modified and optimized McFadden hydraulic force-loader systems (Ref.2).A custom digital-control interface allows for comprehensive adjustment of the controller's static and dynamic characteristics.Force-loader characteristics may be varied during simulated flight as necessary for studying pilot cueing concepts using inceptors.A variety of aircraft manipulators, ranging from the regular columnand-wheel type to conventional rotorcraft controls and side sticks, are available and may be combined with the force-loader systems.A Rockwell-Collins EPX-5000 image generation system creates the out-the-window visual scene and provides a high-resolution and complex visual environment at update rates exceeding 60Hz.Fifteen channels are available, allowing up to three different simulation experiments to be conducted simultaneously in the facility.In-house graphics expertise is used to customize the visual databases to meet widely varying simulation requirements.Separate graphics processors generate the content for the primary flight displays, head-up displays, sensor imagery, etc, which can be fully customized.All the essential elements of the simulation are linked with the host environment through a dedicated network and the simulation is managed from a fully equipped control room.The flexible simulation architecture makes it convenient to interface and evaluate custom software and hardware modules.This capability may be used to evaluate sensors, vehicle dynamic models, flight control systems, etc. +A. Motion System DescriptionThe VMS motion system, shown in Fig. 2, is an uncoupled, six-degree-of-freedom, combined electromechanical/electro-hydraulic servo system (Ref.3).It is located in, and partially supported by, a specially constructed 120-ft tower.The motion system includes a beam structure, called the vertical platform, which spans the width of the tower.The vertical platform is mounted on two columns, called equilibrators which extend down into 75-ft deep shafts under the tower floor.Wheel assemblies, which ride along vertical guide rails attached to the tower walls, restrain the vertical platform at both ends and the center of its span. +Figure 2. VMS Motion SystemThe two equilibrators act as pneumatic counterweights.The hollow equilibrator columns slide over inner columns so that the two, along with a gas-tight seal between them, form a cylinder/piston arrangement.Nitrogen gas, supplied by a special storage system, pressurizes the equilibrators such that the pressure forces balance the weight of the 140,000-lb cab and platform structure.This counterbalancing force reduces the power requirement of the vertical drive motors and results in a linear motion response in both directions of vertical travel.An added benefit is that if vertical drive power is lost, the motion system will float to an equilibrium position towards the center of the tower.Eight mechanically coupled, 150-hp direct-current servomotors power the vertical motion through reduction gearboxes and a rack-and-pinion drive system with the racks mounted on the equilibrator columns (Fig. 2).Four 40hp direct-current servomotors power the lateral carriage along the vertical platform using reduction gearboxes and a rack-and-pinion drive system with the rack mounted on the top of the vertical platform.A linear hydraulic actuator powers the longitudinal carriage, located atop the lateral carriage.A 48-inch diameter roller bearing provides the yaw motion, which is mounted on top of longitudinal carriage that is driven by another linear hydraulic actuator.The yaw bearing supports a conical center structure, which has a two-axis gimbal on top that provides roll and pitch motion using two linear hydraulic actuators.A unique feature of the VMS is that the yaw actuator may be attached at two different locations, 90° apart, allowing the longer ±20 ft.translational displacement to be used as either the aircraft's longitudinal or lateral axis, as required for the particular simulation. +B. Motion System PerformanceTable 1 summarizes the VMS motion capability.Included are two sets of limits: system limits, that represent the absolute maximum levels attainable under controlled conditions; and operational limits, that represent attainable levels for normal piloted operations.The operational limits listed in Table 1 include the effects of all the system limiters, both hardware and software.The motion system also incorporates a parabolic limiter, which is not reflected explicitly in Table 1.When triggered, the parabolic limiter commands a maximum acceleration opposite to the direction of travel so that the simulator stops just short of a displacement limit.The motion drive dynamics may be modeled as equivalent time delays ranging from 90 msec in pitch and roll to approximately 130 msec in the yaw and translational axes (Ref.4).The VMS motion system includes digital feedforward compensators (motion lead compensators) in each degree of freedom that may be used to alter and improve the overall motion system dynamics, within limits.The motion lags are typically larger in the translational axes than documented (when they are documented) for small hexapod simulators, which one would expect when considering the relative scale difference between a small hexapod and the VMS.These inherent lags can be effectively eliminated, if a particular task deems it necessary, by modifying the math model so that lags due to actuators and digital effects are removed in exchange for the motion system lag so that the overall equivalent delay in the simulated vehicle is maintained for the evaluation (Ref.5). +C. Motion Washout FiltersThe cockpit motion cueing algorithm uses a high-pass (washout) filter and a rotational/translational cross-feed arrangement shown schematically in Fig. 3.The computed pilot station accelerations of the modeled aircraft are high-pass filtered and attenuated before commanding the motion drive system.Turn coordination and induced acceleration compensation account for the cross-coupled motion commands and provide the correct cues at the pilot's station.A low-pass filter tilts the simulator to provide steady-state longitudinal and lateral acceleration cueing at low frequency (Fig. 3). +Residual tilt for low frequency acceleration +III. VMS Motion System DesignThe design of the VMS is based on the experience gained from a series of ground-based flight simulators at NASA Ames beginning in the late 1940's (Ref.6).Some key aspects of the VMS motion system design directly trace to the research experience from three past flight simulators at NASA Ames -the sizing of the translational envelope was based on insights gained from the Height Control Test Apparatus (HCTA) and the Flight Simulator for Advanced Aircraft (FSAA), while the equilibrator design was based on experience with the Six-Degree-of-Freedom Motion Simulator (Fig. 4).The HCTA was a single degree-of-freedom flight simulator with 80 ft. of vertical travel that became operational in 1961.In the early 1970s a study on the HCTA determined the significance of vertical acceleration cues when simulating the visual approach and landing maneuver (Ref.7).The results indicated that vertical motion cues are important for the landing task, particularly for aircraft with marginal longitudinal handling qualities.The study concluded that a simulator must have a vertical excursion capability of at least ±20 ft. to provide realistic pilot cueing for the approach and landing task.The sizing of the VMS vertical travel envelope was based on this finding.The FSAA was a six degree-of-freedom flight simulator with ±40 ft. of lateral travel but limited vertical (±4 ft.) and longitudinal (±3.5 ft.) travel.Since its inception in 1969, the FSAA contributed to several important fixed-wing aircraft research programs, but its small vertical travel limited the ability to simulate VTOL aircraft accurately.Similarly, the longitudinal travel was adequate for conventional aircraft, but more travel was needed for simulating the low-speed maneuvers of VTOL aircraft (Ref. +IV. Handling Qualities StudiesIn the 1970's, the Army needed a handling qualities specification that could guide the development of rotorcraft that would meet the more demanding missions and tactics envisioned in the future.The military specification for rotorcraft handling qualities, MIL-H-8501A, was written in 1952 and was inadequate (Ref.9).Several attempts to update MIL-H-8501A met with little success and were not adopted, primarily due to a lack of background data of adequate quality.To overcome the deficiencies in MIL-H-8501A, the Army used Prime Item Development Specifications (PIDS) for procuring new helicopters (Ref.9).The U. S. Army Aeroflightdynamics Directorate (AFDD), in collaboration with NASA Ames Research Center, began an effort to build a database of handling qualities data and design criteria that could be incorporated into a new handling qualities specification.In 1982, the AFDD began the development of a new rotorcraft handling qualities specification to supersede MIL-H-8501A (Ref.9).The specification, US Army Aeronautical Design Standard -33 or ADS-33 (Ref.10), was completed and published in 1987.Initial development of the specification was for the procurement of the modern light/attack/scout helicopter (LHX) the Army intended to acquire.This program led to the development of a prototype helicopter, the RAH-66 Comanche, but it was subsequently cancelled.The strategy for developing a helicopter handling qualities database of sufficient quality and validity for use in a military specification was to combine high-fidelity simulation with a limited amount of flight test activity.Almost all the simulation data incorporated into ADS-33 came from VMS studies.The process of developing the database is ongoing with future studies filling known gaps in the database and refining others.Table 2 lists the VMS simulation studies that have supported this process thus far.Reference 1 contains a comprehensive summary of the rotorcraft research conducted in the VMS in its first decade of operation.Initial studies focused on two issues of primary concern: 1) rotorcraft design requirements for low-altitude or nap-of-the-earth (NOE) flight, and 2) flight in adverse visual conditions.Not all these studies were documented or published, but their data were used to develop criteria in ADS-33 and are available in the Background Information and Users Guide to ADS-33 (Ref.10).Initial studies investigated the effect of design variations on rotorcraft dynamic characteristics and handling qualities.These included the effect of vertical damping and thrust available (Ref. 11), pitch/roll and collective/yaw cross coupling (Refs.12 and 13), yaw control (Refs.14 and 15), and roll control (Ref.16).Other studies investigated some of the fundamental precepts being developed for the specification -required response type and bandwidth.An innovative concept introduced in ADS-33 is the trade-off between augmentation (which defines response type) and the visual cueing environment.Ground-based and in-flight simulation studies established that when the visual cueing environment is degraded, increased compensation is required to maintain handling qualities at a satisfactory level.Bandwidth is a measure that defines the quickness in the response to control inputs and represented a shift, at the time, from time-domain based performance criteria to a more accurate and discriminating frequency-domain measure.Studies on the VMS that developed these concepts include those reported in Refs.17, 18, and 19.Several studies by the AFDD also investigated helicopter and flight control system design requirements for aggressive air-to-air combat (Refs.20, 21, and 22) at low-altitude.These studies provided data on the effect of response type, pitch/roll/yaw bandwidth, and turn coordination on handling qualities when the rotorcraft itself is used as a pointing platform.Another study investigated the yaw dynamic requirements for a rotorcraft with a turreted gun engaged in low-level air combat operations (Ref.23).The study investigated the potential trade-off between gun slewing angular movement capability and required rotorcraft yaw axis response.A separate series of studies investigated the handling qualities and cockpit interface that would be required for single-pilot operation envisioned, at that time, for the LHX program (Ref.24).These studies indicated that the mission management and demanding NOE flight environment placed an unacceptable workload on a single pilot that may be alleviated by increasing augmentation through the flight control system.An important outcome of these studies was the progressive development of evaluation tasks that were designed to be representative of the mission tasks expected of rotorcraft but also constrained to allow repeatability and promote consistency in handling qualities ratings.These evaluation tasks, or Mission Task Elements (MTEs), were refined over the course of many experiments on the VMS and included in ADS-33 as demonstration maneuvers for evaluating the overall handling qualities of a rotorcraft.Task performance displays for the MTEs were also developed over the course of many experiments on the VMS and these displays have transitioned to flight testing.Finally, a series of eight simulation studies investigated the effect of externally slung loads on helicopter handling qualities.The results of the first five simulation studies by the AFDD are reported in Ref. 25.These simulation studies established a database of handling qualities data for future inclusion in ADS-33 as guidelines for designing rotorcraft that carry external loads.The database on handling qualities will be further expanded by follow-on simulation studies on the VMS to explore heavy cargo rotorcraft operations including the effect of pitch/roll response-type on handling qualities with slung loads.Since its release, ADS-33 has guided the procurement of the CH-47F and CH-53K helicopters, and the UH-60Mu and AH-64D fly-by-wire upgrades by the Army and Navy.It was designed for, and used as, the guiding specification for the RAH-66 Comanche flight control system that is the basis for the full-authority fly-by-wire flight control systems implemented on the UH-60Mu, CH-53K, H-92 Superhawk, and the digital automatic flight control system (DAFCS) on the CH-47F.(The CH-47F flight control system was evaluated on the VMS.)The RAH-66 control system design was, in turn, based on extensive research on the Army's advanced digital optical control system (ADOCS) program that also used the VMS for initial design and development. +V. Flight Control Design StudiesTable 3 lists the VMS simulation studies that evaluated flight control system concepts and designs.All these development efforts were closely linked with the development of a handling qualities database.To evaluate new flight control concepts in preparation for the procurement of a new attack helicopter (LHX), the Army initiated the ADOCS program to investigate modern flight control system concepts and pilot-vehicle interfaces.A series of simulation studies on the VMS evaluated modern control laws, flight control system architectures, and cockpit interface concepts including multi-axis side sticks and advanced pilot displays (Ref.26).Outcomes of these studies were recommendations for the design of multiaxis pilot controllers, flight control system designs with advanced response types; control law mode switching logic; automatic stick force trimming; and helmet mounted display.These concepts were implemented and flight-tested on the ADOCS UH-60 helicopter.The designs translated well from simulation to flight and only minor parametric changes were required in flight to optimize handling qualities (Ref.27).The program provided invaluable data on advanced flight control system design.The fundamental flight control concepts and architectures developed in the program have influenced all the rotorcraft fly-by-wire flight control systems designed since that time.The difficulty of landing rotorcraft in blowing sand, known as brownout, became evident in the first Gulf War.Conducting operations under cover of darkness using night vision goggles (NVG) was also becoming more prevalent as the quality of the devices improved.At that time, there was significant evidence that control augmentation could alleviate handling qualities deficiencies in degraded visual environments such as brownout and NVGs.A series of simulation studies on the VMS investigated the possibility of modifying the existing limited authority flight control system on the UH-60 Black Hawk to provide added augmentation to improve handling qualities when flying in degraded visual environments (Refs.28 -30).The simulation studies explored different methods for adding this increased augmentation to the existing Black Hawk flight control system and evaluated the effect of saturating the stability augmentation system for larger pilot inputs.Based on the favorable results of these studies, increased augmentation was proposed for the Black Hawk and tested on the VMS (Ref.31).The CH-47F with its digital automatic flight control system (DAFCS) was the first production Army helicopter acquired using ADS-33 requirements.The DAFCS included advanced response types to improve handing qualities in degraded visual environments.Following development and initial testing by the manufacturer, the system was refined and evaluated by Army pilots on the VMS (Ref.32).The simulation investigations showed that the DAFCS significantly improved handling qualities in day and night (using NVGs) visual conditions when compared with a CH-47D.Findings from the VMS study also led to improvements in mode transitions between response types in the DAFCS. +VI. Studies on Specific Rotorcraft ProgramsThe high-fidelity simulation capability of the VMS was used to evaluate and test new rotorcraft concepts and configurations such as tilt-wing and tilt-rotor as well as to support existing programs such as Comanche and Apache by implementing and evaluating simulation models (Ref.5).Table 4 lists the simulation studies that fall in this category.Other programs include several unique rotorcraft configurations including a tilt-wing, X-wing, and the Piasecki Vectored Thrust Combat Agility Design (VTCAD).The tilt-wing studies ranged over a period of 10 years and investigated novel control methods such as the geared flap, that were necessary to control the aircraft during wing tilt (Ref.35), and the effect of augmentation on handling qualities during approach and landing (Ref.36).The VTCAD concept involved the substitution of a ducted fan with thrust vectoring capability in lieu of a tail rotor on an AH-64 helicopter to increase its speed and agility.The AFDD evaluated the efficacy of this design in two VMS investigations.The X-wing program was a joint Army/NASA project to investigate high-speed rotorcraft that culminated in the development of the prototype Rotor Systems Research Aircraft (RSRA) to test rotor and propulsion concepts.The complex fly-by-wire flight control system for the X-wing required control transitions as the aircraft transitioned from a rotorcraft to a fixed-wing aircraft (Ref.37).These control laws were refined and evaluated on the VMS.Several simulation studies also supported the initial development of the Army/NASA Rotorcraft variable stability Black Hawk known as the Rotorcraft Aircrew Systems Concepts Airborne Laboratory (RASCAL).RASCAL is operated by the U.S. Army's Aeroflightdynamics Directorate.A particular concern during development was controllability of the helicopter if there was a failure in the research flight control system that operated in parallel with the production flight control system.Several studies investigated and quantified the effect of failure transients and assessed the handling qualities following a failure (Ref.39) and evaluated RASCAL safety systems.A pilot failure rating scale was developed to assess the safety and handling qualities requirements associated with dynamic failures that could occur on a full-authority flight control system as envisioned for RASCAL (Ref.40).Recent studies on the VMS focused on modifying the RASCAL safety systems to allow flight testing at lower altitudes. +VII. Simulation Fidelity StudiesA host of experiments aimed at understanding simulation fidelity requirements for helicopters were performed on the VMS since its inception including those listed in Table 5.Since the VMS represented the most capable simulator in the world, the first natural question to ask was how good is it compared to flight.While answering critical questions necessary for advancing simulation-based flight training, the results of these studies were also fundamental to the validity of the studies conducted in the other streams of research outlined in this paper.The first attempt to compare how well the VMS could simulate helicopter flight was conducted cooperatively among several organizations (Ref.41).Pilot opinion (HQRs) were Level 1 in flight but Level 2 in simulation.These differences led to investigations to improve the rotor model, servo dynamics, and engine.Improvements to the math model, visual systems, motion configurations, and experimental tasks and protocols led to experiments documented in Ref. 42.This experiment compared performance and pilot opinion using a UH-60 operated out of NASA Ames.Extensive frequency-domain identification compared the simulation performance with flight, with the conclusion that the model was a reasonable representation of the flight vehicle.This was an improvement from the 1984 simulation study but deficiencies in visual and heave motion fidelity were noted.Early research on the VMS showed how far away the simulated cues in a typical helicopter are from the real world (Ref.43) and exposed the need for systematic investigations with objective measures to determine fidelity requirements.Early emphasis was on the relative timing between the visual and motion cues that was a suspected cause of simulator sickness.Using four different motion conditions, McCauley (Ref.44) found that simulator sickness increased with both time and the level of maneuvering.Using a UH-60 model, Sharkey (Ref.45) subsequently found that false motion cues had an adverse effect similar to having no motion cues at all.The effects of asynchrony in the motion and visual cues were also investigated for handling qualities effects.Mitchell and Hart (Ref.4) examined variations in visual time delays and motion washout filters.They suggested tailoring the motion system to the task and minimizing the mismatch between motion and visual delays.Chung and Schroeder (Ref. 46) studied the motion and visual synchrony among roll, and lateral axes using a predominantly lateral axis task and recommended that the equivalent time delay mismatch between the roll and lateral motion cues not exceed 40 msec.Their work also suggested that the equivalent delays in the motion cuing could exceed the equivalent visual delay without a resulting degradation in handling qualities ratings.When simulating motion in a ground-based simulator where motion displacement is limited, there is a necessary trade-off between the desired initial, or short-term acceleration and the desired sustaining, or long-term, acceleration.Mitchell (Ref. 47) showed that the addition of motion improved pilot opinion ranging from ½ to 2 HQR points.For precision tasks, sustained acceleration cues were preferred (reduced washout natural frequencies of the motion filter), while, for aggressive tasks, short-term acceleration cues were preferred.Schroeder (Ref.48) used pilot describing function measurements to examine a variety of motion gains and motion washout filter variations on a classical single-axis compensatory tracking task.The results showed that motion cues allowed the pilot to generate lead compensation and improve target tracking phase margins with increasing filter gain or decreased natural frequency.The study also showed no effects for any of the pure yaw motion configurations, which led to subsequent investigations.To help answer the question on what characteristics a motion filter should have so that simulation is a reasonable representation of flight, Schroeder (Ref.49) evaluated the proposed Sinacori motion fidelity criteria (Ref.8).Objective and subjective results showed that the original criteria could be relaxed, and these criteria (shown in Fig. 5) are still used today as a guide when configuring the VMS motion system as well as other simulators.When compared against these criteria, the gain and phase mismatch for the VMS motion system spans the "like flight" and "different from flight" regions, depending on how the motion filter settings are optimized (Fig. 5). +Figure 5. Motion Fidelity CriteriaA question that often arises when configuring the motion system is the level of fidelity required in each axis.Specifically, the earlier study that showed little effect of yaw cueing led to a more detailed study (Ref. 50).This study evaluated a helicopter in a single degree of freedom hovering yaw task.Four variations in the motion cueing were studied: full motion, only lateral translational motion, only yaw rotational motion, and no motion.The study found that the lateral acceleration cue was of predominant importance in both performance and opinion.This suggested that if you had strong lateral translational cues, then the yaw rotational cues might be redundant and unnecessary.In a study that considered visual cueing aspects as well as motion, Johnson (Ref.51) investigated how the displayed level-of-detail changes as one gets closer or further away from an object for a height control task.The results showed that changing the level-of-detail to maintain constant optical density as the altitude changed, like that of the real world, improved altitude awareness.Texture at this time was only beginning to be used in the VMS.Separately, adding platform motion improved speed regulation and altitude perception.Further systematic changes in visual scene, via changing spatial frequency with alternating black-and-white stripes, and motion cues were evaluated in the vertical axis (Ref.52).The variations in visual scene evaluated had no effect, while the motion configurations did have an effect.These configurations were subsequently analyzed using a structural pilot model (Ref.53).The intent was to develop and calibrate a model that would predict pilot opinion for a given rotorcraft and task, and the model's predictions correlated well, in a ranking sense, with the subjective ratings.The VMS was used to evaluate cueing effects on autorotation in two studies separated by more than a decade (Refs.54 and 55).In the first study, autorotation task performance decreased with degraded motion cues, yet acceptable performance could be attained as long as there were adequate visual cues.With improved visual cueing technology, the second study evaluated the effect of visual texture and motion variations on autorotation task performance.Visual texture affected all measures, but the finest texture did not perform the best, debunking a myth that more texture is better.This result was also supported by a fixed-base psychophysics study (Ref.56).Visual detail affected only pilot subjective opinion.Pilot performance as well as opinion of motion fidelity improved with increased motion cueing. +VIII. Guidance and Display StudiesThe high-level of motion and cockpit fidelity offered by the VMS made it an ideal platform to assess and prototype rotorcraft guidance and display concepts (Table 6).Early studies on the VMS evaluated the rotorcraft handling characteristics necessary for low-level nap-of-the-earth missions envisioned for the next generation of military rotorcraft.These studies showed that the workload for such missions was unacceptable for a single pilot.A series of studies over several years beginning in 1988 conducted research to develop an automated helicopter flight capability for potential application in the U.S. Army light, attack, scout helicopter (LHX) program (Fig. 6).Several simulation studies on the VMS developed and refined the critical components of a guidance system for terrain and obstacle detection, path prediction, guidance displays and symbology, and automatic control concepts for terrain following/terrain avoidance (TF/TA) (Refs.57, 58, and 59).The resulting guidance system was implemented on the Army's UH-60 System Testbed for Avionics Research (STAR) and tested in flight.The outcome of the project included several fundamental concepts for displaying flight information superimposed on sensor imagery.Another series of studies by the AFDD investigated methods for predicting helicopter maneuver limits and communicating this information to the pilot via tactile cueing through the pilot inceptor (Ref.60).The results showed that tactile cueing through a conventional inceptor or a sidestick enabled the pilots to easily track helicopter structural and power-train limits and avoid exceedances while maneuvering aggressively without referring to flight instruments.The cueing allowed the pilots to extract maximum maneuvering performance from the helicopter without risk of damage.This technology was developed for testing on an AH-64 Apache helicopter under the Army's Helicopter Active Control Technology program (Ref.61).The increased operation of helicopters in urban areas brought with it the risk of an accident, particularly in an emergency situation such as an engine failure.Updated procedures for recovering from an engine failure when taking off or landing in confined areas were needed.Two simulation studies on the VMS by NASA and the FAA investigated methods to reduce pilot workload and increase safety for rotorcraft terminal area operations (Ref.62).The studies examined the benefits of optimal procedures, cockpit displays, and alternate cueing methods for safe terminal area procedures with one-engine inoperative.Results showed that an integrated display reduced pilot workload and improved safety when compared with conventional instruments. +Table 6. Research into guidance and display concepts +SimulationIn the area of head-mounted displays, a study on the VMS investigated the handling qualities benefits that could be realized using new display law design methods for hover displays (Ref.63).The display law design was applied to the Apache helmet-mounted display format, using the Apache vehicle dynamics to tailor the dynamics of the velocity predictor symbol.The new symbol dynamics improved the pilots' ability to maneuver about hover in poor visual cuing environments and improved pilot opinion.More recently, the AFDD conducted a simulation study to examine the performance of the Comanche Contact Analog world-referenced symbology displayed on the Comanche's helmet-mounted display when compared with a compressed symbology design similar to that specified by the former MIL-STD-1295 (Ref.64).Pilot opinion showed a preference for the MIL-STD-1295 symbology.The study recommended specific improvements to the Contact Analog symbology.Another AFDD study, reported in Ref. 65, examined ways to optimize the alerting effectiveness of helmet display symbology.The research investigated two approaches to increasing the effectiveness of alerts -using the entire display surface and including information about the required response in the alert itself.Helmet display symbology was based on the AH-64's pilot night vision system (PNVS), cruise mode symbology.The data showed a small benefit from both the fullscreen alert and the partial information alert. +IX. Tilt-Rotor StudiesIn 1981 the Deputy Secretary of Defense directed that the services reviewed V/STOL technology with the intent of establishing a joint rotary wing aircraft development program to satisfy service lift requirements for medium lift V/STOL aircraft in the 1990s and beyond.This would take advantage of the advanced, but mature, tilt-rotor technology already in place by the early 1970s.The joint NASA/Army XV-15 Tilt-Rotor Research Aircraft program had already begun in the early 1970's and became the foundation for the full-scale development of the JVX, later designated as the V-22 Osprey.A number of simulation studies conducted at the VMS from 1980 to 1985 validated the JVX math model and evaluated the flight control system characteristics.Previous studies on the FSAA had led to the evaluation and selection of Bell to build the XV-15 prototype.The continued development and flight-testing of the military V-22 Osprey prepared the way for the introduction of a civil tilt-rotor transport.The potential introduction of a civil tilt-rotor transport into the National Airspace System presented challenges and opportunities for vertical flight solutions to airspace congestion.The Federal Aviation Administration (FAA) had developed a Vertiport design guide based on helicopter capabilities and projected civil tilt-rotor transport performance.As a complement to the FAA guide, NASA undertook an effort to develop controls and display technology to fully utilize the capability of this rotorcraft.This effort took the form of a series of ten piloted simulation studies conducted at the VMS (from 1988 to 2001) to investigate tilt-rotor terminal operations and certification issues (Fig. 7).The general objectives of these simulation studies were: 1) to develop design guidance for safe, all-weather, low-noise flight operations, 2) to develop controls and cockpit displays to support tilt-rotor transport operations, and 3) to develop tilt-rotor transport terminal area procedures.All these objectives were met.Initial studies evaluated steep instrument (IFR) approaches to confined spaces.These steep approaches would reduce the required obstruction-free approach zone and could significantly reduce the noise footprint of terminal operations.The studies investigated two display concepts to provide guidance for steep IFR approaches (Refs.66 and 67) with glide slopes ranging from the nominal 3-degrees up to 25-degrees.The next series of experiments further evaluated these steep IFR approaches under One-Engine-Inoperative (OEI) conditions.This was followed by handling qualities evaluations of noise abatement landing approaches and comparing them with acoustic measurements from flight tests using the XV-15 aircraft flying similar trajectories (Ref.68).A potential twosegment approach with initial deceleration at a three-degree glide slope converting to final approach along a ninedegree glide slope was also investigated.The final experiments were full mission simulation studies that evaluated operation in congested airspace and led to the development and use of pursuit displays, particularly for the transition from the airplane type cruise configuration to the helicopter configuration for final approach and landing (Refs.69 and 70).In another study, NASA evaluated the effectiveness of the V-22 Osprey tilt-rotor in one-on-one air combat maneuvering on the VMS, for the Marine Corps (Ref.71).The study showed that the unique speed and maneuvering characteristics of the V-22 enhanced its survivability against both fixed-wing and rotorcraft aggressors. +X. Concluding RemarksOver the past three decades, the NASA Vertical Motion Simulator has provided a wealth of data and knowledge to further rotorcraft technology and safety.The collaboration between NASA and the Army Aeroflightdynamics Directorate at the Ames Research Center has been, and continues to be, a primary reason for the prolific output of valuable research from this facility.This collaboration has developed a database of knowledge on a variety of interacting disciplines on rotorcraft including handling qualities, simulation fidelity, guidance and displays, and design.This database is now used in military procurement and in civil applications on training simulation and guidance displays.NASA programs as well as those in collaboration with the FAA have also contributed significantly to the development of the tilt-rotor aircraft and its civil derivative.Specific contributions include:• Data on rotorcraft handling qualities that formed a basis for the current military specification on rotorcraft handling qualities, ADS-33E;• Understanding on human motion and visual cueing, and developing guidelines for configuring simulation cueing environments;• Designing and evaluating novel rotorcraft configurations including the tilt-wing and tilt-rotor;• Designing and evaluating production flight control systems including the CH-47F;• Designing and evaluating rotorcraft guidance and display concepts for low-level terrain following and helmet-mounted displays; and• Civil tilt rotor operation and certification.Figure 3 .3Figure 3. VMS Motion Algorithm Schematic +1).A study on the FSAA determined that at least ±16 ft. of lateral travel is required for helicopter flight research (Ref.8).The sizing of the VMS lateral-travel envelope and the ability to orient the cockpit with either the lateral or longitudinal axis were based on these findings.The Six DOF Motion Simulator was the first flight simulator to use equilibrators, instead of counterweights to help offset gravitational effects and, thereby, improve vertical dynamic performance (Ref.6).It became operational in 1964 and had ±9 ft. of travel in all translational axes and ±45 degrees in all rotational axes.Experience with the equilibrators led to the improved design used in the VMS. +Figure 4 .4Figure 4.Primary Influences on the VMS Motion System Design +Figure 6 .6Figure 6.Automated NOE cockpit and test course +Figure 7 .7Figure 7. Civil tilt-rotor cockpit (with HUD) and typical visual scene + + +Table 1 . VMS motion system performance limits (from Ref. 3) Displacement Velocity Acceleration Degree of Freedom System Limits Operational Limits System Limits Operational Limits System Limits1OperationalLimits +Table 2 . Rotorcraft handling qualities simulation studies2Simulation DescriptionYear(s)Heavy-Lift Rotorcraft (HLR) stability margin/handling qualities2008Handling qualities with external slung loads (8 simulations)1994 -2008Civil handling qualities specification2001LHX/RAH 66 handling qualities (4 simulations)1982 -1992Helicopter cross-coupling studies (3 simulations)1986 -1994Handling qualities for helicopter air combat (6 simulations)1984 -1992Handling qualities for yaw control1984Handling qualities for roll control (2 simulations)1984, 1985Single/dual pilot advanced cockpit and handling qualities (2 simulations)1986Handling qualities for shipboard landing1984Handling qualities for vertical control1983-1984Effect of rotorcraft design on handling qualities (3 simulations)1980-1982 +Table 3 . Flight control systems research simulations3Simulation DescriptionYear(s)CH-47F digital automatic flight control system (2 simulations)2004 -2005UH-60 MCLAWS evaluation2002Partial authority flight control systems (4 simulations)1991 -1998Advanced Digital Optical Control System (ADOCS) program (4 simulations)1981 -1985 +Table 4 . Simulation studies supporting specific rotorcraft programs438)orsky study on the VMS compared Variable Diameter Tilt Rotor (VDTR) concept with a conventional fixed diameter tilt-rotor (Ref.38).Pilots from government and industry evaluated the VDTR in regular and emergency (engine failure) operations.The study demonstrated the enhanced performance potential of the VDTR and identified areas for further study.Simulation DescriptionYear(s)RASCAL safety systems (3 simulations)1989 -2008Joint Shipboard Helicopter Integration Process (JSHIP) (2 simulations)2000 -2001RAH-66 Comanche1997Tilt-Wing/Advanced Theater Transport (ATT) (4 simulations)1991 -2002Variable Diameter Tilt Rotor (VDTR)1996AH-64 Apache (4 simulations)1988 -1993Piasecki VTCAD (2 simulations)1991 -1993X-Wing/RSRA (4 simulations)1984 -1987SH-2F1982A joint NASA/ +Table 5 . Simulation fidelity research5Simulation DescriptionYear(s)Visual scene height perception -PsychoPath (2 simulations)2001 -2006Autorotation Cues -AutoCue (2 simulations)2000 -2001Computation situational awareness model --SAMSIM2000Simulation fidelity requirements (7 simulations)1996 -1999Motion and visual evaluation -MOTIVE1993Simulation validation -SIMVAL (3 simulations)1990 -1993Visual/motion synchronization -VISMOSYNC (2 simulations)1990 -1992Visual and motion delay -SIMVAC1992Blackhawk validation1989Simulator sickness study1989Helicopter autorotation (2 simulations)1984 -1985 + + + + + + + + + Rotorcraft Handling-Qualities Design Criteria Development + + EWAiken + + + VJLebacqz + + + RTChen + + + DLKey + + SEE N88-16632 09-01 + + + Materials and Structures, Propulsion and Drive Systems, Flight Dynamics and Control, and Acoustics + Washington + + 1988 + 2 + + + + NASA + Aiken, E. W., Lebacqz, V. J., Chen, R. T., and Key, D. L., "Rotorcraft Handling-Qualities Design Criteria Development," NASA/Army Rotorcraft Technology. Volume 2: Materials and Structures, Propulsion and Drive Systems, Flight Dynamics and Control, and Acoustics, SEE N88-16632 09-01; NASA, Washington, 1988, p 948-998. + + + + + Optimizing the Performance of the Pilot Control Loaders at the NASA Vertical Motion Simulator + + RAMueller + + + + AIAA Paper 2008-6349, AIAA Modeling and Simulation Technologies Conference + Honolulu, HI + + Aug. 2008 + + + Mueller, R. A., "Optimizing the Performance of the Pilot Control Loaders at the NASA Vertical Motion Simulator," AIAA Paper 2008-6349, AIAA Modeling and Simulation Technologies Conference, Honolulu, HI, Aug. 2008. + + + + + Vertical Motion Simulator Familiarization Guide + + GeorgeLDanek + + + May 1993 + + + NASA TM 103923 + Danek, George L., "Vertical Motion Simulator Familiarization Guide," NASA TM 103923, May 1993. + + + + + Effects of Simulator Motion and Visual Characteristics on Rotorcraft Handling Qualities + + DGMitchell + + + DCHart + + + + American Helicopter Society Conference on Piloting Vertical Flight Aircraft + San Francisco, CA + + Jan. 1993 + + + Mitchell, D.G. and Hart, D.C., "Effects of Simulator Motion and Visual Characteristics on Rotorcraft Handling Qualities," American Helicopter Society Conference on Piloting Vertical Flight Aircraft, San Francisco, CA, Jan. 1993. + + + + + Identification and Simulation Evaluation of an AH-64 Helicopter Hover Math Model + + JASchroeder + + + DCWatson + + + MBTischler + + + MMEshow + + + + AIAA Paper 91-2877, AIAA Atmospheric Flight Mechanics Conference + New Orleans, LA + + Aug. 1991 + + + Schroeder, J. A., Watson, D. C., Tischler, M. B., and Eshow, M. M., "Identification and Simulation Evaluation of an AH-64 Helicopter Hover Math Model," AIAA Paper 91-2877, AIAA Atmospheric Flight Mechanics Conference, New Orleans, LA, Aug. 1991. + + + + + Historical Review of Piloted Simulation at NASA Ames + + SethBAnderson + + + + Proceedings of the AGARD FVP Symposium on "Flight Simulation -Where are the Challenges? + the AGARD FVP Symposium on "Flight Simulation -Where are the Challenges?Braunschweig, Germany + + + Anderson, Seth B., "Historical Review of Piloted Simulation at NASA Ames," Proceedings of the AGARD FVP Symposium on "Flight Simulation -Where are the Challenges?" Braunschweig, Germany. + + + + + Vertical Motion Requirements for Landing Simulation + + RichardSBray + + + Feb 1973 + 236 + + + NASA TM X-62 + Bray, Richard S., "Vertical Motion Requirements for Landing Simulation," NASA TM X-62,236, Feb 1973. + + + + + The Determination of Some Requirements for a Helicopter Flight Research Simulation Facility + + JBSinacori + + Inc. TR 1097-1 + + + Systems Technology + + Sep. 1977 + + + Sinacori, J. B., "The Determination of Some Requirements for a Helicopter Flight Research Simulation Facility," Systems Technology, Inc. TR 1097-1, Sep. 1977. + + + + + Handling Qualities Specifications for U.S. Military Helicopters + + DLKey + + + + AIAA Journal of Aircraft + + 19 + 2 + Feb. 1982 + + + Key, D. L., "Handling Qualities Specifications for U.S. Military Helicopters," AIAA Journal of Aircraft, Vol. 19, no. 2, Feb. 1982. + + + + + Handling Qualities Requirements for Military Rotorcraft + + Anon + + + + Aeronautical Design Standard-33 (ADS-33E-PRF), US Army Aviation and Missile Command + + Mar. 2000 + + + Anon., "Handling Qualities Requirements for Military Rotorcraft," Aeronautical Design Standard-33 (ADS-33E-PRF), US Army Aviation and Missile Command, Mar. 2000. + + + + + Effects of Rotor Inertia and RPM Control on Helicopter Handling Qualities + + LDCorliss + + + CLBlanken + + + KNelson + + + + AIAA Paper 1983-2070, AIAA Atmospheric Flight Mechanics Conference + Gatlinburg, TN + + Aug. 1983 + + + Corliss, L. D., Blanken, C. L., and Nelson, K., "Effects of Rotor Inertia and RPM Control on Helicopter Handling Qualities," AIAA Paper 1983-2070, AIAA Atmospheric Flight Mechanics Conference, Gatlinburg, TN, Aug. 1983. + + + + + An Investigation of the Effects of Pitch-Roll Cross Coupling on Helicopter Handling Qualities for Terrain Flight + + DCWatson + + + EWAiken + + + + AIAA Conference on Guidance, Navigation, and Control + Monterey, CA + + Aug. 1987 + + + Watson, D. C., and Aiken, E. W., "An Investigation of the Effects of Pitch-Roll Cross Coupling on Helicopter Handling Qualities for Terrain Flight," AIAA Conference on Guidance, Navigation, and Control, Monterey, CA., Aug. 1987. + + + + + An Investigation of the Effects of Pitch-Roll (De)coupling on Helicopter Handling Qualities + + CLBlanken + + + HJPausder + + + CJOckier + + NASA-TM-110349 + + May 1995 + + + Blanken, C. L., Pausder, H. J., and Ockier, C. J., "An Investigation of the Effects of Pitch-Roll (De)coupling on Helicopter Handling Qualities," NASA-TM-110349, May 1995. + + + + + A Simulation Investigation of Scout/Attack Helicopter Directional Control Requirements for Hover and Low-Speed Tasks + + CCBivens + + + JGGuercio + + NASA-TM-86755 + + Mar. 1987 + + + Bivens, C. C., and Guercio, J. G., "A Simulation Investigation of Scout/Attack Helicopter Directional Control Requirements for Hover and Low-Speed Tasks," NASA-TM-86755, Mar. 1987. + + + + + A Piloted Simulation Investigation of Yaw Attitude Quickness in Hover and Yaw Bandwidth in Forward Flight + + MSWhalley + + + + NASA TM-103948 + + Sep. 1992 + + + Whalley, M. S., "A Piloted Simulation Investigation of Yaw Attitude Quickness in Hover and Yaw Bandwidth in Forward Flight," NASA TM-103948, Sep. 1992. + + + + + + RKHeffley + + + SMBourne + + + HCCurtiss + + + Jr + + + WSHindson + + + RAHess + + NASA-CR-177404 + Study of Helicopter Roll Control Effectiveness Criteria; Final Report + + Apr. 1986 + + + Heffley, R. K., Bourne, S. M., Curtiss, H. C., Jr., Hindson, W. S., and Hess, R. A., "Study of Helicopter Roll Control Effectiveness Criteria; Final Report," NASA-CR-177404, Apr. 1986. + + + + + Helicopter Control Response Types for Hover and Low-Speed Near-Earth Tasks in Degraded Visual Conditions + + CLBlanken + + + DCHart + + + RHHoh + + + + AHS 47th Annual Forum + Phoenix, AZ + + May 1991 + + + Blanken, C. L., Hart, D. C., and Hoh, R. H., "Helicopter Control Response Types for Hover and Low- Speed Near-Earth Tasks in Degraded Visual Conditions," AHS 47th Annual Forum, Phoenix, AZ, May 1991. + + + + + An Investigation of the Use of Bandwidth Criteria for Rotorcraft Handling Qualities Specifications + + CLBlanken + + + CCBivens + + + MSWhalley + + + + AHS International Conference on Rotorcraft Basic Research + Research Triangle Park, NC + + Feb. 1985 + + + Blanken, C. L., Bivens, C. C., and Whalley, M. S., " An Investigation of the Use of Bandwidth Criteria for Rotorcraft Handling Qualities Specifications," AHS International Conference on Rotorcraft Basic Research, Research Triangle Park, NC, Feb. 1985. + + + + + Investigation of the Effects of Bandwidth and Time Delay on Helicopter Roll-Axis Handling Qualities + + Heinz-JuergenPausder + + + CLBlanken + + + + Piloting Vertical Flight Aircraft: A Conference on Flying Qualities and Human Factors (SEE N94-13294 02-08) + + Jul.1993 + + + + Pausder, Heinz-Juergen, and Blanken, C. L., "Investigation of the Effects of Bandwidth and Time Delay on Helicopter Roll-Axis Handling Qualities," NASA Ames Research Center, Piloting Vertical Flight Aircraft: A Conference on Flying Qualities and Human Factors (SEE N94-13294 02-08), Jul.1993, p 91-110. + + + + + Piloted Simulation of One-on-One Helicopter Air Combat at NOE Flight Levels + + MSLewis + + + EWAiken + + NASA-TM-86686 + + Apr. 1985 + + + Lewis, M. S., and Aiken, E. W., "Piloted Simulation of One-on-One Helicopter Air Combat at NOE Flight Levels," NASA-TM-86686, Apr. 1985. + + + + + A Piloted Simulation of Helicopter Air Combat to Investigate Effects of Variations in Selected Performance and Control Response Characteristics + + MSLewis + + + MHMansur + + + RTChen + + USAAVSCOM TM-87-A-3 + + Aug. 1987 + + + Lewis, M. S., Mansur, M. H., and Chen, R. T., "A Piloted Simulation of Helicopter Air Combat to Investigate Effects of Variations in Selected Performance and Control Response Characteristics," USAAVSCOM TM-87-A-3, Aug. 1987. + + + + + A Piloted Simulation Investigation of Forward Flight Handling Qualities Requirements for Helicopter Air-to-Air Combat + + MSWhalley + + + WRCarpenter + + + + NASA TM-103919 + + May 1992 + + + Whalley, M. S., and Carpenter, W. R., "A Piloted Simulation Investigation of Forward Flight Handling Qualities Requirements for Helicopter Air-to-Air Combat," NASA TM-103919, May 1992. + + + + + A Piloted Simulation Investigation of Yaw Dynamics Requirements for Turreted Gun Use in Low-Level Helicopter Air Combat + + WMDecker + + + PMMorris + + + JNWilliams + + + + 44th AHS Annual Forum + + June 1988 + Washington, DC + + + Decker, W. M., Morris, P. M., and Williams, J. N., "A Piloted Simulation Investigation of Yaw Dynamics Requirements for Turreted Gun Use in Low-Level Helicopter Air Combat," 44th AHS Annual Forum, Washington, DC, June 1988. + + + + + Advanced Helicopter Cockpit And Control Configurations for Helicopter Combat Missions + + LAHaworth + + + AAtencio + + + Jr + + + CBivens + + + RShively + + + DDelgado + + + Dec. 1987 + + + NASA-TM-100017 + Haworth, L. A., Atencio, A., Jr., Bivens, C., Shively, R., and Delgado, D., "Advanced Helicopter Cockpit And Control Configurations for Helicopter Combat Missions," NASA-TM-100017, Dec. 1987. + + + + + Development of Handling Qualities Criteria for Rotorcraft with Externally Slung Loads + + RogerHHoh + + + RobertKHeffley + + + DavidGMitchell + + + + NASA CR + + 2006-213488, Oct. 2006 + + + Hoh, Roger H.; Heffley, Robert K.; and Mitchell, David G., "Development of Handling Qualities Criteria for Rotorcraft with Externally Slung Loads," NASA CR-2006-213488, Oct. 2006. + + + + + Development of ADOCS Controllers and Control laws + + KHLandis + + + SIGlusman + + NASA-CR-177339 + + Mar. 1985 + 1 + + + Executive Summary + Landis, K. H., and Glusman, S. I., "Development of ADOCS Controllers and Control laws, Volume 1: Executive Summary," NASA-CR-177339, Mar. 1985. + + + + + Evaluation of ADOCS Demonstrator Handling Qualities; Advanced Digital Optical Control System + + SIGlusman + + + CDabundo + + + KHLandis + + + + AHS, 43rd Annual Forum + Saint Louis, MO + + May 1987 + + + Glusman, S. I., Dabundo, C., and Landis, K. H., "Evaluation of ADOCS Demonstrator Handling Qualities; Advanced Digital Optical Control System," AHS, 43rd Annual Forum, Saint Louis, MO, May 1987. + + + + + Optimizations of Partial Authority Automatic Flight Control Systems for Hover/Low-Speed Maneuvering in Degraded Visual Environments + + MWhalley + + + JHowitt + + + + Journal of the American Helicopter Society + + 47 + 2 + Apr. 2002 + + + Whalley, M., and Howitt, J., "Optimizations of Partial Authority Automatic Flight Control Systems for Hover/Low-Speed Maneuvering in Degraded Visual Environments," Journal of the American Helicopter Society, Vol. 47, No. 2, Apr. 2002. + + + + + + DLKey + + + RKHeffley + + NASA-CR-2002-211391 USAAMCOM-TR-02-A- 003 + Piloted Simulator Investigation of Techniques to Achieve Attitude Command Response with Limited Authority Servos + + Jan. 2002 + + + Key, D. L., and Heffley, R. K., Piloted Simulator Investigation of Techniques to Achieve Attitude Command Response with Limited Authority Servos, NASA-CR-2002-211391 USAAMCOM-TR-02-A- 003; Jan. 2002. + + + + + Increased Stabilization for UH-60A Blackhawk Night Operations + + DGMitchell + + + BLAponso + + + AAtencio + + + DLKey + + + RHHoh + + USAAVSCOM TR-92-A-007 + + Nov. 1992 + + + Mitchell, D. G., Aponso, B. L., Atencio, A., Key, D. L., and Hoh, R. H., "Increased Stabilization for UH- 60A Blackhawk Night Operations," USAAVSCOM TR-92-A-007, Nov. 1992. + + + + + Piloted Evaluation of Modernized Limited Authority Control Laws in the NASA-Ames Vertical Motion Simulator (VMS) + + VSahasrabudhe + + + EMelkers + + + Faynberg + + + CLBlanken + + + + AHS 59 th Annual Forum + Phoenix, AZ + + May 2003 + + + Sahasrabudhe, V., Melkers, E., Faynberg, and Blanken, C. L., "Piloted Evaluation of Modernized Limited Authority Control Laws in the NASA-Ames Vertical Motion Simulator (VMS)," AHS 59 th Annual Forum, Phoenix, AZ, May 2003. + + + + + ADS-33E Predicted and Assigned Low-Speed Handling Qualities of the CH-47F with Digital AFCS + + JGIrwin + + + PGEinthoven + + + DGMiller + + + CLBlanken + + + + AHS 63 rd Annual Forum + Virginia Beach, VA + + May 2007 + + + Irwin, J. G., Einthoven, P. G., Miller, D. G., and Blanken, C. L., "ADS-33E Predicted and Assigned Low- Speed Handling Qualities of the CH-47F with Digital AFCS," AHS 63 rd Annual Forum, Virginia Beach, VA, May 2007. + + + + + DIMSS -JSHIPs Modeling and Simulation Process for Ship/Helicopter Testing and Training + + MFRoscoe + + + CHWilkinson + + + + AIAA Modeling and Simulation Technologies Conference and Exhibit + Monterey, CA + + Aug. 2002 + + + Roscoe, M. F., and Wilkinson, C. H., "DIMSS -JSHIPs Modeling and Simulation Process for Ship/Helicopter Testing and Training," AIAA Modeling and Simulation Technologies Conference and Exhibit, Monterey, CA, Aug. 2002. + + + + + Simulation and Evaluation of the SH-2F Helicopter in a Shipboard Environment Using the Interchangeable Cab System + + CHPaulk + + + Jr + + + DLAstill + + + STDonley + + NASA-TM-84387 + + Aug. 1983 + + + Paulk, C. H., Jr., Astill, D. L., and Donley, S. T., "Simulation and Evaluation of the SH-2F Helicopter in a Shipboard Environment Using the Interchangeable Cab System," NASA-TM-84387, Aug. 1983. + + + + + Initial Piloted Simulation Study of Geared Flap Control for Tilt-Wing V/STOL Aircraft + + LMGuerrero + + + Corliss + + + LD + + + + NASA TM-103872 + + Oct. 1991 + + + Guerrero, L. M., and Corliss, L. D., "Initial Piloted Simulation Study of Geared Flap Control for Tilt-Wing V/STOL Aircraft," NASA TM-103872, Oct. 1991. + + + + + Evaluation of Flying Qualities and Guidance Displays for an Advanced Tilt-Wing STOL Transport Aircraft in Final Approach and Landing + + CRFrost + + + JAFranklin + + + GHHardy + + + + Biennial International Powered Lift Conference and Exhibit + Williamsburg, Virginia + + 2002. Nov. 2002 + + + AIAA Paper 2002-6016 + Frost, C. R., Franklin, J. A., and Hardy, G. H., Evaluation of Flying Qualities and Guidance Displays for an Advanced Tilt-Wing STOL Transport Aircraft in Final Approach and Landing," AIAA Paper 2002-6016, 2002 Biennial International Powered Lift Conference and Exhibit, Williamsburg, Virginia, Nov. 2002. + + + + + RSRA/X-Wing Flight Control System Development -Lessons Learned + + LDCorliss + + + WRDunn + + + MAMorrison + + + + AHS 45th Annual Forum + + May 1989 + Boston, MA + + + Corliss, L. D., Dunn, W. R.; and Morrison, M. A., "RSRA/X-Wing Flight Control System Development - Lessons Learned," AHS 45th Annual Forum, Boston, MA, May 1989. + + + + + Simulation of a Variable Diameter Tiltrotor + + KStudebaker Fletcher + + + WADecker + + + DCMatuska + + + PMMorris + + + MTSmith + + + Vms + + + + AHS 53 rd Annual Forum + Virginia Beach, VA + + May 1992 + + + Studebaker Fletcher, K., Decker, W. A., Matuska, D. C., Morris, P.M., and Smith, M. T, VMS Simulation of a Variable Diameter Tiltrotor," AHS 53 rd Annual Forum, Virginia Beach, VA, May 1992. + + + + + + MHMansur + + + JASchroeder + + An Investigation of the Ability to Recover from Transients Following Failures for Single-Pilot Rotorcraft + + May1988 + + + NASA-TM-100078 + Mansur, M. H., and Schroeder, J. A., "An Investigation of the Ability to Recover from Transients Following Failures for Single-Pilot Rotorcraft," NASA-TM-100078, May1988. + + + + + A Pilot Rating Scale for Evaluating Failure Transients in Electronic Flight Control Systems + + WSHindson + + + MMEshow + + + JASchroeder + + + + AIAA Paper 90-2827, AIAA Atmospheric Flight Mechanics Conference + Portland, OR + + Aug. 1990 + + + Hindson, W. S., Eshow, M. M., and Schroeder, J. A., "A Pilot Rating Scale for Evaluating Failure Transients in Electronic Flight Control Systems," AIAA Paper 90-2827, AIAA Atmospheric Flight Mechanics Conference, Portland, OR, Aug. 1990. + + + + + Assessment of Simulation Fidelity Using Measurements of Piloting Technique in Flight + + WFClement + + + WBCleveland + + + DLKey + + + + AGARD Conference Proceedings No + + 359 + May 1984 + Monterey, CA + + + Clement, W.F., Cleveland, W.B., and Key, D.L., "Assessment of Simulation Fidelity Using Measurements of Piloting Technique in Flight, AGARD Conference Proceedings No 359, Monterey, CA, May 1984. + + + + + Fidelity Assessment of a UH-60A Simulation on the NASA Ames Vertical Motion Simulator + + AAtencioJr + + 93-A-005 + + + NASA TM 104016, USAATC + + Sept. 1993 + + + Tech. Report + Atencio, Jr., A. "Fidelity Assessment of a UH-60A Simulation on the NASA Ames Vertical Motion Simulator," NASA TM 104016, USAATC Tech. Report 93-A-005, Sept. 1993. + + + + + Visual and Motion Cuing in Helicopter Simulation + + RSBray + + + Sept. 1985 + + + NASA TM-86818 + Bray, R. S., "Visual and Motion Cuing in Helicopter Simulation," NASA TM-86818, Sept. 1985. + + + + + The Effects of Simulator Visual-Motion Asynchrony on Simulator Induced Sickness + + MEMccauley + + + LJHettinger + + + TJSharkey + + + JBSinacori + + + + AIAA Flight Simulation Technologies Conference and Exhibit + Dayton, OH + + Sept. 1990 + + + McCauley, M.E., Hettinger, L.J., Sharkey, T.J., and Sinacori, J.B., "The Effects of Simulator Visual- Motion Asynchrony on Simulator Induced Sickness," AIAA Flight Simulation Technologies Conference and Exhibit, Dayton, OH, Sept. 1990. + + + + + Does a Motion Base Prevent Simulator Sickness? + + TJSharkey + + + MEMccauley + + + + AIAA/AHS Flight Simulation Technologies Conference + Hilton Head, SC + + Aug. 1992 + + + Sharkey, T.J. and McCauley, M.E., "Does a Motion Base Prevent Simulator Sickness?" AIAA/AHS Flight Simulation Technologies Conference, Hilton Head, SC, Aug. 1992. + + + + + Visual and Roll-Lateral Motion Cueing Synchronization Requirements for Motion-Based Flight Simulations + + WWChung + + + JASchroeder + + + + 53 rd Annual Forum Proceedings + Virginia Beach, VA + + American Helicopter Society + Apr. 1997 + + + Chung, W.W. and Schroeder, J.A., "Visual and Roll-Lateral Motion Cueing Synchronization Requirements for Motion-Based Flight Simulations" American Helicopter Society 53 rd Annual Forum Proceedings, Virginia Beach, VA, Apr. 1997. + + + + + Ground Based Simulation Evaluation of the Effects of Time Delays and Motion on Rotorcraft Handling Qualities + + DGMitchell + + + RHHoh + + + AAtencioJr + + + DLKey + + AVSCOM-TR-91-A-010 + + Aug. 1990 + + + Mitchell, D.G., Hoh, R.H., Atencio Jr., A., and Key, D.L., "Ground Based Simulation Evaluation of the Effects of Time Delays and Motion on Rotorcraft Handling Qualities," AVSCOM-TR-91-A-010, Aug. 1990. + + + + + Simulation Motion Effect on Single Axis Compensatory Tracking + + JASchroeder + + + + AIAA Flight Simulation Technologies Conference + Monterey, CA + + 1993 + + + Schroeder, J.A. "Simulation Motion Effect on Single Axis Compensatory Tracking," AIAA Flight Simulation Technologies Conference, Monterey, CA, 1993. + + + + + Evaluation of Simulation Motion Fidelity Criteria in the Vertical and Directional Axes + + JASchroeder + + + + Journal of the American Helicopter Society + + 41 + 2 + + Apr. 1996 + + + Schroeder, J.A., "Evaluation of Simulation Motion Fidelity Criteria in the Vertical and Directional Axes," Journal of the American Helicopter Society, Vol. 41, No. 2, Apr. 1996, pp. 44-57. + + + + + AGARD CP-577, Flight Simulation -Where are the Challenges? + + JASchroeder + + + WWJohnson + + + Apr. 1996 + Braunschweig, Germany + + + Yaw Motion Cues in Helicopter Simulation + Schroeder, J.A. and Johnson, W.W., "Yaw Motion Cues in Helicopter Simulation," Paper No. 5, AGARD CP-577, Flight Simulation -Where are the Challenges?," Braunschweig, Germany, Apr. 1996. + + + + + Visual-Motion Cueing in Altitude and Yaw Control + + WJohnson + + + JASchroeder + + + + 38 th Annual Human Factors and Ergonomics Society Meeting + Nashville, TN + + Oct. 1994 + + + Johnson, W. and Schroeder, J.A., "Visual-Motion Cueing in Altitude and Yaw Control," 38 th Annual Human Factors and Ergonomics Society Meeting, Nashville, TN, Oct. 1994. + + + + + Evaluation of a Motion Fidelity Criterion with Visual Scene Changes + + JASchroeder + + + WW YChung + + + RAHess + + + + Journal of Aircraft + + 37 + 4 + + July-August, 2000 + + + Schroeder, J.A., Chung, W.W.Y., and Hess, R.A., "Evaluation of a Motion Fidelity Criterion with Visual Scene Changes," Journal of Aircraft, Vol. 37, No. 4, July-August, 2000, pp. 580-587. + + + + + Modeling Human Pilot Cue Utilization with Applications to Simulator Fidelity Assessment + + YZeyada + + + RAHess + + + + Journal of Aircraft + + 37 + 4 + + July-Aug., 2000 + + + Zeyada, Y. and Hess, R.A., "Modeling Human Pilot Cue Utilization with Applications to Simulator Fidelity Assessment," Journal of Aircraft, Vol. 37, No. 4, July-Aug., 2000, pp. 588-597. + + + + + Pilot Use of Simulator cues for Autorotation Landings + + WADecker + + + CFAdam + + + RMGerdes + + + June 1986 + American Helicopter Society's 42 nd Annual Forum + Washington, DC + + + Decker, W.A., Adam, C.F., and Gerdes, R.M., "Pilot Use of Simulator cues for Autorotation Landings," American Helicopter Society's 42 nd Annual Forum, Washington, DC, June 1986. + + + + + Effects of Visual Texture, Grids, and Platform Motion on Unpowered Helicopter Landings + + MGDearing + + + JASchroeder + + + BTSweet + + + MKKaiser + + + + AIAA Modeling and Simulation Conference + Montreal, Canada + + 2001 + + + Paper 2001-4251 + Dearing, M.G., Schroeder, J.A., Sweet, B.T., and Kaiser, M.K., "Effects of Visual Texture, Grids, and Platform Motion on Unpowered Helicopter Landings," Paper 2001-4251, AIAA Modeling and Simulation Conference, Montreal, Canada, 2001. + + + + + Runway Texture and Grid Pattern Effects on Rate-of-Descent Perception + + JASchroeder + + + MGDearing + + + BTSweet + + + MKKaiser + + + AAAtencioJr + + + + Paper 2001-4307, AIAA Modeling and Simulation Conference + Montreal, Canada + + 2001 + + + Schroeder, J.A., Dearing, M.G., Sweet, B.T., Kaiser, M.K., and Atencio, Jr., A.A., "Runway Texture and Grid Pattern Effects on Rate-of-Descent Perception," Paper 2001-4307, AIAA Modeling and Simulation Conference, Montreal, Canada, 2001. + + + + + Simulation Evaluation of Display/FLIR Concepts for Low-Altitude Terrain-Following Helicopter Operations + + HNSwenson + + + CHPaulk + + + RLKilmer + + + FGKilmer + + NASA TM-86779 + + 1985 + + + Swenson, H. N., Paulk, C. H., Kilmer, R. L., and Kilmer, F. G., "Simulation Evaluation of Display/FLIR Concepts for Low-Altitude Terrain-Following Helicopter Operations," NASA TM-86779, 1985. + + + + + Rotary Wing Aircraft Terrain Following/Terrain Avoidance System Development + + DWDorr + + + + NASA TM-88322 + + 1986 + + + Dorr, D. W., "Rotary Wing Aircraft Terrain Following/Terrain Avoidance System Development," NASA TM-88322, 1986. + + + + + Simulation Evaluation of a Low-Altitude Helicopter Flight Guidance System Adapted for a Helmet-Mounted Display + + HNSwenson + + + REZelenka + + + GHHardy + + + MGDearing + + NASA TM-103883 + + Feb. 1992 + + + Swenson, H. N., Zelenka, R. E., Hardy, G. H., and Dearing, M. G., "Simulation Evaluation of a Low- Altitude Helicopter Flight Guidance System Adapted for a Helmet-Mounted Display," NASA TM-103883, Feb. 1992. + + + + + A Comparison of Active Sidestick and Conventional Inceptors for Helicopter Flight Envelope Tactical Cueing + + MSWhalley + + + + AHS 56 th Annual Forum + Virginia Beach, VA + + May 2000 + + + Whalley, M. S., "A Comparison of Active Sidestick and Conventional Inceptors for Helicopter Flight Envelope Tactical Cueing," AHS 56 th Annual Forum, Virginia Beach, VA, May 2000. + + + + + Helicopter Active Control Technology Demonstrator Program + + MSWhalley + + + JKeller + + + RBuckanin + + + JRoos + + + + AHS 57 th Annual Forum + + May 2001 + Washington, DC + + + Whalley, M. S., Keller, J., Buckanin, R., and Roos, J., "Helicopter Active Control Technology Demonstrator Program," AHS 57 th Annual Forum, Washington, DC, May 2001. + + + + + Simulator Investigation of Pilot Aids for Helicopter Terminal Area Operations with One Engine Inoperative + + LIseler + + + RChen + + + MDearing + + + WDecker + + + EWAiken + + + + AGARD Flight Vehicle Integration Panel Spring 1996 Symposium + Ottawa, Canada + + May 1996 + + + Iseler, L., Chen, R., Dearing, M., Decker, W., and Aiken, E. W.; "Simulator Investigation of Pilot Aids for Helicopter Terminal Area Operations with One Engine Inoperative," AGARD Flight Vehicle Integration Panel Spring 1996 Symposium, Ottawa, Canada, May 1996. + + + + + Improvements in Hover Display Dynamics for a Combat Helicopter + + MMEshow + + + JASchroeder + + + + Piloting Vertical Flight Aircraft: A Conference on Flying Qualities and Human Factors + San Francisco, CA + + 1993 + + + Eshow, M. M., and Schroeder, J. A., "Improvements in Hover Display Dynamics for a Combat Helicopter," Piloting Vertical Flight Aircraft: A Conference on Flying Qualities and Human Factors, San Francisco, CA, 1993. + + + + + + TSTurpin + + + SRDowell + + + ZSzoboszlay + + + + Comanche Helmet-Mounted Display Symbology Simulation; Final Report + + 2006-212834, Oct. 2006 + + + Turpin, T. S., Dowell, S. R., and Szoboszlay, Z., "Comanche Helmet-Mounted Display Symbology Simulation; Final Report," NASA CR 2006-212834, Oct. 2006. + + + + + Evaluation of Helmet Mounted Display Alerting Symbology + + JDemaio + + + MRutkowski + + + + NASA TM + + 2000-209603, Sep. 2000 + + + DeMaio, J., and Rutkowski, M., "Evaluation of Helmet Mounted Display Alerting Symbology," NASA TM 2000-209603, Sep. 2000. + + + + + Piloted Simulator Investigations of a Civil Tilt-Rotor Aircraft on Steep Instrument Approaches + + WADecker + + + + American Helicopter Society 48 th Annual Forum + Washington, D.C. + + Jun. 1992 + + + Decker, W.A., "Piloted Simulator Investigations of a Civil Tilt-Rotor Aircraft on Steep Instrument Approaches," American Helicopter Society 48 th Annual Forum, Washington, D.C., Jun. 1992. + + + + + Evaluation of Two Cockpit Display Concepts for Civil Tiltrotor Instrument Operations on Steep Approaches + + WADecker + + + RBray + + + RCSimmons + + + GETucker + + + + Piloting Vertical Flight Aircraft: A Conference on Flying Qualities and Human Factors + San Francisco, CA + + American Helicopter Society + Jan. 1993 + + + Decker, W.A., Bray, R., Simmons, R.C., and Tucker, G.E., "Evaluation of Two Cockpit Display Concepts for Civil Tiltrotor Instrument Operations on Steep Approaches," American Helicopter Society Piloting Vertical Flight Aircraft: A Conference on Flying Qualities and Human Factors, San Francisco, CA, Jan. 1993. + + + + + Handling Qualities Evaluation of XV-15 Noise Abatement Landing Approaches Using a Flight Simulator + + WADecker + + + + 57 th Annual Forum + Washington, D.C. + + American Helicopter Society + May 2001 + + + Decker, W.A., "Handling Qualities Evaluation of XV-15 Noise Abatement Landing Approaches Using a Flight Simulator," American Helicopter Society 57 th Annual Forum, Washington, D.C., May 2001. + + + + + Pursuit Display Review and Extension to a Civil Tilt Rotor Flight Director + + GHHardy + + + + AIAA Paper 2002-4925, AIAA Guidance, Navigation, and Control Conference and Exhibit + Monterey, CA + + Aug. 2002 + + + Hardy, G.H., "Pursuit Display Review and Extension to a Civil Tilt Rotor Flight Director," AIAA Paper 2002-4925, AIAA Guidance, Navigation, and Control Conference and Exhibit, Monterey, CA, Aug. 2002. + + + + + Civil Tiltrotor Transport Procedure and Requirement Development Using a Flight Simulator + + WADecker + + + GHHardy + + + + AIAA Paper 2002-4530, AIAA Guidance, Navigation, and Control Conference and Exhibit + Monterey, CA + + Aug. 2002 + + + Decker, W.A., and Hardy, G.H., "Civil Tiltrotor Transport Procedure and Requirement Development Using a Flight Simulator," AIAA Paper 2002-4530, AIAA Guidance, Navigation, and Control Conference and Exhibit, Monterey, CA, Aug. 2002. + + + + + + WADecker + + + DIsleib + + + Major + + + JJohn + + A Simulator Investigation of Air-to-Air Combat Maneuvering for Tilt-Rotor Aircraft + New Bern, NC + + Sep. 1989 + + + AHS National Technical Specialists' Meeting on Tactical V/STOL + Decker, W. A., Isleib, D., Major, and John, J., "A Simulator Investigation of Air-to-Air Combat Maneuvering for Tilt-Rotor Aircraft," AHS National Technical Specialists' Meeting on Tactical V/STOL, New Bern, NC, Sep. 1989. + + + + + + diff --git a/file17.txt b/file17.txt new file mode 100644 index 0000000000000000000000000000000000000000..6f088606c72a865a77b7a909ccce596ed4b403b7 --- /dev/null +++ b/file17.txt @@ -0,0 +1,387 @@ + + + + +I. IntroductionCurrently, traffic management coordinators (TMCs) react to flow constraints in the National Airspace System (NAS) by implementing air traffic management initiatives in order to safely and efficiently route air traffic through the NAS.One method of handling traffic is to reroute air traffic around regions of airspace affected by convective weather.Acceptable reroutes are issued by TMCs based on their understanding of weather conditions and their previous experience dealing with similar conditions.However, some of this responsibility of suggesting reroutes is now being given to airlines.A new system for pre-departure rerouting, Collaborative Trajectory Options Program CTOP, is currently in limited use. 1 This program allows airlines to submit multiple preferred routes to the FAA through a trajectory option set (TOS) when severe weather will affect their flights' planned routes.A decision support tool that uses historical data to inform future decisions along with metrics of efficiency will benefit both the FAA and airlines as they develop and evaluate their rerouting solutions.Previous and ongoing work is focused on finding similarities between days based on weather conditions or NAS performance metrics. 2,3,4 Wth a vast amount of weather information, the challenge here is to condense the data in such a way that operationally relevant features are preserved (or highlighted) in order to compare days.Operationally relevant features are often difficult to determine and depend heavily on the objective.In Ref. 4, days are clustered base on the Weather Impacted Traffic Index (WITI) values at each Air Route Traffic Control Center (ARTCC or Center).Once similar days are discovered, the use of particular advisory reroutes used on similar days are quantified.Previous research has also involved creating or selecting reroutes for flights which avoid airspace affected by severe weather, for example Refs.5, 6, and 7.However, little work has been done to predict the use of specific routes in the presence of convective weather.Here, rather than identifying similar scenarios based on weather conditions, we apply reroute usage as labels for supervised learning models and allow the model to discover patterns in the weather data indicating the use of a particular reroute.This work is a stepping stone in the development of a tool that could be used to inform the TMC and airline reroute decision process.This paper describes and shows the results of determining if a flight followed a reroute and the associated convective weather scenario.A distance metric was used to determine if flight followed a specific reroute.The reroutes are then clustered using the k -means algorithm 8 and this distance metric.A model is then developed using the reroute and convective information to predict whether or not a route will be used based on convective weather data.There are two major challenges in developing predictive models for this problem.First, the number of weather features available, many of which are irrelevant, is large compared to the number of observations, making it difficult for the model to identify relevant weather features.Second, each reroute is used infrequently making it difficult for the model to extract similarities between observations.Models are at the early stages of development and results are given here to illustrate the use of machine learning methods to provide a starting point for more detailed model development.Detailed discussions of the process of selecting and issuing reroutes, the CTOP program and methods of comparing routes are given in section II.An overview of the methodology is given in Section III.The route distance metric is presented in Section IV.Reroute implementation and use is quantified in Section V and model development and preliminary results are described in Section VI.Several features were calculated from the raw weather data and an ensemble method 9 was used to create models for clusters of advisory routes. +II. BackgroundIn this section, some background information is given on the use of reroutes and CTOP.Previous work in the areas of route comparison and route use analysis is discussed in more detail.This section ends with a discussion of methods designed to predict the use of reroutes.Currently, reroutes used in response to weather conditions are specified by TMCs using reroute advisories.These advisories are based on plays published by the FAA in the National Severe Weather Playbook. 10Each play in the Playbook is a set of routes frequently used to react to severe weather.The Playbook currently consists of 237 plays.Advisories, as implemented, are usually based on a play, however, TMCs can and often do, modify the play.Having a set of plays available for use by TMCs aids in communication among traffic managers when planning for a specific event.However, TMCs have the flexibility to modify plays as they see fit, thus creating somewhat custom built advisories.For example, during the study period from June to August 2011, 1,669 weather required advisories were issued, with 735 unique names.These advisories consisted of a total of 34,247 routes with 2,770 unique origin and destination pairs.In an effort to give airlines a means to state their preference for reroute options, a traffic management initiative known as the Collaborative Trajectory Options Program (CTOP) is currently being developed. 1TOP allows airlines to specify and rank acceptable routes for each of their flights through a Trajectory Options Set (TOS).The implementation of this program requires that airlines evaluate and file reroute options which can be used to route around convective weather.Airlines are at different stages in developing their procedures and tools required for their participation in CTOP.Airlines would like to identify geographically diverse and operationally feasible routes as potential members of a TOS.Each potential route must be evaluated based on airline objectives.A tool which can suggest previously used routes for similar weather would assist airlines in identifying potential routes which will likely be acceptable by the FAA.There is a wealth of information available in historical records of advisory reroutes and flight tracks that can be leveraged for use in a variety of applications.This work often involves comparing routes to find similarities, form clusters, or identify outlier trajectories.Given widely varying spatial and temporal descriptions of routes and flight tracks, each method involves a representation of a route or flight track and an associated distance metric.Distance metrics that have been used for similar problems could be applied here.De Armon et al. 11 focus on building a queueing model of the NAS and group routes in order to limit the number of paths between a given origin and destination in an effort to simplify the model.They describe each route by the sequence of airspace sectors through which the route passes and use the edit distance between these paths as the distance metric.The edit distance quantifies differences in sequences of "words" (in this case, sector identifiers) by finding the minimum number of manipulations (delete, shift or replace) of words required to morph one sequence into the other.This route representation is at a courser scale than would be useful for the current application.Enriquez, et al. 12 parameterize flight tracks by time, scaling the duration of the flight to span the interval [0, 1] and interpolating between points to get some fixed number of positions evenly spaced throughout the time interval.The distance between two routes is the sum of the Euclidean distances between corresponding points.In this work, we are not concerned with a time component of the trajectories.The method of 12 could be modified to generate points along each route that are evenly spaced spatially and then compared using the Euclidean distance.However, a small "dogleg" in one of two nearly identical routes causes a shift in the location of these points throughout the length of the route.This shift results in a distance defined by this metric that could be quite large when in fact, the two routes overlap for most of their lengths.Distance metrics have been developed which are based on the longest common subsequence (LCS) of the sequences of points used to describe two routes.The calculation of the LCS between two sequences is similar to the edit distance in that it is a measure of difference in the words used in each sequence.However, unlike the edit distance, an LCS method identifies ordered subsequences of words that are common to both sequences.Gariel, et al. 13 focus on identifying typical flight paths in the TRACON (Terminal Radar Approach Control) area based on historical flight paths.As a first step, turning points in trajectories are identified by finding heading changes.These turning points are clustered using the Euclidean distance as the metric to form waypoints.Each route is described as the sequence of waypoints over which the route passes.Routes are then compared to one another using an LCS-based distance metric.A similar method is presented in Vlachos, et al. 14 however, the route description is a time series of locations and is not transformed in any way.A method is presented to compare routes using a LCS method while allowing for spatial and temporal shifts in routes.In this work, we use a spatially uniform description of routes and use a LCS based metric to quantify overlap between routes.This is quite similar to the method used in 13, however, we do not need to perform the steps of identifying and clustering turning points or identifying the waypoints through which each route passes.Instead, we divide the NAS into a grid and represent each route as the sequence of grid elements through which it passes. +III. MethodologyThe main objective of this work is to predict whether or not an advisory route was used during specific weather scenarios.There are several intermediate pre-processing steps required to get to a point at which models can be developed to predict advisory route usage.At a high level, these steps involve analyzing the use of reroutes, clustering reroutes and extracting relevant features from weather data.These steps are described in more detail in the following paragraphs.First, we would like to understand reroute use.It is possible for a route to be issued as part of an advisory but not used by a flight during the valid time range of the advisory.We will refer to routes that are issued simply as issued routes.We focus on routes that were flown with the intent of identifying operationally relevant routes.Those will be called used routes.A route is considered to be used if at least one flight which departs during the valid time range of the advisory follows that route.Flight plans could be used to identify flights that follow a particular route, however, flight plans can be amended throughout the flight.Rather than checking each flight plan for conformance to a route throughout the flight, the actually flown flight track is used to determine adherence to the route.Given the difference in spatial resolution of the advisory route description and flight track, this comparison is not straight forward.The method of describing and comparing routes is given in Section IV.Generally, the route and track descriptions are transformed into the same spatially uniform grid.Then a distance metric is defined based on the fraction of overlapping segments of the flight track and route advisory.As mentioned previously, many of the routes used are spatially similar.When looking for trends in routing decisions, we are less concerned with the particular route used than we are with the path the route defines.That is, we are interested in predicting the use of spatially similar routes.Groups of similar routes are identified by clustering routes.Routes are first grouped by origin and destination Center or airport pairs.The routes of each group are then clustered using the k -means algorithm.The problem is then transformed into predicting whether or not a cluster of routes is used during a given time window.A route cluster is considered to be used during a time window if any of its members was used during that time window.We would then like to predict the use of route clusters.Advisory routes inbound to ZNY from June to August 2011 were used for this work as part of a NASA project to improve arrival operations in that area.This time range is broken into 2,208 one-hour time windows.Routing decisions are made hours in advance of the time that weather is predicted to affect airspace and advisories have ranges of one to fifteen hours.Thus, working with one-hour intervals is an appropriate resolution given the prediction lead times and valid durations of advisories.Each of these one-hour time windows is considered an observation.Each observation consists of weather data, referred to as features, and an indication of whether or not each route cluster was used during that time window, referred to as the class of the observation.If a route cluster is used, the observation is said to be a positive observation for that cluster, otherwise it is said to be a negative observation.The problem is then posed as a classification problem in which machine learning methods are used to predict the class (positive or negative) of each observation.A challenge of working with this type of data is that there is a large imbalance of positive and negative observations.Routes used during this time period were grouped into 253 clusters.The top twenty most frequently used clusters were used during 50 to 240 one-hour time windows.The number of positive observations is much less than the total number of observations.The class label that appears most frequently in an imbalanced data set is referred to as the majority class while the underrepresented class is referred to as the minority class.In this case, negative observations are the majority class and positive observations are the minority class.Classification problems with imbalanced data are challenging as machine learning algorithms rely on statistics of a large amount of data to be able to extract patterns and accurately classify data.Given a large imbalance of examples of each class, machine learning algorithms will not be able to identify the minority class accurately.A minority class over-sampling technique is used to address this imbalance.Weather information is used as the features of each observations.The features make up a set of data for each observation that is used by a machine learning algorithm to generate models to predict the class of the observation.Weather predictions and actual data for the NAS are available at high spatial and temporal resolutions making the size of available data impractical for use as is in model generation.For this work, observed cloud echo top height measurements are used to describe convective weather conditions.Those data can be used to quantify flight deviations based on the comparison of echo top and the flight altitude of the aircraft.The number of echo top readings which cover the NAS is 57,540.In order to extract high level trends in the data, the echo data are first down-sampled spatially and averaged over the one hour time windows.This reduces the number of weather features covering the NAS to 1,000.The number of weather features is still quite large, with many being irrelevant to the use of a particular cluster.Eighteen hand selected features are then calculated from these data with an attempt to reveal weather conditions at locations that are key to determining whether or not a route cluster was used.Finally, with the above organization of cluster use and weather features, machine learning models are developed.A model is built for each cluster individually which takes weather features of observations as input and predicts the class of each observation.The models are built using labeled data, that is, a set of observations including weather features and class label. +IV. Comparison of PathsAdvisory reroutes are based on plays, but TMCs do have the flexibility to modify these routes.Given that there are various ways to specify and implement rerouting, there may be hundreds of unique routes used to react to similar whether conditions, but their differences may be small enough to be considered practically equivalent.An integral part of finding trends in the methods used to react to similar weather is having a means to identify similar routes and flights that follow those routes.A first step in this work is to develop a distance metric which can be used to compare routes to one another and routes to flight tracks.Since this comparison can be made between advisory routes and flight tracks, routes and tracks will be referred to generically as paths.Routes are described in advisories and plays by a text description of the waypoints, fixes and jet routes used to define the route.A comparison of these text descriptions can be used to identify identical or similar routes.However, even if the text descriptions do not match, two routes may be spatially similar.Since traffic is routed around certain regions of airspace to avoid severe weather, the geographical location of routes is important.Thus, the physical proximity of two routes should be quantified, regardless of the text description used to define the routes.In Section IV.A gives more detailed descriptions of routes and flight tracks.The distance metric used is presented in Section IV.B. +IV.A. Path DescriptionEach reroute advisory or play consists of several routes.The routes are described by an origin, destination and route between the origin and destination.Origins and destinations are specified as an airport or Air Route Traffic Control Center (ARTCC or Center).The route consists of a series of departure fixes, waypoints, jet routes and arrival fixes which are specified by alphanumeric identifiers.Jet routes themselves are defined as a series of fixes and waypoints.Each fix and waypoint has a location specified by latitude and longitude coordinates.During the months of June through August 2011, 1,669 weather required advisories were issued, with 735 unique names.These advisories consisted of a total of 34,247 routes with 2,770 unique origin and destination pairs.Each advisory route is described by up to 100 coordinates.There are 3,234 unique coordinates used in the description the routes issued during the study period.Flight track data are available in Aircraft Situation Display To Industry (ASDI) data.RADAR systems are used to track flights and the speed, altitude and position of flights are calculated.Updates are provided every 12 seconds for flights in enroute airspace and every 5 seconds for flights within a TRACON.This information is then made available through ASDI.Both advisory routes and flight tracks can be described by a sequence of points, specified by latitude and longitude coordinates.The mathematical description of paths used in the development of the distance metric is given in Appendix A.A. The points used to describe a given path are not necessarily evenly spaced.An overlapping stretch of two paths may be described by a different number of points, thus a point-by-point comparison of two paths would not yield a meaningful distance.Furthermore, when tracks are compared to paths, the problem is further complicated due to the continuous nature of aircraft flight position measurements. +IV.B. Distance MetricThe objective in the development of a distance metric for this application is to design a distance metric which:1. Provides a meaningful measure of spatial distance between paths 2. Is robust to path specification 3. Can be used to assess partial overlap between paths.In order to capture spatial properties of each path while being robust to the path description, a grid based representation of paths is used.The path description is spatially uniform, with airspace divided into a grid.A path is described by the sequence of grid elements through which it passes.Distances between paths are based on the overlapping portions of the paths.This can be used to test for full or partial overlap of paths.The grid size is specified by the user and affects the tolerance of the metric.The specific grid size choice used for the problem addressed here is discussed in Section V.An illustrative example is shown in Fig. 1 in which two paths are plotted along with their respective grid representations.The overlapping grid elements are highlighted in green.The mathematical development of the distance metric calculation is given in Appendix A.B. +IV.C. Characteristics of MetricAn advantage of this route description and distance metric comes from the spatially uniform description of paths.However, the choice of ∆lat and ∆lon will affect the distances between paths.Larger ∆lat and ∆lon will allow larger deviations to be considered "identical" while smaller ∆lat and ∆lon requires "identical" routes to be closer together.A downside of this technique is that two parallel path segments may be separated by an amount less than ∆lat or ∆lon but have grid representations that do not overlap.The user is given the choice of the denominator in the fractional difference calculation to allow for different objectives.A flight track is likely to extend beyond the reroute description, thus, we would like to find the fraction of the flight track overlapping the advisory route, and not vice-versa.The minimum length of the two paths compared is used as the denominator for this case.If routes do not use overlapping grid elements, the distance is 1, thus, there is no distinction between routes that are close and parallel and routes that may be on opposite sides of the country.To address this issue, this metric could be used at different levels, that is, start with a relatively large ∆lat, ∆lon to identify roughly close routes.Then, a smaller grid ∆lat, ∆lon can be used to get a more precise distance between routes identified as close using the coarser grid description. +V. Advisory Route Use AnalysisHere, flight tracks are matched to advisory routes.The results presented here are for weather required advisories issued from June 2011 through August 2011 that have their destination within New York Center (ZNY).The distance metric described above is used to compare advisory routes with flight tracks.The procedure used to identify matching flights and routes is detailed below, followed by analysis results.For each flight, all advisories active at the time of departure are identified.The origin and destination airports of the flight are then compared to each active advisory route origin and destination.Recall that the origin and destination of a route may be a Center or an airport.If the origin (destination) is specified as an airport, the flight origin (destination) matches the route origin (destination) if the airports match.If the origin (destination) is specified as a Center, the flight origin (destination) matches the route origin (destination) if the flight origin (destination) airport is within this Center.Flights and active routes which match both the origin and destination are referred to as origin-destination matches.Next, a coarse distance is calculated between each flight and each advisory route in its set of origindestination matches using grid elements of size 1.50 • latitude by 1.50 • longitude, roughly 90 nmi by 70 nmi.If the coarse distance is less than 0.5, a finer distance is calculated using a smaller grid size.An adaptive grid size is used as the smaller grid size, varying from 0.25 • latitude by 0.25 • longitude, roughly 15 nmi by 12 nmi, to 1.50 • latitude by 1.50 • longitude.If the advisory route consists of 10 or fewer grid elements in the large grid representation, the small grid size is used for the second comparison.If the advisory route consists of 25 or more grid elements in the large grid representation the second comparison is not made.Grid size is varied linearly between the extremes based on the number of large grid elements needed to describe the advisory route.The extreme values of the grid size and the function used to calculate the grid size for the finer comparison are user specified.For this study, values were chosen based on visual inspection of various routes and comparisons performed with various grid sizes.The extremes of the grid element sizes are illustrated in Fig. 2. Two advisory reroutes are shown.Figures 2a and2b show the waypoint descriptions of two advisory routes overlaid with their large grid representations and adaptive grid representations.The route shown in 2a is described by 25 large grid elements, with a corresponding adaptive grid size equal to the large grid size.The route shown in 2b is described by 10 large grid elements, with a corresponding adaptive grid size equal to the small grid size.If the distance between the advisory route and the flight tracks (using the minimum number of grid elements used to describe advisory route or the flight tracks as the denominator) is less than a user defined threshold then the flight is considered a possible match for that advisory route.Adjusting the threshold will affect the results of matching flights to advisory routes and will be considered in future work, as illustrated in Fig. 3. Adaptive grid path distances were only calculated when the large grid path distance was at or below 0.5, thus, the asymptote of the adaptive grid curve is the fraction of flights that match at least 50% For the purposes of this work, this threshold was set to 0.15 (marked in Fig. 3) for the adaptive grid distance, requiring that a flight track overlap an advisory route for at least 85% of the route to be considered a potential match.Each flight is matched to at most one route, selected based on the distance between the flight and all of the potential route matches.These matches are referred to as path matches.Routes with at least one path match are considered used routes.Flight tracks and advisory matching was performed using the above procedure for flights inbound to ZNY, with results shown in Table 1.Results are presented for flights, advisories and routes.Of the flights inbound to ZNY during the period of the study, 3.71% are an origin-destination match to at least one advisory route.Of those origin-destination matches, 29.07%are also path matches of the advisory route.Looking at advisories, 92.47% of the issued advisories consist of at least one route which is an origin-destination match of a flight.Of these advisories, 78.29% consisted of at least one route matched by a flight.This indicates that, while a path match may not be found for every route within an advisory, most advisories consisted of routes that were matched by at least one flight.Of the individual routes issued, 44.17% have an origin-destination match, and only 20.22% of these routes are matched by a flight.In summary, we found that roughly 29% of flights which match the origin and destination of an advisory route actually use an advisory route and that only 20% of routes between an origin destination pair matching at least one flight are actually used.There are several factors that could contribute to the resulting low conformance of flights to advisory routes.Here, we consider a flight track to match a route if their paths overlap for more than 85% of the shorter path.Segments of routes can be omitted for flights departing from particular airports, allowing these flights to join up with the route at some intermediate point.Additionally, advisories can be canceled or modified and thus, comparisons made here could include inactive routes. +VI. Modeling the Use of Advisory ReroutesThe focus of this section is on the development of a predictive model of advisory routes inbound to ZNY from June to August of 2011.The time range is broken into 2,208 one-hour time windows.The objective is to predict the use of specific routes during a given hour based on weather conditions.Each time window is considered an observation, consisting of a set of features and a class label.The features consist of weather data.The class label is positive if the advisory cluster was used at least once during the time window, otherwise, the label is negative.There are two main challenges in developing a model for this particular data: the large number of features available compared to the number of observations and the low ratio of positive to negative examples.When the number of features is large compared to the number of observations, models tend to capture random error and can easily be over-fit to the training data.The number of times each route is used throughout this time period is much less than the number of times it is not used.Building models from an imbalanced data set tends to result in large error rates in the prediction of the minority class. +VI.A. Clustering RoutesOver 4,000 advisory routes were issued during the time frame of interest.Many of these routes are identical or practically identical.In order to understand reroute use, we would like to first group identical or similar reroutes.The first step in the clustering is to group routes by origin and destination.Given that the path portion of the advisory reroute description may lead from a Center rather than a specific airport, several routes have paths which overlap but originate in different Centers.Airports within 100 nmi of each other were considered to be identical for the purpose of this grouping.A total of 72 groups were formed at this stage.The gridbased metric with the large grid size is then used to compare paths.Clustering within each origin-destination group was performed using the k -means algorithm.The total number of resulting clusters is 253.The top 20 most frequently used clusters are used during 50 to 240 of the 2,208 one-hour time windows.Formulating the prediction problem as predicting the use of route clusters rather than individual routes increases the number of positive examples in the data set.However, even after clustering, the imbalance between positive an negative observation for each cluster is large.This imbalance is addressed in the model formulation. +VI.B. Weather FeaturesNext, we process weather data to create features.The specific weather data used in this work are measurements of convective cloud top altitude, referred to as echo tops.These values are estimates of the top of clouds based on radar measurements.The values are discrete altitude levels, 0 ft to 50,000 ft at 5,000 ft intervals for a total of 11 levels.Readings of echo tops are available every 2.5 minutes at a spatial resolution of 4.65 nmi.The number of echo top readings which cover the NAS is 57,540.If echo top values at this spatial and temporal density were naively used as features in a predictive model it would likely lead to over-fitting the training data.In order to reduce the number of features, echo top values are averaged and then used to create a smaller set of features.Values were averaged at the center of grid elements of size 1.25 • latitude by 1.25 • longitude, roughly 75 nmi by 58 nmi (1,000 total grid elements of this size cover the NAS), over one-hour time windows.These values were then binned into 11 bins to mimic the echo top data.This is a large number of features, most of which are likely unrelated to the use of a particular cluster of routes.In order to reduce the number of features used to generate the predictive models and attempt to extract information relevant to the use of a particular cluster, several features were calculated from the gridded echo top values.These features were created with the intent of condensing the large feature space of the echo top data to fewer, more relevant features.The focus was on capturing echo top information along the route cluster and direct from origin to destination.The motivation behind this is to create features which can be used to compare weather conditions on the direct path to an optional reroute.If the weather along the direct path is mild, there would be no need to issue a weather required advisory between that origin and destination.Here, we use the direct path between origin and destination.Details of the calculation of the features used are given in Appendix B. +VI.C. Model FormulationAt this point, we have organized weather and reroute use information into observations consisting of weather features with class labels for each reroute cluster and we are ready to create models of reroute cluster use.Models were generated for each advisory route cluster with the objective of determining the time windows in the test set in which the advisory cluster was used.We use a random forest method 9 for this classification problem.A random forest method is an example of an ensemble method.Ensemble methods consist of many simple classifiers that may result in poor classification results individually.The predictions of these weak classifiers are then given a weighted vote to give the final prediction of the ensemble method.The predictions of each weak learner are weighted based on the fraction of misclassified examples in the training set.The specific type of classifier used as the weak learners in a random forest method are decision trees.The weak learners are developed using only a subset of the training observations and a subset of the available features.Observations are selected uniformly at random from the training set for use in the development of each weak learner.The observations are selected in a way that increases the ratio of positive to negative examples compared to the ratio found in the full training set.However, oversampling of the positive examples in this way can lead to over-fitting the model to the training set.We use Synthetic Minority Oversampling Technique (SMOTE) 15 to create additional positive examples.In SMOTE, features of additional positive observations are created by selecting a positive observation and taking a convex combination of that observation and one if its nearest neighbors.These observations are referred to as "synthetic" since they do not naturally appear in the data.This process can be repeated using other neighbors of the selected positive observation.This forces the decision region of the positive examples to be more general and thus less likely to over-fit the positive examples in the training set than simply oversampling the positive observations.In this work, synthetic positive observations are created from the features listed in section VI.B.Each of the weak learners exhibits poor performance.The predictions from these models are given a weight α ∈ R based on their respective prediction errors ε.Here, we use a function of ε to calculate α.This function of ε has a value of 0 at ε = 0.5 and increases as ε decreases.Learners with prediction errors ε ≥ 0.5 are given a weight of α = 0. Using this weighting function, weak learners with low error are given high weights.The ensemble prediction for the test data set is a weighted vote from each of the weak learners.There are several parameters of the algorithm which the user must specify.These parameters are: the number of synthetic positive observations to be created from some number of nearest neighbors, the fraction of positive examples in the training set for each weak learner, the number of features used for each weak learner, and the tree depth of each weak learner.Typically, tree depths of one or two are used as deeper trees tend to over-fit the test data.In order to test the performance of each ensemble model, we need to apply it to a testing set of observations that were not used in the development of the model.Here, we use K-fold cross-validation method to assess the performance of the ensemble models.In this method, the original data set is divided into K equal sized subsets, where K is a user specified value.The cross-validation is then performed K times for k = 1, 2, . . .K. For each k, the k th subset is held out from the rest of the subsets.The holdout set is referred to as the testing set, the other observations make up the training set.The ensemble model is built using only the data available in the training set.Once the model has been built, it is used to predict the classes of the observations in the testing set.The testing set prediction results for each value of k are used to describe the performance of the modeling method. +VI.D. ResultsPredictive models were developed for each route cluster using the method described in Section VI.C. Models were generated for each cluster individually using the same parameters of the algorithm for each model.The particular parameters used were hand selected through trial and error.The data were divided into three sets for three-fold cross validation.The same test and training sets were used for the ensemble models of each cluster.SMOTE was used to increase the number of positive examples in the training set by a factor of nine.Positive and negative examples were selected uniformly at random for the sub-training set with a ratio of 3:2.Ten features were selected uniformly at random for each weak learner.Decision trees of depth two were used as the weak learners.Prediction results are shown in Fig. 4 for models which do not make use of SMOTE and models that do.True positive rates (TPR) and true negative rates (TNR) are plotted for each of the top 20 most frequently used clusters.TPR indicates the fraction of positive observations that were correctly classified as positive and TNR indicates the fraction of negative observations that were correctly classified as negative.A perfect classifier would have both TPR and TNR equal to one.The top 20 most frequently used clusters were selected based on the number of time windows in which a cluster was used.These particular clusters were used in 50 to 240 time windows.Figure 4a shows results for classifiers built to predict the use of advisory routes in a given time window without using SMOTE, while figure 4b shows corresponding results from classifiers built using SMOTE.Results shown in fig.4a indicate that prediction performance degrades considerably as the number of positive observations in the data set decreases.Including additional positive features using SMOTE greatly improved the classification results.A total of 15 of 20 clusters have average TPR and TNR above 0.6 for the classifiers which made use of SMOTE compared to only 8 for models which did not make use of SMOTE.While TPR and TNR over 0.6 is only marginally better than random guessing, this does indicate that the use of machine learning methods for this problem could provide useful results.These results show that even a relatively simple predictive model is able to extract patterns and trends from the data to predict whether or not a route cluster will be used.To better understand the models and prediction results, we will take a more detailed look at a specific cluster.Members of the ninth most frequently used cluster and its cluster center are shown in Fig. 5 Figure 6a shows the distribution of prediction results for each weak learner used in the ensemble classifier for one of the cross validation sets.Notice that both true positive and true negative rates vary greatly and are quite poor in certain instances.However, the ensemble model which uses the weighted vote of the predictions of each of the weak learners achieves a TPR and TNR of 0.69 and 0.79, respectively.As mentioned above, there are several parameters which need to be set in order to generate models of this type.Varying any one of these parameters will affect the prediction results.Here, we examine one of these parameters in further detail.The fraction of positive examples used in the sub-training set for the models used to generate the prediction results shown in Fig. 4 was 0.6.We examine the affect of varying this value by building models with various values of this fraction.The fraction of positive examples in the sub-training set were varied from 0.1 to 0.9, with results shown in Fig. 6b.When the fraction of positive to negative observations is 0.1, little information is given to the classifier to distinguish positive observations from negative.With more negative observations in the training set, the classifier is biased towards predicting the negative class.Looking at Fig. 6b, it can be seen that for low values of this fraction, TNR is close to 1 while TPR is 0. As this ratio is increased, the classifier is better able to distinguish positive and negative observations and more balanced prediction results are seen.At the other extreme, when the ratio of positive to negative observations is 0.9, most examples are positive and the classifier has little information about negative observations, thus, the classifier is biased toward positive predictions.Understanding how TPR and TNR vary with this parameter could help a user tune this parameter to achieve desired results.For example, if the mis-classification of a positive example is costly or dangerous, it may be desirable to select a higher ratio of positive examples in the sub-training set to correctly classify more of the positive examples if the increase in incorrectly classified negative examples is acceptable. +VII. Conclusions and Future WorkA method of processing advisory reroute data, flight track data and weather data was presented and a model was developed to predict weather or not particular advisory reroutes were used given weather conditions.While the prediction of true positive and negative rates are lower than what would be useful in the generation of a decision support tool, they do indicate that this line of research is worth pursuing with modifications.While we have data describing weather and traffic conditions, there are many factors being considered by the TMC that are not represented in this data and are not well documented.Furthermore, the translation of the weather data that we do have into an operationally relevant form is an art requiring knowledge of the domain and factors influencing routing decisions.While several hand selected features were used in this work, a more systematic method of data reduction should be developed.A method of comparing advisory routes, and flight paths was presented and used to analyze the historical use of advisory reroutes.A machine learning technique was then developed to predict the use of reroute advisory clusters based on convective weather information.Model development involved addressing two fundamental challenges inherent in working with the reroute and weather data, namely the abundance of features and relatively low number of positive examples in the data set.Fifteen out of the twenty clusters for which models were built result in average true positive and true negative rates above 60%.While these results are preliminary, they do indicate that applying machine learning techniques to the problem of predicting advisory reroute use could give meaningful and useful results.Several ways in which analysis and modeling could be modified to improve prediction results are discussed in the following paragraphs.The path comparison method could be improved to better indicate the proximity of paths when they do not have grid elements in common but are nevertheless close spatially.Modifying this distance metric based on the techniques presented in Ref. 14 or the using the Hough transform 16 could add more flexibility in route matching.The choice of the threshold used to specify the distance between paths that are considered a match should be further studied.Partial matches may be operationally significant if, for instance, a close match occurs near the departure or arrival airport where traffic density is higher than in enroute airspace.Features were created from weather data in order to highlight convective weather along the routes and along the direct path from origin to destination.Further work can be done to generate more features, focusing on the locations of important fixes or jet routes along the reroute and along typical direct routes.Additionally, a finer grid size could be used at key locations along the route, such as arrival fixes or within the TRACON of the destination airport.This finer resolution would help in determining which among several routes would be most appropriate to react to weather conditions.There are many types of weather data and forecasts available.In this work, the convective weather data used was actual echo tops.These values indicate the tops of could.While this is important in determining if flights are able to fly above a storm or around it, more information is needed to understand the impact of the storm on flights.Vertically Integrated Liquid (VIL) combined with echo tops gives a better indication of the intensity of convective weather.The Convective Weather Avoidance Model (CWAM) 17 model makes use of both echo and VIL to model the probability of pilot deviation around weather.The use of CWAM should give a better indication of weather or not a flight would follow a specific path.The use of weather forecast data to predict the implementation and use of reroutes is a necessary step in the development of a tool that could be used to select reroutes as reactions to specific weather forecasts.TMCs use the Collaborative Convective Forecast Product (CCFP) during planning.CCFP is a forecast that combines weather predictions from various sources.The forecast indicates the predicted intense convection activity and is updated every 2 hours with forecasts made for 2, 4, and 6-hour periods.Using the CCFP forecast that TMCs were using to make reroute advisory decisions will be useful in predicting whether or not a reroute was issued.There are a variety of ways in which the machine learning method could be improved.There are several parameters of the model that the user must specify.Future work will include a more systematic approach to setting these parameters for the models of each cluster individually.Minority class oversampling was performed using SMOTE to address the imbalance of the data set.Expanding the data set to include summer months of additional years would add more positive observations and could also improve the modeling results.Rather than predicting whether or not a route cluster will be used, one could develop models to predict how likely it is that a route cluster would be used.For instance, if many flights follow the same route cluster in a given hour, it may be more likely that a route from that cluster would be used during similar weather conditions.The number of flights matching a path along with a varying threshold for path matching to distinguish the level of conformance to a path could be used in models designed to predict the probability of route use. +A. Path Comparison +A.A. Path DescriptionEach path can be described by a sequence of points, specified by latitude and longitude coordinates, for exampleP =      p 1 p 2 . . . p m      =      p 11 p 12 p 21 p 22 . . . p m1 p m2     where m is the total number of points used to describe the path.The latitude and longitude coordinates of each waypoint are rounded to the nearest multiple of ∆lat, ∆lon, effectively dividing the NAS into a grid of size ∆lat by ∆lon.Grid elements between waypoints are included by assuming a straight line trajectory between waypoints.The grid based description of route P isP =      p1 p2 . . . p m      =      p11 p12 p21 p22 . . . p m1 p m2      , such that pi+1 = pi + [u i ∆lat v i ∆lon] , ∀ i = 1, 2, . . . , m -1where u i ∈ {-1, 0, 1}, v i ∈ {-1, 0, 1} and |u i + v i | = 1. +A.B. Distance MetricThe distance between two paths is based on the overlap of the grid representations of the paths.The distance is defined as the fractional difference between the number of grid elements in common to both paths and the number of grid elements used to describe one of the paths being compared.The calculation of the distance between two routes, P and Q, according to this metric can be described in steps, as follows.First, find the grid representations P and Q of routes P and Q. Next, find subsets of P and Q that consist of sequential points within the original vectors and are common to both routes.A sequential subset of route P is described as       +C. Model DevelopmentAn ensemble model was built for each route cluster individually.Before describing the procedure used to create the ensemble model, we need to introduce several terms.The number of weak learners used in each ensemble model is M .The set of all observations is denoted by S. The set of all calculated features is denoted by F. The vector x i consists of the feature values of observation i.Let x Fm i denote the features in F m ⊆ F of observation i.And finally, y i is the class label of observation i. Model development was performed for each route cluster using the following procedure.Each of the weak learners exhibits poor performance.The predictions from these models are given a weight α based on their respective prediction errors ε.If the prediction error is less than 0.5, α is calculated using Step 14.The particular choice of α comes from a formulation of the adaptive boost algorithm. 9This function of ε has a value of 0 at ε = 1 and increases as ε decreases.Using this weighting function, weak learners with low error are given high weights.The ensemble prediction for the test data set is a weighted vote from each of the weak learners.Figure 1 :1Figure 1: Example paths, grid representations and overlapping grid elements. +Figure 2 :2Figure 2: Figures 2a and 2b show the waypoint descriptions of two advisory routes overlaid with their large grid representations and their respective adaptive grid representations. +Figure 3 :3Figure 3: The fraction of flights matching the origin and destination of an advisory route that also match the path as a function of the cutoff threshold for path matching. +Test set prediction results for models built to predict the use of advisory route clusters without the use of Test set prediction results for models built to predict the use of advisory route clusters with the use of SMOTE. +Figure 4 :4Figure4: The mean and standard deviation over each cross validation set of true positive rates (TPR) and true negative rates (TNR) are plotted for the top 20 most frequently used clusters.The results are ordered such that the results for the most frequently used cluster are on the left and results for the least frequently used cluster are on the right. +Figure 5 :5Figure 5: Example reroute advisory cluster.Member routes are shown in red with the cluster center shown in black cluster origin Center is outlined in green. +Intermediate TPR and TNR for the weak learners for one cross validation set and the resulting TPR and TNR of the ensemble model.True positive and negative rates as a function of the ratio of positive to negative observations used in the sub-training sets for each of the weak learners. +Figure 6 :6Figure 6: Additional prediction results for advisory reroute cluster shown in Fig. 5. +3 .3some a, l ∈ N such that 1 ≤ a ≤ a + l ≤ m.The longest common sequential subsets of P and Q are common( P , Q) = {S 1 , S 2 , . . ., S R } where R ∈ N is the number of common sequential subsets andS i = Mean of grid element echo values in G c µi (G c diri ) 4.Item 2 divided by Item 3 5.Maximum echo value of grid elements in G µi (G diri ) 6. Maximum echo value of grid elements in G c µi (G c diri ) 7. Item 5 divided by Item 6 8. Standard deviation of echo values of grid elements in G µi (G c diri ) 9. Number of grid element values in G µi (G diri ) at each binned echo top value 10.K nearest neighbor label based on Euclidean distance between the above features for each observationThe mean and maximum values along the reroute and direct route give an indication of the severity of weather along those paths.The ratios of these values with the values calculated for the complement of these sets gives an indication of the conditions along the paths compared to conditions in the rest of the NAS.The number of grid element values along the reroute and direct route provides an indication of the convective weather coverage of the path.The K nearest neighbors label is used to provide a rough guess at the class label of the observation based on the similarity of the weather features in this observation compared to observations in the training set. +1 : 6 : 12 : 13 : 14 :16121314Divide S into S train and S test sets based on dates of observations 2: P train = {i ∈ S train |y i = +} 3: Generate synthetic positive observations from P train using SMOTE 4: S train = S train ∪ {synthetic positive observations} 5: for m = 1 to M do Subdivide the training data S train into S sub-train and S sub-test by days 7: P sub-train = {i ∈ S sub-train |y i = +} 8: N sub-train = {i ∈ S sub-train |y i = -} 9: S sub-test = {i|i ∈ S sub-test ∩ S} 10: Create S mod-train by selecting p observations from P sub-train and n observations from N sub-train , uniformly at random 11: Create F m by selecting f features uniformly at random from F Build a decision tree, C m (•) using observations S mod-train and features F m Calculate the error of the prediction results, ε m = 1 N j∈S sub-test δ C m (x Fm j ) = y j where N = |S sub-test | and δ (•) is 1 if the argument is true, and 0 otherwise If ε m < 0.5, set the importance factor α m according to α m = 1 2 ln 1-εm εm , otherwise α m = 0 15: end for 16: Predict the class of observation i for some i ∈ S test according to C * (x i ) = argmax y M m=1 α m δ(C m (x Fm i ) = y) +Table 1 :1Flights and Advisories inbound to ZNY.TotalOrigin -DestinationPathMatched % Matched Matched % MatchedFlights190,0467,0603.71%2,05229.07%Advisories27925892.47%20278.29%Advisory routes4,4761,97744.17%90520.22% +16Ballard, D., "Generalizing the Hough Transform to Detect Arbitrary Shapes," Pattern Recognition, Vol. 13, No. 2, 1981, pp.111-122. 17DeLaura, R., Robinson, M., Pawlak, M., and Evans, J., "Modeling convective weather avoidance in enroute airspace," 13th Conference on Aviation, Range, and Aerospace Meteorology, AMS, New Orleans, LA, Citeseer, 2008. 18Bergroth, L., Hakonen, H., and Raita, T., "A survey of longest common subsequence algorithms," String Processing and Information Retrieval, 2000.SPIRE 2000.Proceedings.Seventh International Symposium on, 2000, pp.39-48. + of 15 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on July 2, 2015 | http://arc.aiaa.org| DOI: 10.2514/6.2015-3395 + of 15 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on July 2, 2015 | http://arc.aiaa.org| DOI: 10.2514/6.2015-3395 + of 15 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on July 2, 2015 | http://arc.aiaa.org| DOI: 10.2514/6.2015-3395 + of 15 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on July 2, 2015 | http://arc.aiaa.org| DOI: 10.2514/6.2015-3395 + Downloaded by NASA AMES RESEARCH CENTER on July 2, 2015 | http://arc.aiaa.org| DOI: 10.2514/6.2015-3395of the route using the large grid. + of 15 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on July 2, 2015 | http://arc.aiaa.org| DOI: 10.2514/6.2015-3395 + of 15 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on July 2, 2015 | http://arc.aiaa.org| DOI: 10.2514/6.2015-3395 + of 15 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on July 2, 2015 | http://arc.aiaa.org| DOI: + 10.2514/6.2015-3395 + American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on July 2, 2015 | http://arc.aiaa.org| DOI: 10.2514/6.2015-3395 + Downloaded by NASA AMES RESEARCH CENTER on July 2, 2015 | http://arc.aiaa.org| DOI: 10.2514/6.2015-3395 + + + + +AcknowledgmentsI would like to thank Michael Bloem, Shon Grabbe, Deepak Kulkarni, Avijit Mukherjee, and Banavar Sridhar for conversations and recommendations regarding this work. + + + +such that, for i = 1, 2, . . ., R, s ij = pai+j-1 = qbi+j-1 , for j = 1, 2, . . ., l i and the common subsets do not overlap, that iswhere m and n are the number of grid elements of P and Q, respectively.There are various algorithms available to compute the longest common subsequence amongst a set (or in this case, pair) of sequences. 18he distance between P and Q is distwhere c may be min( m, n) or max( m, n), depending on the objective of the comparison.Note that the resulting distance value is unitless and represents the fraction of overlapping grid elements.If c = min ( m, n) is chosen, the distance between P and Q will be zero if the two paths overlap for the entire length of the shorter path, regardless of whether the longer path extends beyond the shorter.Alternatively, c could be chosen as max ( m, n) to compare the shorter path to the longer.An advisory reroute is compared to two different flight tracks and shown in Fig. 7. Using this distance metric, the overlap error of the flight track in Fig. 7a is 0, while the overlap error of the two paths in Fig. 7b is 0.21.Both difference are calculated with respect to the minimum number of grid elements required to describe each path. +B. Weather FeaturesFor each cluster of advisory routes grid elements along the route specified as the cluster center µ were identified.For a given cluster i, this set of grid elements is G µi Additionally, grid elements along the direct route between the origin and destination of the cluster center were identified.For a given cluster i, this set of grid elements is G diri The complement of these sets, that is, all grid elements not in each of these sets, will be represented by G c µi and G c diri , respectively.The full list of features used in model generation are:1.The hour of the beginning of the time window 2. Mean of grid element echo values in G µi (G diri ) + + + + + + + Collaborative Trajectory Options Program (CTOP) + + + + "Collaborative Trajectory Options Program (CTOP)," http://www.nbaa.org/ops/airspace/tfm/tools/ctop.php. + + + + + A Cluster Analysis to Classify Days in the National Airspace System + + BHoffman + + + JKrozel + + + SPenny + + + ARoy + + + KRoth + + + + AIAA Guidance, Navigation and Control Conference + + 2003 + 11 + + + Hoffman, B., Krozel, J., Penny, S., Roy, A., and Roth, K., "A Cluster Analysis to Classify Days in the National Airspace System," AIAA Guidance, Navigation and Control Conference, 2003, p. 11. + + + + + A clustering approach for analysis of convective weather impacting the NAS + + MAsencio + + + + Integrated Communications, Navigation and Surveillance Conference (ICNS) + + 2012. April 2012 + + + + Asencio, M., "A clustering approach for analysis of convective weather impacting the NAS," Integrated Communications, Navigation and Surveillance Conference (ICNS), 2012 , April 2012, pp. N4-1-N4-11. + + + + + Classification of Days Using Weather Impacted Traffic in the National Airspace System + + AMukherjee + + + SGrabbe + + + BSridhar + + + + Transportation Research Board 92nd Annual Meeting + Washington DC + + January 2013 + + + Mukherjee, A., Grabbe, S., and Sridhar, B., "Classification of Days Using Weather Impacted Traffic in the National Airspace System," Transportation Research Board 92nd Annual Meeting, Washington DC, January 2013. + + + + + Sequential Congestion Management with Weather Forecast Uncertainty + + CWanke + + + DGreenbaum + + No. AIAA 2008-6327 + + + AIAA Guidance, Navigation and Control Conference and Exhibit + + August 2008 + + + Wanke, C. and Greenbaum, D., "Sequential Congestion Management with Weather Forecast Uncertainty," AIAA Guid- ance, Navigation and Control Conference and Exhibit, No. AIAA 2008-6327, August 2008. + + + + + Dynamic Generation of Operationally Acceptable Reroutes + + CTaylor + + + CWanke + + No. 2009-7091 + + + Proceedings of 9th AIAA Aviation Technology, Integration, and Operations Conference + 9th AIAA Aviation Technology, Integration, and Operations Conference + + AIAA + September 2009 + + + Taylor, C. and Wanke, C., "Dynamic Generation of Operationally Acceptable Reroutes," Proceedings of 9th AIAA Aviation Technology, Integration, and Operations Conference (ATIO), No. 2009-7091, AIAA, September 2009. + + + + + Dynamic weather routes: a weather avoidance system for near-term trajectory-based operations + + DMcnally + + + KSheth + + + CGong + + + JLove + + + CHLee + + + SSahlman + + + JCheng + + + 2012 + Brisbane + + + 28th international congress of the aeronautical sciences (ICAS + McNally, D., Sheth, K., Gong, C., Love, J., Lee, C. H., Sahlman, S., and Cheng, J., "Dynamic weather routes: a weather avoidance system for near-term trajectory-based operations," 28th international congress of the aeronautical sciences (ICAS), Brisbane, 2012. + + + + + Some methods for classification and analysis of multivariate observations + + JMacqueen + + + + Proceedings of the fifth Berkeley symposium on mathematical statistics and probability + the fifth Berkeley symposium on mathematical statistics and probabilityOakland, CA, USA + + 1967 + 1 + + + + MacQueen, J. et al., "Some methods for classification and analysis of multivariate observations," Proceedings of the fifth Berkeley symposium on mathematical statistics and probability, Vol. 1, Oakland, CA, USA., 1967, pp. 281-297. + + + + + Introduction to Data Mining + + P.-NTan + + + MSteinbach + + + VKumar + + + 2005 + China Machine Press + + + Tan, P.-N., Steinbach, M., and Kumar, V., Introduction to Data Mining, China Machine Press, 2005. + + + + + National Severe Weather Playbook + + + 2013 + + + "National Severe Weather Playbook," http://www.fly.faa.gov/PLAYBOOK/pbindex.html, 2013. + + + + + Air Route Clustering for a Queuing Network Model of the National Airspace System + + JSDearmon + + + CPTaylor + + + TMasek + + + CRWanke + + + + 14th AIAA Aviation Technology, Integration, and Operations Conference + + American Institute of Aeronautics and Astronautics + 2015/05/15 2014 + + + DeArmon, J. S., Taylor, C. P., Masek, T., and Wanke, C. R., "Air Route Clustering for a Queuing Network Model of the National Airspace System," 14th AIAA Aviation Technology, Integration, and Operations Conference, American Institute of Aeronautics and Astronautics, 2015/05/15 2014. + + + + + A Simple and Robust Flow Detection Algorithm Based on Spectral Clustering + + MEnriquez + + + CKurcz + + + + Submitted to 5th International Conference on Research in Air Transportation + + 2012. 2012 + + + Enriquez, M. and Kurcz, C., "A Simple and Robust Flow Detection Algorithm Based on Spectral Clustering," Submitted to 5th International Conference on Research in Air Transportation (ICRAT 2012), 2012. + + + + + Trajectory Clustering and an Application to Airspace Monitoring + + MGariel + + + ASrivastava + + + EFeron + + + + IEEE Transactions on + + 12 + 4 + + Dec 2011 + + + Intelligent Transportation Systems + Gariel, M., Srivastava, A., and Feron, E., "Trajectory Clustering and an Application to Airspace Monitoring," Intelligent Transportation Systems, IEEE Transactions on, Vol. 12, No. 4, Dec 2011, pp. 1511-1524. + + + + + Discovering similar multidimensional trajectories + + MVlachos + + + GKollios + + + DGunopulos + + + + Proceedings. 18th International Conference on + 18th International Conference on + + 2002. 2002 + + + + Vlachos, M., Kollios, G., and Gunopulos, D., "Discovering similar multidimensional trajectories," Data Engineering, 2002. Proceedings. 18th International Conference on, 2002, pp. 673-684. + + + + + SMOTE: Synthetic Minority Over-sampling Technique + + NVChawla + + + KWBowyer + + + LOHall + + + WPKegelmeyer + + + + Journal Of Artificial Intelligence Research + + 16 + + 2002 + + + Chawla, N. V., Bowyer, K. W., Hall, L. O., and Kegelmeyer, W. P., "SMOTE: Synthetic Minority Over-sampling Technique," Journal Of Artificial Intelligence Research, Vol. 16, 2002, pp. 321 -357. + + + + + + diff --git a/file18.txt b/file18.txt new file mode 100644 index 0000000000000000000000000000000000000000..39afe7c867c28b2850d8b5dbad752ab68efaf135 --- /dev/null +++ b/file18.txt @@ -0,0 +1,536 @@ + + + + +Weight applied to ground holding in the calculation of total cost.w a Weight applied to airborne holding in the calculation of total cost. +I. IntroductionWhen the capacity of en route airspace is limited due to convective weather or other factors, the traffic through this area must often be delayed to prevent capacity from being exceeded.Weather conditions in recent years have caused approximately 65% of delays, 1 so this is an important problem.The term used to refer to a region of airspace with reduced capacity is Flow Constrained Area (FCA).Flights scheduled to fly through the FCA are identified and given an option of routing around the FCA.Flights that choose to travel through the FCA must be controlled so that the FCA capacity is not exceeded.Recent research has focused on the development of algorithms for scheduling ground and airborne delay. 2,3 ith the inherent uncertainty in weather forecasts, and thus in FCA capacity predictions, an important aspect of any scheduling algorithm is the way in which this uncertainty is handled.Scheduling algorithms have accounted for this uncertainty in a variety of ways.Bertsimas and Stock Patterson assume a deterministic capacity model in Ref. 4 and note that more work must be done to account for the uncertainty in en route capacities dependent on weather.Hoffman et al. 2 present a method of scheduling flights for use in an AFP that is based on en route resource allocation.Here, as in Ref. 4, the authors mention considering for future work the stochastic effects of uncertainty in airspace capacity.In Ref.3, Krozel et al. assume that the weather forecast will be accurate for some amount of time, beyond which the forecast is uncertain.The authors propose rescheduling at a fixed frequency that ensures a reschedule occurs before the planning horizon extends into the uncertain time.In testing their algorithm, they use more or less severe weather predictions, as compared to the actual weather, beyond the accurate forecast horizon to obtain a rough measure of the robustness of their algorithm to uncertainty in weather predictions.However, their algorithm does not systematically consider various weather forecasts and their relative probabilities.In this work, probabilistic forecasts of FCA capacity are incorporated into scheduling.Furthermore, a cost-benefit analysis is proposed to determine when departure rate replanning should occur.A discrete time aggregate flow model, similar to the cell transmission model, 5 describes the flow of aircraft through a network of airspace.This solution method is iterative.It involves solving two linear programs (LPs).One solves for new departure rates and for airborne holding.The other fixes departure rates at their current values and solves for airborne holding.Updated departure rates from the first LP are implemented only when they lead to a significant cost reduction.Both LPs consider probabilistic airspace constraint forecasts.The cost function is the expected value of a combination of the ground holding cost and the cost of airborne holding.These LPs are solved repeatedly throughout the day with updated FCA capacity forecasts.Background is provided in Section II.The problem is stated and the solution method is then described in Section III.Simulations using synthetic data compare this approach to simpler approaches that use only one capacity profile to represent predicted FCA capacity.Simulations designed to study the effects of changing the cost of replanning on the total cost as compared to replanning at a fixed frequency were performed as well.Simulation results are reported in Section IV. Conclusions and future work are discussed in Section V. +II. BackgroundCurrently, the maximum flow through an FCA is set by air traffic controllers based on their experience controlling traffic in that region during similar weather events.A more systematic use of probabilistic forecasts of the capacity of an FCA may allow for better utilization of airspace and greater FCA throughput.This possibility is investigated in this research.Therefore, it is focused on the use of probabilistic airspace capacity forecasts as opposed to the generation of these forecasts.Some probabilistic weather forecasts exist and are a current topic of research.For instance, the National Convective Weather Forecast (NCWF-2) provides a probabilistic forecast for the NAS with a 0 to 2 hour range.A version of this forecast, NCWF-6, extends the forecast to 6 hours.The Collaborative Convective Forecast Product (CCFP) is a 2, 4, and 6 hour forecast product that is probabilistic in the sense that each predicted weather polygon is assigned a low or high confidence. 6A more recent product known as the Localized Aviation MOS Program (LAMP) provides probabilities of convection in 20 km by 20 km grid cells up to 25 hours in the future. 7eather forecasts must also be converted into predictions of flight deviations to be useful for traffic flow management (TFM) decision making.The Convective Weather Avoidance Model (CWAM) 8 uses deterministic weather forecasts from the Corridor Integrated Weather System (CIWS) 9 to predict pilot deviations around weather.More specifically, CWAM produces polygons containing weather and specifies the proba-bility that pilots will deviate around each polygon.Another aspect of determining airspace capacity due to weather is finding not just the probability of flight deviations, but also a deterministic or probabilistic forecast of the resulting amount of traffic flow that can pass through a particular type of airspace.This sort of computation involves the structure of the airspace that is not passable due to weather.For example, the use of a scenario-based stochastic weather model to generate expected capacity and a probability distribution of capacity for an FCA is proposed in Ref. 10.For each weather scenario, the authors use a mincut algorithm, similar to the algorithm presented in Ref. 11, to generate a set of routes through the FCA which avoid convective weather.A capacity forecast and probability are associated with the weather avoidance routes generated for each weather scenario.Thus probabilistic capacity profiles for weather avoidance routes through the FCA are generated from probabilistic weather forecasts.As suggested by the authors of Ref. 10, other route generation algorithms, such as the Flow-Based Route Planner 12 which generates routes through an FCA with deterministic constraints or the scanning method, 13 could be extended to work with stochastic weather models.The research in this paper is built on an approach similar to that presented in Ref. 10, in which uncertainty in weather predictions gets translated into several different possible capacity profiles for each weather avoidance route through an FCA, each with an associated probability of occurring.An example set of capacity profiles for a weather avoidance route through an FCA during a weather event is shown in figure 1. Associated with the capacity forecast is an accurate forecast horizon, T .The accurate forecast horizon is the length of time over which the weather forecast is assumed accurate, and thus the predicted capacity of the weather avoidance route is assumed accurate.All capacity profiles are identical over the accurate forecast horizon.Beyond this horizon, the capacity profiles may differ.Each profile has a probability that corresponds to uncertainty in the weather forecast.Finally, much research has been done on both static and dynamic and both deterministic and stochastic management of Ground Delay Programs (GDPs), 14,15 which constrain arrivals to weather-impacted airports while also considering equity.This line of research has not yet been extended to the management of weather-induced congestion in airspace, even though much of it may be directly applicable.Additionally, some work has been done to find scenarios for airport acceptance rates with corresponding probabilities from historical traffic data, 16 which could also prove useful in generating FCA acceptance rate scenarios. +III. Scheduling using a Probabilistic Capacity ForecastWith the background given in Section II, the formal problem statement is given in Section III.A.The proposed approach to solving this problem is outlined in Section III.B, with details of this method following in Sections III.C through III.G.Concluding this section, advantages and disadvantages are summarized in Section III.H. +III.A. Problem StatementUsing the capacity forecast framework discussed in Section II, the problem can be stated as: Given several possible capacity profiles for weather avoidance routes through an FCA with associated probabilities, find airport departure rates and airborne holding to satisfy capacity constraints at the FCA while minimizing the expected value of the delay cost. +III.B. ApproachThe schedule of departures and airborne holding for flights going to an FCA should be updated throughout the day to react to changing weather conditions.Most scheduling methods, e.g.Ref. 3, replan at a fixed frequency.A contribution of this paper is that a cost-benefit analysis determines when a departure replan should be implemented.In fact, AFPs are currently only updated if the update saves at least 10 minutes per flight, so a variation of this cost-benefit approach is already in use.Repeated changes to departure times cause aggravation for passengers and airlines and costly logistical changes.The value Γ is the reduction of cost that is required for a departure replan to be implemented.It can be chosen to ensure that any departure rate update results in average delay savings greater than some specified minimum.Replanning is done before the end of the accurate forecast horizon to ensure the satisfaction of constraints.At each time step, the weather forecast is updated and two separate linear programs (LPs), which will be formally defined later, are solved.Problem 1 generates departure rates and airborne holding.Problem 2 uses a fixed departure rate and generates airborne holding.As discussed in Section III.E, the cost that is to be minimized by both LPs is the expected value of the weighted sum of ground holding cost and airborne holding cost.The planning horizon ranges from time step K I to K F .Both problems find solutions starting at the current time step K, where K ≥ K I , through K F .They are solved at some frequency, K R , where K R is less than or equal to the accurate forecast horizon, K.Both K R and K have units of time steps.The algorithm proceeds as described here:Input: FCA capacity profiles with associated probabilities, flight schedule, accurate forecast horizon K, and replan frequency K R (where K R ≤ K) Output: Departure rates and airborne holding while K ≤ K F do Solve Problem 1 for departure rates and airborne holding for the time period K to K F ; Solve Problem 2 for airborne holding only for the time period K to K F , using departure rates from previous loop (K R time steps ago); if cost(Problem 1)+Γ < cost(Problem 2) then return departure rates and airborne holding from Problem 1; end else return departure rates from previous iteration and airborne holding from Problem 2;end K = K + K R ; endThe integration of the proposed departure rate and airborne holding scheduling method with weather prediction methods and the NAS is depicted in a flow chart in figure 2. The proposed scheduling method is represented by the red block titled "Generate TFM Control."Terms c s and p s represent the FCA capacity forecast and its associated probability of occurring, respectively, for each weather scenario, s = 1, . . ., S. The controlled departure rate and airborne holding control denoted by d and u, respectively.The departure rates are constrained to be the same for each weather scenario both during and beyond the accurate forecast horizon.This is done so that estimated departure times can be assigned for flights scheduled to depart after the accurate forecast horizon.The airborne holding solution for each weather scenario is constrained to be identical for each weather scenario for the duration of the accurate forecast.Beyond the accurate forecast horizon, the airborne holding solutions for each weather scenario are allowed to differ.This results in the generation of departure rates and airborne holding that can actually be implemented for the accurate forecast horizon.Beyond this horizon the capacity of the FCA is uncertain, so the solution includes finding airborne holding for each weather scenario, which is then incorporated into the expected value of the cost function.In this way, the cost of future control is incorporated into the cost function and thus affects the departure rates and airborne holding calculated at the current time step. +III.C. Airspace Model and Variable DefinitionsThe model framework, associated variables and nomenclature are introduced in this section to provide context for the variables that will be used in the formal statements of Problems 1 and 2. This section gives a high level overview of network dynamics and variable names.Equations relating these variables are stated as constraints in Problems 1 and 2 in Section III.D.These constraints ensure that the flow of aircraft through the network of sections leading to the FCA is physically achievable and feasible for the given set of FCA capacity constraints.Details of these constraints will be given in Sections III.F and III.G.Airspace and flights are described mathematically by a discrete time, aggregate flow model.One of the advantages of using an aggregate model is that it can be described with linear equations.The equations can be used as constraints in a convex optimization problem, as is done in this work.Several aggregate flow models have been proposed with four such models compared in Ref. 17.The model used for this research is similar to the cell transmission model. 5In the model, routes between airports and the FCA are divided into sections.These sections could be existing sectors or other volumes of airspace.Figure 3 shows a simple example.The length of a time step is represented by ∆t.All aircraft are assumed to travel at the same average speed within each section, and thus the traversal times, τ i , i = 1, . . ., N , for each section specify how many time steps all flights require to traverse each section.Scheduled airborne holding can be thought of as requiring certain flights to remain in a given section longer than the average flight time.Airborne holding can be realized in the real system by decreasing flight speed, commanding a vector for spacing which increases the length of the flight path, or placing flights into holding patterns.Aircraft take off from the departure airport at the controlled rate d and enter section 1. Aircraft flow from one section to the next sequentially within the route.Aircraft exiting the final section of the route, N , enter the weather avoidance route through the FCA.The term c s (k) represents the predicted flow rate capacity of the weather avoidance route during time step k under weather scenario s.The flow of aircraft through a typical section of the route is presented graphically in figure 4. Subscripts of variables represent the index of a section.Superscripts represent the index of a capacity forecast.The time step is indicated within the parentheses.The quantities x and h represent the number of aircraft in a section and holding at a section, respectively.The quantities u and q represent the number of aircraft flowing from the given section into or out of holding (u) and from one section to another (q).For example, the term x s i (k) represents the number of aircraft in section i during time step k under weather scenario s.It is assumed that both section and holding capacities are limited, with xi the maximum allowable capacity for section i and hi the maximum allowable holding capacity at section i.Equations which describe the flow of material between sections and holding and ensure the feasibility of solutions will be given as constraints of Problems 1 and 2 and will be explained in more detail in Sections III.F and III.G. +III.D. Statement of LPsIn this section, Problems 1 and 2 are formally defined.These problems are formulated as linear programs designed to minimize delay subject to uncertain airspace capacity constraints.Here, the uncertainty in airspace capacity constraints is explicitly incorporated by generating airborne holding beyond the accurate forecast horizon for each possible weather scenario.The airborne holding schedule determined for each scenario is included in the cost by taking the expected value of the airborne holding cost over all possible scenarios.More details of the formulation of the cost function are given in Section III.E.This cost is minimized subject to dynamic and feasibility constraints.Dynamic constraints describe the physical flow of aircraft between sections and holding areas.The feasibility constraints ensure that control actions are achievable and that airspace capacity constraints are not violated.Detailed descriptions of dynamic and feasibility constraints are given in Sections III.F and III.G.The control inputs available using this model framework are holding control at each section under each weather scenario, u s i (•) for all i and s, and the departure rate at the airport, d(•).Problems 1 and 2 find u s i (•) and d(•) to minimize the expected value of delay costs while satisfying dynamic and feasibility constraints. +Problem 1Givenc s (k), k = K, . . . , K F , ∀ s D(k), k = K I , . . . , K F xi , hi , ∀ i Find d(k), k = K, . . . , K F u s i (k), k = K, . . . , K F , ∀ i, ∀ sMinimize Cost (see Section III.E for details)w g   K F k=K I k l=K I [D(l) -d(l)]   ∆t + w a S s=1 p s   K F k=K I N i=K I h s i (k)   ∆t (1)Subject To For each s = 1, . . ., S Dynamic Constraints (see Section III.F for details)h s i (k + 1) = h s i (k) + u s i (k + 1), i = 1, . . . , N, k = K, . . . , K F(2)q s 1 (k + 1) = d(k + 1 -τ 1 ) -u s 1 (k + 1), k = K, . . . , K F(3)q s i (k + 1) = q s i-1 (k + 1 -τ i ) -u s i (k + 1), i = 2, . . . , N, k = K, . . . , K F (4) x s 1 (k + 1) = x s 1 (k) + d(k + 1) -d(k + 1 -τ 1 ) k = K, . . . , K F(5)x s i (k + 1) = x s i (k) + q s i-1 (k + 1) -q s i-1 (k + 1 -τ i ), i = 2, . . . , N k = K, . . . , K F(6)Feasibility Constraints (see Section III.G for details)k l=K0 D(l) - k l=K0 d(l) ≥ 0, k = K, . . . , K F(7)u s i (k) = u 1 i (k), k = K, . . . , K + K, ∀ i (8) -h s i (k) ≤ u s i (k + 1) ≤ q s i-1 (k + 1 -τ i ), k = K, . . . , K F , i = 2, . . . , N(9)-h s 1 (k) ≤ u s 1 (k + 1) ≤ d(k + 1 -τ 1 ), k = K, . . . , K F (10) 0 ≤ x s i (k) ≤ xi , k = K, . . . , K F , ∀ i (11) 0 ≤ h s i (k) ≤ hi , k = K, . . . , K F , ∀ i(12)q s N (k) ≤ c s (k), k = K, . . . , K F(13)Problem 2 Givenc s (k), k = K, . . . , K F , ∀ s D(k), k = K I , . . . , K F d(k), k = K, . . . , K F xi , hi , ∀ i Find u s i (k), k = K, . . . , K F , ∀ i, ∀ s Minimize Equation (1). Subject To For each s = 1, . . . , S Dynamic Constraints Equations (2) to (6). Feasibility Constraints Equations (8) to (13).Problem 2 is similar to Problem 1 with only the parameters and variables differing between the two problems.Note that constraint (7) is not included in Problem 2 since this constraint is assumed to be satisfied by the given departure rate, d(•).The number of variables in each of these problems depends on the number of sections used to model the airspace, the number of airports, and the number of weather scenarios considered.However, all equations are linear, which leads to less complex problems than those encountered when using Lagrangian methods.Also, dual decomposition methods can be used to solve these optimization problems relatively quickly. 18inally, since an AFP is designed to control air traffic scheduled to fly through a specific region of airspace, only a subsection of the NAS needs to be modeled.Actually, the solutions to these problems must be integers because only integer number of aircraft can depart or be in a region of airspace at a given time.This constraint is ignored in this research, but if this approach were implemented in a more realistic simulation of the NAS, such as the Future ATM Concepts Evaluation Tool (FACET), 19 in which control inputs must be generated for individual flights, rounding heuristics could be applied or the problems could be posed as integer programming problems. +III.E. CostTypically, the cost used for similar TFM problems is the weighted sum of ground and airborne delay. 3,4 ighted delay is used to account for the additional costs associated with airborne delay, such as the cost of additional fuel consumed.In Problems 1 and 2 a single departure rate profile is found for all capacity profile scenarios, resulting in a deterministic ground holding cost.Since airborne holding is generated for each scenario, a deterministic airborne holding cost does not exist.Therefore, the overall cost is an expected value that is expressed as where K I and K F are the initial and final time steps of the full plan, respectively, w g and w a are the ground and airborne holding weights, respectively, p s is the probability that weather scenario s will occur, and S is the total number of weather scenarios given for the weather avoidance route.Note that it is required that S s=1 p s = 1 and p s ≥ 0 for all s.w g   K F k=K I k l=K I [D(l) -d(l)]As is commonly done in the literature, e.g.Ref. 3, values of w g = 1 and w a = 2 roughly represent the fact that airborne holding is more expensive.Airborne holding costs airlines both time and money due to additional fuel consumption.Most TFM literature has used a w a value of 2 but without providing justification.Airline operating cost data from table 4-1 of Ref. 20 suggests that this ratio was 1.18 in 2002.More recent research 21 suggests that, due to increases in fuel costs, the ratio was about 1.96 in 2006.If fuel costs continue to increase or if airlines are charged for emissions, the ratio could become even higher. +III.F. Dynamic ConstraintsIn this section, the details of the dynamic constraints, equations ( 2) - (6), will be discussed.The dynamics of a typical section (index i = 2, . . ., N ) under weather scenario s are described first.The dynamics of section 1 are slightly different than a typical section since the inflow to this section depends on the departure rate.Thus, section 1 dynamics are described separately.For a specific weather scenario, s, the number of aircraft in section i during time step k is x s i (k).The number of aircraft flowing into section i from an upstream section i -1 during time step k + 1 is q s i-1 (k + 1).The number of aircraft flowing out of section i during time step k + 1 is the number of aircraft that entered that section during time step k + 1 -τ i , i.e., q s i-1 (k + 1 -τ i ).Of the aircraft exiting section i, some number may enter holding at section i while others will either enter section i + 1 if i = 2, . . ., N -1 or the FCA if i = N .If u s i (k + 1) > 0, the control variable u s i (k + 1) represents the net number of aircraft exiting section i that are sent into holding at section i.If u s i (k + 1) < 0, this value represents the net number of aircraft in holding at section i that are removed from holding and sent on to section i + 1 if i = 2, . . ., N -1 or to the FCA if i = N .The control input of aircraft sent into or removed from holding at a given section represents the net change in the number of aircraft holding at that section.For example, even if u s i (k + 1) = 0, some aircraft exiting section i may be sent into holding while the same number of aircraft are removed from holding and sent on to section i + 1, resulting in a net change in aircraft holding at section i of 0 aircraft.The variable h s i (k + 1) represents the number of aircraft holding at section i during time step k + 1, the dynamics of which are described by Eq. ( 2).The number of aircraft exiting section i or leaving holding at section i and entering either the subsequent section if i = 2, . . ., N -1 or the FCA if i = N during time step k + 1 is described by Eq. ( 4).Thus the number of aircraft in section i at the end of time step k + 1 is given by Eq. ( 6).The flow of aircraft into and out of section 1 is defined by the departure rate at the airport.Therefore, Eq. ( 3) describes the number of aircraft exiting section 1 or leaving holding at section 1 and entering section 2 during time step k + 1.Similarly, Eq. ( 5) gives the number of aircraft in section 1 at the end of time step k + 1. +III.G. Feasibility ConstraintsThe feasibility constraints, equations ( 7) -( 13), of Problems 1 and 2 are discussed in more detail in this section.These constraints ensure that control actions are feasible, section and holding capacities are not exceeded, and the FCA flow rate capacity is not exceeded.Equation ( 8) forces the airborne holding control to be identical for every weather scenario as long as the weather is assumed to be accurate ( K).These controls can then be implemented until the next replan occurs, which is required to happen before the end of the accurate weather prediction horizon.The total number of controlled departures at the end of time step k must not exceed the total number of aircraft originally scheduled to depart by the end of time step k.This avoids assigning a departure time that is earlier than the time at which this flight will be available for departure.This constraint can be represented mathematically by the inequality in Eq. ( 7), where D(l) represents the number of aircraft originally scheduled to depart from the airport during time step l.The number of aircraft that are placed into holding at section i during time step k + 1 must not exceed the outflow of that section, q s i-1 (k + 1 -τ i ), and the number of aircraft that are removed from holding during time step k + 1 must not exceed the total number of aircraft holding in that section during time step k, h s i (k).These constraints are captured by the inequalities in Eq. ( 9) and (10).Since the state of each section represents the number of aircraft in that section, the value of x s i (k) must be nonnegative for all i, k, and s.Additionally, due to limitations in the number of aircraft that can be safely routed through any particular section, a maximum capacity constraint can be enforced for each section.Both of these constraints are captured in the inequality in Eq. ( 11), where xi is the maximum allowable capacity of section i.Note that xi could be made dependent on the weather scenario but that in this work the weather scenarios are assumed to only impact the FCA, not the airspace around the FCA.Similarly, the term h s i (k) represents the number of aircraft in holding at section i during weather scenario s at the end of time step k and must be a nonnegative value.Note that sinceu s i (k + 1) = h s i (k + 1) -h s i (k)by a rearrangement of Eq. ( 2), the constraints in Eq. ( 9) and ( 10) include the constraint that the number of aircraft holding at each section must be nonnegative, i.e. h s i (k + 1) ≥ 0. The maximum holding capacity at each section represents the number of aircraft that can safely be held at each section.This constraint is represented by the inequality in Eq. ( 12), where hi is the maximum allowable holding capacity at section i.This capacity is set to zero at sections where holding is not possible.Also, it is possible to define an upper bound on the sum of x s i (k) and h s i (k) rather than bounding them individually, but this was not done in this research.The values of the maximum section and holding capacities can be chosen based on constraints specific to the region of airspace affected by the AFP and weather conditions of the AFP.Thus the selection of these capacities provides an additional level of control design.For each weather scenario, the number of aircraft exiting section N and entering the FCA under weather scenario s during time step k must not exceed the predicted flow rate capacity of the weather avoidance route through the FCA under weather scenario s.This flow rate constraint is captured by Eq. ( 13), where c s (k) represents the predicted capacity of the weather avoidance route during time step k under weather scenario s. +III.H. Advantages and DisadvantagesSeveral properties of this proposed solution method are worth noting.As discussed previously, this approach requires that a change in departure rates induce cost savings of more than Γ.The value of Γ may be chosen to ensure that the delay cost savings are not exceeded by implementation costs.This would require an estimate of airline specific costs, which may be difficult or impossible to calculate, depending on operation cost information that each airline is willing to share.The strategy will not exceed the FCA capacity.The probabilistic cost of future control is considered.More likely weather scenarios will impact the control actions more than less likely scenarios.These features can be incorporated into solutions for many TFM problems, and they are considered contributions of this research.A limitation of this approach is that intra-airline flight substitutions in the AFP are only guaranteed to be permissible if the two flights being substituted originate at the same airport and enter the FCA at the same weather avoidance route.These substitutions give airlines flexibility to use the arrivals to an FCA as they prefer.If flights originating at different airports or entering the FCA at different weather avoidance routes exchange their arrival times, the constraints on the capacity of the airspace sections and the weather avoidance routes may be violated.In order to accommodate flight substitutions of this type, each substitution would have to be evaluated and permitted if it does not lead to any constraint violations. +IV. Simulations and ResultsResults are presented comparing the proposed probabilistic approach to methods using different deterministic capacity profiles.Results for a one airport problem are presented in Section IV.B and results for a two airport problem are given in Section IV.C. Simulations were also performed to study the effects of varying the replanning frequency and replanning cost, Γ, with results presented in Section IV.D.Simulations are performed for a number of weather events, where each event consists of a set of three possible capacity profiles and associated probability of occurring.In Section IV.A, the generation of these weather events and creation of deterministic capacity profiles for these events are described.Additionally, parameters common to each simulation are given at the end of the section. +IV.A. Simulation Parameters and SetupThe process of generating synthetic capacity predictions for the FCA proceeds as follows.Each synthetic capacity profile consists of three stages, a decreasing capacity stage, constant low capacity stage and increasing capacity stage.The low capacity value of each profile was randomly selected from a uniform distribution of the integers from 2 to 8 aircraft per time step.The low capacity values for each profile were required to be distinct.The duration of the decreasing and increasing stages were independently randomly chosen from a uniform distribution of the integers from 4 to 7 time steps.The duration of the minimum stage was randomly selected from a uniform distribution of the integers from 7 to 9 time steps.The start time of the decreasing stage was randomly selected from a uniform distribution of integer values while ensuring that the entire weather event occurred within a 24 time step block starting at time step 11.A typical set of three capacity profiles for the weather avoidance route through the FCA is given in figure 1.As labeled in figure 1, the profiles will be referred to as the high, mid, and low capacity profiles.In an effort to compare the proposed probabilistic approach, simulations were performed using the iterative approach discussed in Section III.B with approaches that use single profile FCA capacity forecasts for planning.When using a single capacity profile to represent the predicted capacity of the weather avoidance route, the expected value of the cost used in Problems 1 and 2 becomes a deterministic value.The results of these simulations are given in Sections IV.B and IV.C.For each weather event used in these simulations, airport departure rates and airborne holding were planned using the method presented in Section III using one of the following capacity forecast methods: high, mid, low, expected value, probabilistic and constant.The high, mid, and low capacity forecast methods simply use the high, mid or low capacity profile as the predicted capacity.The expected value forecast uses the expected value of the capacity at each time step based on the three capacity profiles and their associated probabilities of occurring; an example is shown in figure 5.The method presented in Section III of using all three capacity profiles and minimizing the expected value of the cost is referred to as the probabilistic method.The constant capacity profile is constant throughout the duration of the weather event.This constant value is the average of all three capacity profiles for the average duration of the weather event; an example profile is shown in figure 5.The use of a constant capacity is motivated by current AFP practices in which the default procedure is to set a fixed constant capacity.Changes to the FCA predicted capacity are made only if significant cost reduction can be achieved by adjusting the FCA capacity.When solving the problem using one of these single profile methods, the set of possible FCA capacities, c s (•) for s = 1, . . ., S consisted of a single profile, thus S = 1.When using the probabilistic method, all three profiles were used, each of which had an equal probability of occurring.A set of 10 randomly generated weather events was used in these simulations in order to capture a small variety of weather conditions.When solving the problem using a specific weather event, one of three possible capacity profiles was chosen to be the "actual capacity profile."That is, one capacity profile was chosen to be the actual capacity profile of the FCA.Since all capacity profiles have equal probability, for each weather event, three simulations were performed.Each capacity profile, high, mid and low, was used as the actual weather profile in one of these three simulations.This was done to generate average results over all capacity profiles that could occur.Future research should consider the more realistic scenario where the actual capacity of the FCA does not exactly match any forecast.For each solution, departure rates and airborne holding were scheduled.For each actual weather scenario, there is an optimal solution within the modeling and control framework used.The optimal solutions are obtained when the predicted FCA capacity matches the actual FCA capacity exactly, i.e. the predicted FCA capacity is accurate for the entire planning horizon.The optimal solutions use ground holding only and the FCA capacity is matched precisely throughout the planning horizon.As a method of normalizing results, costs are reported in terms of a percent of this optimal cost.The parameters common to each of the simulations are given in table 1. Parameters specific to each simulation will be given in the appropriate results sections. +IV.B. One Airport Problem ResultsUsing the problem setup consisting of a single airport and single weather avoidance route through the FCA, shown in figure 3, several simulations were performed to study the behavior of the proposed solution method.The problems use the parameters listed in table 1.Additionally, the originally scheduled departure rate from the airport, D, is 10 aircraft per time step for the first 36 time steps (or 1 aircraft per minute for 6 hours), dropping off to 0 aircraft per time step for the rest of the planning horizon.As shown in figure 3, the route leading from the airport to the FCA consists of 10 sections.The replanning cost is 0, so new departure rates are implemented whenever calculated.Simulations were performed to compare the proposed probabilistic approach to several single capacity profile prediction methods, each of which is described in Section IV.A.The results are presented in table 2. For a given capacity prediction method and actual capacity profile, the "Average Cost as Percent of Optimal" is the average cost incurred divided by the average optimal cost for the 10 events simulated.The "Overall Average Cost as Percent of Optimal" for each prediction method is the average cost, over all 10 events using each of the 3 capacity profiles as the actual capacity profile, divided by the average optimal cost for the same set of events.Note that using the high capacity prediction method with the actual capacity being the high profile, the resulting average cost as a percent of optimal is 100%.This is because the capacity prediction matched the actual capacity for the entire planning horizon and the optimal solution was found.The "Average Cost as a Percent of Optimal" results are presented graphically in figure 6(a).From this chart, it can be seen that the probabilistic forecast method achieves a cost reduction of approximately 13% of the optimal cost over the constant capacity method, which corresponds to an average total delay reduction of 5.8 minutes per flight.It can also be seen that the high capacity profile prediction method outperforms the probabilistic method with a cost reduction of about 1.3% over the probabilistic method.It is interesting to note that when using the high prediction method, a larger portion of the cost is derived from airborne holding than any other method.This is because this prediction is optimistic and will allow more flights to depart which then must be held in the air when the actual FCA capacity is lower than originally predicted.All other capacity prediction methods incur the largest cost penalties when the high capacity profile is the actual capacity profile.However, comparing the high capacity prediction method to the probabilistic prediction method when the mid and low capacity profiles are the actual capacity profiles, there is a significant advantage, 11% and 12% of the optimal cost, respectively, to using the probabilistic prediction method.The corresponding average total delay reductions per flight are 1.0 and 2.9 minutes, respectively.While costs tend to be lower when more capacity is available, the cost penalty for not planning to fully take advantage of that capacity is a relatively large fraction of the cost that is achieved if all the capacity is used.It can be concluded that, if probabilistic capacity forecasts are available, the use of the proposed scheduling scheme will lead to a cost reduction over the use of most of the methods studied here.The only single capacity profile method which out performs, on average, the proposed probabilistic method is the high capacity prediction method.However, in circumstances in which the actual capacity profile is the mid or low, the use of the probabilistic method can significantly reduce delay costs.Additionally, the high capacity prediction method assigns over twice the amount of airborne holding than the probabilistic method, on average.Additional simulations were performed using randomly generated probabilities that each of the three scenarios would occur, as opposed to each scenario occurring with equal probability.For each weather event, the real weather scenario was selected using the associated probability distribution.The results were found to be similar to those presented in figure 6. +IV.C. Two Airport ProblemSimulations similar to those involving the one airport problem presented in Section IV.B were performed for a problem involving two airports and merging flow.A diagram of the specific network considered is shown in figure 7. The departure rate at airport A is 7 aircraft every 10 minutes and 3 aircraft every 10 minutes at airport B. Departure rates at these airports were chosen so that the time history of the demand at the FCA is the same as that of the one airport problem.The same set of 10 events, with equal probability that each of the three capacity profiles will be the real capacity profile, as used in one airport problem were used for the two airport problem.The average cost as a percent of optimal for each capacity prediction method is presented in figure 8(a).The data here exhibit the same trends as seen in the one airport problem, figure 6(a).However, for each capacity prediction method, costs are lower for the two airport problem than for the one airport problem.Solutions to the two airport problem use less airborne holding than solutions to the one airport problem.This is due to the fact that the departure rate at airport B can be adjusted to react to changes in the FCA capacity prediction over a shorter time scale since flights departing from airport B have a shorter nominal flight time to the FCA than those departing from airport A. In figure 8(b) average ground holding cost as a percent of optimal per flight is shown for each departure airport.The issue of equitably distributing delay among flights arises in both AFPs and GDPs involving multiple departure airports.Several different scheduling paradigms for assigning delays to flights in an AFP and their effects on the solutions, including delay equity, are discussed in Ref. 2. At the aggregate level, equitable solutions will assign the same average delay per flight for all flights involved in the AFP, regardless of the respective distances from departure airports to the FCA.Notice that, although the high capacity prediction method results in an average cost which is lower than the average cost using the probabilistic method, the probabilistic method is more equitable, distributing ground holding more evenly to flights leaving from each of the two departure airports.Using the high capacity prediction method, average ground delay for flights departing from airport B is 65% longer than flights departing from airport A, as opposed to only 20% when using the probabilistic capacity prediction method. +IV.D. Replanning Cost and Frequency ComparisonIn this section, replanning departure rates at a fixed frequency is compared to using a cost-benefit analysis to determine when departure rates should be updated.The one airport problem of figure 3 was used for this study.In addition to the parameters listed in table 1, the accurate forecast horizon is 120 minutes.Replanning frequencies of once every 30, 60, 90, and 120 minutes were used for the fixed replanning frequency simulations.For these simulations, the replanning cost was set to 0. Thus, new departure rates were implemented whenever calculated.Replanning costs, Γ, of 0, 500, 1000, and 1500 were used with Problems 1 and 2 solved once every 30 minutes.In these simulations, airborne holding control could be updated every 30 minutes while departure rates were recalculated every 30 minutes, but actually implemented only if they resulted in a cost benefit greater than the replanning cost.As described in Section IV.A, a set of three possible capacity profiles were randomly generated for each of ten weather events.Each capacity profile was given an equal probability of occurring.The probabilistic capacity prediction method was used for all simulations.As described in Section IV.A, three simulations were performed for each weather event in which each capacity profile, high, mid and low, was used as the actual capacity profile in one of these three simulations.The average cost as percent of optimal was generated for these simulations in the manner described in Section IV.B.Additionally, for each solution, the actual number of departure replans performed throughout the weather event were found.Note that even when replanning at a fixed frequency with no replanning cost, departure rates may not change from one replanning cycle to the next.Thus, when calculating the average number of departure replans, an actual departure replan is one in which the new departure rate plan differs from the previous departure rate plan.In figure 9, the average cost as a percent of the optimal is plotted versus the average number of departure replans for both the fixed replanning frequency method and the cost-benefit replanning method for cases in which the actual weather is the high, mid and low profile.There is a slight (1% to 2%) improvement in cost when using the cost-benefit method to reschedule departure rates for the low and medium capacity actual weather profiles.For the high capacity profile, the cost savings as a percentage of the optimal cost are as high as 20% in some cases.This is partially because the optimal cost for the high capacity situation is lower to begin with, but also because replanning is not likely to be needed to avoid airborne holding as is the case when the lower capacity profiles occur.Also, note that if replan costs are added to the overall cost values, then the benefits of this cost-benefit replanning method would be more profound.As mentioned earlier, it may be difficult to determine the appropriate value of Γ to be used.However, the 10 minute per flight minimum delay reduction that is currently required for an AFP rate change could be used as a starting point.The operational use of this 10 minute threshold and the fact that an appropriate Γ will reduce workload for all participants without sacrificing any cost suggest that it should be possible to arrive at a consensus for the Γ value. +V. Conclusions and Future WorkA method for scheduling ground delay and airborne holding for flights scheduled to fly through an FCA with uncertain capacity has been proposed.The contributions of this approach are that it explicitly considers probabilistic constraint information, it only implements changes to the departure schedule when the resulting cost savings are sufficient to justify the change, and it takes advantage of new constraint information as it becomes available to ensure that capacity constraints are not violated.However, this approach does impose some restrictions on flight substitutions for flights involved in an AFP.A variety of simulations have demonstrated the behavior of this approach.In these simulations, the constant capacity estimation method performed at 129% of the optimal cost level, while this approach was able to operate at 116% of the optimal cost.Similar performance is observed when this approach is applied to a two airport problem.Although the approach of assuming that the capacity will be high performs better than the probabilistic approach proposed here when considering cost only, there are some advantages to using the probabilistic approach.In the one airport problem, the high capacity prediction method assigns over twice the amount of airborne holding than the probabilistic method.The probabilistic method assigns delay more equitably in the two airport problem.Finally, a parametric study demonstrates that the cost-benefit feature of the proposed approach reduces the costs of the fixed planning frequency approach by between 1% and 20%, depending on which capacity profile occurs.Future work can be divided into theoretical, modeling and simulation work.More work must be done to understand why the approach of assuming that the capacity will be high performs better than the probabilistic approach in the simulations presented here.The primary future modeling work involves more complex networks of flows and multiple weather avoidance routes through FCAs.When considering these more complex problems, methods for disaggregating aggregate TFM controls into individual flight controls must be developed.When the flow network between airports and the FCA involves both merging and diverging, a mechanism for deciding which flights should take which network links must be developed.Similarly, when multiple weather avoidance routes through an FCA exist, a method for assigning flights to routes must be determined.Such methods should consider individual flight preferences.The approach should be simulated with real traffic data from an actual AFP.Such a simulation would require implementation decisions such as the appropriate size for the airspace sections in the model and the disaggregation of control actions.Sun et al. present a method for disaggregation in Ref. 22 which addresses some of the issues that will be encountered when simulating the proposed control method with real traffic data.of Illinois, Urbana-Champaign) for discussions regarding this research.This work was funded in part by a California Space Grant and NASA Aeronautics Scholarship Program Fellowship awarded to Heather Arneson.Figure 1 .1Figure 1.Example of capacity prediction for a single weather avoidance route through the FCA given as three possible capacity profiles with associated probabilities.The FCA capacity has units of number of aircraft per time step. +Figure 2 .2Figure 2. Chart showing the flow of information and TFM actions between weather forecast methods, capacity prediction methods, the NAS and the proposed method of scheduling departure rates and airborne holding control (represented by the block outlined in red). +Figure 3 .3Figure 3.One airport problem: Aircraft depart from a single airport and fly to the FCA using a single route. +Figure 4 .4Figure 4. Variables and dynamics associated with the flow of aircraft at a typical section. +Figure 5 .5Figure 5. Predicted FCA capacity profile using the expected value approach (left) and the constant capacity approach (right). +Table 1 .1Parameter values used for simulations discussed in Sections IV.B, IV.C and IV.D. Parameter Value Time step: ∆t = 10 minutes Traversal time: τ i = 1 time step, ∀ i Flight time from airport to FCA: 10 time steps Total number of departing flights: 360 Total simulation time: 72 time steps Section and holding capacity: not enforced +Average results by real weather scenario type. +Figure 6 .6Figure 6.Cost comparison for several different capacity prediction methods for the one airport problem.In figure6(a), average results over all events and all real weather scenarios are presented.In 6(b), average results over all events are broken down by the real weather profile type. +Figure 7 .7Figure 7. Two airport problem: Aircraft take off from airports A and B and fly towards the FCA using routes 1 and 2 respectively, and merge into route 3 before entering the FCA. +Ground holding cost comparison. +Figure 8 .8Figure 8. Cost comparison for several different capacity prediction methods for the two airport problem.In figure 8(a), average results over all events and all real weather scenarios are presented.In 8(b), costs are shown per flight departing from airport A and airport B. +Figure 9 .9Figure 9. Cost of replanning departure rates at a fixed frequency (red circles) and based on a cost-benefit analysis (blue squares) as a function of the average number of departure replans.Note the variation in y-axis scales. +Table 2 .2Average results from capacity model comparison simulations.CapacityActualAverageAverage Average Average Cost Overall AveragePredictionCapacityGroundAirborneCostas PercentCost as PercentMethodProfileHoldingHoldingof Optimalof OptimalHigh11155011155100.0HighMid16074200320080116.5114.9Low20264359627456120.9High1776622718220163.3MidMid17230017230100.0122.5Low22991207827146119.6High2349721823933214.6LowMid2436658625537148.2141.3Low22701022701100.0ExpectedHigh170786817214154.3ValueMid1849659319682114.2122.2Low21950178025510112.4High1615712916415147.2ProbabilisticMid1717652318222105.8116.2Low20909190024710108.8High172073417276154.9ConstantMid1960483821280123.5129.3Low23251212427500121.1 + + + + +AcknowledgmentsThe authors would like to thank Mark Evans (consultant), Al Mahilo (Supervisory Traffic Management Coordinator, Cleveland ARTCC), and David Foyle (Manager System Operations, Oakland ARTCC) for their insight into current AFP practices.Shon Grabbe and William Chan helped the authors understand current research related to weather-induced reductions in airspace capacity.The authors would also like to thank Cédric Langbort (Assistant Professor in the Department of Aerospace Engineering at the University + + + + + + + + + Resource Allocation Principles for Airspace Flow Control + + RHoffman + + + JBurke + + + TLewis + + + AFuter + + + MBall + + + + AIAA Guidance, Navigation, and Control Conference and Exhibit + + 2005 + + + Hoffman, R., Burke, J., Lewis, T., Futer, A., and Ball, M., "Resource Allocation Principles for Airspace Flow Control," AIAA Guidance, Navigation, and Control Conference and Exhibit, 2005. + + + + + An Algorithmic Approach for Airspace Flow Programs + + JKrozel + + + RJakobovits + + + SPenny + + + + Air Traffic Control Quarterly + + 14 + 3 + + 2006 + + + Krozel, J., Jakobovits, R., and Penny, S., "An Algorithmic Approach for Airspace Flow Programs," Air Traffic Control Quarterly, Vol. 14(3), 2006, pp. 203-229. + + + + + The Air Traffic Flow Management Problem with Enroute Capacities + + DBertsimas + + + SSPatterson + + + + Operations Research + + 46 + 3 + + May -June 1998 + + + Bertsimas, D. and Patterson, S. S., "The Air Traffic Flow Management Problem with Enroute Capacities," Operations Research, Vol. 46, No. 3, May -June 1998, pp. 406 -422. + + + + + Multicommodity Eulerian-Lagrangian Large-Capacity Cell Transmission Model for En Route Traffic + + DSun + + + AMBayen + + + + AIAA Journal of Guidance, Control, and Dynamics + + 31 + 3 + + May -June 2008 + + + Sun, D. and Bayen, A. M., "Multicommodity Eulerian-Lagrangian Large-Capacity Cell Transmission Model for En Route Traffic," AIAA Journal of Guidance, Control, and Dynamics, Vol. 31, No. 3, May -June 2008, pp. 616 -628. + + + + + Collaborative Decision Making, Federal Aviation Administration + + + + Collaborative Convective Forecast Product + Collaborative Decision Making, Federal Aviation Administration, "Collaborative Convective Forecast Product," . + + + + + LAMP CCFP Hybrid + + Avmet + + + + + Avmet, "LAMP CCFP Hybrid," . + + + + + Modeling Convective Weather Avoidance in Enroute Airspace + + RDelaura + + + MRobinson + + + MPawlak + + + JEvans + + + + Proc. of 13th Conference on Aviation, Range, and Aerospace Meteorology + of 13th Conference on Aviation, Range, and Aerospace MeteorologyNew Orleans, LA + + 2008 + + + DeLaura, R., Robinson, M., Pawlak, M., and Evans, J., "Modeling Convective Weather Avoidance in Enroute Airspace," Proc. of 13th Conference on Aviation, Range, and Aerospace Meteorology, New Orleans, LA, 2008. + + + + + Corridor Integrated Weather System + + JEEvans + + + ERDucot + + + + MIT Lincoln Laboratory Journal + + 16 + 1 + + 2006 + + + Evans, J. E. and Ducot, E. R., "Corridor Integrated Weather System," MIT Lincoln Laboratory Journal, Vol. 16, No. 1, 2006, pp. 59-80. + + + + + Airspace Throughput Analysis Considering Stochastic Weather + + JMitchell + + + VPolishchuk + + + JKrozel + + + + AIAA Guidance, Navigation, and Control Conference and Exhibit + Keystone, CO + + August 2006 + + + Mitchell, J., Polishchuk, V., and Krozel, J., "Airspace Throughput Analysis Considering Stochastic Weather," AIAA Guidance, Navigation, and Control Conference and Exhibit, Keystone, CO, August 2006. + + + + + Maximum Flow Rates for Capacity Estimation in Level Flight with Convective Weather Constraints + + JKrozel + + + JS BMitchell + + + VPolishchuk + + + JPrete + + + + Air Traffic Control Quarterly + + 15 + 3 + + 2007 + + + Krozel, J., Mitchell, J. S. B., Polishchuk, V., and Prete, J., "Maximum Flow Rates for Capacity Estimation in Level Flight with Convective Weather Constraints," Air Traffic Control Quarterly, Vol. 15, No. 3, 2007, pp. 209-238. + + + + + + JPrete + + + JS BMitchell + + Safe Routing of Multiple Aircraft Flows in the Presence of Time-Varying Weather + + + Prete, J. and Mitchell, J. S. B., "Safe Routing of Multiple Aircraft Flows in the Presence of Time-Varying Weather + + + + + + Data + + AIAA Guidance, Navigation, and Control Conference and Exhibit + + August 2004 + + + Data," AIAA Guidance, Navigation, and Control Conference and Exhibit, August 2004. + + + + + Airspace Availability Estimation for Traffic Flow Management Using the Scanning Method + + AKlein + + + LCook + + + + Proc. of AIAA/IEEE Digital Avionics Systems Conference + of AIAA/IEEE Digital Avionics Systems ConferenceSt. Paul, MN + + October 2008 + + + Klein, A. and Cook, L., "Airspace Availability Estimation for Traffic Flow Management Using the Scanning Method," Proc. of AIAA/IEEE Digital Avionics Systems Conference, St. Paul, MN, October 2008. + + + + + A Stochastic Integer Program with Dual Network Structure and Its Application to the Ground-Holding Problem + + MOBall + + + RHoffman + + + AROdoni + + + RRifkin + + + + Operations Research + + 51 + 1 + + January-February 2003 + + + Ball, M. O., Hoffman, R., Odoni, A. R., and Rifkin, R., "A Stochastic Integer Program with Dual Network Structure and Its Application to the Ground-Holding Problem," Operations Research, Vol. 51, No. 1, January-February 2003, pp. 167-171. + + + + + A Dynamic Stochastic Model for the Single Airport Ground Holding Problem + + AMukherjee + + + MHansen + + + + Transportation Science + + 41 + 4 + 2007 + + + Mukherjee, A. and Hansen, M., "A Dynamic Stochastic Model for the Single Airport Ground Holding Problem," Trans- portation Science, Vol. 41, No. 4, 2007. + + + + + Scenario-based air traffic flow management: From theory to practice + + PLiu + + + MHansen + + + AMukherjee + + + + Transportation Research Part B + + 42 + + 2008 + + + Liu, P., Hansen, M., and Mukherjee, A., "Scenario-based air traffic flow management: From theory to practice," Trans- portation Research Part B , Vol. 42, 2008, pp. 685-702. + + + + + Comparison of the Performance of Four Eulerian Network Flow Models for Strategic Air Traffic Management + + DSun + + + ISStrub + + + AMBayen + + + + Networks and Heterogeneous Media + + 2 + 4 + + December 2007 + + + Sun, D., Strub, I. S., and Bayen, A. M., "Comparison of the Performance of Four Eulerian Network Flow Models for Strategic Air Traffic Management," Networks and Heterogeneous Media, Vol. 2, No. 4, December 2007, pp. 569 -595. + + + + + Large-scale modeling and optimization of en route air traffic flow + + DSun + + + August 2008 + + + University of California, Berkeley + + + PhD thesis + Sun, D., Large-scale modeling and optimization of en route air traffic flow , PhD thesis, University of California, Berkeley, August 2008. + + + + + FACET: Future ATM Concepts Evaluation Tool + + KBilimoria + + + BSridhar + + + GChatterji + + + KSheth + + + SGrabbe + + + + Europe Air Traffic Management R&D Seminar + + 2000 + + + 3rd USA/ + Bilimoria, K., Sridhar, B., Chatterji, G., Sheth, K., and Grabbe, S., "FACET: Future ATM Concepts Evaluation Tool," 3rd USA/Europe Air Traffic Management R&D Seminar , 2000. + + + + + Economic Values for FAA Investment and Regulatory Decisions, A Guide + + IncorporatedGra + + No. DTFA 01-02-C00200 + + + FAA Office of Aviation Policy and Plans + + December 2004 + Washington, DC + + + Tech. Rep. Contract + GRA, Incorporated, "Economic Values for FAA Investment and Regulatory Decisions, A Guide," Tech. Rep. Contract No. DTFA 01-02-C00200, FAA Office of Aviation Policy and Plans, Washington, DC, December 2004. + + + + + An Estimation of the Benefits of Air Traffic Flow Management + + JDearmon + + + JHoffman + + + THolden + + + JMayo + + + GSolomos + + + PKuzminski + + + AChambliss + + + + Proc. of International Council of the Aeronautical Sciences + of International Council of the Aeronautical SciencesAnchorage, AK + + September 2008 + + + DeArmon, J., Hoffman, J., Holden, T., Mayo, J., Solomos, G., Kuzminski, P., and Chambliss, A., "An Estimation of the Benefits of Air Traffic Flow Management," Proc. of International Council of the Aeronautical Sciences, Anchorage, AK, September 2008. + + + + + Traffic Flow Management using Aggregate Flow Models and the Development of Disaggregation Methods + + DSun + + + BSridhar + + + SGrabbe + + + + AIAA Guidance, Navigation and Control Conference and Exhibit + Chicago, IL + + August 2009 + + + Sun, D., Sridhar, B., and Grabbe, S., "Traffic Flow Management using Aggregate Flow Models and the Development of Disaggregation Methods," Accepted for presentation at AIAA Guidance, Navigation and Control Conference and Exhibit, Chicago, IL, August 2009. + + + + + + diff --git a/file19.txt b/file19.txt new file mode 100644 index 0000000000000000000000000000000000000000..7d94f8853243ece24ddb5698c65fe56b0e19f427 --- /dev/null +++ b/file19.txt @@ -0,0 +1,676 @@ + + + + +I. IntroductionCurrently, Traffic Managers (TMs) assess weather predictions and issue traffic management initiatives (TMIs) reducing flow rates through regions of airspace affected by severe weather.Flow rates can be reduced below nominal levels by holding flights on the ground or routing flights around these regions.TMs make these decisions based on their experience with similar weather conditions.Flow rates through sectors or flow constrained areas are reduced based on heuristics, the experience of the TM, and the experience of controllers at key sectors.Capacities are reduced with broad strokes and may be overly conservative, or vary under similar scenarios, based on individual TM preferences.Airlines are unable to reliably predict what the TM will prescribe in a given situation, causing delays in their own decision processes, which results in a lack of operational predictability.A new program for dealing with airspace constraints, Collaborative Trajectory Options Program 1 (CTOP), is currently in limited use.This program allows airlines to submit multiple preferred routes to the Federal Aviation Administration (FAA) through a trajectory options set (TOS) when severe weather is predicted to affect their flights' planned routes.With this shift towards greater airline participation in pre-departure routing during convective weather events, the need for a predictable system of constraint selection becomes apparent.This need for improved predictability is addressed in the plan outlined by the National Aeronautics and Space Administration (NASA) Aeronautics Research Mission Directorate (ARMD) in the NASA Aeronautics Strategic Implementation Plan. 2 The plan calls for the development of concepts and tools to realize trajectory-based operations (TBO), including such concepts as system modeling to improve predictive capabilities.Predictable constraint selection would require a tool to identify viable routing options and include flow rate constraints along each route.There are a variety of route and rate choices that can satisfy constraints caused by a particular convective weather scenario.Whereas one might think to choose the optimal solution among feasible solutions, there is no one metric that can be used to determine optimality.Each airline has its own metric based on its individual business case.The FAA has its own objectives that may vary in relative importance across scenarios.Rather than attempt to find "optimal" solutions, this system should provide feasible solutions along with estimated quantitative performance values calculated using various metrics.Historical data can be leveraged to generate models to find usable routes given weather information.Such a tool would benefit both the user and the FAA.Airlines can benefit from knowing if previous trajectory-based routes were effective to inform future TOS generation in similar weather conditions.Air navigation service providers will also benefit from a tool that suggests historical solutions.This will allow TMs to make use of solutions used by other TMs, thus allowing an individual TM to learn from the prior experience of many TMs, not just his or her own experience.This would lead to more informed and consistent decision making, the latter of which would increase predictability for airlines.The objective of this work is to quantify the weather impact on routing structures commonly observed in the absence of convective weather.Ideally, we would also like to identify conditions under which reroutes that do not conform to the commonly used route structure are used.However, work done to date that has attempted to model the use of non-conforming routes based on weather conditions has proven difficult.Challenging aspects of this work include taking the large amount of weather data available and creating a set of features that are relevant to the TM decision process, and the fact that a wide range of responses can be and are implemented by TMs during similar weather scenarios.For the former challenge, TMs have a variety of weather products available to them, each of which has strengths and weaknesses, and interpretation by each individual TM, in consultation with weather experts, is nuanced and difficult to articulate to researchers.The latter challenge makes the problem of finding patterns in the TM response difficult.In this work, we focus on flights from Fort Worth Air Route Traffic Control Center (ZFW) to New York Air Route Traffic Control Center (ZNY) and find the dominant routes used in the absence of convective weather.This involves identifying clear-weather days and clustering flight tracks from those days.Details of this process are given in Section III.In Section IV, we develop a metric to quantify convective weather impact on these routes.Historical flow rates along these dominant routes are found by identifying tracks flown in convective weather conditions that conform to these clusters.These historical flow rates are examined and modeled as a function of this convective weather impact quantity.This relationship is modeled using an exponential curve which can then be used to predict flow rates based on given weather conditions.These results are given in Section V. Results show that the flow rate does not depend uniformly on convective weather found throughout the route; however, convective weather on the final third of the route was found to have a greater impact on the flow rate restriction than other portions of the route. +II. Background4][5] With a vast amount of weather information, the challenge is to condense the data in such a way that operationally relevant features are preserved (or highlighted) in order to compare days.It is often difficult to identify operationally relevant features, which typically depend heavily on the objective.In Ref. [5], days are clustered based on the Weather Impacted Traffic Index (WITI) values at each Air Route Traffic Control Center (ARTCC or Center).Once similar days are discovered, the use of particular advisory reroutes on similar days is quantified.Previous research [6][7][8] has also involved creating or selecting reroutes for flights that avoid airspace affected by severe weather.However, little work has been done to predict the use of specific routes in the presence of convective weather.Previous work 9 focused on predicting reroute advisories put in place in response to convective weather.However, airline dispatch may make routing choices to avoid convective weather that do not conform to an issued reroute.Additionally, the paradigm shift from the historical use of reroutes issued by the FAA to airline participation, for example, via CTOP, is making the study of flown flight tracks important.The goal of this work is to capture that behavior by focusing on previously flown flight tracks, making this problem more difficult than that of predicting reroute advisories.Advisories are put in place for several hours (typically 4 to 8), and thus represent an aggregate control of many flights.However, looking at individual flights, aggregation is left to the researcher, which requires determining appropriate spatial and temporal ranges over which flights are to be grouped.This is extremely difficult when weather can change drastically at the time scale of quarter hours, whereas only a handful of flights traveling through the affected area depart per hour.These factors make the problem of predicting the use of a particular route more difficult than predicting the implementation of an advisory route.Traffic from ZFW to ZNY is studied in this work.1][12][13][14][15][16] This particular region of airspace was chosen as the focus for various research studies because operations in this area have been subject to chronic delay, problems in this region are complex due to the close proximity of high volume airports, and problems in this region have far-reaching impacts throughout the National Airspace System (NAS).In order to complement that research, this work is focused on traffic with destinations within the New York metropolitan area.In order to gain understanding of pre-departure routing choices, we selected ZFW as the origin Center since it is close enough to the New York metropolitan area such that a reasonably accurate two-hour forecast would be sufficient to predict en route weather conditions but far enough that significantly different alternate routes exist.Flights departing from Dallas/Fort Worth and Dallas Love Field make up the majority of this traffic with flight times from these airports to New York metroplex airports between 2 hours and 2 hours 45 minutes.While it is impossible to completely decouple the traffic in any region of the NAS from all other traffic in the NAS, this region of airspace is heavily used by flights departing from ZFW and landing within ZNY.Forty-two percent of all traffic that traverses ZFW and subsequently traverses ZNY is traffic that originates and terminates in ZFW and ZNY, respectively.In this paper we focus on analyzing the impact of convective weather on this traffic specifically, which has the opportunity for pre-departure rerouting based on reasonably accurate two-hour weather predictions.Future work could include an investigation into the impact of the shift of other fights into this airspace on the capacity available for traffic originating in ZFW and terminating in ZNY. +III. Clear-weather flight track analysisThe first objective is to identify flight paths commonly used in the absence of convective weather.Days with little or no convective weather activity throughout the NAS will be referred to as "clear-weather days."An understanding of en route routing options during clear-weather days will allow us to quantify impact of weather on nominal routing, and provide insight into the response of airlines and TMs to convective weather events.This process includes the identification of clear-weather days, collection of tracks flown on these days, and the removal of any bad data from this set of tracks.The portions of the flight tracks near the departure and arrival airports vary and are removed so as not to be quantified as a difference between tracks that may overlap significantly in the en route portion of the flights.Clusters are then identified and "lanes" are formed to describe the clear-weather routing structure.These steps are described in detail in the remainder of this section.Step 1: Identify clear-weather days Clear-weather days were identified using total daily Center WITI, which is the sum of Center WITI for all 20 Centers across the NAS.Note that a Center WITI value is an indication of the convective weather impact on a Center, and does not take into account other weather features, such as winds or turbulence.Days with low total daily Center WITI values (below 2.5% of the maximum observed value) are considered to be clear-weather days.A total of 78 days in 2014 were identified as clear-weather days.Tracks for flights departing from any airport in ZFW destined for any airport in ZNY were then analyzed.In total, there were 2,984 flights from ZFW to ZNY on these 78 days.The airport pairs making up 90% of the traffic from ZFW to ZNY were selected to represent the dominant flows.The breakdown of the number of flights for each of these origin and destination pairs is given in Table 1.Step 2: Identify invalid flight tracks LGA 94Flight track information was obtained from Aircraft Situation Display to Industry 17 (ASDI) data.RADAR systems are used to track flights and the speed, altitude, and position of flights are recorded.Updates are provided every one minute for flights in en route airspace.This information is then made available through ASDI.Flight tracks can be described by a sequence of points, specified by latitude, longitude, and altitude coordinates.As with any data set, there is likely to be missing or invalid information.Before clustering the clear-weather tracks selected in the previous step, flight tracks with invalid values should be removed.Invalid flight tracks are identified by inter-track point distance and total path length.These values are calculated for all flight tracks and compared within origin-destination city pairs.Flight tracks that are found to have inter-track point distances greater than the 90 th percentile value or total path length below the 10 th percentile value or above the 90 th percentile value for the corresponding origin-destination city pair are removed from the data set.A total of 20% of the flight tracks were found to be invalid tracks.This is a large portion of the total flight tracks, however, their removal aggressively cleans the clear weather track data in order to extract trends. +Step 3: Isolate en route portion of flight tracksWe are interested in routing choices for the en route airspace.Flight tracks may vary greatly within the departure and arrival segments of the flights based on, for example, departure and arrival procedures and runway configuration, which would not otherwise affect the flight trajectory.We remove these segments so variations in these portions of the routes are not quantified as major differences between tracks if they are otherwise co-located on the en route portions.The portions of tracks just after departure and just prior to landing are characterized by turns while flights follow departure and arrival procedures.Flight tracks become more uniform and similar to other flight tracks as the distance from the origin or destination airport increases.To identify the point at which a track becomes smooth enough to cluster, we calculate the rate of change of the heading angle and find the distance from the airport at which this value is within some threshold.We define this threshold to be the average rate of change of the heading angle 30 minutes after departure and 30 minutes prior to arrival.Using the departure portion of the tracks as an example, this value was found to be 8 degrees per minute.In Fig. 1, the rate of change of the heading angle versus distance is shown.From this we can see that for this specific example, at a distance of 72 nmi from the origin airport, the rate of change of the heading angle falls below 8 degrees per minute for 95% of the flight tracks.For each track, we cut off the points in the track whose distance to the origin airport is less than 72 nmi.The same procedure is used to remove the final portion of the flight tracks, which also tends to vary significantly.These results are shown in Fig. 2.For the purpose of comparing flight tracks in the subsequent steps, only the en route portions of the flight tracks are used.Figure 3 shows the full tracks with en route and arrival or departure portions indicated.Step 4: Remove outliers and cluster flight tracksIn this work, we are interested in finding flight tracks that are similar in the horizontal spatial plane.Whereas it would be useful to take altitude and elapsed time since departure into consideration, we simply do not have enough data to segregate tracks further and still have enough information to find statistically significant trends.This will become evident in Section V when results are presented.Focusing on similarity in the horizontal plane, we need to choose a metric with which to compare flight tracks.Several distance metrics were assessed for use in this application.A more detailed discussion of the requirements for distance metrics with respect to this type of application are given in Ref. [9].The Frechét distance is a good candidate for this application.The Frechét distance is often described as the length of the smallest leash that is required to connect a dog and its owner as they walk along two separate paths.The calculation of this metric is computationally intensive.Similar results were found using Euclidean distance between corresponding track points, as was done in Ref. [18].One advantage of using the Euclidean distance is that it can can be calculated faster than the Frechét distance.In order to make use of the Euclidean distance metric, each flight track must be represented by the same number of coordinates.Here, we choose to take each track and describe it by 100 evenly spatially distributed points.These latitude and longitude coordinates are then transformed to the Lambert azimuthal equal-area coordinate system.Since this is an equal-area projection, the Euclidean distance can be used to compare relative distances between points near the reference point.The challenge of unsupervised clustering is to determine the appropriate number of clusters.Metrics such as the silhouette score or inter-cluster versus intra-cluster distances can indicate that clusters are tight and well separated.The existence of outliers in a data set results in low scores on these metrics, thus possibly artificially indicating that a given clustering is not appropriate.If outliers are removed from the data set before clustering, these metrics will more accurately indicate the quality of the resulting clusters.Prior to clustering the tracks, we first remove outlier tracks from the data set.An outlier is identified by examining the nearest neighbor distances for each track in each origin-destination pair.Tracks with few nearest neighbors are removed from the data set before clustering.This outlier detection method is described in detail in Ref. [19,20].All flight tracks of interest and those identified as outliers are shown in Fig. 4.Finally, the flight tracks are clustered using the hierarchical clustering technique. 21The silhouette score for each track, which is a value that indicates how well each track fits in its assigned cluster, and visual inspection were used to select the total number of clusters.Silhouette scores range from -1 to 1, with a high value indicating that the cluster member is close to other members of the cluster and distant from members of other clusters.If all tracks have high silhouette scores, this is an indication that clusters are tight and well separated and thus an appropriate clustering of the data.The silhouette scores for each track are shown in Fig. 5.The final flight track clusters are shown in Fig. 6.Step 5: Create clear-weather track cluster lanesIn the next section, flight tracks flown during the convective weather season from April to September will be analyzed.Each track flown during this season will be identified as either conforming to a clear-weather cluster or not.Additionally, weather impact on the clear-weather track clusters must be quantified.To facilitate these processes, a "lane" is created for each clear-weather cluster.One potential method of classifying a flight track as conforming to a cluster is based on cluster metrics.Each new flight track would be considered a member of each cluster and cluster metrics could be calculated with this trial member.For example, the silhouette score of the new track could be compared to others in the cluster or the changes to inter-cluster and intra-cluster distances as a result of the addition of this track could be calculated.Either of these values could be used to indicate a conforming or non-conforming track.This process would require all of the clear-weather flight tracks to be saved and each new track to be compared to all of the clear-weather flight tracks.Depending on the distance metric used, this could become a computationally intensive procedure.As an alternative to this method, we chose to create lanes for each cluster based on the dispersion of flight track coordinates of the members of the clusters.Using these lanes, a new flight track can be classified as conforming to a given lane based on whether or not the new track lies within the lane.This procedure is described in Section IV.These lanes will also be used to quantify convective weather impact on the clear-weather track clusters.We will use Convective Weather Avoidance Model 22 (CWAM) forecasts for this calculation.We break each lane up into segments of roughly 15 minute flight time per segment so that CWAM forecasts at 15 minute intervals can be used to quantify the interaction of convective weather with the clear-weather lanes.More details will be provided in Section IV.The most commonly used clear-weather track cluster, cluster 3, will be used as an example throughout the explanation of lane formulation.Each track (with beginning and end portions cut) is re-sampled to be described by coordinates spaced by 15 minutes of flight time.With en route flight times between 115 minutes and 165 minutes, each flight track is then described by 8 to 12 points.Most flight tracks in cluster 3 were described by 9 segments.Only flights described by this number of points are then used to develop segments of the lane.The 15 minute interval description of each of these tracks for this example is shown in Fig. 7.For this cluster, 775 tracks out of the total 1343 tracks are described by 9 points spaced 15 minutes apart.The end points of each 15 minute traversal time segment is found by averaging locations of the corresponding points of all tracks in this group.The width of the lane is then based on the lateral spread of points at each end point.The lane width is twice the mean distance of each point to this mean location, plus two standard deviations.We now focus on the use of the commonly-used clear-weather clusters from ZFW to ZNY identified above during the convective weather season.The convective weather season for this region runs from April through September.Data was collected for convective weather seasons in both 2014 and 2015.Tracks for flights departing between 7 AM and 10 PM, EDT were used in the analysis.A total of 8,334 flights fit this criteria. +IV. Convective weather season flight track analysisIn order to identify trends in the data, we must aggregate information over some time period.We choose to aggregate over one-hour long time intervals.For each hour long time interval, we find the number of departing flights using each commonly used clear-weather route cluster.We quantify the convective weather impacts along each cluster.We found that during any one-hour time window, a maximum of 12 aircraft using a particular cluster take off.These processes are described in the following two steps.Step 1: Label tracks In order to gain insight into the use of each of the clear-weather route lanes during the convective weather season, we need a method of identifying whether a given flight track conforms to a clear-weather route cluster or not.First, the departure and arrival portions of each convective weather season track are removed using the distance from origin and destination airport group found in Section III.Next, we compare a given flight track to each cluster center to find the closest cluster center.We then classify the track as conforming, marginally conforming, or nonconforming to that cluster based on the fraction of flight track points lying within the segments making up the closest cluster's lane and "extended lane."The extended lane is 40% wider than the original lane.If 90% of the track points are within the lane, the track is labeled conforming.If more than 10% of the track points are outside the extended lane, the track is labeled nonconforming.Otherwise, the track is labeled marginally conforming.Labeled tracks for cluster 3 are shown in Fig. 8. +Marginal Cluster 3 TracksNon-conforming Cluster 3 Tracks Step 2: Generate weather featuresAs discussed in Section II, the major challenge of this work is to select, or calculate, appropriate features to describe the weather impact on commonly used routes.In this work, the Convective Weather Avoidance Model 22 (CWAM) is used.This product consists of polygons defined at various flight levels and forecast times.Each polygon represents a region of airspace that a specific percentage of pilots are expected to avoid.Contours are given for probability of pilot deviation of 60%, 70% and 80%.To model the probability of pilot deviation around weather, CWAM makes use of both the measured heights of the tops of clouds and the amount of Vertically Integrated Liquid (VIL) within the cloud.CWAM forecast extends up to two hours, making this product most useful for planning short haul flights.With flight times from ZFW to ZNY of roughly two and a half hours, this product is useful for our purposes.However, extending a similar technique to more distant origin and destination pairs will involve the selection of a weather product more appropriate for strategic planning of long haul flights.We quantify the weather impact on a route cluster lane based on the predicted total time that a flight traveling along that route cluster would spend within CWAM polygons.This will be referred to as "dwell time" within CWAM polygons.In order to simplify calculations, 70% probability deviation polygons at the most common flight level observed in the clearweather data (37 thousand feet) are used.Figure 9 shows the flight levels observed in the data.As previously described, each route cluster lane is broken down into segments of roughly 15 minute traversal time.For a given segment, 10 flow lines are drawn, with beginning and end points spaced evenly on the respective upstream and downstream edges of the segment.The CWAM data available at the middle of the hour of interest are used.That is, if we are interested in flights departing between 9 AM and 10 AM, the CWAM file available at 9:30 AM is used.With segments of roughly 15 minute traversal times, we can calculate the expected time that a flight taking off in the middle of this hour would reach each segment.The CWAM forecast associated with this time is used to quantify predicted dwell time within a given segment.The dwell time of each flow line through each CWAM polygon is calculated.That is, the total time that a flight traveling along a particular flow line would be within a CWAM polygon is found.This value can be calculated for each probability of deviation threshold.The lane, segment and flow lines of cluster 3 are shown in Figure 10 overlaid with the CWAM polygons of the prediction used to calculate estimated dwell times in segment 4. For each segment, we then calculate 10 values: (10 flow lines)×(1 altitude)×(1 probability of deviation thresholds).This is calculated for each segment of each cluster.There are a total of 350 dwell time values used to describe the predicted interaction of flights taking off within a certain hour long time window with CWAM polygons. +V. ResultsWith the analysis of Section IV complete, we can now look for trends in clear-weather route cluster use given the calculated convective weather features.We are interested in pre-departure routing choices and thus focus on the number of flights which take off during an hour long time window using each of the clusters.Fifty-six percent of the hour long time windows examined had nonzero CWAM dwell times and only these time windows were used in the following analysis.Figure 11 shows a plot of the number of flights taking off and using (conforming or marginally conforming to) cluster 3 during an hour long time window versus the associated predicted average dwell time in CWAM polygons at or above a 70% probability of deviation during a traversal of this cluster.For each number of take-offs, the 95 th percentile dwell time is found and plotted in red.An exponential curve is fit to these values.The exponential decrease in departing flights with an increase in dwell time is quite evident.Note that the estimated dwell time is not necessarily the same as the actual dwell time for any given flight.Each flight has room to maneuver while still being categorized as using a particular route.Thus, actual dwell time may be lower for a particular flight as it can fly around regions of severe convective weather.This may be the major cause in the exponential decay of route cluster use as a function of dwell time, as opposed to a linear relationship, or a simple "open" or "closed" status for each cluster separated by some critical value of dwell time.An exponential model was fit to take-offs for each cluster versus predicted dwell time along the beginning segments, middle segments, and end segments of the respective clusters.These results are shown in Fig. 12. Data points that fall below the 95 th percentile curve could be due to a variety of factors such as lower overall demand, shift in traffic to an alternate route cluster predicted to have lower CWAM dwell times, or increased over flight traffic.It is interesting to note that the number of departures drops off steepest as a function of the dwell time in the final segments of the cluster for all clusters.This is an indication that the predicted weather conditions on the arrival portion of the flight have a strong influence on routing selection.In practice, these results could be used to inform TM decisions.For example, given a specific CWAM forecast, the dwell time values could be calculated for each route cluster.These values could be input into the respective models of the number of take offs per hour as a function of dwell time.The resulting number of departures per hour could be used to set flow rate constraints along heavily used route clusters in a CTOP.Results shown here are for CWAM, which has a forecast of up to two hours.For longer flight times, a weather product with a longer forecast horizon should be chosen. +VI. Conclusions and future workIn this paper, a method of processing flight tracks from 78 clear-weather days in 2014 and clustering these tracks to identify commonly used flight paths in the absence of convective weather was presented.We then identified whether or not flight tracks from convective weather seasons of 2014 and 2015 conformed to these clear-weather flight paths.Weather impact on these clear-weather paths was calculated based on predicted dwell time of a flight using each cluster during convective weather conditions.Here, the CWAM weather product was used.Given predicted dwell times and labeled flight tracks in convective weather conditions, we were able to model the decay in the use of each cluster as an exponential function of the dwell time along the full route, and also along the beginning, middle, and end portions of the route.Results show that the use of the commonly used paths falls exponentially with dwell time.Thus, any machine learning predictive model that relies solely on linear combinations of dwell times will likely not perform as well as one that uses this exponential function to calculate a capacity for commonly used paths during convective weather events.The relations between dwell time and path use found here could be used in a CTOP application to set reduced flow rates along commonly used flight paths in the presence of convective weather.Some work has been done to model the use of commonly used paths in convective weather with predictive accuracy of roughly 60%.Future work will involve incorporating this calculated capacity feature in machine learning models in an attempt to improve prediction accuracy.At this point, the models developed are binary classifiers, that is, the prediction is simply whether or not a given commonly used flight path is used during a specific time window.Prediction accuracy may improve if either regression or multi-class classification methods are applied to the problem.Furthermore, the method of identifying commonly used flight paths in clear convective weather will be applied to additional origin centers, with flights bound for airports in the New York Center.This analysis will hopefully identify structure in airspace and allow researchers to better understand routing options in regions of airspace with dense traffic.Figure 1 :1Figure 1: (left) Rate of change of heading angle for each track up to 100 nmi from mean location of origin airports.The red lines indicate the mean en route rate of change of the heading angle, and the black dashed line indicates the distance at which 95% of the flight tracks have a rate of change of heading angle that drops below the mean rate of change of en route heading angle.(right) Departure portion of flight tracks.The blue indicates the departure portion that is not included in the flight track comparison, green indicates en route portion (beyond 72 nmi from origin). +Figure 2 :2Figure 2: (left) Rate of change of heading angle for each track up to 100 nmi from mean location of destination airports.The red lines indicate the mean en route rate of change of the heading angle, and the black dashed line indicates the distance at which 95% of the flight tracks have a rate of change of heading angle below the mean rate of change of en route heading angle.(right) Arrival portion of flight tracks.The blue indicates the arrival portion that is not included in the flight track comparison, green indicates en route portion (beyond 42 nmi from destination). +Figure 3 :3Figure 3: Full flight tracks with green indicating en route portions of tracks and blue indicating departure or arrival portions of tracks. +Figure 4 :4Figure 4: Flight tracks from ZFW to ZNY (top 5 origin destination pairs only). +Figure 5 :5Figure 5: Silhouette scores for each track.Figure 6: Flight track clusters from ZFW to ZNY. +Figure 6 :6Figure 5: Silhouette scores for each track.Figure 6: Flight track clusters from ZFW to ZNY. +Figure 7 :7Figure 7: Example of clear-weather track cluster lane segments for cluster 3.Each colored grouping of points were used to define the end points and widths of each segment. +Figure 8 :8Figure 8: Labeled 2014 convective weather season tracks for cluster 3: 2,484 conforming, 129 marginal, 147 nonconforming tracks. +Figure 9 :9Figure 9: Mode value of flight levels observed during en route portion of each flight track. +Figure 10 :10Figure 10: (left) Segments and flow lines of cluster 3, with segment 4 highlighted in blue overlaid with CWAM polygons.The colored CWAM polygons are from the 45 minute forecast in order to estimate flight interaction with CWAM polygons when a flight departing in the middle of the hour of interest is predicted to enter segment 4. (right) A closer view of segment 4 overlaid with the 45 minute forecast CWAM polygons.CWAM dwell times for each flow line in segment 4 are calculated using these CWAM polygons and an estimated segment traversal time of 15 minutes. +Figure 11 :11Figure 11: Each data point plotted represents the number of flights that depart during a particular hour long time interval and follow cluster 3 that are predicted to encounter CWAM dwell time of a particular value.For each number of departures, the 95 th percentile dwell time is plotted in red. +time [min] Number of take offs Full route (1.22) Beginning of route (0.55) Middle of route (1.48) End of route (1.33) +Figure 12 :12Figure 12: Exponential fit of the number of departures in hour long intervals as a function of the estimated CWAM dwell time at the beginning, middle, and end of the route and throughout the full route.Values in parentheses following each legend label indicate the mean error in number of flights of each trend. +Table 1 :1Origin and destination city pairs constituting 90% of the traffic departing from airports within ZFW to destination airports within ZNY.OriginDestinationNumberAirportAirportof FlightsDFWLGA1531DFWEWR749↑DFWJFK18990% of trafficDALTEB120↓DAL + of 12 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 9, 2017 | http://arc.aiaa.org| DOI: 10.2514/6.2017-3593 + of 12 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 9, 2017 | http://arc.aiaa.org| DOI: 10.2514/6.2017-3593 + of 12 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 9, 2017 | http://arc.aiaa.org| DOI: 10.2514/6.2017-3593 + of 12 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 9, 2017 | http://arc.aiaa.org| DOI: 10.2514/6.2017-3593 + of 12 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 9, 2017 | http://arc.aiaa.org| DOI: 10.2514/6.2017-3593 + American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 9, 2017 | http://arc.aiaa.org| DOI: 10.2514/6.2017-3593 + Downloaded by NASA AMES RESEARCH CENTER on June 9, 2017 | http://arc.aiaa.org| DOI: 10.2514/6.2017-3593 + + + + +AcknowledgementsThe authors would like to thank Antony Evans and Paul Lee for their feedback at various stages of development of this work. + + + + + + + + + Collaborative Trajectory Options Program (CTOP) + + + + NASA Aeronautics Research Mission Directorate + Washington D.C. + + 2016 + + + NASA Aeronautics Strategic Implementation Plan + "Collaborative Trajectory Options Program (CTOP)," http://www.nbaa.org/ops/airspace/tfm/tools/ctop.php. 2 National Aeronautics and Space Administration, "NASA Aeronautics Strategic Implementation Plan," NASA Aeronau- tics Research Mission Directorate, Washington D.C., 2016. + + + + + A Cluster Analysis to Classify Days in the National Airspace System + + BHoffman + + + JKrozel + + + SPenny + + + ARoy + + + KRoth + + + + AIAA Guidance, Navigation and Control Conference + + 2003 + 11 + + + Hoffman, B., Krozel, J., Penny, S., Roy, A., and Roth, K., "A Cluster Analysis to Classify Days in the National Airspace System," AIAA Guidance, Navigation and Control Conference, 2003, p. 11. + + + + + A clustering approach for analysis of convective weather impacting the NAS + + MAsencio + + + + Integrated Communications, Navigation and Surveillance Conference (ICNS) + + 2012. April 2012 + + + + Asencio, M., "A clustering approach for analysis of convective weather impacting the NAS," Integrated Communications, Navigation and Surveillance Conference (ICNS), 2012 , April 2012, pp. N4-1-N4-11. + + + + + Classification of Days Using Weather Impacted Traffic in the National Airspace System + + AMukherjee + + + SGrabbe + + + BSridhar + + + + Transportation Research Board 92nd Annual Meeting + Washington DC + + January 2013 + + + Mukherjee, A., Grabbe, S., and Sridhar, B., "Classification of Days Using Weather Impacted Traffic in the National Airspace System," Transportation Research Board 92nd Annual Meeting, Washington DC, January 2013. + + + + + Sequential Congestion Management with Weather Forecast Uncertainty + + CWanke + + + DGreenbaum + + No. AIAA 2008-6327 + + + AIAA Guidance, Navigation and Control Conference and Exhibit + + August 2008 + + + Wanke, C. and Greenbaum, D., "Sequential Congestion Management with Weather Forecast Uncertainty," AIAA Guid- ance, Navigation and Control Conference and Exhibit, No. AIAA 2008-6327, August 2008. + + + + + Dynamic Generation of Operationally Acceptable Reroutes + + CTaylor + + + CWanke + + No. 2009-7091 + + + Proceedings of 9th AIAA Aviation Technology, Integration, and Operations Conference + 9th AIAA Aviation Technology, Integration, and Operations Conference + + AIAA + September 2009 + + + Taylor, C. and Wanke, C., "Dynamic Generation of Operationally Acceptable Reroutes," Proceedings of 9th AIAA Aviation Technology, Integration, and Operations Conference (ATIO), No. 2009-7091, AIAA, September 2009. + + + + + Dynamic weather routes: a weather avoidance system for near-term trajectory-based operations + + DMcnally + + + KSheth + + + CGong + + + JLove + + + CHLee + + + SSahlman + + + JCheng + + + 2012 + Brisbane + + + 28th international congress of the aeronautical sciences (ICAS + McNally, D., Sheth, K., Gong, C., Love, J., Lee, C. H., Sahlman, S., and Cheng, J., "Dynamic weather routes: a weather avoidance system for near-term trajectory-based operations," 28th international congress of the aeronautical sciences (ICAS), Brisbane, 2012. + + + + + Initial Analysis of and Predictive Model Development for Weather Reroute Advisory Use + + HArneson + + + + Proceedings of the 15th AIAA Aviation Technology, Integration, and Operations Conference + the 15th AIAA Aviation Technology, Integration, and Operations ConferenceDallas, TX + + June 2015 + + + 9 Arneson, H., "Initial Analysis of and Predictive Model Development for Weather Reroute Advisory Use," Proceedings of the 15th AIAA Aviation Technology, Integration, and Operations Conference, Dallas, TX , June 2015. + + + + + Nextgen operations in a simulated NY area airspace + + NMSmith + + + BParke + + + PLee + + + JHomola + + + CBrasil + + + NBuckley + + + CCabrall + + + EChevalley + + + CLin + + + SMorey + + + FOmar + + + DRein-Weston + + + H.-SYoo + + + + 2013 IEEE/AIAA 32nd Digital Avionics Systems Conference (DASC) + + Oct 2013 + + + + Smith, N. M., Parke, B., Lee, P., Homola, J., Brasil, C., Buckley, N., Cabrall, C., Chevalley, E., Lin, C., Morey, S., Omar, F., Rein-Weston, D., and Yoo, H.-S., "Nextgen operations in a simulated NY area airspace," 2013 IEEE/AIAA 32nd Digital Avionics Systems Conference (DASC), Oct 2013, pp. 5D5-1-5D5-15. + + + + + Reducing Departure Delays at LaGuardia Airport with Departure-Sensitive Arrival Spacing (DSAS) Operations + + PULee + + + NMSmith + + + CBrasil + + + EChevalley + + + JHomola + + + BParke + + + H.-SYoo + + + NBienert + + + ABorade + + + NBuckley + + + CCabrall + + + FOmar + + + CGabriel + + + + Air Traffic Control Quarterly + + 23 + 4 + + 2017/04/26 2015 + + + Lee, P. U., Smith, N. M., Brasil, C., Chevalley, E., Homola, J., Parke, B., Yoo, H.-S., Bienert, N., Borade, A., Buckley, N., Cabrall, C., Omar, F., and Gabriel, C., "Reducing Departure Delays at LaGuardia Airport with Departure-Sensitive Arrival Spacing (DSAS) Operations," Air Traffic Control Quarterly, Vol. 23, No. 4, 2017/04/26 2015, pp. 245-273. + + + + + Exploring management of arrival spacing using route extensions with terminal spacing tools + + BParke + + + NBienert + + + EChevalley + + + FOmar + + + NBuckley + + + CBrasil + + + HSYoo + + + ABorade + + + CGabriel + + + PLee + + + JHomola + + + NSmith + + + + IEEE/AIAA 34th Digital Avionics Systems Conference (DASC) + + 2015. Sept 2015 + + + + Parke, B., Bienert, N., Chevalley, E., Omar, F., Buckley, N., Brasil, C., Yoo, H. S., Borade, A., Gabriel, C., Lee, P., Homola, J., and Smith, N., "Exploring management of arrival spacing using route extensions with terminal spacing tools," 2015 IEEE/AIAA 34th Digital Avionics Systems Conference (DASC), Sept 2015, pp. 3E1-1-3E1-12. + + + + + Analyzing Double Delays at Newark Liberty International Airport + + ADEvans + + + PULee + + + + 16th AIAA Aviation Technology, Integration, and Operations Conference + + 2016 + 3456 + + + Evans, A. D. and Lee, P. U., "Analyzing Double Delays at Newark Liberty International Airport," 16th AIAA Aviation Technology, Integration, and Operations Conference, 2016, p. 3456. + + + + + A Human-in-the-loop Evaluation of a Coordinated Arrival/Departure Scheduling Operations for Managing Departure Delays at LaGuardia Airport + + PULee + + + NMSmith + + + NBienert + + + CBrasil + + + NBuckley + + + EChevalley + + + JHomola + + + FOmar + + + BParke + + + H.-SYoo + + + + 16th AIAA Aviation Technology, Integration, and Operations Conference + + 2016 + + + + Lee, P. U., Smith, N. M., Bienert, N., Brasil, C., Buckley, N., Chevalley, E., Homola, J., Omar, F., Parke, B., and Yoo, H.-S., "A Human-in-the-loop Evaluation of a Coordinated Arrival/Departure Scheduling Operations for Managing Departure Delays at LaGuardia Airport," 16th AIAA Aviation Technology, Integration, and Operations Conference, 2016, p. 3457. 15 + + + + + Integrated Demand Management: Coordinating Strategic and Tactical Flow Scheduling Operations + + NSmith + + + CBrasil + + + PLee + + + NBuckley + + + CGabriel + + + CMohlenbrink + + + FOmar + + + BParke + + + CSperidakos + + + H.-SYoo + + + + 16th AIAA Aviation Technology, Integration, and Operations Conference + + June 2016 + + + Smith, N., Brasil, C., Lee, P., Buckley, N., Gabriel, C., Mohlenbrink, C., Omar, F., Parke, B., Speridakos, C., and Yoo, H.-S., "Integrated Demand Management: Coordinating Strategic and Tactical Flow Scheduling Operations," 16th AIAA Aviation Technology, Integration, and Operations Conference, June 2016. + + + + + Predicting & Quantifying Risk in Airport Capacity Profile Selection for Air Traffic Management + + JCJones + + + RDelaura + + + MPawlak + + + STroxel + + + NUnderhill + + + + Twelfth USA/Europe Air Traffic Management Research and Development Seminar + + 2017. 2017 + + + Jones, J. C., DeLaura, R., Pawlak, M., Troxel, S., and Underhill, N., "Predicting & Quantifying Risk in Airport Capac- ity Profile Selection for Air Traffic Management," Twelfth USA/Europe Air Traffic Management Research and Development Seminar (ATM2017), 2017. + + + + + Aircraft Situation Display To Industry: Functional Description and Interface Control Document for the XML Version + No. DTFAWA-04-C-00045 + + + 100 Decadon Drive Egg Harbor Township + + April 15 2011 + 08234 + + + Computer Sciences Corporation Federal Sector -Civil Group + + + Version 1.8 Report + "Aircraft Situation Display To Industry: Functional Description and Interface Control Document for the XML Version," Version 1.8 Report No. DTFAWA-04-C-00045, Computer Sciences Corporation Federal Sector -Civil Group, 100 Decadon Drive Egg Harbor Township, NJ 08234, April 15 2011. + + + + + Trajectory Clustering and Classification for Characterization of Air Traffic Flows + + MC RMurca + + + RDelaura + + + RHansman + + + RJordan + + + TReynolds + + + HBalakrishnan + + + + AIAA Aviation Technology, Integration, and Operations Conference + Washington, D.C. + + June 2016 + + + + Murca, M. C. R., DeLaura, R., Hansman, R., Jordan, R., Reynolds, T., and Balakrishnan, H., "Trajectory Clustering and Classification for Characterization of Air Traffic Flows," AIAA Aviation Technology, Integration, and Operations Conference, Washington, D.C., 13-17 June 2016. + + + + + Automated Route Clustering for Air Traffic Modeling + + ABombelli + + + ASegarra Torne + + + ETrumbauer + + + KDMease + + + + AIAA Modeling and Simulation Technologies Conference + Grapevine, TX + + January 2017 + 1318 + + + Bombelli, A., Segarra Torne, A., Trumbauer, E., and Mease, K. D., "Automated Route Clustering for Air Traffic Mod- eling," AIAA Modeling and Simulation Technologies Conference, Grapevine, TX , January 2017, p. 1318. + + + + + Strategic Air Traffic Planning with Fréchet Distance Aggregation and Rerouting + + ABombelli + + + LSoler + + + ETrumbauer + + + KDMease + + + + Journal of Guidance, Control, and Dynamics + + 40 + 5 + + 2017 + + + Bombelli, A., Soler, L., Trumbauer, E., and Mease, K. D., "Strategic Air Traffic Planning with Fréchet Distance Aggre- gation and Rerouting," Journal of Guidance, Control, and Dynamics, Vol. 40, No. 5, 2017, pp. 1117-1129. + + + + + The elements of statistical learning : data mining, inference, and prediction + + TJHastie + + + RJTibshirani + + + JHFriedman + + + + Springer series in statistics + New York + + Springer + 2009. 2013 + + + Autres impressions : 2011. e corr. + Hastie, T. J., Tibshirani, R. J., and Friedman, J. H., The elements of statistical learning : data mining, inference, and prediction, Springer series in statistics, Springer, New York, 2009, Autres impressions : 2011 (corr.), 2013 (7e corr.). + + + + + Modeling convective weather avoidance in enroute airspace + + RDelaura + + + MRobinson + + + MPawlak + + + JEvans + + + + 13th Conference on Aviation, Range, and Aerospace Meteorology + New Orleans, LA, Citeseer + + AMS + 2008 + + + DeLaura, R., Robinson, M., Pawlak, M., and Evans, J., "Modeling convective weather avoidance in enroute airspace," 13th Conference on Aviation, Range, and Aerospace Meteorology, AMS, New Orleans, LA, Citeseer, 2008. + + + + + + diff --git a/file2.txt b/file2.txt new file mode 100644 index 0000000000000000000000000000000000000000..5f74c59676872ecae028a9fc5d348ce36226357b --- /dev/null +++ b/file2.txt @@ -0,0 +1,1061 @@ + + + + +I. IntroductionAviation is undergoing rapid transformation, with increasing levels of automation and new technology and operational concepts continually being introduced.A key factor common to all is operational safety.Safe separation between aircraft has been and continues to be a key driving force in air traffic management and related research.In today's airspace, separation is maintained by controllers using generous separation criteria and by pilots using highly subjective means dependent on visual acuity.While this has proven sufficient in the past, increasing airspace density and technical innovations are pushing the limits of the system.Introduction of Unmanned Aerial Systems (UAS) into the airspace would further complicate the situation because unmanned aircraft cannot "see and avoid" other traffic.Integration of UAS in the National Airspace System (NAS) therefore requires a robust separation capability to meet FAA requirement to "see and avoid" and, when passing proximate traffic, to remain "well clear" as mandated for manned aircraft in 14 CFR §91.113, 1 "Rules of the Road".The Detect-and-Avoid (DAA) concept was introduced to satisfy this requirement for unmanned aircraft.DAA systems are intended to enable UAS to remain "well clear" and avoid collisions with other airborne traffic, [2][3][4] and to do so they require an objective definition of "well clear" (referred to as DAA Well Clear or DWC).DAA is required to provide detection and guidance to maintain DWC and, where DWC is lost, to provide recovery guidance to regain it.DAA systems are envisioned as a means to maintain a low level of risk of Near Mid-Air Collisions 6 (NMAC).To accomplish this goal, it is essential to investigate a number of objective definitions of Well Clear 7 and to quantify the levels of risk achieved by proposed DAA concepts.To that end, NASA has developed a portable software system, Java Architecture for DAA Extensibility and Modeling (JADEM), that supports various DAA algorithms and can interface with different fast-time and real-time simulation tools and live systems. 8The target users of JADEM are researchers who need to simulate the main features of proposed DAA systems before those systems are fully developed by manufacturers.The availability of a general purpose, flexible, and fast conflict detection and resolution algorithm was critical for the purpose.The Generic Resolution Advisor and Conflict Evaluator (GRACE) 8 proved suitable for that role.In addition to its core alerting and resolution features, GRACE powers JADEM's implementation of the recently introduced bands and well clear recovery guidance concepts (see Ref. 8).0][11][12][13][14][15][16][17][18] The complex and rapidly evolving requirements of these studies demanded and created ample opportunities for extensive evaluation of GRACE in a broad range of operational scenarios and applications.The goal of this paper is to describe GRACE with sufficient algorithmic detail to be of interest to researchers involved in air traffic simulations and to developers of practical DAA systems for manufacture and deployment.The next section provides an overview of previous work related to development of conflict detection and resolution algorithms.Section III describes GRACE features and algorithmic details.Section IV provides sample results of GRACE evaluation.Finally, concluding remarks are presented in Section V. +II. BackgroundA number of conflict detection and resolution algorithms have been proposed and actively used for decades for various research projects (see Ref. 4, 19 and the references therein).Several such algorithms were inspired by the Traffic Alert and Collision Avoidance System (TCAS) 20 that has been very successful in preventing mid-air collisions for manned aircraft.These systems rely on the cooperative behavior of aircraft, which actively communicate their state and intent data, 21,22 hence said systems cannot be used with non-cooperative aircraft.Some algorithms were specifically designed for certain types of onboard sensors, such as radars 21 or optical sensors. 23Other algorithms, such as Jointly Optimal Collision Avoidance, are in principle sensoragnostic, 24 but are tuned for a specific aircraft response model.Another next-generation algorithm, the Airborne Collision Avoidance System (ACAS), proposed to replace TCAS, 25 is more flexible and can be adapted to different aircraft, including UAS, by updating the probabilistic lookup table driving the ACAS threat logic.However, these algorithms do not use separation standards that would be needed to maintain Well Clear as defined by RTCA Special Committee 228. 8Note that DAA support is currently being added to ACAS-Xu (the version of ACAS supporting UAS).Other algorithms, more suitable for Separation Assurance in Air Traffic Control, include the Profile Selector En Route (PFS-E) component of the Center / TRACON Automation System, 26 the Advanced Airspace Concept Autoresolver, and the Tactical Separation-Assured Flight Environment (TSAFE). 27However for use in DAA, these algorithms would require non-trivial modifications to accommodate smaller look-ahead times and spatial separation standards, introduction of temporal separation, the lack of trajectory intent information, and more frequent update rates.More recent algorithms, "inspired by nature," include "force field" methods, [28][29][30] the closely related "light propagation" model, 31 and "navigation function-based" algorithms that model aircraft as particles moving in a potential field formed by other aircraft with the same "charge" and hence generating a "repulsive force." 32,33 hese algorithms demonstrated their effectiveness in guaranteed collision avoidance, but they can lead to overcosts and large deviations from nominal routes. 35 number of algorithms are based on optimization methods, which provide a very general framework for non-linear systems (see Ref. 4 and references therein).These methods typically use a mathematically rigorous Hamilton-Jacobi approach or heuristic optimization algorithms, such as simulated annealing 34 or genetic algorithms. 35These methods are efficient enough for strategic flight planning, but they are still not sufficiently fast for real-time applications that require update rates on the order of one second.Another drawback of these methods is that they model smooth trajectories suitable for future Flight Management System abilities, which are not (currently) practical for use in UAS DAA systems.The Generic Resolution Advisor and Conflict Evaluator (GRACE), the subject of this paper, was inspired by the ideas of force field [28][29][30] and complexity theory, 36 and it leverages the computational efficiency of grid-based methods. 37,38 RACE provides the benefits of flexibility, robustness, and good computational performance, which make it more suitable for evolving requirements of research and modeling of future DAA systems.It was therefore adopted in 2013 to provide the core alerting and guidance functions of NASA's Java Architecture for DAA Extensibility and Modeling (JADEM). 8t should be noted that JADEM was expressly designed to evaluate DAA concepts using different algorithms.For example, the Detect-and-Avoid Alerting Logic for Unmanned Systems (DAIDALUS), 39 released to open source in 2015, is being integrated into JADEM as an alternate alerting and guidance module. +III. Generic Resolution Advisor and Conflict Evaluator (GRACE)As implemented in JADEM, GRACE is an aircraft-centric algorithm that works to predict and prevent collision threats to a single unmanned aircraft (the "ownship") from all other aircraft ("intruders").Using the taxonomy defined in Ref. 19, GRACE can be characterized as: Dimensions: Horizontal and Vertical planes (3D) Detection: Explicit conflict detection threshold (but user-defined) Resolution: Optimized Maneuvers: Turns, Vertical maneuvers, and Speed changes Multiple:Global GRACE is a combination of two loosely coupled algorithms.The Generic Conflict Evaluator (GCE) provides a customizable implementation of conflict detection functionality.The Generic Resolution Advisor (GRA) is a fast general purpose conflict resolver.GRACE does not make any assumptions regarding temporal or spatial scales, performance capabilities of aircraft, or its sensor and communication systems.This flexibility is achieved by using customizable separation standards in GCE and a cost function in GRA, and by reliance on an external trajectory predictor for aircraft flight modeling. 8In particular, this allows GRACE to be used for rotorcraft or other unconventional UAS. +III.A. Generic Conflict Evaluator (GCE)GCE is an efficient deterministic algorithm for assessing intruder threats based on user-defined "separation standards" specified by logical conditions that can include almost any combination of the following (along with corresponding thresholds and filtering conditions): 8,40 • horizontal separation;• vertical separation;• tau-separation based on "simple tau" or "modified tau"; • horizontal miss distance;• time to Closest Point of Approach (CPA); • time to first loss (violation) of separation.In this paper, the term "violation" (of separation) will be used to indicate that a logical condition that defines a separation standard is violated.GCE can support several different "levels" of threat detection, each level being defined by its own userspecified separation standard.For instance, a separation standard for the highest severity threat can be set to NMAC as defined in TCAS (500 × 100 ft). +III.A.1. Dynamic Grid Mapping and Threat DetectionGCE can be used in two different roles:1. as initial conflict detection to find the threats that may require resolution; in this role it is called once for each call to GRACE; 2. in GRA to assess whether candidate maneuvers will result in a conflict-free trajectory; in this role GCE can be called many times (once per candidate maneuver); for this reason, its computational performance is critical.GCE computation time is essentially linear with the number of intruders.Therefore a key performance challenge is to efficiently identify aircraft that can potentially create threats prior to making any geometrical computations, which can be time-consuming.This needs to be done for arbitrary aircraft positions, which can deviate from the nominal route (based on intent) or from previously proposed resolutions.GCE uses a fast algorithm that avoids distance calculations.The algorithm is based on an airspace model consisting of identical discrete elements, or cells (the model is referred to as the "grid map").The grid map provides a quick way to find intruders close to any arbitrary ownship state by "mapping" all intruder trajectories predicted within a specified "look-ahead time" to a sequence of 2D grids, one grid per each time step. 8This can be seen as a deterministic variant of the method proposed in Ref. 37.The GCE algorithm involves three main operations:1. Predicting intruder trajectories at discrete time steps up to a specified look-ahead time.2. Mapping intruders by locating them in grid cells using simple comparisons between the intruder coordinates and cell boundaries.Note that the grid map is stored efficiently as a time sequence of hash maps populated only with "occupied" cells.3. Detecting threats from mapped intruders.This uses simple comparisons between cell boundaries and a bounding box with a side 2δ around the ownship's position to identify potential intruders, where δ is a horizontal distance deduced from the separation standards (Fig. 1).Intruders found within this bounding box are passed to the state-based threat detection logic, which uses specified separation standards to evaluate threats.A threat is considered "detected" if it is found to violate separation standards at least at one timestep.If more than one threat is "detected", GCE returns the first threat with the earliest conflict start time, which is defined as the time of the first state that violated the separation standard.Note that the first operation, which is expensive, is only done once, the resulting grid mapping subsequently being used in conflict resolution.This dramatically reduces GRACE computation time. +III.B. Generic Resolution AdvisorThe Generic Resolution Advisor (GRA) is the conflict resolution component of GRACE.It relies on the output from GCE (including the highest priority declared threat and mapping intruders to grid cells).The main output of GRA is a resolution that includes a recommended avoidance maneuver and a corresponding ownship predicted trajectory.GRA uses fast analytical transformations of linearized ownship trajectories before calling GCE for rigorous re-evaluation of candidate resolutions.It relies on user-defined cost functions for selecting the best maneuver.This promotes high computational efficiency without sacrificing flexibility or quality of proposed resolutions.The following six Standard Maneuvers (in the remainder of the paper we refer to them simply as "maneuvers") can be used:1. Turn Right; 2. Turn Left; 3. Vertical Up: increased vertical speed / flight path angle (faster climb or slower descent); 4. Vertical Down: reduced vertical speed / flight path angle (slower climb or faster descent); 5. Speed Down (decelerate); 6. Speed Up (accelerate).Finding the best maneuver starts from a local transformation of the CPA, for the highest priority threat, into a Trajectory Change Point (TCP) such that it satisfies the following conditions:1. it should not violate any separation criteria; 2. it should constitute a CPA for the transformed trajectory.The first condition is self-explanatory.The second condition is needed to ensure that after passing the TCP an aircraft could return to its initial route without the risk of encountering the same threat.This is important to improve the stability of proposed resolutions and to help pilots determine when they can safely execute a "recapture" maneuver returning to mission flight plan.Effectively, the TCP defines the end of maneuver in GRA.GRA starts by finding a linear trajectory calculated by extrapolating the state at CPA backwards to the desired start time (see section III.B.2 for details).Each of maneuvers listed above is applied in user-defined increments (steps), such as 5 degrees for heading change and 1 degree for change in flight path angle.At every step:1. a new TCP is obtained by perturbing the linearized trajectory in the direction dictated by the applied maneuver and analytically estimating the CPA along the new trajectory; 2. the candidate TCP is evaluated to determine whether it is "locally conflict-free" (i.e.does not violate separation criteria); 3. if the answer is "yes", a candidate solution is re-evaluated by calling GCE, which checks for conflicts with all intruders in grid map to verify that the solution is "globally conflict-free" within the specified lookahead time; this involves generating a new ownship trajectory with the TCP added as a new constraint; 4. if re-evaluation did not confirm that the solution is "globally conflict-free", these operations are repeated, in the next step, for the same maneuver until the predefined limit in change of control variable (heading, flight path angle, or speed) is reached.If the new solution is "better" than the previous "best" candidate, it replaces it as the new "best".Note that first two operations use fast analytic calculations.The third operation takes advantage of GCE optimization provided by grid map, but still requires computationally intensive calls to Trajectory Predictor for generating a new ownship trajectory.However these calls are made typically only once for each maneuver rather than at every step, because in most cases reevaluation confirms that solution is conflict-free and immediately terminates iterations.This helps greatly reduce computation time.This sequence of operations is shown in Fig. 2, which is further clarified in the remainder of this section.a CPA for the original trajectory (black solid curve in Fig. 3) and a predicted intruder trajectory (red curve) is known from the output of GCE, then the CPA for any perturbed ownship trajectory (blue curve) can be quickly estimated from an analytical solution based on linearization of predicted ownship and intruder trajectories near the CPA as illustrated in Fig. 3.As a first step, the states of ownship and intruder at CPA (shown as arrows) are extrapolated backward to a maneuver start time that may include an anticipated delay in maneuver execution (referred to herein as "algorithm delay"), as shown by dashed black and red lines, respectively.This approximation is justified because the accurate representation of trajectory around a CPA is most important for evaluation of candidate resolutions.It also simplifies modeling of manuevers by applying them to the extrapolated ownship state rather than to its actual initial state.Therefore, the perturbed ownship trajectory is approximated by a straight line, shown as a dashed blue line in Fig. 3.Then, a candidate TCP can be defined as the CPA for the perturbed trajectory, shown as an arrow on the blue line.Note that this approach is based on the predicted trajectories from GCE, therefore it does not require any assumptions regarding the availability of intent data for ownship and intruders.The CPA is found from the time to minimal distance between ownship and intruder positions on their linearized trajectories t CPA , assuming that they are moving with constant speeds (see Ref. 8)t CPA = max 0, -k ∆x k ∆v k k (∆v k ) 2 ) (1)where k represents a cartesian component and the summation occurs over all three components, with∆x k = (x I -x O ) k , ∆v k = (v I -v O ) k -intruder +III.B.3. Fast CPA Evaluation Using State-based Threat DetectorTo determine whether a candidate solution is "locally conflict-free", the horizontal and vertical separations at CPA are checked against their threshold values for corresponding separation standard.These separations can be easily found from ownship and intruder positions at CPA given by:(x O ) k + (v O ) k • t CPA(2)(x I ) k + (v I ) k • t CPA(3) +III.B.4. Re-evaluation of SolutionsOnce a fast check has determined that a solution is "locally conflict-free," quick evaluation is repeated at the next step to account for possible uncertainty.If the solution is still conflict-free or a limit of control variable (e.g.heading or flight path angle) is reached, the solution is re-evaluated using a trajectory predictor of sufficient fidelity to account for aircraft performance and to check for conflicts with other intruders.This step involves recalculation of the ownship's predicted trajectory starting with actual rather than extrapolated initial state and using constraints modified by adding the new TCP.This trajectory then is rigorously evaluated for conflicts with all intruders as described in section III.A.Note that intruder trajectories do not need to be re-calculated since they are not perturbed. +III.B.5. Limits of Control VariablesGRA provides configurable "operational limits" of control variables for maneuvers.For instance, users can limit the maximum heading change for turns by 90 degrees, and the maximum change in flight path angle for vertical maneuvers by 5 degrees.In addition to operational limits, GRA estimates "dynamic limits" of control variables, based on aircraft performance.GRA stops incrementing the control variable for each maneuver if it reaches either its dynamic or operational limit. +III.B.6. Selection of ResolutionGRA can run in two user configurable "control modes".In the so called "UseFirst" control mode, GRA returns the first conflict-free solution it finds.In this case GRA works faster, guarantees that it will find a conflict-free solution if it exists, and provides solutions that are almost always stable but not necessarily optimal.On the other hand, if a conflict-free resolution does not exist, or if GRA is configured for the so called "UseBest" mode, GRA will find the best resolution using the following rules:1. if the best solution was not defined yet, the first candidate solution becomes the best solution; 2. if a candidate solution is conflict-free and the old best solution is not conflict-free, the candidate solution becomes the new best solution; 3. if a candidate solution and the old best solution are both conflict-free or both not conflict-free, and the cost of candidate solution is lower than the cost of the old best solution, the candidate solution becomes the new best solution; 4. in all other cases the best solution does not change.Note that, even though a TCP estimated from linearized trajectories is approximate, a candidate resolution is always based on the predicted ownship trajectory, generated using all known information about the ownship flight constraints (including the TCP) and aircraft performance.GRA can use any externally defined cost function to select the best resolution maneuver.The cost function can rank maneuver types by an order of preference (e.g. to support right of way rules) and penalize or suppress specific maneuver types, too aggressive maneuvers, too frequent maneuver type changes within an encounter, and maneuvers that would result in too small separation at predicted CPA. 8 +IV. Evaluation of GRACE PerformanceThe most obvious potential uses of GRACE are as a fast conflict detection algorithm (GCE), as an automatic guidance algorithm for UAS (GRA), or as a model of such guidance in simulations of UAS.GRACE provided the core algorithms for JADEM's alerting and guidance tools that have been evaluated and refined in a number of HITL studies and flight tests. 8In all these studies GRACE was used to provide alerting to pilots. GRACE was also used to provide directive guidance in earlier studies and was used to compute bands and recovery guidance for the most recent studies.GRACE also has been used for modeling UAS DAA systems in a number of parametric/factorial studies 41,42 and NAS-wide simulations. 13,43,44 Ts section describes the methodology and sample results of such studies obtained using two approaches, 1. Parametric: uses 180 encounter geometries to test the performance of the algorithm under demanding conditions regardless of likelihood of occurrence in real life.2. NAS-wide: uses projected UAS mission profiles developed under prior work 45 and recorded VFR traffic data 46 for a more realistic encounter model.Three different scenario configurations were used in the parametric evaluation, namely, 1. Baseline: uses "perfect" surveillance data without simulated sensor errors and executes GRACE resolution maneuvers immediately; the mission profile is recaptured at GRACE recommended times following each maneuver; 2. Noisy Surveillance: similar to the baseline case but uses airborne radar model paired with a tracker; 8 this approximates a real-world fully automated concept; 3. Pilot Delay: similar to the baseline case but uses a model of pilot delayed action; this essentially represents a pilot-in-the-loop concept of operation.In contrast, a single configuration was used for the NAS-wide case with "perfect" surveillance data, automatic execution of resolution maneuvers, and mission recapture with no delays.The remainder of this section provides a summary of the metrics used in evaluations of GRACE performance, followed by results for the parametric and then for the NAS-wide evaluation. +IV.A. Performance MetricsJADEM fast-time simulator calculates a number of metrics that can be used to evaluate system performance.The metrics most relevant to GRACE are listed below. +Conflicts number of conflicts (encounters) Resolutionstotal number of resolutions proposed by GRE +Changes Per Encounternumber of changes in resolution type, such as a right turn followed by a left turn or by a vertical maneuver, divided by the number of encounters (conflicts) Predicted Violations number of times when GRE predicted a violation for any separation standard within a specified look-ahead time Actual Violations number of times when GRE detected a violation; this indicates that an avoidance maneuver failed to resolve a conflict Failure Rate a ratio of number of failures to resolve a conflict to number of detected conflicts in % Predicted NMAC number of times when GRE predicted a NMAC within a specified lookahead time Actual NMAC number of times when GRE detected an NMAC; this indicates a failure of DAA system to prevent an imminent NMAC S NMAC S NMAC = max( RNMAC RCPA , ZNMAC ZCPA ) × 100, where S is a simple measure of severity, R is the range, and Z is the vertical separation.S NMAC exceeds 100% in the case of an Actual NMAC, and lower values of S NMAC correspond to larger separations at CPA relative to the NMAC zone. +IV.B. Parametric Evaluation of GRACE PerformanceGRACE performance was evaluated for perfect and noisy non-cooperative sensors with 8nmi detection range. 8A theoretical omnidirectional sensor was used to study the effect of pilot delay on the performance of GRACE, while onboard radar and tracker models were used to evaluate the effect of noise.All test cases in this sub-section were generated for a 50-minute ownship flight following a multi-turn mission plan typically used in the HITL simulations described in Ref. 8.Intruders cross the ownship's trajectory at five different points, which may occur before, after, or within turns.Encounters are created at each of these points for four headings, three ground speeds, and three vertical speeds (level, climb, descent) for a total of 180 encounters, all of which are designed to result in NMAC.Each of these encounters was processed separately and independently from others, and a summary of statistics for all encounters was generated.This is equivalent to repeating the same ownship flight for 180 different intruders.GRACE was configured to provide guidance for preventing violation (loss of DAA Well Clear) with 0.66nmi miss distance threshold, 450ft vertical separation threshold, and 35sec modified tau threshold with tau computed using 0.66nmi distance modifier. 7,8 f loss of Well Clear could not be prevented, GRACE works to avoid NMAC and maximize separation at CPA.In all these tests GRACE is called every second with two minutes prediction horizon (look-ahead time), which is typical for DAA applications.Table 1 summarizes the results for three sensor models, namely, 1. Omnidrectional: a theoretical perfect sensor that can detect all intruders within a geometric cylinder with a range of 8 nmi , a height of ±5000 ft; 2. Directional Perfect: a tracker and sensor model of airborne radar, 8 using truth ownship and intruder states without sensor noise and navigation errors; 3. Directional Noisy: the same directional tracker and sensor model of airborne radar, which uses perturbed ownship and intruder states with added sensor noise and navigation errors.With unlimited look-ahead time and sensor Field-of-Regard, any algorithm can be reasonably expected to avoid all violations of DWC.Table 1 demonstrates that with an 8nmi sensor range and a 2min look-ahead time, GRACE was still able to prevent most violations.Moreover, GRACE avoided NMAC in all cases when Well Clear violations could not be prevented.Note that, in the table, the number of predicted violations exceeds the number of encounters for two reasons: first, return to mission plan may generate a secondary conflict, and second, for the directional cases, multiple resolutions may be commanded for each encounter (this is particularly evident for the noisy case).In the omnidirectional sensor case, only one conflict was not resolved.Analysis of this conflict showed that it was an artifact of JADEM recapture logic combined with a mismatch in handling turns.JADEM's flight simulator always executes turns in flyover mode (i.e. after crossing a waypoint), whereas GRACE was configured in this study to use flyby mode.This discrepancy induced a secondary conflict on recapture after the primary conflict was successfully resolved.In essence, this is a problem that needs to be addressed in JADEM.The failure rate appears significantly higher in the case of directional perfect and noisy sensors.Closer examination revealed that all these failures could be attributed to late surveillance detections with intruders being detected when they already violated Well Clear.Moreover, in all these cases the detected trajectories were already diverging after they passed CPA, so no maneuvers were needed to improve the situation.These late detections are an artifact of sensor performance, which was not a focus of this study.For both perfect and noisy sensors GRACE was able to resolve all conflicts that were not detected too late.The number of changed resolutions per encounter for perfect sensors was very low (less than 0.06).For noisy sensors the number of changed resolutions per encounter increased to 1.1, which is attributed to large vertical errors of non-cooperative radar sensor.For concepts that require a pilot to evaluate and execute DAA guidance, a delay is incurred before the maneuver is finally flown by the UAS control system.DAA algorithms should ensure that recommended maneuvers remain valid at that time.GRACE allows for delayed response by introducing a delay parameter (the aforementioned algorithm delay) and computing guidance starting at a point on ownship's trajectory corresponding to this delay (keeping the intervening segment "frozen" as it were; see section III.B.2).It should be noted, however, that pilot response delays are not known in advance and vary from one pilot to another and from one encounter to another.Therefore, it is important to know how sensitive GRACE performance is to a mismatch between algorithm delay and actual response time.Table 2 illustrates the effect of algorithm delays when the total pilot response time is 10 seconds.These delays are chosen based on results of previous HITL studies.For all algorithm delays varying between 5 and 15 seconds GRACE resolutions avoided all NMACs and prevented almost all Well Clear violations.3 illustrates the effect of pilot response time given a conservative 15-second algorithm delay.For all pilot delays GRACE ensured that all NMACs could be avoided with comfortable safety margin (S NMAC below 20%).The number of encounters that resulted in Well Clear violations increased with pilot delays as expected, but remained below 4% of total number of encounters even for total pilot delay as high as 20 seconds.The number of changed resolutions per encounter was below 1.3 for all combinations of algorithm and pilot delays.These results clearly indicate that GRACE remains robust even when pilot delays are large and differ from delays anticipated by the algorithm.The effectiveness of GRACE in mitigating collision threats was validated in a 24-hour NAS-wide fast-time study of simulated UAS traffic and recorded radar Visual Flight Rules (VFR) traffic.Instrument Flight Rules (IFR) traffic was not included, since separation of IFR traffic will presumably be maintained by Air Traffic Control.Eighteen UAS mission profiles developed under prior work were used in this study to simulate a variety of UAS aircraft conducting an assortment of possible future UAS missions, including: point-to-point transport, regional mapping/monitoring, and patrol. 13,45 he 24-hour scenario included more than 17,000 UAS flights with about 10,000 total flight hours, mostly in transitional class E airspace in proximity to VFR traffic. 45It was assumed that all unmanned aircraft were equipped with cooperative ADS-B and MODE-C sensors and a non-cooperative directional sensor with 8-nmi range typical for onboard radars.VFR traffic was modeled as a mix of flights with and without cooperative sensors.The study leveraged JADEM's NAS-wide simulation capability and used a kinematic performance model for flying UAS missions (honoring commanded resolutions) and for evaluating resolution candidates in GRACE.UAS-to-UAS encounters were not considered in this study, and algorithm performance was evaluated in aggregate over all airspaces.Note that, while the study used perfect surveillance sensors, the VFR data is itself inherently noisy.The study compares two simulations: "unmitigated" and "mitigated."The unmitigated simulation, for which GRACE resolutions are neither commanded nor executed, represents the baseline scenario for UAS without active DAA systems.In contrast, the mitigated simulation commands and executes GRACE resolutions.Moreover, JADEM's flight simulator "recaptures" the UAS nominal flight plan after successful avoidance maneuvers.This recapture can in turn result in secondary conflicts.This study did not model communication failures, latency, or delayed pilot responses.Work to include more realistic models of pilot behavior with random delays is currently underway.Table 4 compares the results of the two simulations and shows that mitigation reduced the number of predicted violations by a factor of eight and the number of actual violations by a factor of five.GRACE failed to prevent actual violations for only 2.5% of conflicts with predicted violations.More importantly, mitigation reduced the number of predicted NMAC events by a factor of 33, and eliminated all actual NMACs.The frequency of changed resolutions averaged at the level of 0.4 per encounter.Changes in resolution types and failures to resolve conflicts can both be attributed to late detections, with intruders first detected when they are too close to ownship.The situation is further complicated by unknown intruder intent, with GRACE having to rely on extrapolated (dead-reckoned) intruder trajectories.These results show that GRACE, used as an automatic guidance algorithm, was able to prevent almost all violations of separation.In cases where this was not possible because of late detection or unexpected intruder maneuver, GRACE was still able to avoid the NMAC. +V. ConclusionsSafe integration of UAS into the NAS requires development and validation of DAA systems as a means to comply with the FAA-mandated "see-and-avoid" requirement for human pilots.Despite the diversity of algorithms that could potentially provide DAA functions, NASA recognized the need for a fast and flexible "generic" alerting and resolution algorithm that could help reduce the complexity of DAA system modeling.To fill that need, the Generic Resolution Advisor and Conflict Evaluator (GRACE) was implemented in NASA's Java Architecture for DAA Extensibility and Modeling to provide alerting, directive guidance, trial planning capabilities, bands, and DWC recovery guidance.GRACE was designed without any assumptions regarding performance capabilities of an aircraft or its sensor and communication systems.This made it suitable for various applications and DAA guidance concepts.In particular, GRACE can be used in prototyping various decision support tools for ground pilots.Furthermore, GRACE makes no assumptions about degree of autonomy.This allows it to be used in fully autonomous UAS DAA operations, in remotely piloted unmanned aircraft, and potentially even in manned flights.The new algorithm was validated and used in a number of human-in-the-loop experiments, in flight tests with live aircraft, in parametric studies with diverse encounter geometries, and in full-day NAS-wide simulations.All these tests demonstrated the ability of GRACE to reduce the frequency and severity of Losses of Well Clear and to prevent NMAC in interactions with other aircraft flying by VFR.Current and future work includes using GRACE to improve robustness of DAA resolutions, taking into account uncertainties of intruder positions and intent.Figure 1 .1Figure 1.Grid-based threat detection +Figure 2 .2Figure 2. Generic Resolution Advisor algorithm +Figure 3 .3Figure 3. Linearization of the ownship and intruder trajectories near TCP +'s coordinates and velocities relative to ownship, (x O ) k , (x I ) k -initial ownship's and intruder's positions, (v O ) k , (v I ) k -ownship's and intruder's velocities.Although the Eq. 1 for time-to-CPA is the same as in Ref.40, uses positions and velocities from the states extrapolated backward CPA for original trajectories predicted by GCE and not from the observed initial states.This ensures more accurate TCP if a CPA for perturbed trajectory does not deviate too much from the original CPA. +Table 1 .1GRACE performance in simulated encounters using different sensorsSensorPredicted ViolationsPredicted NMACChanges per EncounterActual ViolationsFailure Rate (%)S NMAC (%) Mean MaxOmnidirectional2091740.05610.61515Directional perfect2461490.00695.134.339.5Directional noisy38041.11137.331.846.8 +Table 2 .2GRACE performance with pilots in the loop for 10-second total response timeAlgorithmPredictedPredictedChanges perActualFailureS NMAC (%)DelayViolationsNMACEncounterViolationsRate (%)Mean Max5 seconds234918490.7331.713.413.810 seconds234618490.6400--15 seconds231618490.6010.615.615.6Table +Table 3 .3GRACE performance with pilots in the loop for 15-second algorithm delayTotal PilotPredictedPredictedChanges perActualFailureS NMAC (%)DelayViolationsNMACEncounterViolationsRate (%)Mean Max10 seconds231618490.6410.615.615.615 seconds341926361.0063.3812.520 seconds468634631.2773.99.914.7IV.C. Effectiveness in NAS-wide Mitigated Studies +Table 4 .4Effect of mitigation on a full-day simulated UAS traffic over the NASSimulationPredicted ViolationsActual ViolationsPredicted NMACActual NMACS NMAC (%) Mean MaxUnmitigated115409189422204429.4546.3Mitigated1439435968020.292.2 + of 13 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 9, 2017 | http://arc.aiaa.org| DOI: 10.2514/6.2017-4485 + Downloaded by NASA AMES RESEARCH CENTER on June 9, 2017 | http://arc.aiaa.org| DOI: 10.2514/6.2017-4485 + of 13 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 9, 2017 | http://arc.aiaa.org| DOI: 10.2514/6.2017-4485 + of 13 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 9, 2017 | http://arc.aiaa.org| DOI: 10.2514/6.2017-4485 + of 13 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 9, 2017 | http://arc.aiaa.org| DOI: 10.2514/6.2017-4485 + of 13 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June + 9, 2017 | http://arc.aiaa.org| DOI: 10.2514/6.2017-4485 + of 13 American Institute of Aeronautics and Astronautics Downloaded by NASA AMES RESEARCH CENTER on June 9, 2017 | http://arc.aiaa.org| DOI: + 10.2514/6.2017-4485 + + + + +AcknowledgmentsThe authors want to express gratitude to Eric Mueller and Heinz Erzberger for fruitful discussions and encouragement, Gilbert Wu for his work on integrating the sensor and tracker models and for contributing a description to said models, and the whole DAA team at NASA and our partners from Honeywell for all their help and support. + + + + + + + + + Code of Federal Regulation + sec. 91.113 + + + General Operating and Flight Rules + + 14 + + + Code of Federal Regulation, General Operating and Flight Rules, title 14, sec. 91.113. + + + + + Sense and Avoid (SAA) for Unmanned Aircraft Systems (UAS) + + + FAA Workshop + + Oct. 2009 + + + "Sense and Avoid (SAA) for Unmanned Aircraft Systems (UAS)," FAA Workshop, Oct. 2009. + + + + + Sense and Avoid in UAS + + PAngelov + + + + Research and Applications + + 38 + 2012 + Wiley + + + Angelov, P., Sense and Avoid in UAS. Research and Applications, Wiley, 2012, p. 38. + + + + + Collision analysis for an UAV + + ADesilles + + + HZidani + + + ECruck + + + + Proceedings of the AIAA Guidance, Navigation, and Control Conference + the AIAA Guidance, Navigation, and Control Conference + + Sept. 2012 + + + Desilles, A., Zidani, H., and Cruck, E., "Collision analysis for an UAV," Proceedings of the AIAA Guidance, Navigation, and Control Conference, Sept. 2012. + + + + + RTCA DO-185B "MOPS for TCAS II + + AZeitlin + + + + Progress on Requirements and Standards for Sense & Avoid, The MITRE Corporation + + Aug. 2010. June 19, 2008 + I + + + Zeitlin, A., Progress on Requirements and Standards for Sense & Avoid, The MITRE Corporation, Aug. 2010. 6 RTCA DO-185B "MOPS for TCAS II," Version 7.1, Volume I, June 19, 2008. + + + + + Defining Well Clear for Unmanned Aircraft Systems + + SCook + + + DBrooks + + + RCole + + + DHackenburg + + + VRaska + + AIAA 2015-0481 + + American Institute of Aeronautics and Astronautics + + + Paper presented to the + Cook, S., Brooks, D., Cole, R., Hackenburg, D., and Raska, V., "Defining Well Clear for Unmanned Aircraft Systems," Paper presented to the American Institute of Aeronautics and Astronautics (AIAA 2015-0481). + + + + + The Generic Resolution Advisor and Conflict Evaluator (GRACE) for Unmanned Aircraft Detect-And-Avoid Systems + + MAbramson + + + MRefai + + + CSantiago + + NASA/TM-2017-219507 + + 2017 + In Press + + + Abramson, M., Refai, M., Santiago, C., "The Generic Resolution Advisor and Conflict Evaluator (GRACE) for Unmanned Aircraft Detect-And-Avoid Systems," 2017, (In Press), NASA/TM-2017-219507. + + + + + UAS measured response: The effect of GCS control mode interfaces on pilot ability to comply with atc clearances + + RCRorie + + + LFern + + + + Proceedings of the Human Factors and Ergonomics Society Annual Meeting + the Human Factors and Ergonomics Society Annual Meeting + + 2014 + 58 + + + + Rorie, R.C. and Fern, L., "UAS measured response: The effect of GCS control mode interfaces on pilot ability to comply with atc clearances," Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 2014, Vol. 58, No. 1, pp. 64-68. + + + + + Air Traffic Controller Acceptability of Unmanned Aircraft System Detect and Avoid Thresholds + + ERMueller + + + DIsaacson + + + DStevens + + NASA TM-2015-219392 + + 2015 + + + Mueller, E.R., Isaacson, D., and Stevens, D., "Air Traffic Controller Acceptability of Unmanned Aircraft System Detect and Avoid Thresholds," 2015, NASA TM-2015-219392. + + + + + Pilot Evaluation of a UAS Detect-and-Avoid System's Effectiveness in Remaining Well Clear + + CSantiago + + + RMueller + + + + 10th USA/Europe ATM R & D Seminar (ATM2015) + Lisbon, Portugal + + 2015 + + + Santiago, C., and Mueller, R., "Pilot Evaluation of a UAS Detect-and-Avoid System's Effectiveness in Remaining Well Clear," 10th USA/Europe ATM R & D Seminar (ATM2015), Lisbon, Portugal, 2015. + + + + + UAS Integration in the NAS Project: Flight Test 3 Data Analysis of JADEM-Autoresolver Detect and Avoid System + + CGong + + + GMinghong + + + Wu + + + CSantiago + + NASA/TM-2012-216051 + + 2012 + + + Gong, C., Minghong G. Wu, and Santiago, C., "UAS Integration in the NAS Project: Flight Test 3 Data Analysis of JADEM-Autoresolver Detect and Avoid System," 2012, NASA/TM-2012-216051. + + + + + Evaluating Alerting and Guidance Performance of a UAS Detect-And-Avoid System + + SMLee + + + CPark + + + DPThipphavong + + + DRIsaacson + + + CSantiago + + NASA-TM-2016-219067 + + Feb. 2016 + + + Lee, S. M., Park, C., Thipphavong, D. P., Isaacson, D. R., and Santiago, C., "Evaluating Alerting and Guidance Performance of a UAS Detect-And-Avoid System," NASA-TM-2016-219067 , Feb. 2016. + + + + + UAS Pilot Evaluations of Suggestive Guidance on Detect-and-Avoid Displays + + KJMonk + + + ZRoberts + + + + Proceedings of the Human Factors and Ergonomics Society 60th Annual Meeting + the Human Factors and Ergonomics Society 60th Annual MeetingWashington, DC + + 2016 + + + Monk, K.J. and Roberts, Z., "UAS Pilot Evaluations of Suggestive Guidance on Detect-and-Avoid Displays," Proceedings of the Human Factors and Ergonomics Society 60th Annual Meeting, Washington, DC, 2016. + + + + + Piloted Well Clear Performance Evaluation of Detect-and-Avoid Systems with Suggestive Guidance + + ERMueller + + + CSantiago + + + SWatza + + NASA TM-2016-219396 + + 2016 + + + Mueller, E. R., Santiago, C., and Watza, S., "Piloted Well Clear Performance Evaluation of Detect-and-Avoid Systems with Suggestive Guidance," 2016, NASA TM-2016-219396. + + + + + Flight Test Overview for UAS Integration in the NAS Project + + JMurphy + + + PHayes + + + SKim + + + WBridges + + + MMarston + + + + AIAA Atmospheric Flight Mechanics Conference (AIAA-2016-1756) + San Diego, CA + + 2016 + + + Murphy, J., Hayes, P., Kim, S., Bridges, W., and Marston, M., "Flight Test Overview for UAS Integration in the NAS Project," AIAA Atmospheric Flight Mechanics Conference (AIAA-2016-1756), San Diego, CA, 2016. + + + + + Maintain and Regain Well Clear: Maneuver Guidance Designs for Pilots Performing the Detect-and-Avoid Task + + KJMonk + + + ZRoberts + + + + Proceedings of the International Conference on Applied Human Factors and Ergonomics + the International Conference on Applied Human Factors and ErgonomicsLos Angeles + + CA (In Press + 2016 + + + Monk, K.J. and Roberts, Z., "Maintain and Regain Well Clear: Maneuver Guidance Designs for Pilots Performing the Detect-and-Avoid Task," Proceedings of the International Conference on Applied Human Factors and Ergonomics, Los Angeles, CA (In Press), 2016. + + + + + Validation of minimum display requirements for a UAS detect and avoid system + + RCRorie + + + LFern + + + KMonk + + + CSantiago + + + RJShively + + + ZRoberts + + + + AIAA Aviation Conference + Denver, CO + + In Press + 2017 + + + Rorie, R.C., Fern, L., Monk, K., Santiago, C., Shively, R.J., and Roberts, Z., "Validation of minimum display requirements for a UAS detect and avoid system," AIAA Aviation Conference, Denver, CO (In Press), 2017 + + + + + A Review of Conflict Detection and Resolution Modeling Methods + + JKKuchar + + + LCYang + + + + IEEE Transactions on Intelligent Transportation Systems + + 1 + 4 + 2000 + + + Kuchar, J. K. and Yang, L. C. "A Review of Conflict Detection and Resolution Modeling Methods." IEEE Transactions on Intelligent Transportation Systems, 1(4), 2000. + + + + + TCAS: A System for Preventing Midair Collisions + + WHHarman + + + + Lincoln Laboratory Journal + + 2 + 3 + 1989 + + + Harman, W. H., "TCAS: A System for Preventing Midair Collisions," Lincoln Laboratory Journal, 1989, vol. 2, no. 3. + + + + + Unmanned Aircraft Collision Avoidance System Using Cooperative Agent-Based Negotiation Approach + + BMAlbaker + + + NARahim + + + + Int. J. Simulation, Syst. Sci. Technol + + April 2010 + + + Albaker, B. M. and Rahim, N. A., "Unmanned Aircraft Collision Avoidance System Using Cooperative Agent-Based Negotiation Approach," Int. J. Simulation, Syst. Sci. Technol., April 2010. + + + + + A Cooperative Multi-agent Approach to Free Flight + + JCHill + + + FRJohnson + + + JKArchibald + + + RLFrost + + + WCStirling + + + + Proceedings of the 4th International Joint Conference on Autonomous Agents and Multi Agent Systems + the 4th International Joint Conference on Autonomous Agents and Multi Agent Systems + + July 2005 + + + Hill, J. C., Johnson, F. R., Archibald, J. K., Frost, R. L., and Stirling, W. C., "A Cooperative Multi-agent Approach to Free Flight," Proceedings of the 4th International Joint Conference on Autonomous Agents and Multi Agent Systems, July 2005. + + + + + Performance Characteristics of a Complete Vision Only Sense and Avoid System + + BVanek + + + TPeni + + + AZarandy + + + JBokor + + + TZsedrovits + + + TRoska + + + W.-ZChen + + + LWong + + + JKay + + + VMRaska + + + + + Proceedings of the AIAA Guidance, Navigation, and Control Conference + the AIAA Guidance, Navigation, and Control Conference + + Sept. 2012 + + + MP-SCI-202-28 + Autonomous Sense and Avoid (SAA) for Unmanned Air Systems (UAS) + Vanek, B., Peni, T., Zarandy, A., Bokor, J., Zsedrovits, T., and Roska, T., "Performance Characteristics of a Complete Vision Only Sense and Avoid System," Proceedings of the AIAA Guidance, Navigation, and Control Conference, Sept. 2012. 24 Chen, W.-Z., Wong, L., Kay, J., , and Raska, V. M., "Autonomous Sense and Avoid (SAA) for Unmanned Air Systems (UAS)," MP-SCI-202-28.doc available at ftp://ftp.rta.nato.int/PubFullText/RTO/MP/RTO-MP-SCI-202/, 2011. 25 + + + + + Robust Airborne Collision Avoidance through Dynamic Programming + + MJKochenderfer + + + JPChryssanthacopoulos + + ATC-371 + + 2011 + + + Massachusetts Institute of Technology, Lincoln Laboratory + + + Project Report + Kochenderfer, M. J. and Chryssanthacopoulos, J. P., "Robust Airborne Collision Avoidance through Dynamic Program- ming," Massachusetts Institute of Technology, Lincoln Laboratory, 2011, Project Report ATC-371. + + + + + Design of a Conflict Detection Algorithm for the Center/TRACON Automation System + + DIsaacson + + + HErzberger + + + + Proc. 16th Digital Avionics Systems Conf + 16th Digital Avionics Systems ConfIrvine, CA + + 1997 + 27 + + + Isaacson, D. and Erzberger, H. "Design of a Conflict Detection Algorithm for the Center/TRACON Automation System," Proc. 16th Digital Avionics Systems Conf., Irvine, CA, 1997, 27 + + + + + The Automated Airspace Concept + + HErzberger + + + + Proceedings of the Fourth USA/Europe Air Traffic Management R & D Seminar + the Fourth USA/Europe Air Traffic Management R & D Seminar + + Dec. 2001 + + + Erzberger, H., "The Automated Airspace Concept," Proceedings of the Fourth USA/Europe Air Traffic Management R & D Seminar , Dec. 2001. + + + + + Advances in Force Field Conflict Resolution Algorithms + + WKelly + + + MEby + + + + Proceedings of the AIAA Guidance, Navigation, and Control Conference + the AIAA Guidance, Navigation, and Control Conference + + Aug. 2000 + + + Kelly, W. and Eby, M., "Advances in Force Field Conflict Resolution Algorithms," Proceedings of the AIAA Guidance, Navigation, and Control Conference, Aug. 2000. + + + + + Optimal Coordinated Maneuvers for Three-Dimensional Aircraft Conflict Resolution + + JHu + + + MPrandini + + + ANilim + + + SSastry + + + + Proceedings of the AIAA Guidance, Navigation, and Control Conference + the AIAA Guidance, Navigation, and Control Conference + + Aug. 2001 + + + Hu, J., Prandini, M., Nilim, A., and Sastry, S., "Optimal Coordinated Maneuvers for Three-Dimensional Aircraft Conflict Resolution," Proceedings of the AIAA Guidance, Navigation, and Control Conference, Aug. 2001. + + + + + Generation of Conflict Resolution Maneuvers for Air Traffic Management + + JKosecka + + + CTomlin + + + GPappas + + + SSastry + + + + 1997 International Conf. on Robotics and Intelligent Systems + + Sept. 1997 + + + Kosecka, J., Tomlin, C., Pappas, G., and Sastry, S. "Generation of Conflict Resolution Maneuvers for Air Traffic Man- agement," 1997 International Conf. on Robotics and Intelligent Systems, Sept. 1997. + + + + + A light propagation model for aircraft trajectory planning + + NDougui + + + DDelahaye + + + SPeuchmorel + + + MMongeau + + + + Journal of Global Optimization + + 56 + 3 + 2013 + + + Dougui, N., Delahaye, D., Peuchmorel, S., and Mongeau, M. "A light propagation model for aircraft trajectory planning," Journal of Global Optimization, 56(3), 2013. + + + + + Control of multiple non-holonomic air vehicles under wind uncertainty using model predictive control and decentralized navigation functions + + GRoussos + + + GChaloulos + + + KKyriakopoulos + + + JLygeros + + + 2008 + IEEE + + + Roussos, G., Chaloulos, G., Kyriakopoulos, K., Lygeros, J. "Control of multiple non-holonomic air vehicles under wind uncertainty using model predictive control and decentralized navigation functions," IEEE Conference on Decision and Control. IEEE , 2008 + + + + + Towards constant velocity navigation and collision avoidance for autonomous nonholonomic aircraft-like vehicles + + GRoussos + + + KKyriakopoulos + + + + th Chinese Control Conference. Proceedings of the 48th IEEE Conference + Shanghai + + 2009. 2009. 2009 + 28 + + + Decision and Control + Roussos, G., Kyriakopoulos, K. "Towards constant velocity navigation and collision avoidance for autonomous non- holonomic aircraft-like vehicles." Decision and Control, 2009 held jointly with the 2009 28th Chinese Control Conference. Proceedings of the 48th IEEE Conference, Shanghai, 2009 + + + + + Deconflicting Wind-Optimal Aircraft Trajectories in North Atlantic Oceanic Airspace + + ORodionova + + + DDelahaye + + + BSridhar + + + HKNg + + + + Advanced Aircraft Efficiency in a Global Transport System (AEGATS '16) + Paris, France + + 2016 + 35 + + + Rodionova O., Delahaye D., Sridhar B., Ng H. K. "Deconflicting Wind-Optimal Aircraft Trajectories in North Atlantic Oceanic Airspace." Advanced Aircraft Efficiency in a Global Transport System (AEGATS '16), Paris, France, 2016 35 + + + + + Aircraft conflict resolution by genetic algorithm and B-spline approximation + + DDelahaye + + + CPeyronne + + + MMongeau + + + SPeuchmorel + + + + 2nd ENRI International Workshop on ATM/CNS + + 2011 + + + Delahaye, D., Peyronne, C., Mongeau, M., and Peuchmorel, S. "Aircraft conflict resolution by genetic algorithm and B-spline approximation." 2nd ENRI International Workshop on ATM/CNS , 2011 + + + + + Modeling Systemic Phenomena in the National Airspace System + + BSawhill + + + + NASA Aeronautics Airspace Systems Program Technical Interchange Meeting + + March 2011 + + + Sawhill, B., "Modeling Systemic Phenomena in the National Airspace System," NASA Aeronautics Airspace Systems Program Technical Interchange Meeting, March 2011. + + + + + A Method for Detecting and Avoiding Flight Hazards + + HVon Viebahn + + + JSchiefele + + + + Proceedings of the SPIE Meeting on Enhanced Synthetic Vision + the SPIE Meeting on Enhanced Synthetic Vision + + April 1997 + + + Von Viebahn, H. and Schiefele, J., "A Method for Detecting and Avoiding Flight Hazards," Proceedings of the SPIE Meeting on Enhanced Synthetic Vision, April 1997. + + + + + Stochastic Reachability for Discrete Time Systems: An Application to Aircraft Collision Avoidance + + OWatkins + + + JLygeros + + + 2003 + + + Watkins, O. and Lygeros, J., "Stochastic Reachability for Discrete Time Systems: An Application to Aircraft Collision Avoidance," IEEE Conference on Decision and Control, 2003. + + + + + DAIDALUS: Detect and Avoid Alerting Logic for Unmanned Systems + + CMuñoz + + + ANarkawicz + + + GHagen + + + JUpchurch + + + ADutle + + + MConsiglio + + + JChamberlain + + + + Proceedings of the 34th Digital Avionics Systems Conference (DASC 2015) + the 34th Digital Avionics Systems Conference (DASC 2015)Prague, Czech Republic + + 2015 + + + Muñoz, C., Narkawicz, A., Hagen, G., Upchurch, J., Dutle, A., Consiglio, M., and Chamberlain, J., "DAIDALUS: Detect and Avoid Alerting Logic for Unmanned Systems," Proceedings of the 34th Digital Avionics Systems Conference (DASC 2015), Prague, Czech Republic, 2015 + + + + + A TCAS-II Resolution Advisory Detection Algorithm + + CMuñoz + + + ANarkawicz + + + JChamberlain + + + + AIAA Guidance, Navigation, and Control Conference + + 2013 + + + Muñoz, C., Narkawicz, A., and Chamberlain, J., "A TCAS-II Resolution Advisory Detection Algorithm," AIAA Guid- ance, Navigation, and Control Conference, 2013. + + + + + UAS Well Clear Recovery against Non-Cooperative Intruders using Vertical Maneuvers + + ACone + + + DPThipphavong + + + SMLee + + + CSantiago + + + + 17th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference + + (In Press + 2017 + + + Cone, A., Thipphavong, D.P., Lee, S.M., and Santiago, C., "UAS Well Clear Recovery against Non-Cooperative Intruders using Vertical Maneuvers," 17th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference , (In Press), 2017 + + + + + Ensuring Interoperability between UAS Detect-and-Avoid and Manned Aircraft Collision Avoidance + + DThipphavong + + + ACone + + + + 12th USA/Europe ATM R & D Seminar (ATM2017) + + (In Press + 2017 + + + Thipphavong, D. and Cone, A.,"Ensuring Interoperability between UAS Detect-and-Avoid and Manned Aircraft Collision Avoidance," 12th USA/Europe ATM R & D Seminar (ATM2017), (In Press), 2017 + + + + + Characteristics of a Well Clear Definition and Alerting Criteria for Encounters between UAS and Manned Aircraft in Class E Airspace + + MJohnson + + + EMueller + + + CSantiago + + + + 10th USA/Europe ATM R & D Seminar (ATM2015) + Lisbon, Portugal + + June 2015 + + + + Johnson, M., Mueller, E., and Santiago, C., "Characteristics of a Well Clear Definition and Alerting Criteria for Encounters between UAS and Manned Aircraft in Class E Airspace," 10th USA/Europe ATM R & D Seminar (ATM2015), Lisbon, Portugal, 23-26 June 2015. + + + + + Downstream Effects of Separation Assurance on Encounters between UAS and Manned Aircraft + + DPThipphavong + + + MAJohnson + + + MSRefai + + + JWSnow + + + + Journal of Air Transportation + + 2017 + AIAA), (In Press + + + Thipphavong, D.P., Johnson, M.A., Refai, M.S., and Snow, J.W., "Downstream Effects of Separation Assurance on Encounters between UAS and Manned Aircraft," Journal of Air Transportation (AIAA), (In Press), 2017 + + + + + Unmanned Aircraft System Demand Generation and Airspace Performance Impact Prediction + + SAyyalasomayajula + + + FWieland + + + ATrani + + + NHinze + + + + Proceedings of the 32nd IEEE Digital Avionics Systems Conference + the 32nd IEEE Digital Avionics Systems ConferenceSyracuse, NY + + IEEE + October 2013 + + + Ayyalasomayajula, S., Wieland, F., Trani, A., and Hinze, N., "Unmanned Aircraft System Demand Generation and Airspace Performance Impact Prediction," Proceedings of the 32nd IEEE Digital Avionics Systems Conference, IEEE. Syracuse, NY, October 2013. + + + + + Radar Data Tracking Using Minimum Spanning Tree-Based Clustering Algorithm + + CPark + + + HLee + + + BMusaffar + + + + 11th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference (AIAA-2011-6825) + Virginia Beach, VA + + Sept. 2011 + + + Park, C., Lee, H., and Musaffar, B., "Radar Data Tracking Using Minimum Spanning Tree-Based Clustering Algo- rithm," 11th AIAA Aviation Technology, Integration, and Operations (ATIO) Conference (AIAA-2011-6825), Virginia Beach, VA, Sept. 2011. + + + + + + diff --git a/file20.txt b/file20.txt new file mode 100644 index 0000000000000000000000000000000000000000..cbaf41f68ebec621ab57265be7ee5b3916c6df85 --- /dev/null +++ b/file20.txt @@ -0,0 +1,364 @@ + + + + +I. Introductionlight delays result when the capacity of critical resources, such as airspace or airports, is insufficient to meet demand.Better matching between demand and available capacity can lead to efficiency improvements, with increased throughput, reduced delays, and more efficient flight trajectories.However, because of uncertainties in both demand and capacity (e.g., due to weather), such matching can be difficult.Mismatches in demand and capacity can be addressed by using a strategic pre-departure traffic flow management capability, such as the Collaborative Trajectory Options Program (CTOP), to pre-condition demand into the more tactical Time-Based Flow Management System (TBFM).This would enable TBFM to better manage delivery of demand to capacity constrained resources.CTOP adjusts demand in two ways.First, it can delay flights on the ground, reducing the arrival rate at the constraint resource.Second, it can reroute flights to avoid capacity constrained airspace or airports by selecting from a userspecified list of acceptable reroutes for each flight.A wide variety of open questions exist addressing benefits of airline participation in CTOP and sensitivity of TBFM airborne delay to various sources of uncertainty.A number of human-in-the-loop (HITL) simulation studies have been carried out with experienced air traffic manager and controller participants 1,2 to better understand the interaction of CTOP and TBFM.While these experiments have produced important results, the resources and time required to conduct real-time studies with human participants limits the number of HITL simulations that can be conducted and thus the number of experimental conditions that can be tested.In this paper, an automated simulation capability, 3 called Traffic Management Initiative Automated Simulation (TMIAutoSim), is used to study the impact of airline participation in CTOP on ground delay and departure uncertainty on TBFM delay.TMIAutoSim can closely mimic the HITL simulation capabilities while automating both the human components and the collaboration between operational systems. 3TOP and TBFM are described in more detail in Section II.In Section III, a variety of relevant open questions which need to be addressed are enumerated, followed by the specific objectives of this work.Section IV and Section V describe the two studies that were performed within the TMIAutoSim environment to meet these objectives.Section IV outlines a study that was conducted to quantify the impact of varying the levels of airline participation in CTOP on delay.Section V presents the results from a study used to quantify the impact of departure uncertainty on TBFM delays.Concluding remarks and notes for future work are given in Section VI. +II. BackgroundAs described in Ref. 1 and Ref. 3, demand for capacity-constrained airspace or airports can be controlled by a series of Traffic Management Initiatives (TMIs), which use departure and airborne delays, as well as pre-departure and airborne reroutes, to manage access to the constrained resources.Two systems exist in current and planned future operations to address imbalances between demand and capacity, but in different operational contexts and different timeframes.One is more strategic in nature, with a planning horizon of hours and geographical reach that spans across the NAS.The second is more locally targeted to terminal airspace surrounding airports and has a shorter, more tactical time horizon.The capabilities comprising these two systems address different parts of the capacity management problem without close coordination.Integrated Demand Management (IDM) is a NASA concept that aims to integrate and coordinate demand management across these systems to work together more effectively. 1Within the IDM concept framework, the strategic system may redistribute demand from an overloaded resource to one with reserve capacity, or allocate predeparture delay so that it is calibrated to variations in demand, and better matched to the demand that the tactical system can accommodate.This should allow the strategic system to manage delivery to the capacity-constrained destination more efficiently.For strategic flow management, the IDM concept utilizes a new TMI called the CTOP.CTOP-specific capabilities support a characterization of the reduced capacity problem using multiple constraints, called Flow Constrained Areas (FCAs).CTOP can be initiated by the Air Traffic Control System Command Center when the capacity at a particular airport or region of airspace is expected to fall well below expected demand.When airspace is constrained, CTOP allows airlines to route their flights out of the associated FCA, or across a different FCA that has available capacity, by submitting a preference-weighted set of alternative routes called a Trajectory Option Set (TOS) from which the program can select.CTOP distributes traffic to reduce demand to manageable levels by delaying departures and allocating trajectories from airline TOSs to flights scheduled to pass through the specified FCAs.Departure delay is assigned via the Expect Departure Clearance Time (EDCT).The mechanism by which each alternative route is weighted is a Relative Trajectory Cost (RTC), in minutes, submitted by airlines, indicating the amount of ground delay the airline is willing to accept before preferring the next, lower ranked trajectory option.Trajectories and ground delays for each flight are calculated using the CTOP TOS allocation algorithm, using a ratio-by-schedule approach, and considering the RTC values assigned by each airline to each trajectory option. 4,5lthough the objective of CTOP is to set departure times and routes such that maximum rates at capacity constrained resources are not exceeded, several factors impact a flight's actual arrival time at the resource, including departure error, winds, and changes in trajectory due to weather conditions.Thus, using only strategic tools, there is no guarantee that resource constraints will be satisfied.For tactical flow management, TBFM is an arrival scheduling tool that was developed to improve traffic delivery into capacity-constrained airports or airspace resources. 1,6As described in Ref. 3, TBFM is generally used to schedule flights arriving into a constrained airspace or airport.In this study, we focus on a specific example in which flights destined for Newark Liberty International Airport (EWR) are subject to arrival rate constraints at the runway threshold.TBFM creates a schedule for all inbound flights which provides a scheduled time of arrival for each flight at the runway threshold.Arrivals are sequenced by a "first-come first-served" scheduling algorithm according to their respective estimated arrival time at the runway threshold.A scheduled time of arrival is then assigned to each flight to satisfy the required inter-arrival spacing for a sequence of fixes through to the runway threshold.The arrival sequence and assigned scheduled time of arrival for each flight continues to change as the flight's estimated time of arrival changes, up until the flight crosses a specified TBFM freeze horizon, roughly 400 nmi upstream of the runway threshold.At that point, the flight's scheduled time of arrival is "frozen" and the target arrival time at the runway threshold is fixed.Air traffic controllers are then responsible for managing aircraft to absorb delay in the air using delay vectors and speed clearances to meet these assigned scheduled times of arrival.Operationally, there is a limit to the amount of airborne delay that can be absorbed within the TBFM freeze horizon.The problem becomes progressively more difficult as the amount of delay that must be absorbed increases.While this will depend on a variety of factors, such as distance from the TBFM freeze horizon to the runway threshold and airspace complexity, subject matter experts can be consulted to classify ranges of airborne delay for individual flights as acceptable, marginal, and unacceptable.Understanding the conditions under which airborne delay is pushed into the marginal or unacceptable realm is an open question that can be addressed by a parametric study.Flights that depart from airports within the TBFM freeze horizon are referred to as "internal departures."These flights can receive delay on the ground in order to ensure that they fit into the overhead stream being controlled by TBFM.Their departure times are obtained using a TBFM departure scheduling function that assigns a departure time corresponding to an available arrival slot.Upon take off, flights can receive further airborne delay from TBFM, but this is typically small because the departure scheduling process ensures that these flights are departing into gaps in the overhead stream.Internal departure scheduling is performed in one of two ways.In the first scheduling paradigm, internal departures are given priority and are inserted into the overhead stream ahead of airborne flights.This may require assigning some delay to airborne flights in order to create a gap large enough to fit the internal departure.In the second, priority is given to airborne flights, with internal departures held on the ground until a large enough gap exists in the overhead stream.The prioritization of internal departures is a key enabler of the IDM concept, reducing the incidence of significant 'double delays,' in which internal departures receive high ground delays from CTOP, followed by high scheduling delays from TBFM because they cannot easily fit into the overhead stream.When internal departures are prioritized, this leads to an increase in airborne delay compared to using the scheduling paradigm in which airborne flights are prioritized.If the number of flights receiving marginal or unacceptable delay is too high when internal departures are prioritized, switching to the prioritization of airborne flights will reduce airborne delay in favor of greater ground delay for internal departures. 3ost of the software tools used in the HITL simulations were adopted, with some modification, for use by TMIAutoSim.Because the version of TBFM used in the HITL simulations is not suitable for use without human input, an alternative TBFM emulator was developed for TMIAutoSim based on an arrival scheduler developed for the Optimized Route Capability (ORC) 7 tool.Functionality of the newly developed TBFM emulator including the implementation of the two TBFM scheduling paradigms described above (prioritizing airborne flights and prioritizing internal departures) and the automation of internal departure scheduling in accordance with TBFM procedures are described in detail in Ref. 3.An emulation of CTOP, called nCTOP (NASA CTOP) 1 and constructed for the HITL simulations, was used to emulate the key functions of the operational CTOP system.nCTOP contains some added functionality from the original CTOP concept that may be introduced in the operational system in the future.The Multi-Aircraft Control System (MACS), 8 a high-fidelity air traffic control simulation environment for prototyping scheduling systems and simulating air traffic, was used to simulate the air traffic.More information about the development and details of TMIAutoSim can be found in Ref. 3. +III. ObjectivesA variety of parametric studies could be performed to improve understanding of the IDM concept and help researchers develop guidelines for use of the strategic and tactical tools under various conditions.These questions can be broken down into two categories: benefits of airline participation in CTOP and sensitivity of TBFM airborne delay to various sources of uncertainty.While CTOP is an operational FAA TMI, first introduced in 2014, 4 it is still in limited operational use due to a variety of factors.For example, airlines need to do a significant amount of work to develop tools and procedures to build TOSs.Furthermore, this would impose an increased workload on airline dispatchers to review and approve trajectory options before submission into CTOP.This takes a considerable amount of research and development.Thus, airlines are coming online with CTOP capabilities at various rates.Some airlines with a greater capacity for research and development have been developing these tools while other carriers have not advanced as far.Quantifying the effects of varying levels of CTOP participation on ground delays and flight times serves two purposes.First, this analysis will provide an indication of potential benefits of CTOP participation to airlines.And second, the analysis will provide insights into the relationships between CTOP participation and the effectiveness of CTOP to reduce average ground delays to reasonable levels, given capacity constraints.Sources of uncertainty in proposed IDM operations include: (1) flight departure error, when flights destined for the constrained airport do not take off exactly at their assigned departure time; (2) FCA capacity forecast error, when there is a mismatch between the predicted FCA rates used strategic planning with CTOP and the actual rates that are later used in tactical planning by TBFM; (3) airport arrival rate forecast error, a special case of FCA capacity forecast error when an FCA is located at the destination airport; and (4) the impact of weather and other traffic on flight time.Example research questions include identifying uncertainty values for each of these parameters at which the IDM concept is no longer effective, and whether they vary with the constraint capacity; whether or not the robustness of the concept can be improved by giving flights required times of arrival to meet at or upstream of the TBFM freeze horizon; and to what extent demand should be strategically set above forecast capacity in order to compensate for uncertainties, maintaining throughput.From this list of parametric studies, this paper has two objectives: 1. Explore the impact of varying levels of airline participation in CTOP on ground delay, and 2. Explore the impact of departure uncertainty on TBFM delays.These studies were chosen as they have been addressed to a limited extent on the HITL platform.The studies presented here are meant to augment the results found in those HITL studies and highlight the benefit of using TMIAutoSim in conjunction with HITL experiments to gain greater insight into the operational feasibility of the IDM concept. +IV. Sensitivity to Airline Participation in CTOPIn this section, experiments are described that estimate the potential benefits, both in terms of overall system efficiency and airline/flight specific benefits of different levels of CTOP participation by airlines.Airlines are considered to be participating in CTOP if they provide TOSs with more than one trajectory option per flight.The potential benefits are studied with a limited number of HITL simulations in Ref. 9. In these HITL simulations, the portion of airline participation was varied from none up to 100% participation.The results indicate that nearly all of the overall system benefit can be achieved with only 50% CTOP participation. 12However, these HITL results are limited to the specific cases that were tested and their generalizability is unknown.Because TMIAutoSim does not require participant input during execution, a greater number of simulations can be run, allowing the analysis to be extended to generate statistically significant results which have improved generalizability.In this paper, benefits are estimated across a range of CTOP participation levels both for airlines that participate in CTOP and those that do not.However, the experiments presented also aim to explore the CTOP participation levels required to make a CTOP TMI effective at reducing average ground delay below an acceptable threshold, given the capacity constraint.Alternatively, given a CTOP participation level, these experiments aim to identify the capacity constraint values for which CTOP would be an effective choice of TMI. +A. Experiment DetailsA common traffic scenario, describing origin airports, flight plans, and scheduled departure times for flights destined for Newark Liberty International Airport (EWR) was used for all results presented in the paper.This scenario consists of 195 flights, departing 66 origin airports.A total of 42 flights are airborne at the start of the simulation, of which 10 are international flights.All pre-departure flights are scheduled to depart within 4.5 hours of the start of the scenario.Flight times for flights that have not yet departed at the start of the simulation range from 30 minutes to over 4.5 hours.CTOP is used to schedule departures to meet arrival rate constraints at FCAs set at each of the three arrival fixes for EWR, as shown in Figure 1.A total of 39% of flights are originally filed through the West fix, 24% through the North fix, and 37% through the South fix.The arrival rate constraints at these arrival fixes are varied.This experiment only involves the CTOP solution to the problem and thus MACS simulation of air traffic and simulated TBFM metering and scheduling were not required.Airlines simulated to participate in CTOP specify a set of acceptable trajectories for each flight, in the form of a TOS, along with RTCs, which indicate airline trajectory preferences.TOSs were generated, as in Ref. 9, using a combination of inputs from subject matter experts, and clustering of 3 months of historical flight plan data from June to August 2015.Hierarchical clustering was used, with the dissimilarity between different routes described by Euclidean distance. 10Outlier detection was used prior to clustering to remove extreme routes, before the appropriate number of clusters was chosen by comparing average silhouette scores. 11Example TOSs are shown in Figure 2. All TOS routes went through one of the three meter fix FCAs.RTCs were calculated based on differences in flight time between each trajectory option, and the most preferred route, with an assumption of flight delay being 1.5 times the cost of ground delay, as in Ref. 12.Airlines that do not participate in CTOP were still included in the program in that they were simulated to submit a single-option TOS (i.e., a TOS that consists of only their preferred route).Flights with a single-option TOS were only assigned ground delay by CTOP and were not rerouted.The experimental setup used in these studies differ from that used in the HITL simulation presented in Refs.12, and 9.As in Ref. 9, the North and South flows controlled by the arrival meter fix FCAs at the North and South arrival fixes, were set at clear-weather capacity (7 and 6 aircraft per quarter hour, respectively).However, the FCA flow rate for the West flow, controlled by the arrival meter fix FCA at the West arrival fix, was varied from 2 aircraft per quarter hour (40% of average demand) to a clear weather capacity of 5 aircraft per quarter hour (100% of average demand) (in Ref. 12 the FCA flow rate for the West flow is fixed at 3 aircraft per quarter hour).Furthermore, in contrast to Ref. 12, where an arrival rate of 11 aircraft per quarter hour was set for the arrival runway, the runway was left unconstrained here.This was done so that the effect of the constraints on the arrival fixes, and particularly the West Flow, could be isolated, without complicating impacts from the runway constraint.It is also important that only one flow (the West flow) is constrained, to ensure that trajectory options exist to other arrival fixes, on which there is available capacity.For each capacity constraint, nCTOP was run with CTOP participation varied from 1% to 72% (because some flights were considered exempt, TOS options were only calculated for 72% of flights).Benefits are measured using average delay for flights on the West flow, for aircraft submitting and not submitting TOSs. +B. ResultsFigure 3 shows results for the case in which the West flow FCA capacity was limited to 60% of average demand, while the North and South flows were at clear-weather capacity.As the percentage of CTOP participation increases, average ground delays (blue curve) assigned to aircraft going through the FCA decreases.However, as the percentage CTOP participation increases to 40%, further decrease in delay cost is minimal.The orange curve in Figure 3 shows the average RTC of aircraft that are routed out (i.e., given a trajectory option that is different from their preferred route, because the ground delay associated with their preferred route exceeded the specified RTC).This cost depends primarily on the RTCs associated with the alternate trajectories and is mostly independent of the percent CTOP participation.The difference between the blue and orange curves indicates the relative benefit, in terms delay minutes, of filing a TOS -airlines not filing a TOS would, on average, receive the ground delay shown (blue), whereas airlines filing a TOS would, on average, receive delay that is equivalent to the RTC delay cost (orange), because any ground delay greater than the RTC would result in the flight being routed out, incurring extra flight time of similar cost to the RTC. Figure 3 indicates that the benefit TOS filers have relative to non-TOS filers can be as much as 60 minutes, on average, in situations in which the percentage of TOS filers is small.This relative benefit decreases as the percentage of TOS filers increases.When the percentage of TOS filers is greater than 40%, TOS filers as a group have only a small benefit over those not filing TOSs (in the order of 10 minutes), although there is still a benefit.Thus, as the percentage of CTOP participation increases, overall system performance improves while relative benefits to TOS filers declines.The percentage of CTOP participation at which delay cost plateaus for TOS filers was further investigated by varying the FCA capacity constraint applied to the West flow.We define capacity deficit to be the percentage difference between the available capacity and demand, i.e. capacity demand deficit is 100 × $ %&'()%*+(,(+-./%&'()% 0. Figure 4 shows that the impact of CTOP participation on ground delays depends on the degree of capacity demand deficit.In all scenarios with non-zero capacity deficit, ground delays decrease as the percentage of CTOP participation increases.The larger the capacity deficit, the higher the CTOP participation beyond which there is no benefit to increased CTOP participation -at a 60% capacity deficit, there is still benefit to increasing CTOP participation, but at 0% or 20% capacity deficit, there is little benefit to increasing CTOP participation beyond 20% or 30%, respectively.This is expected because after the percentage of CTOP participation is large enough to make up the capacity demand imbalance, further increase in the percentage of CTOP participation only results in a small decrease in average delays.According to feedback from subject matter experts, if average West flow delay is over an hour, stakeholders are likely to consider alternative TMIs such as required advisory reroutes.By defining a 1-hour threshold of delays as acceptable, our analysis can be used to identify the minimum CTOP participation required for CTOP to be a suitable TMI option.As seen in Figure 4, what constitutes minimum required CTOP participation depends on the degree of capacity deficit.Given the likely level of CTOP participation, one can determine the type of capacity deficit scenarios in which CTOP use would be appropriate.Thus, in Figure 4, if acceptable delays are as indicated (1 hour), 30% CTOP participation may mean that CTOP can be used effectively in scenarios with capacity deficit of 40% or less.These delays would depend on other factors such as duration of the CTOP TMI, as well as the specific scenario simulated, so further analysis is required to confirm this result. +V. Departure Error SensitivityDeparture uncertainty is one of the major sources of uncertainty in the TBFM system.While the exact departure time affects the estimated arrival time at the TBFM freeze horizon, it also affects the order in which flights reach the freeze horizon and thus the order in which they are scheduled to reach the runway threshold.Slight changes in this order can lead to significant changes in delay assigned to particular aircraft.The time of departure for each flight is assigned by CTOP as an EDCT.For internal departures, a call-for-release time is then assigned by TBFM, delaying the flight beyond its EDCT.A flight is considered to be EDCT compliant if it departs within plus or minus five minutes of its EDCT, and call-for-release compliant if it departs between two minutes before and one minute after its call-forrelease time.However, in today's operations, many flights depart outside these compliance limits. 12In this section we examine the impact of departure error on assigned TBFM delay.This study is divided into three parts.After describing some scenario details that are common across all parts, we look at the repeatability of results using TMIAutoSim in Section V-A.In section V-B we then examine the differences in TBFM delay assigned as the mean of the departure error shifts.Finally, we study the impact of changes in EDCT and call-for-release compliance variability on TBFM delay in Section V-C.While total delay is a good metric to understand the overall efficiency of the scheduling algorithms, operationally it is more important to understand how individual flight delays are distributed in the simulation.Air traffic controllers can relatively easily control flights to absorb a small amount of delay in the air, without turning off the TBFM system or putting flights into airborne holding.The problem becomes progressively more difficult as the amount of delay that must be absorbed increases.Based on subject matter expert feedback for the airspace simulated, TBFM airborne delay is considered operationally acceptable if it is below 7 minutes, marginal if it is between 7 and 14 minutes, and unacceptable if it is greater than 14 minutes.The traffic scenario presented in Section IV-A is used for all experiments in this section.However, instead of FCAs at each meter fix, in these experiments a single FCA is applied at the runway threshold, with a rate of 11 aircraft per quarter hour, representing single runway operations on runway 22L.Given the initial scheduled departure times, the arrival rate at EWR would have been as shown in Figure 5, labeled 'Baseline'.EDCTs for pre-departure flights were calculated using nCTOP in order to meet the runway arrival rate constraint (labelled 'AAR' in Figure 5).Because some flights are airborne at the start of the simulation, and others are exempted from CTOP, the resulting rate (labelled 'CTOP' in Figure 5) does not meet the desired rate precisely.It is therefore necessary for TBFM to assign some amount of airborne delay in order to achieve the desired arrival rate at EWR.This is done by simulating flights in MACS, before using the TBFM emulator to simulate TBFM operations in post-processing.This applied additional inter-arrival spacing at the final approach fix of 0.4 nmi, which achieves as close to 11 aircraft per quarter hour as possible.The TBFM freeze horizon was set at approximately 400 nmi from the EWR runway threshold, as in current TBFM operations (shown in Figure 1).Aircraft departing from airports close to EWR (roughly within the freeze horizon) are considered internal departures and have their scheduled times of arrival at the runway threshold frozen as they depart. +A. Simulation RepeatabilityAs shown in Ref. 3, running the simulation multiple times with identical at real-time speed does not produce precisely identical results -total flight time varies by up to 5 seconds.In order to distinguish between the impact of departure error versus variations in simulation performance, experiments were performed to quantify the variations in TBFM delay caused by slight changes in simulation output over several executions of the scenario, with each run simulating exactly the same set of departure error values. +Experiment DetailsThe common traffic scenario was executed using the EDCTs generated by nCTOP.It was run five times with no departure error and five times with a fixed set of departure errors (i.e., the same departure errors were used in each of the 5 runs).This set of departure errors was extracted from empirical departure error distributions generated from observed departure errors from historical Ground Delay Programs and periods when TBFM internal departure scheduling was active, from June to August 2010.In these distributions, external flights had a mean departure error of 1.1 minute early and standard deviation of 14.5 minutes, while internal departures had a mean departure error of 0.6 minutes early, and standard deviation of 5.7 minutes.This set of departure errors has been used in previous work, 1 ,2,3,12 and is used here for consistency.Both TBFM scheduling paradigms (i.e., prioritizing airborne flights and prioritizing internal departures) were applied in post processing on all ten of these simulation results. +ResultsWhen the simulation is run using EDCTs from nCTOP and no departure error, TBFM must impose a total of nearly 2,000 minutes of delay (airborne and ground), on average across the 5 runs, in order to meet the runway arrival rate constraints.The introduction of departure error results in an increase in the amount of airborne delay required to meet runway arrival constraints.When airborne flights are prioritized, on average, 285 additional minutes of airborne delay are required across all flights when departure error is introduced (186 minutes when internal departures are prioritized).The standard deviations of total airborne delay assigned across all simulations for each scheduling paradigm, with and without departure error, ranges from 9.5 to 23.4 minutes.This indicates that errors due to simulation repeatability are small in comparison to total system delay differences caused by departure errors on the order of minutes.TBFM airborne delay per flight for the paradigms in which priority is given to airborne flights and priority is given to internal departures is shown in Figure 6 and Figure 7, respectively.The bars indicate the average number of flights assigned a given delay value.The black lines at the top of the bar indicate the standard deviation of the number of flights assigned delay in that one-minute range over the five executions of the same scenario.Standard deviations are up to 1.2 aircraft when prioritizing airborne flights, and 0.6 aircraft when prioritizing internal departures.While not insignificant, the variability can be seen to be generally small in comparison to the numbers of flights plotted in each bin, indicating that the simulation repeatability is acceptable.Note that, while airborne delay increases when priority is given to internal departures, it remains below unacceptable levels and internal departure ground delay decreases significantly (shown in Refs. 1, 2 and 3) making it a key enabler of the IDM concept.Figure 7. Priority given to internal departures. +B. Impact of Departure Error Offset on TBFM DelayAs demonstrated in Ref. 3 and shown in Figure 6 and Figure 7, a greater portion of assigned delay is categorized as marginal or unacceptable when internal departures are prioritized.Figure 6 and Figure 7 also show that, in both cases, delay is shifted from acceptable to marginal, and even unacceptable, when departure error is introduced.This indicates that, under the capacity constraints simulated, significant departure error may make the use of TBFM with priority given to internal departures, and even TBFM for arrival metering generally, impractical.The level of acceptable departure error that would allow feasible TBFM metering is considered in the following sections, first in terms of mean error (a fixed offset) in this section and then error variability in Section V-C. +Experiment DetailsFive different experimental runs were completed to explore the impact of departure error mean on TBFM delay.The departures errors simulated were based on those used in previous work, 1,2,3,9 but with the distribution mean shifted by a constant offset of -10, -5, 0, 5 and 10 minutes, for each of the 5 runs, respectively.After the offset constant was applied to the departure errors, the resulting departure errors for some flights were truncated to ensure that flights scheduled to depart early in the scenario would not depart before the start of the simulation.For the baseline distribution, with no shift to the mean departure error, 87% of flights were EDCT compliant with departure error within five minutes of the assigned EDCT.With a shift in departure error of +/-5 minutes, just over 58% of flights were EDCT compliant.Roughly 30% of flights were EDCT compliant after a departure error shift of +/-10 minutes.Both TBFM scheduling paradigms (i.e., prioritizing airborne flights and prioritizing internal departures) were applied in post-processing on all five of simulation outputs from MACS. +ResultsThe resulting required TBFM airborne delay for each run is shown in Figure 8 and Figure 9.As the mean of the departure errors is shifted away from zero, marginal and unacceptable delays increase, as expected.The data labeled "No error" and "Error with 0 min offset" are results from the repeatability study for the cases in which no departure error is injected into the scenario, and in which a fixed set of departure errors is injected into the scenario, respectively.Looking at the results for the TBFM scheduling paradigm that prioritizes internal departures in Figure 9, the number of flights assigned marginal delay is quite high, even at +/-5 minutes shift in departure error (i.e., 2.3 times and 1.8 times the number of flights given marginal offset when no departure error was assigned, respectively).At +/-10 minutes shift in departure error, unacceptable delay spikes to 16 and 33 flights, respectively.In an operational setting, controllers would likely discontinue the use of TBFM under the scheduling paradigm in which internal departures are prioritized at just +/-5-minute shift in departure errors and prioritize airborne flights in order to reduce the amount of delay that must be absorbed in the air.This, in turn, would push more of the required delay to ground delay for internal departures.Airborne delay is reasonable using the paradigm in which airborne flights are prioritized up to 5-minute error shifts.When also considering TBFM ground delay (not shown in the figures), with priority given to airborne flights the total TBFM delays are highest under +10-minute mean departure error, and lowest under -5-minute mean departure error.This indicates that, given uncertainty in EDCT compliance, allowing flights to depart before their assigned EDCT could result in lower total TBFM delay.This effect is not present when priority is given to internal departures.In this case, total TBFM delays are lowest under zero mean departure error.This suggests that, with priority given to internal departures, it is important that both early and late departures are minimized, and not only late. +C. Impact of Departure Error Variability on TBFM Delay Experiment DetailsGaussian mixture distributions were fitted to the previously generated empirical distributions of departure error, 1,2,3,9 for external flights and internal departures separately.Two-component Gaussian mixtures were used rather than simple Gaussian distributions because the mixture distributions were found to better fit the empirical distributions.Experiments were conducted to explore the impact of departure error variability, in the form of standard deviation in departure error, on TBFM delay.The departure error distributions for external and internal departures that were derived from the empirical data were modified to simulate improvement and degradation in the conformance to EDCT and call-for-release compliance to study the impact on TBFM assigned delay.Figure 10 and Figure 11 show these adjusted distributions.These distributions were generated by multiplying the standard deviations of the original distributions, represented by s0, by 0.5, 1.0, 2.0, 3.0 and 4.0.For these distributions, the percentage of external flights departing with errors outside ±5 minutes is 31%, 48%, 65%, 74% and 80%, respectively.The percentage of internal flights departing with errors outside -2 to +1 minutes is 9%, 30%, 58%, 72% and 78%, respectively.When samples were drawn, they were drawn for external and internal departures separately, but from the corresponding distribution (e.g., if external departure errors were sampled from "3.0 s 0", internal departure errors were sampled from "3.0 s0").Simulations were performed using departure errors sampled from each of these distributions for external and internal departures, respectively.When sampling from these distributions to generate departure errors for the various experiments, departure errors greater than 60 minutes were discarded and a new error was sampled from the distribution.Five sample sets were drawn from each of the five distribution pairs, for a total of 25 simulation runs.Additional departure errors were generated by increasing these sampled departure errors by 5 minutes, for a further 25 simulation runs with a 5-minute offset and a total of 50 experiments. +ResultsBoth TBFM scheduling paradigms (i.e., prioritizing airborne flights and prioritizing internal departures) were applied in post-processing on all thirty of the simulation runs from MACS.Results for zero-offset departure error are shown in Figure 12 and Figure 13, with corresponding results for departure error offset by 5 minutes shown in Figure 14 and Figure 15. Figure 12 and Figure 14 show results for the paradigm in which priority is given to airborne flights, while Figure 13 and Figure 15 show results for the paradigm in which priority is given to internal departures.Looking at TBFM airborne delay for the case in which departure error offset is zero minutes (Figure 12 and Figure 13) there is a general trend of delay shifting towards marginal values as the standard deviation of the departure error increases -the number of flights with acceptable delay decreases, while the number with marginal delay increases for 2.0s0, 3.0s0 and 4.0s0 over the relatively lower number of flights given marginal delay at 0.5s0 and 1.0s0.This is true both for the scheduling paradigm in which airborne flights are prioritized (Figure 12) and in the scheduling paradigm which internal departures are prioritized (Figure 13).In the paradigm in which airborne flights are prioritized (Figure 12), relatively few aircraft are assigned marginal delay -31.6 flights on average for a standard deviation in departure error of 4.0s0.However, in the case in which internal departures are prioritized (Figure 13), the marginal delays affect 68.8 flights on average for a standard deviation in departure error of 4.0s0.Under the scheduling paradigm in which airborne flights are prioritized (Figure 12), the number of flights assigned marginal delay stays relatively constant with changes in error variability.This is because required delay is absorbed by internal departures on the ground.Under the scheduling paradigm in which internal departures are prioritized (Figure 13), the increased number of flights assigned marginal delay is more pronounced.This suggests that, in the conditions simulated, high variability in departure error may indeed limit the feasibility of prioritizing internal departure scheduling in TBFM.When departure errors are increased by a constant 5-minute offset (Figure 14 and Figure 15), the number of flights with acceptable delay, across all runs, decreases relative to the 0-departure error offset results (Figure 12 and Figure 13).This trend is also shown in Figure 6 and Figure 7 above.Correspondingly, the number of flights with marginal and unacceptable delay increase, across all runs.When airborne flights are prioritized the number of flights incurring acceptable, marginal and unacceptable delay, respectively, remains relatively constant over variations in the standard deviation of departure error.However, when internal departures are prioritized, the number of flights incurring acceptable delay decreases as the departure error standard deviation increases.The total decrease in the number of flights given acceptable delay is 25.8, on average, as the variability in departure error increases from 0.5s0 to 4.0s0.There is a corresponding increase in flights with both marginal (by 17.4 flights) and unacceptable (by 8.4 flights) delay.These results further suggest that, in the conditions simulated, high variability in departure error may indeed limit the feasibility of prioritizing internal departure scheduling in TBFM.However, comparing to the results in Figure 12 and Figure 13, these results suggest that the impact of offset errors on airborne delay may dominate over the impact of error variability. +VI. Conclusions and Future WorkWhile there is a wide range of parametric studies that could be conducted in order to gain greater insight into the management of demand under capacity constrained resources using TMIs, two of them were addressed here, namely: one to quantify the impact of varying levels of airline participation in the user negotiation of route options, presented in Section IV, and a second to quantify the impact of departure uncertainty on tactical TBFM delays, presented in Section V. TMIAutoSim, a simulation tool developed to support parametric studies and existing HITL capabilities, was used to conduct these studies.The airline participation study may help incentivize airlines to participate in CTOP.The results of the departure error studies can be used to either support EDCT and call-for-release conformance ranges or be used to determine when to switch between TBFM scheduling paradigms that prioritize internal departures or airborne flights.Experimental results suggest that, for the scenario studied, as CTOP participation increases in the form of airlines submitting TOSs with multiple trajectory options (as opposed to a single option TOS), average ground delays assigned to aircraft going through the FCA decreases.This supports the findings of Ref. 12 that there are significant benefits to airlines by participating in CTOP.A threshold in CTOP participation, which varies with the constraint capacity, is also identified beyond which there is relatively little further reduction in average ground delays.Given the likely level of CTOP participation, the capacity reduction for which CTOP would be an appropriate TMI, resulting in acceptable levels of delay, is also identified.Experimental results from the paper suggest that high average departure errors (either positive or negative), and high variability in departure error, can make the prioritization of internal departures in TBFM departure metering and scheduling infeasible.Results presented here show that average departure errors of +/-10 minutes causes marginal airborne delay for 31.6 and 68.8 flights, respectively, when internal departures are prioritized.When departure errors of these magnitudes are observed, a switch from prioritizing internal departures to prioritizing airborne flights would decrease unacceptable airborne delay.Alternatively, these results could be used to justify requiring average departure delays to be within +/-5 minutes to ensure the feasibility of TBFM scheduling which prioritized internal departures.Further work will extend the capabilities of TMIAutoSim.This will include, particularly, (1) to simulate the absorption of TBFM airborne delay using speed control, and (2) to simulate TBFM extended metering, which includes multiple uncoupled schedules for each flow, to a series of metering points, each with its own freeze horizon.Extended metering is another capability that forms a key component of the IDM concept, and is applied in the ongoing HITL simulations supporting IDM.Automatic simulation of TBFM airborne delay absorption would allow aircraft to meet scheduled times of arrival specified by TBFM, which is required for simulating extended metering, or other constraints, such as meet miles-in-trail requirements.The latter would allow for the simulation of current day operations, allowing benefits of the IDM concept to be quantified across a range of traffic and capacity scenarios.TMIAutoSim will also be developed to enable fast-time simulation, which would contribute to studying a wider range of parameters than is currently possible in real-time.Figure 1 .1Figure 1.Location of FCAs at arrival meter fixes (North, South, and West) and/or at the EWR runway threshold and TBFM freeze horizon. +Figure 2 .2Figure 2. Example TOS, for flights from Chicago O'Hare (KORD), San Francisco (KSFO), Dallas-Fort Worth (KDFW), and Atlanta Hartsfield (KATL) International Airports, to Newark Liberty International Airport (KEWR). +Figure 3 .3Figure 3. Delay cost as a function of CTOP participation, with West flow FCA capacity 60% of average demand (capacity deficit of 40%). +Figure 4 .4Figure 4. Average delay as a function of CTOP participation under various West flow FCA capacity constraints. +Figure 5 .5Figure 5. Baseline demand indicates arrival demand resulting from original schedule with no departure delay.CTOP demand indicates demand given CTOP EDCTs with no departure delay.Airport Arrival Rate (AAR) is the desired arrival rate. +Figure 6 .6Figure 6.Priority given to airborne flights.Figure7.Priority given to internal departures. +Figure 8 Figure 989Figure 8 Airborne delay assigned by TBFM using the scheduling paradigm in which airborne flights are priortized.Flights depart with no departure error and fixed depararture error shifted by a variety of offsets.Figure 9 Airborne delay assigned by TBFM using the scheduling paradigm in which internal departures are priortized.Flights depart with no departure error and fixed depararture error shifted by a variety of offsets. +Figure 10 .10Figure 10.Departure error distributions for external departures modified from the distribution fit to empirical data. +Figure 11 .11Figure 11.Departure error distributions for internal departures modified from the distribution fit to empirical data. +Figure 12 .12Figure 12.Airborne delay assigned by TBFM using the scheduling paradigm in which airborne flights are priortized.Flights departed with error sampled from distributions given in Figure 10 and Figure 11, and means of 0 minutes. +Figure 13 .13Figure 13.Airborne delay assigned by TBFM using the scheduling paradigm in which internal departures are priortized.Flights departed with error sampled from distributions given in Figure 10 and Figure 11, and means of 0 minutes. +Figure 14 .14Figure 14.Airborne delay assigned by TBFM using the scheduling paradigm in which airborne flights are priortized.Flights departed with error sampled from distributions given in Figure 10 and Figure 11, and offset by 5 minutes. +Figure 15 .15Figure 15.Airborne delay assigned by TBFM using the scheduling paradigm in which internal departures are priortized.Flights departed with error sampled from distributions in Figure 10 and Figure 11, and offset by 5 minutes. + + + +AcceptableMarginal Unacceptable eTBFM Airborne Delay Ranges + + + + + + + + NSmith + + + CBrasil + + + PULee + + + NBuckley + + + CGabriel + + + Mohlenbrink + + + FOmar + + + BParke + + + CSperidakos + + + HYoo + + Integrated Demand Management: Coordinating Strategic and Tactical Flow Scheduling Operations, 16th AIAA Aviation Technology, Integration, and Operations Conference + + 2016 + + + + Smith, N., Brasil, C., Lee, P.U., Buckley, N., Gabriel, C., Mohlenbrink., C, Omar, F., Parke., B., Speridakos, C., and Yoo, H., Integrated Demand Management: Coordinating Strategic and Tactical Flow Scheduling Operations, 16th AIAA Aviation Technology, Integration, and Operations Conference, 2016, AIAA 2016-4221. + + + + + + HSYoo + + + CBrasil + + + NBuckley + + + CMohlenbrink + + + CSperidakos + + + BParke + + + GHodell + + + PULee + + + NMSmith + + AIAA 2017-4100 + Integrated Demand Management: Minimizing Unanticipated Excessive Departure Delay while Ensuring Fairness from a Traffic Management Initiative, 17th AIAA Aviation Technology, Integration, and Operations Conference + + 2017 + + + Yoo, H.S., Brasil, C., Buckley, N., Mohlenbrink, C., Speridakos, C., Parke, B., Hodell, G., Lee, P.U., and Smith, N.M., Integrated Demand Management: Minimizing Unanticipated Excessive Departure Delay while Ensuring Fairness from a Traffic Management Initiative, 17th AIAA Aviation Technology, Integration, and Operations Conference, 2017, AIAA 2017-4100. + + + + + Development and validation of an automated simulation capability in support of Integrated Demand Management + + HArneson + + + ADEvans + + + JLi + + + MYWei + + + + Royal Aeronautical Society Flight Simulation Conference + + AIAA + November 2017 + + + Arneson, H., Evans, A. D., Li, J., Wei, M.Y., Development and validation of an automated simulation capability in support of Integrated Demand Management, Royal Aeronautical Society Flight Simulation Conference, November 2017, RAeS, AIAA. + + + + + Operating in a CTOP (Collaborative Trajectory Options Program) Environment + + PSmith + + + EStellings + + + + CDM Flow Evaluation Team, Tech. Rep + + 2014 + + + P. Smith, E. Stellings, "Operating in a CTOP (Collaborative Trajectory Options Program) Environment", CDM Flow Evaluation Team, Tech. Rep., 2014. + + + + + TFMS Functional Description, Appendix C: Traffic Management Initiative (TMI) Algorithms + CSC/TFMM-13/1744 + + 2014 + + + Tech. Rep + "TFMS Functional Description, Appendix C: Traffic Management Initiative (TMI) Algorithms", CSC/TFMM-13/1744, Tech. Rep., 2014. + + + + + + RTyo + + + CPressler + + + Industry Day Briefing: Time-Based Flow Management (TBFM) + + April, 2014. June 12, 2017 + + + Federal Aviation Administration Industry Day + Tyo, R., and Pressler, C., Industry Day Briefing: Time-Based Flow Management (TBFM), Federal Aviation Administration Industry Day, April, 2014. from: http://www.faa.gov/about/office_org/headquarters_offices/ato/service_units/pmo/industry_day/media/day_1_-_tbfm_bob_tyo_- _chris_pressler_-_final.pdf [Retrieved June 12, 2017]. + + + + + Optimized Route Capability (ORC) Intelligent Offloading of Congested Arrival Routes + + SZelinski + + + MXue + + + PBassett + + + + 16th AIAA Aviation Technology, Integration, and Operations Conference + + 2016 + + + + 7 Zelinski, S., Xue, M., and Bassett, P., Optimized Route Capability (ORC) Intelligent Offloading of Congested Arrival Routes, 16th AIAA Aviation Technology, Integration, and Operations Conference, 2016, AIAA 2016-4357. + + + + + Exploring the many perspectives of distributed air traffic management: The Multi Aircraft Control System MACS + + TPrevot + + + + Proceedings of the HCI-Aero + the HCI-Aero + + 2002 + + + + Prevot, T., Exploring the many perspectives of distributed air traffic management: The Multi Aircraft Control System MACS, In Proceedings of the HCI-Aero, 2002, pp. 149-154. + + + + + Impact of Different Trajectory Option Set Participation Levels within an Air Traffic Management Collaborative Trajectory Option Program + + Hyo-SangYoo + + + ConnieLBrasil + + + NathanBuckley + + + GitaSHodell + + + ScottNKalush + + + PaulULee + + + NancyMSmith + + + + 18th AIAA Aviation Technology, Integration, and Operations Conference + + + 9 Hyo-Sang Yoo, Connie L. Brasil, Nathan Buckley, Gita S. Hodell, Scott N. Kalush, Paul U. Lee, Nancy M. Smith. "Impact of Different Trajectory Option Set Participation Levels within an Air Traffic Management Collaborative Trajectory Option Program." In 18th AIAA Aviation Technology, Integration, and Operations Conference. + + + + + Trajectory clustering and classification for characterization of air traffic flows + + CondeRocha Murca + + + MDelaura + + + RHansman + + + RJJordan + + + RReynolds + + + TBalakrishnan + + + H + + + + 16th AIAA Aviation Technology, Integration, and Operations Conference + + 2016 + 3760 + + + Conde Rocha Murca, M., DeLaura, R., Hansman, R.J., Jordan, R., Reynolds, T. and Balakrishnan, H., 2016. Trajectory clustering and classification for characterization of air traffic flows. In 16th AIAA Aviation Technology, Integration, and Operations Conference (p. 3760). + + + + + Automated route clustering for air traffic modeling + + ABombelli + + + ASegarra Torne + + + ETrumbauer + + + KDMease + + + + AIAA Modeling and Simulation Technologies Conference + + 2017 + 1318 + + + Bombelli, A., Segarra Torne, A., Trumbauer, E. and Mease, K.D., 2017. Automated route clustering for air traffic modeling. In AIAA Modeling and Simulation Technologies Conference (p. 1318). + + + + + Required time of arrival as a control mechanism to mitigate uncertainty in arrival traffic demand management, 35 th IEEE/AIAA Digital Avionics Systems Conference + + HSYoo + + + CMohlenbrink + + + CBrasil + + + NBuckley + + + AGlobus + + + NMSmith + + + PULee + + + 2016 + IEEE + + + Yoo, H.S., Mohlenbrink, C., Brasil, C., Buckley, N., Globus, A., Smith, N.M. and Lee, P.U., Required time of arrival as a control mechanism to mitigate uncertainty in arrival traffic demand management, 35 th IEEE/AIAA Digital Avionics Systems Conference, 2016, IEEE. + + + + + + diff --git a/file21.txt b/file21.txt new file mode 100644 index 0000000000000000000000000000000000000000..3c57d9e66fe3015c94bb27d0b7f4539daee0dfb8 --- /dev/null +++ b/file21.txt @@ -0,0 +1,339 @@ + + + + +I. Introductionntegrated Demand Management (IDM) is a near-to mid-term NASA concept that is intended to contribute to future user-negotiated routing of aircraft to enable the flexible management of aircraft and National Airspace System (NAS) resources based on reliable predictions of future states. 1 IDM will develop more powerful, integrated operations and tools for managing flight trajectory constraints by leveraging existing systems as well as new automation tools and methods.A central component of IDM is the user-negotiated routing of aircraft, which plays an increasingly important and effective role in the flexible management of aircraft and NAS resources based on reliable predictions of future states.Flight delays result when the capacity of critical resources, such as airspace or airports, is insufficient to meet demand.Better matching between demand and available capacity can lead to efficiency improvements, with increased throughput, reduced delays and more efficient flight trajectories.However, because of uncertainties in both demand and capacity (e.g., due to weather), such matching can be difficult.IDM proposes to address mismatches in demand and capacity by using strategic flow management capabilities with the Traffic Flow Management System (TFMS) toolset to pre-condition demand into the more tactical Time-Based Flow Management System (TBFM), which would enable TBFM to better manage delivery of demand to capacity constrained resources.A number of human-in-the-loop (HITL) simulation studies have been carried out with experienced air traffic manager and controller participants in the NASA Ames Airspace Operations Laboratory (AOL) to verify and develop the IDM concept. 1,2While these experiments have produced important results, the resources and time required to conduct real-time studies with human participants limits the number of experimental conditions that can be tested as well as the overall number of HITL simulations that can be conducted.IDM concept development would benefit from the evaluation of a wider variety of capabilities and concepts across a broader range of possible conditions.An automated simulation capability that can closely mimic the HITL simulation capabilities, automate both the human I components and the collaboration between operational systems, and speed up the real-time aircraft simulations could thus provide real benefit.Such a capability would allow for parametric studies to be conducted that complement the HITL simulations, for example, by identifying breaking points and parameter values at which significant changes in system behavior occur.Such a capability could also be used to simulate background traffic during HITL simulations, improving realism.NASA is also developing the Shadow Mode Assessment using Realistic Technologies for the NAS (SMART-NAS) Test Bed (SNTB)a platform that is designed to enable high-fidelity HITL and automation-in-the-loop simulations by allowing operational systems to be integrated with high-fidelity models. 3,4The SNTB is envisioned to leverage cloud-based services to provide cost-effective scalability for large, multi-facility, multi-organization simulations, and will ultimately enable complex, gate-to-gate HITL experiments for concepts such as IDM.Therefore, the automated simulation capability will be built to fit within the SNTB framework.This paper describes the development of such a capability, its initial validation, and an example application, the results of which are compared to results from similar HITL simulations. +II. BackgroundIn Traffic Flow Management (TFM), demand for capacity-constrained airspace or airports is controlled by a series of Traffic Management Initiatives (TMIs), which use departure delays and airborne delays, as well as pre-departure and airborne reroutes, to manage access to the constrained resources.Two key TFM systems that exist in both current and planned future operations are the TFMS and TBFM systems, both of which are described below.Both systems address imbalances between demand and capacity, but in different operational contexts and different timeframes: whereas TFMS is more strategic in nature, with a planning horizon of hours and geographical reach that spans across the NAS, TBFM is more locally targeted to terminal airspace surrounding airports, and has a shorter, more tactical time horizon.The capabilities comprising these two systems are very powerful, but address different parts of the capacity management problem without close coordination.IDM aims to integrate and coordinate demand management across these systems to work together more effectively. 1Within the IDM concept framework, TFMS may redistribute demand from an overloaded resource to one with reserve capacity, or allocate delay so that it is calibrated to variations in demand, and better matched to the demand that TBFM can accommodate.This should allow TBFM to manage delivery to the capacity-constrained destination more efficiently.A short overview of TFMS, TBFM and IDM is given below.A more detailed description of these concepts is available in Ref. 1. +A. Traffic Flow Management System (TFMS)TFMS is a set of planning tools and capabilities that support NAS-wide traffic flow management and monitoring.In current operations, this is achieved primarily through Ground Delay Programs (GDPs) and Airspace Flow Programs (AFPs).GDPs and AFPs are TMIs put in place by the Air Traffic Control System Command Center (ATCSCC) when the capacity at a particular airport or airspace, respectively, is expected to fall well below the expected demand.These programs delay the departure of flights destined for the affected airport or airspace to reduce demand to manageable levels.Departure delay is assigned through TFMS at least 45 minutes before departure and often many hours before the flight will arrive at the constrained airport or airspace.Operators may choose to re-route flights around the constraints associated with an AFP, but their options are often restricted to a re-route advisory specified by the ATCSCC.The IDM concept utilizes a new TMI called the Collaborative Trajectory Options Program (CTOP).New CTOPspecific capabilities within TFMS support a more complex characterization of the reduced capacity problem using multiple constraints, called Flow Constrained Areas (FCAs).Like GDPs and AFPs, CTOP is initiated by the ATCSCC when the capacity at a particular airport or airspace is expected to fall well below expected demand.When airspace is constrained, CTOP allows for greater flexibility than an AFP by allowing airlines to route their flights out of the associated FCA, or across a different FCA that has capacity available.CTOP allows flight operators to submit a preference-weighted set of alternative routes called a Trajectory Option Set (TOS) from which the program can select.CTOP distributes traffic to minimize congestion and reduce demand to manageable levels by delaying departures and allocating trajectories from the TOS to flights scheduled to pass through the specified FCAs.Departure delay is assigned via the TFMS's Expect Departure Clearance Time (EDCT).Like GDPs and AFPs, CTOP provides operators with advance notice of constraints and some opportunity to "swap slots" (exchange assigned arrival times) among their flights to reduce the impact on their operations.Although the objective of these TFMS tools is to set departure times and routes such that maximum rates at capacity constrained resources are not exceeded, several factors impact a flight's actual arrival time at the resource.For example, EDCTs may not be precisely met.A flight is considered to be EDCT compliant if it departs within +/-5 minutes of the EDCT, but in today's operations, many flights depart outside of EDCT compliance limits. 6dditionally, winds and changes in trajectory due to weather conditions will affect flight time.Thus, using only TFMS tools, there is no guarantee that resource constraints will be satisfied. +B. Time-Based Flow Management (TBFM)As described by Refs. 1 and 5, TBFM is an arrival scheduling tool that was developed to improve traffic delivery into capacity-constrained airports or airspace resources.TBFM creates a schedule for all inbound flights which provides a Scheduled Time of Arrival (STA) for each flight at specific constraint satisfaction points (CSPs), such as the runway threshold, final approach fix, or meter fix located near the boundary between en-route airspace which is controlled by an Air Route Traffic Control Center (referred to as a Center) and the terminal area around the airport which is controlled by the Terminal Radar Approach Control (TRACON).Arrivals are sequenced by a "first-come first-served" scheduling algorithm according to the Estimated Time of Arrival (ETA) of each flight at the relevant CSP.An STA is then assigned to each flight to satisfy the required inter-arrival spacing for a sequence of fixes through to the final CSP (typically the runway threshold or final approach fix).The arrival sequence and assigned STA for each flight continues to change as the flight ETA changes, up until the flight crosses a specified TBFM freeze horizon.This is defined at a particular time horizon or geographical distance from the CSP.At that point, the flight's STA is "frozen" and the target arrival time at the CSP is fixed.Air traffic controllers are then responsible for managing aircraft using delay vectors and speed clearances to meet these assigned STAs.Flights that depart from airports within the TBFM freeze horizon are referred to as "TBFM-internal" departures.These flights can receive delay on the ground in order to ensure that they fit into the overhead stream being controlled by TBFM.Their departure times are obtained using a TBFM departure scheduling function that assigns a departure time corresponding to an available arrival slot.Upon take off, flights can receive further airborne delay from TBFM, but this is typically small because the departure scheduling process ensures that these flights are departing into gaps in the overhead stream.Coupled scheduling and extended metering are two new TBFM features that have been introduced to improve performance.Without these features, TBFM arrival metering uses a single freeze horizon and schedule for each arrival flow.Coupled scheduling and extended metering include multiple coupled or uncoupled schedules for each flow, to a series of CSPs, each with its own freeze horizon.This delivers a more consistent traffic flow to the TRACON boundary by smoothing out errors in trajectory predictions. +C. Integrated Demand Management (IDM)As stated above, IDM research is focused on the integration of the strategic TFMS and tactical TBFM concepts to manage delivery of demand to constrained resources.HITL (e.g., Refs. 1, 2) and human-out-of-the-loop experiments, in which procedures are simplified and subject matter experts are not used to control traffic (e.g., Ref. 6), have been completed at NASA Ames by researchers in the AOL to investigate the operational feasibility of the IDM concept under realistic conditions.In these experiments, air traffic controller participants solve simulated flow management problems using CTOP and TBFM.Air traffic is simulated using the Multi-Aircraft Control System (MACS), which is a high-fidelity air traffic control simulation environment for prototyping scheduling systems and simulating air traffic. 7During a HITL simulation, MACS is used to run various workstation types for pseudo-pilots and air traffic controllers.In conjunction with MACS, an emulation of CTOP, called nCTOP (NASA CTOP), 1 was constructed to emulate the key functions of the TFMS version of CTOP, with some added functionality from the original CTOP concept that may be introduced in the operational system in the future.The CTOP functions modeled include setting capacity constraints at an FCA, automatically allocating trajectories from airline provided Trajectory Options Sets (TOSs) and assigning departure delay to pre-departure flights in order to balance the predicted demand at the FCA according to its capacity limit. 1 Inputs to nCTOP include FCA capacities, departure schedule, and TOSs.Based on the calculated EDCTs and TOS allocation, nCTOP outputs updated routes and departure times to all MACS stations through a simulation manager.MACS stations for each pilot and controller communicate with all other MACS stations in the simulation, updating aircraft positions.TBFM is used for arrival scheduling, and allows HITL simulation participants to schedule departure times for flights that depart from inside their destination's TBFM freeze horizon, called internal departures, to fit into the overhead stream based on calculated Scheduled Times of Arrival (STAs) at the meter fix and runway threshold.This is referred to as internal departure scheduling.The operational TBFM system (versions 4.2.3 and 4.4.1) is used, with NASA modifications that implement Terminal Sequencing and Spacing (TSAS) (in version 4.2.3 only), extended metering, and a NASA developed airport adaptation.This system, including the stated modifications, will henceforth referred to as hTBFM. +III. Automation Functionality and Simulation ImplementationWhereas a wealth of information and insight is gained through HITL simulations, these are expensive and timeconsuming activities, especially for the IDM concept, which requires five hours or more to simulate flights from predeparture at origin airports to arrival at the destination airport.As such, few HITL simulations have been completed and thus only a small subset of potentially interesting problems have been studied.The ability to run additional scenarios in an automated and fast-time fashion would be immensely useful.Such a simulation capability could be used to test various scenarios before using them in a HITL simulation, to perform parameter studies to quantify system performance of the IDM concept under various circumstances, and to simulate background traffic in HITL simulations.Automated simulation of the IDM experimental setup described in Section II.C requires the integration of a CTOP emulator, an air traffic simulator, and a TBFM emulator.These components are shown in the software flow diagram in Figure 1.Although the HITL simulation capabilities include these three components, several factors make automating and speeding up these simulation tools difficult.The hTBFM tool only runs in real time, and has challenges associated with its integration with other software.A number of processes within hTBFM are also designed to be performed manually through the hTBFM interface, including the scheduling of internal departures.The nCTOP tool was designed as a decision support tool for air traffic managers and thus also requires human input. +Figure 1. Flow diagram indicating information flow between components of the automated simulation capability.The automated simulation capability under development utilizes a number of simulation tools already developed at NASA, including MACS, an emulation of CTOP called nCTOP, and an emulation of TBFM developed for the Optimized Route Capability (ORC) tool. 10MACS and nCTOP are also used in the IDM HITL simulations.Although all of these tools exist, none can be used as-is and some amount of modification or enhancement is required.These tools and the associated changes made to them are described in more detail below. +A. CTOP EmulatorThe CTOP emulator used in the IDM HITL simulations, nCTOP, 1 was designed for interactive use by simulation participants.Through the GUI, participants can visualize demand and set capacities.Nominal flight information is sent from MACS to nCTOP and TOS information is specified in a file.Given this information, nCTOP assigns trajectories and EDCTs which can then be sent to MACS to update the simulation in progress.The development of nCTOP has been closely tied to MACS with various Application Programming Interfaces (API) developed to enable communication between the two tools.Given that the IDM concept is under development, nCTOP is continually updated to add functionality and respond to user requests.The automated simulation capability described in this paper must be able to keep up with these frequent changes to remain relevant.The use of MACS is essential to the rapid integration of new versions of nCTOP into the automated simulation capability.Given that nCTOP was designed as a decision support tool for HITL participants, various functions within the tool require automation to be able to run without participant intervention.Communication between MACS and nCTOP is performed through a TCP/IP connection.During a HITL, the TCP/IP server is started by MACS and an nCTOP operator uses the nCTOP GUI to connect to the server.A modification was made to nCTOP to automate the connection to the TCP/IP server.Additionally, nCTOP was modified to input a configuration file to automate the execution of various commands, including: set program start time, set FCA capacities in increments of 15 minutes, run balance algorithm, run TOS selection algorithm, send TOS selection and EDCTs to MACS, and exit nCTOP.These modifications enable fully automated execution of nCTOP commands and transfer of flight plans and EDCTs output from nCTOP to MACS in order to modify a traffic simulation in progress.Development of nCTOP has been focused on real-time simulations.Additional work needs to be done to synchronize nCTOP and MACS when the MACS simulation is run faster than real-time.This will be addressed in future work. +B. Air Traffic SimulatorA variety of air traffic simulation tools have been developed at NASA Ames, each designed with different objectives.Most notably, the Future Air Traffic Management Concepts Evaluation Tool (FACET) 8 and Airspace Concept Evaluation System (ACES) 9 were developed to enable the development and evaluation of advanced ATM concepts.Although either of these tools could be used to simulate air traffic for the automated simulation capability, MACS was chosen for a variety of reasons.First, MACS is the air traffic simulation tool used in IDM HITL simulations at NASA.Given that a primary objective of the automated simulation capability is to automate and speed up the HITL simulations, using MACS provides the easiest comparison between the automated simulation results and HITL results.Furthermore, the generation of scenarios is currently a lengthy and labor-intensive process.By using MACS, the automated simulation capability can make use of the scenario information of HITL simulations, as well as the scenario generation capabilities of the SNTB, which has been developed for MACS scenario generation.Using the same scenario format and files reduces workload and potential errors in translating a HITL scenario to an automated simulation capability scenario.Finally, as discussed above, nCTOP software is closely tied to MACS.In order to automate the HITL process, several user-initiated actions must be automated, specifically: set simulation speed, set simulation run time, open TCP/IP socket for communication with nCTOP, and exit MACS.A batch mode functionality had already been implemented in a command line version of the MACS software and was easily added to the MACS distribution used in the IDM HITL simulations.This capability allows the user to set simulation speed and duration in a configuration file.MACS was also modified to automatically connect to the TCP/IP server required for communication between MACS and nCTOP.In the automated simulation capability, the server is started by a simulation manager, which is described in more detail in Section D. +C. TBFM EmulatorA number of TBFM emulators exist, including one developed for the Optimized Route Capability (ORC), 10 and schedulers in the Center TRACON Automation System (CTAS) and MACS.None of these emulators include the TBFM capability to schedule internal departures.However, the ORC scheduler runs very fast, provides a high-fidelity model of the operational TBFM scheduling system, and is easily accessible for further development.For these reasons, it was chosen as the basis for the TBFM emulator developed in this paper.A number of features of the operational TBFM system are not modeled by the ORC scheduler.The most relevant of these include freezing flight STAs based on a specified freeze horizon, scheduling internal departures pre-departure, extended metering and coupled scheduling.The scheduling of internal departures is a manual process in the operational TBFM system.The underlying ORC scheduler was further developed and incorporated into the TBFM emulator for the automated simulation capability described in this paper (henceforth referred to as eTBFM), with an adaptation for Newark Liberty International Airport (EWR) modified from the EWR TBFM adaptation used in the HITL simulations for use by the ORC scheduler.The ORC scheduler is integrated into the eTBFM wrapper function that takes as an input timelines generated by MACS that provide ETAs at the meter fix, and an indication of which flights have crossed the freeze horizon.This wrapper function automatically schedules internal departures a set time before their expected departure time (typically EDCT), also input from MACS.This allows modeling of TBFM internal departure scheduling in one of two ways.In the first scheduling paradigm, internal departures are fitted into the overhead stream with no priority given to airborne flights (sometimes referred to as 'checkbox on', referring to a checkbox on the TBFM departure scheduling window that enables this functionality).In the second, priority is given to airborne flights, meaning internal departures are held on the ground until a suitable gap exists in the overhead stream (sometimes referred to as 'checkbox off').This is consistent with the scheduling performed manually in the operational TBFM system.eTBFM is also capable of accepting ETAs that change dynamically, in the form of timelines that update meter fix ETAs.Once a flight crosses the freeze horizon, eTBFM considers the calculated STA frozen, and does not recalculate it as its ETAs change.STAs for internal departures are frozen the moment they take off.A flow diagram showing the algorithm used for scheduling internal and external flights, using eTBFM, is shown in detail in In the current version of eTBFM, neither extended metering nor coupled scheduling are modeled.This functionality will be incorporated into eTBFM in future work. +D. Automated Simulation ManagerIn order to run the simulation fully automated and in batch mode, a simulation manager is required to synchronize the components discussed above.One of the simulation manager's functions is to set up the server required for communication between MACS and nCTOP.In the current configuration, such external communications need to be set up manually by specifying an IP address and a port number on both the server side (MACS) and client side (nCTOP and simulation manager) as shown in Figure 2.As shown in Figure 2, the simulation manager is connected directly to eTBFM.MACS generates ETA timelines at specified locations, notably the TBFM freeze horizon, and the arrival meter fixes for the airport of interest.These timelines are passed to eTBFM, which uses them directly as meter fix ETAs As a development tool, a graphical user interface (GUI) was built to facilitate processes such as setting up communication with MACS and controlling MACS simulation.This is shown in Figure 3, and allows the user to perform various tasks, such as: specify the scenario file to be run by MACS; start MACS and run the simulation; start nCTOP and connect with MACS; set MACS simulation run speed (from real-time (1x) to up to 20 times faster than real time (20x)); and start data collection.As mentioned previously, a key feature of the automated simulation capability is the ability to run tens or even hundreds of simulations with varying scenarios and parameters without requiring any manual interruption.To achieve this, the GUI functionality was automated with a command line interface.During command line execution, the TCP/IP sockets are set up and connected with external tools automatically during launch.Next, a number of simulation parameters, such as simulation speed and simulation duration, are automatically loaded into MACS from userspecified batch executable files.The command line interface allows for the sequential and automatic execution of multiple scenarios.A detailed flowchart of the automated simulation software is given in +IV. Verification and Validation +A. Air Traffic SimulatorIn order to conduct fast-time simulations, MACS software that was used in the HITL simulations was run at a faster speed.Although the functionality already existed in MACS, the trajectory integrity was never validated in the fast-time mode.Therefore, flight tracks produced by MACS running at five, ten and twenty-five times faster than real time (5x, 10x, 25x) were compared to flight tracks generated by MACS running in real time (1x).The traffic scenario used is identical to that used in Ref. 6 consisting of flights departing various airports and with destination EWR.The scenario was run both with and without environmental winds.The wind data used are identical to the "mild" and "heavy" wind conditions of the experiments presented in Ref. 6 with average wind speeds at an altitude of 30,000 ft of 45 knots and 60 knots and up to 130 knots and 160 knots maximum, respectively.MACS was run at 1x, generating reference real-time data, and then again at resampled to obtain latitude, longitude and altitude information at 60-second multiples of simulation time.For the given set of 196 flights, over 22,000 such points were generated.For each flight and each simulation speed, comparisons between each of these resampled points and the corresponding resampled points in the reference realtime track were made.Tracks were compared to the reference real-time tracks based on lateral and vertical distance between corresponding track points and total flight time.Lateral distance between corresponding track points is decomposed into components parallel to the real-time track ("along-track") and perpendicular to the real-time track ("cross-track").Box and whisker plots are used to present the comparison results.Edges of the whiskers are at the 10 th and 90 th percentile of the data, lower and upper boundaries of the box are at the 25 th and 75 th percentile, and the 50 th percentile is indicated by a red line within the box.Results of these comparisons are shown in Figure 4 through Figure 6.Results are shown for the scenario without wind only as results for the scenario with wind are similar.Along-track differences at each simulation speed compared to the corresponding reference real-time tracks are shown in Figure 4.Note that the data for 1x are a measure of repeatability of the The differences between the 1x tracks and the reference 1x tracks are small, but nonzero, indicating that the simulation cannot be repeated precisely.It can be seen that at all simulation speeds, the absolute values of the 25 th and 75 th percentiles of along-track distances from 1x tracks are all below 0.25 nmi.Crosstrack distances were also analyzed and found to be quite small with absolute values of the 25 th and 75 th percentiles of the cross-track distances from 1x tracks below 0.01 nmi.Vertical differences in tracks at each simulation speed are compared to the corresponding reference real-time tracks and shown in Figure 5.It was found that most of the differences in vertical position occurred during climb and descent.For this reason, results are presented for the climb, cruise and descent phases of flight separately.These phases were found using the real-time flight tracks with the climb phase extending from departure to the time at which the flight track reaches cruise altitude and descent extending from initial descent from the cruise altitude to the landing.A total of about 1,500, 14,400 and 4,400 resampled track points make up the data sets for the climb, cruise and descent phases of all flights, respectively.During the cruise phase, over 90% of resampled track points are within one foot of the corresponding reference track point (except for the 25x tracks with wind, for which 89% are within one foot).Total flight times were also compared, with results shown in Figure 6.Note that results are not shown for the 25x simulations.This is due to an error in the MACS simulation code which caused many flights in the 25x simulations to not land.Track data up to the landing point are valid, so the position comparisons could be made, No Wind but an accurate simulation landing time could not be obtained for the 25x flights and thus this data is not shown in Figure 6.Whether or not these differences in location and flight time are acceptable depends on the application.For IDM development, the timing of arrivals at constraint satisfaction points (CSPs) is of utmost importance.For the problem discussed in Section IV.C in which flights arriving into EWR are scheduled by CTOP and then TBFM as the flight enters the freeze horizon, maximum arrival capacity at the EWR runway is 44 aircraft per hour.At this rate, the time between arriving flights is about 80 seconds.Spacing at the various CSPs leading into the EWR runway will have inter-arrival spacing of roughly 80 seconds or more as flights on different flows must be spaced properly so that the appropriate rate is achieved when the flows merge.Clearly, errors in CSP arrival time of +/-10 seconds is significant and would affect the resulting TBFM schedule.However, the IDM study takes into account various uncertainties.EDCTs are met with some error that is obtained by sampling from a distribution found in GDP data.Errors of +/-10 seconds are 3.33% of the EDCT compliance window of +/-300 seconds (+/-5 minutes).Although small, this is still noteworthy and could be offset by adjusting the departure error sampling distribution.Given these results, MACS simulation speeds of up to 10x are considered acceptable for IDM studies. +B. TBFM EmulatorAn initial validation of eTBFM was completed by comparing outputs of the emulator and hTBFM (TBFM version 4.4.1, as run in Ref. 2), given dynamic timelines at the freeze horizon and runway threshold, and internal departure EDCTs.One hour of arrival traffic from the distributed traffic scenario described by Ref. 6 was simulated, which captures the arrival of 39 flights to EWR from a variety of origins (domestic and international), operating a variety of aircraft types.No wind is simulated, and inter-arrival spacing of 0.4 nmi is imposed at the final approach fix.Given the demand level, this results in an airport arrival rate of 50 aircraft per houra high rate for EWR.All three primary arrival fixes at EWR are used: PENNS, SHAFF and DYLIN.Inputs to eTBFM include a series of timelines generated by MACS, which consist of meter fix ETAs that can change over time and freeze horizon crossing times, and internal departure EDCTs.Using hTBFM, internal departures are scheduled manually using the paradigm in which internal departures are fitted into the overhead stream with no priority given to airborne flights.A traffic manager monitors the hTBFM internal departure timeline and schedules departures at a suitable time, as determined by the traffic manager, in advance of their EDCTs.In this experiment internal departure scheduling using hTBFM was done (manually) as close to 20 minutes before the flight EDCTs as possible.In contrast, eTBFM schedules internal departures automatically 20 minutes before their EDCT.hTBFM applies extended metering, where each flow includes two sequential uncoupled schedules, first to a metering point upstream of the meter fix, and secondly to the meter fix.Each schedule has its own freeze horizon, defining an extended metering region and a meter fix region.eTBFM does not apply extended metering, including only a schedule to the meter fix.For this experiment the freeze horizon used for eTBFM is that of the meter fix region in hTBFM.Delays, ETAs and STAs are therefore compared for the meter fix region only.Results are presented in Figure 7 and Table 1 below.Figure 7 shows fairly close correlation between the ETAs and STAs calculated by eTBFM and hTBFM at the meter fix (Figure 7a) and runway threshold (Figure 7b).Although eTBFM uses the same airport adaptation as hTBFM, results are not expected to match precisely given that some simplifying assumptions are built into the eTBFM scheduler, and the scheduling of internal departures in hTBFM is manual.Average metering delays calculated by eTBFM and hTBFM (in the meter fix region only), and average differences in ETAs and STAs, are shown in Table 1, with standard deviations included in brackets.Average delays calculated by eTBFM are higher than those of hTBFM at the meter fix and at the threshold (by 63 and 44 seconds respectively).The ETAs generated by eTBFM are, on average, earlier than those generated by hTBFM.The STAs calculated by eTBFM at the threshold are, accordingly, also slightly earlier than those calculated by hTBFM.However, because of the higher meter fix delays calculated by eTBFM, eTBFM calculated STAs at the meter fix are later than those calculated by hTBFM.Standard deviations in metering delay, and in the differences between ETAs and STAs, are large (between 75 and 135 seconds).This is because TBFM is very sensitive to flight sequencing, and especially to when internal departures are scheduled.When TBFM is used operationally, controllers are shown and expected to meter to ETA and STA values that are truncated to the nearest minute, thus errors within a minute are considered acceptable.However, future research will work to reduce the differences between eTBFM and hTBFM shown here.For the 10 internal departures simulated, Figure 7c compares the scheduled times of departure calculated by eTBFM and hTBFM.There are noteworthy differences, primarily because of differences in flight sequencing.Sequencing is greatly affected by the exact timing of when the flight is scheduled, which is a manual process in hTBFM, and was therefore not possible to control precisely.Average scheduling delays calculated by eTBFM and hTBFM are shown in Table 2. Despite the differences in sequencing, eTBFM and hTBFM show very similar results, with an average difference of only 4 seconds.This is a key output of eTBFM, but needs to be confirmed with a validation simulating a larger number of internal departures.The high standard deviations are a direct consequence of the differences in flight sequencing. +V. Example ApplicationThe motivation for the development of the automated simulation capability is to be able to test IDM concepts in an automated fast-time manner.In order for the automated simulation capability to be useful, results must be qualitatively similar to those of HITL simulations.That is, quantitative results should show similar trends, allowing researchers to draw similar conclusions about the effectiveness and impact of IDM concepts whether using the automated simulation capability or HITL simulation.Two experiments were conducted using the automated simulation capability described in this paper to show an example application, and to allow some of the results to be compared to results from similar HITL simulations.Note that because the scenarios used in these experiments are not exactly the same as those used in the previous HITL simulations and HITL simulation participants take actions that are not repeatable, individual flight delays and flight sequences will differ, but average delays should be similar.The experiments include all components of the simulation capability developed and were run automatically at realtime speed.Both experiments simulate a typical 5-hour traffic scenario including only flights destined for EWR, as was done in Ref. 1, 2 and 6.Weather and wind impacts are not included in the simulation, which is one key difference compared to previous HITL simulations.A single FCA is applied at EWR, set at a rate of 44 aircraft per hour.Interarrival spacing of 0.4 nmi is set at the final approach fix in eTBFM, which results in a 44 aircraft per hour rate at the airport in the absence of wind given the demand supplied by nCTOP.In contrast to the eTBFM validation experiment in Section IV.B, for these experiments the freeze horizon used for eTBFM is that of the extended metering region in previous HITL simulations running a similar scenario 1,2,6 .This means that TBFM airborne delays are considered across the region covering both the extended metering and meter fix regions from previous HITL simulations.Departure errors used in these experiments are identical to those used in previous HITL simulations, 1,2,6 which were drawn from historical data.Using these departure errors, 70% of external flights are EDCT compliant (i.e., flights depart within +/-5 minutes of their assigned EDCT) and 100% of internal departures are compliant with their TBFM scheduled departure time, or "call for release" time (i.e., flights depart within +2/-1 minute of their assigned time).The two experiments differ only in the paradigm used for the scheduling of internal departuresprioritizing airborne flights (as is typically done operationally today), and not prioritizing airborne flights.These experiments should confirm the conclusions of the HITL simulations 1,2,6 suggesting that, under the IDM concept, when airborne flights are not prioritized over internal departures, internal departure scheduling delays are reduced, while the level of airborne delay that must be absorbed by TBFM is still operationally feasible.Results are shown below.Figure 8 shows ground delays assigned by nCTOP by flight, as a function of their runway threshold STA.Ground delays assigned by nCTOP are independent of the paradigm used for the scheduling of internal departures.Only one set of delays is therefore plotted.The magnitude of ground delays found in these experiments (with a maximum of 39 minutes) are similar to those reported in Ref. 2 (with a maximum of 46 minutes), which simulates a similar scenario.Table 3 shows the distribution of TBFM airborne delay for the two simulations.Given the airspace over which TBFM metering is simulated, airborne delay is considered operationally acceptable if it is below 7 minutes, marginal if it is between 7 and 14 minutes, and unacceptable if it is greater than 14 minutes.* As can be seen, in both cases very few flights have operationally unacceptable airborne delay (between 0% and 1%).However, with no priority given to airborne flights, the percentage of marginal airborne delays increases (from 5% to 13%).However, the vast majority of flights still maintain acceptable levels of airborne delay (87%).The results with no priority given to airborne flights are consistent with results from Ref. 1 and Ref. 2, which simulate similar scenarios.Ref. 1 reports 1.5% unacceptable airborne delay, 17% marginal, and 82% acceptable, while Ref. 2 reports 7% unacceptable airborne delay, 28% marginal, and 65% acceptable.† The results in Table 3 suggest that the IDM concept simulated may enable TBFM to be run with no priority given to airborne flights, yielding the improvements in internal departure scheduling delay shown in Figure 9.These conclusions are consistent with the conclusions from previous HITL simulations. 1,2 +VI. Conclusions and Future WorkIn this work, an initial automated simulation capability was developed to support IDM research and development.This simulation capability can be used to test various scenarios and parameters in order to identify candidate scenarios for HITL simulations.Although most of the software tools used in the HITL simulations were adopted for use in the automated simulation capability (with the notable exception of the version of TBFM used in the HITL simulations), all had to undergo some amount of modification to enable the automation of the simulation.As mentioned, the version of TBFM used in the HITL simulations is not suitable for use in an automated manner for a variety of reasons.Thus, an alternative TBFM emulator was built for the automation simulation capability starting from an arrival scheduler developed for the Optimized Route Capability (ORC) tool.Functionality of the newly developed TBFM emulator includes the implementation of two scheduling paradigms (prioritizing airborne flights and not prioritizing airborne flights), and the automation of internal departure scheduling in accordance with TBFM procedures.MACS and nCTOP were used to provide the same functionality in the automated simulation capability as in the HITL simulations.A simulation manager was built to coordinate this set of software.The automated simulation capability can be run in batch mode from the command line.Verification and validation was performed on each component of the automated simulation capability separately.From MACS simulations, it was found that 80% of the differences between the real-time and the 5x trajectory points are within 0.25 nmi along-track, 0.02 nmi cross-track, 1 ft vertically in the cruise phase of flight, and 15 seconds in total flight time.Results from the eTBFM scheduler were found to match corresponding results found using hTBFM with average difference between STAs and ETAs at the threshold of 60 seconds and 30 seconds, respectively.Finally, * In Ref. 1 and 2, airborne delays in the extended metering region are considered to be acceptable if they are below 5 min, marginal if between 5 and 10 minutes, and unacceptable if over 10 minutes.Airborne delays in the meter fix region are considered to be acceptable if they are below 2 min, marginal if between 2 and 4 minutes, and unacceptable if over 4 minutes.In this experiment, airborne delays are considered across the region covering both the extended metering and meter fix regions.Hence the thresholds for acceptable, marginal and unacceptable delays used for this work are calculated by adding the thresholds over the extended and meter fix regions in Ref. 1 and Ref. 2. † Results from Ref. 1 and 2 quoted apply to the meter fix region only.In both studies the distribution of airborne delay is more acceptable in the extended metering region.Differences between the results from Ref. 1 and 2 are because of differences in the scenarios simulated, airport adaptation used, and procedures used for human-operator inputs.all components of the automated simulation capability were used together to execute two simulations as a proof-ofconcept.Results for TBFM required airborne delay and internal departure scheduling delay are comparable to those from previous HITL simulations using similar scenarios. 1,2uture work in the development of the automated simulation capability includes incorporating additional functionality and increasing simulation speed to faster than real-time.Increasing simulation speed requires modifications to nCTOP in order to synchronize times between nCTOP and MACS.The TBFM emulator will be further developed to include extended metering and coupled scheduling.In the current automated simulation capability, STAs and TBFM-scheduled internal departure times output from the TBFM emulator are not input to MACS to update the simulation.Future work will enable this communication and MACS speed control functionality will be used to simulate traffic manager actions to delay flights within the TBFM freeze horizon to meet STAs.Ultimately, this automated simulation capability will be integrated into the SNTB.The SNTB is currently in development with plans to be a platform for research, development, and analysis of air traffic management concepts.MACS is used as the primary traffic simulation tool, which contributed to the choice of MACS for use in the automated simulation capabilities.Work done to automate the communication between MACS and other components of the automated simulation capability and run the simulations in batch mode will be useful additions to the SNTB platform.Scenarios involving more than a couple hundred flights require multiple instances of MACS to perform traffic simulation.SNTB has the required architecture to run and coordinate multiple instances of MACS.Once integrated into SNTB, this automated simulation capability will be scalable to NAS scenarios.Figure A.1 in Appendix A. +, and to indicate whether or not flights have crossed the TBFM freeze horizon.As shown in the flow diagram of the eTBFM scheduling algorithm in Figure A.1 in Appendix A, STAs of external flights are frozen once they pass the TBFM freeze horizon.In contrast, STAs of internal flights are frozen as they depart from their respective origin airports. +Figure A.2 in Appendix A. As mentioned in the previous section, the software takes various scenario files and simulation settings as the input.The software then launches MACS, nCTOP and eTBFM sequentially and automatically to generate dynamic arrival flight schedules. +Figure 2 .2Figure 2. MACS external communication Figure 3.A Graphical User Interface (GUI) to set up communication between MACS and nCTOP and set MACS parameters. +1x and at 5x, 10x and 25x.During the simulation, flights are allowed to depart for 4 hours.Beyond this time, trajectories of airborne flights continue to be simulated until the final flight lands at roughly 5 hours and 30 minutes into the simulation.A detailed comparison was made between the fast-time flight tracks and the corresponding real-time flight tracks.Track data were generated at 12-second intervals.Each track was +Figure 4 .4Figure 4. Component of distance from real-time track points to associated fast-time track points along the realtime track.A positive value indicates that the fast-time track is ahead of the real-time track. +Figure 5 .5Figure 5. Vertical distance from real-time track points to associated fast-time track points along the real-time track.A positive value indicates that the fast-time track is above the real-time. +Figure 6 .6Figure 6.Difference in total flight time between the real-time track and fast-time tracks.A positive value indicates that the flight time of the fast-time track is longer than the real-time. +Figure 7 .7Comparison of ETAs and STAs between eTBFM and hTBFM, at (a) the arrival metering fix, and (b) the runway threshold.(c) Comparison of internal departure scheduled departure times calculated by eTBFM and hTBFM. +Figure 8 .8Figure 8. Ground delay minutes assigned by nCTOP by flight. +Figure 99Figure9shows eTBFM internal departure scheduling delay minutes by flight for the two experiments, as a function of their runway threshold STA.As expected, by not prioritizing airborne flights over internal departures, internal departure scheduling delays are reduced.The average internal departure scheduling delay decreases from just over 18 minutes to just under 3 minutes.The average internal departure scheduling delay with no priority given to airborne flights is consistent with results from Ref. 2, which simulates a similar scenario and reports average internal departure scheduling delays of 2 minutes 10 seconds. +0:30 1:00 1:30 2:00 2:30 3:00 3:30 4:00 4:30 5:00 TBFM Scheduling Delay [min] Threshold STA in Simulation Time [hour:minute] 0:30 1:00 1:30 2:00 2:30 3:00 3:30 4:00 4:30 5:00 TBFM Scheduling Delay [min] Threshold STA in Simulation Time [hour:minute] Priority to Airborne Flights NO Priority to Airborne Flights +Figure A. 22Figure A.2 shows the overall flowchart of the automated simulation software. +Figure A. 22Figure A.2 Process flow chart of the automated simulation capability. +Table 1 . TBFM emulator validation results, comparing metering outputs from hTBFM to those of eTBFM. Standard deviations shown in brackets.1Meter FixThreshold[Seconds] +Table 2 . TBFM emulator validation results, comparing internal departure scheduling outputs from hTBFM to those of eTBFM. Standard deviations shown in brackets.2Seconds +Table 3 . Distribution of TBFM Airborne Delay3The average throughput at the runway, over the final four hours of simulation, with priority given to airborne flights, is 43.5 aircraft per hour.With no priority given airborne flights, this increases slightly to 44.3 aircraft per hour, potentially indicating better utilization of the constrained resource.The throughput results also confirm that the target throughput of 44 aircraft per hour was achieved, as in Ref.1 and Ref. 2.SimulationTBFM Scheduling Paradigm% TBFM Airborne% TBFM Airborne% TBFM AirborneDelay AcceptableDelay MarginalDelay Unacceptable(<7 min)(7-14 min)(>14 min)1Priority to Airborne Flights94%5%1%2NO priority to Airborne Flights87%13%0% + + + + +AcknowledgementsThe authors would like to thank Connie Brasil, Nathan Buckley, Gita Hodell, Scott Kalush, George Lawton, Paul Lee, Christoph Mohlenbrink, Nancy Smith, and Hyo-sang Yoo for numerous discussions, software changes and the exchange of data for validation.The authors would also like to thank Shannon Zelinski and Min Xue for the ORC scheduler code. + + + +Appendix A: Algorithm flow diagrams.Figure A.1 shows the algorithm used by eTBFM developed for the scheduling of both internal and external arrival flights to the constrained airport.This includes both scheduling modes, that is, the mode in which internal departures are fitted into the overhead stream with no priority given to airborne flights, and the mode in which priority it given to airborne flights, meaning internal departures are held on the ground until a suitable gap exists in the overhead stream. + + + + + + + + NSmith + + + CBrasil + + + PULee + + + NBuckley + + + CGabriel + + + OmarMohlenbrink + + + FParke + + + BSperidakos + + + CYoo + + + H + + Integrated Demand Management: Coordinating Strategic and Tactical Flow Scheduling Operations, 16th AIAA Aviation Technology, Integration, and Operations Conference + + 2016 + + + + SMITH, N., BRASIL, C., LEE, P.U., BUCKLEY, N., GABRIEL, C., MOHLENBRINK., C, OMAR, F., PARKE., B., SPERIDAKOS, C., and YOO, H., Integrated Demand Management: Coordinating Strategic and Tactical Flow Scheduling Operations, 16th AIAA Aviation Technology, Integration, and Operations Conference, 2016, AIAA 2016-4221. + + + + + + HSYoo + + + CBrasil + + + NBuckley + + + CMohlenbrink + + + CSperidakos + + + BParke + + + GHodell + + + PULee + + + NMSmith + + AIAA 2017-4100 + Integrated Demand Management: Minimizing Unanticipated Excessive Departure Delay while Ensuring Fairness from a Traffic Management Initiative, 17th AIAA Aviation Technology, Integration, and Operations Conference + + 2017 + + + YOO, H.S., BRASIL, C., BUCKLEY, N., MOHLENBRINK, C., SPERIDAKOS, C., PARKE, B., HODELL, G., LEE, P.U., and SMITH, N.M., Integrated Demand Management: Minimizing Unanticipated Excessive Departure Delay while Ensuring Fairness from a Traffic Management Initiative, 17th AIAA Aviation Technology, Integration, and Operations Conference, 2017, AIAA 2017-4100. + + + + + + KPalopo + + + GBChatterji + + + MDGuminsky + + + PCGlaab + + Shadow Mode Assessment using Realistic Technologies for the National Airspace System (SMART NAS) Test Bed Development. 15th AIAA Aviation Technology, Integration, and Operations Conference + + AIAA + 2015. 2015-2794 + + + PALOPO, K., CHATTERJI, G.B., GUMINSKY, M.D. and GLAAB, P.C., Shadow Mode Assessment using Realistic Technologies for the National Airspace System (SMART NAS) Test Bed Development. 15th AIAA Aviation Technology, Integration, and Operations Conference, 2015, AIAA 2015-2794. + + + + + Development of a High-Fidelity Simulation Environment for Shadow-Mode Assessments of Air Traffic Concepts + + IiiRobinson + + + JELee + + + A + + + Lai + + + C + + + + RAeS and AIAA Modelling and Simulation in Air Traffic Management Conference + London + + 14-15 November, 2017 + + + ROBINSON, III, J. E., LEE, A., and LAI, C., "Development of a High-Fidelity Simulation Environment for Shadow-Mode Assessments of Air Traffic Concepts," RAeS and AIAA Modelling and Simulation in Air Traffic Management Conference, London, 14-15 November, 2017. + + + + + + RTyo + + + Pressler + + + C + + + Industry Day Briefing: Time-Based Flow Management (TBFM) + + April, 2014. June 12, 2017 + + + Federal Aviation Administration Industry Day + TYO, R., and PRESSLER, C., Industry Day Briefing: Time-Based Flow Management (TBFM), Federal Aviation Administration Industry Day, April, 2014. from: http://www.faa.gov/about/office_org/headquarters_offices/ato/service_units/pmo/industry_day/media/day_1_-_tbfm_bob_tyo_- _chris_pressler_-_final.pdf [Retrieved June 12, 2017]. + + + + + Required time of arrival as a control mechanism to mitigate uncertainty in arrival traffic demand management, 35 th IEEE/AIAA Digital Avionics Systems Conference + + HSYoo + + + CMohlenbrink + + + CBrasil + + + NBuckley + + + AGlobus + + + NMSmith + + + PULee + + + 2016 + IEEE + + + YOO, H.S., MOHLENBRINK, C., BRASIL, C., BUCKLEY, N., GLOBUS, A., SMITH, N.M. AND LEE, P.U., Required time of arrival as a control mechanism to mitigate uncertainty in arrival traffic demand management, 35 th IEEE/AIAA Digital Avionics Systems Conference, 2016, IEEE. + + + + + Exploring the many perspectives of distributed air traffic management: The Multi Aircraft Control System MACS + + TPrevot + + + + Proceedings of the HCI-Aero + the HCI-Aero + + 2002 + + + + PREVOT, T., Exploring the many perspectives of distributed air traffic management: The Multi Aircraft Control System MACS, In Proceedings of the HCI-Aero, 2002, pp. 149-154. + + + + + FACET: Future ATM concepts evaluation tool + + KBilimoria + + + BSridhar + + + GBChatterji + + + KSheth + + + SGrabbe + + + + Air Traffic Control Quarterly + + 9 + 1 + + January 2001 + + + BILIMORIA, K., SRIDHAR, B., CHATTERJI, G.B., SHETH, K. AND GRABBE, S. FACET: Future ATM concepts evaluation tool. Air Traffic Control Quarterly, January 2001, 9(1), pp.1-20. + + + + + Fast-time simulation system for analysis of advanced air transportation concepts + + DNSweet + + + VManikonda + + + JSAronson + + + KRoth + + + MBlake + + + + Proceedings of the AIAA modeling and simulation technologies conference + the AIAA modeling and simulation technologies conference + + August 2002 + + + + SWEET, D.N., MANIKONDA, V., ARONSON, J.S., ROTH, K. AND BLAKE, M. Fast-time simulation system for analysis of advanced air transportation concepts. In Proceedings of the AIAA modeling and simulation technologies conference August 2002 (pp. 5-8). + + + + + Optimized Route Capability (ORC) Intelligent Offloading of Congested Arrival Routes + + SZelinski + + + MXue + + + PBassett + + + + 16th AIAA Aviation Technology, Integration, and Operations Conference + + 2016 + + + + ZELINSKI, S., XUE, M., and BASSETT, P., Optimized Route Capability (ORC) Intelligent Offloading of Congested Arrival Routes, 16th AIAA Aviation Technology, Integration, and Operations Conference, 2016, AIAA 2016-4357. + + + + + + diff --git a/file22.txt b/file22.txt new file mode 100644 index 0000000000000000000000000000000000000000..4fa1e7c4725d053a919e9e7abb4a5841c81be03f --- /dev/null +++ b/file22.txt @@ -0,0 +1,424 @@ + + + + +As the capabilities and implementation of multiple Unmanned Aerial Systems (UAS) operations increase, the need to develop reliable, verifiable, and high-performing vehicle control algorithms arises.There are several methods for achieving near optimal navigation control in multiple UAS scenarios.However, many approaches have difficulty identifying boundaries in the state-space where discontinuities in the control signal exist.Neural Networks (NNs) have been shown to be universal approximators of these decision boundaries and can therefore classify the space into discrete regions with differing lower level actions.In this study, a control system for a one-on-one UAS Tail-Chase scenario is developed.An NN was used to define a decision boundary for discrete selection of two different Fuzzy Inference Systems (FISs) for navigation and avoidance control.The parameters for both the NN and FISs are found using a Genetic Algorithm (GA) inside a custom simulation environment.After initial training, uncertainty was included in vehicle movements to improve generalization.The simulation results show that the system was successful in all test cases after adding uncertainty and demonstrate the efficacy of this approach. +I. Nomenclature +II. IntroductionA cooperative multiple Unmanned Aerial Systems (UAS) operations become more realizable, there is a need to develop reliable, verifiable, and high-performing algorithms for multiple vehicle control.One particular problem of interest is a Tail-Chase scenario, where one or more vehicles try to reach a desired state relative to another.For example, leader-follower swarming behavior can utilize this to allow flight formations.There are several solutions for Tail-Chase vehicle control in low number scenarios, such as one-on-one, two-on-one, or even one-on-multiple [1][2][3].The main difficulty when solving these types of problems is finding the boundaries in the state-space where switching actions lie.For the application of a Tail-Chase scenario, the UAS controller must continuously decide the direction and magnitude to turn to reach a desired region relative to the other vehicle.One way to partition the space effectively is to use classification methods as a supervisory, outer loop control.The classes then determine which inner loop control system to enact.Neural Networks (NNs) have been shown to be both universal approximators and to be able to perform highly nonlinear classification with one or more hidden layers and nonlinear activation functions [4,5].There are also systems that can be used effectively as both inner and outer loop control, at the cost of complexity.These systems are able to approximate a continuous function to any arbitrary degree of accuracy [6][7][8].However, these systems have less transparency than a hierarchical system that uses supervisory decision-making.In addition, they may have difficulties approximating discontinuities efficiently.As there have been recent efforts to formally verify NNs [9][10][11], the solution presented in this paper could be promising for complex UAS scenarios where high confidence in correct behavior is required.Additionally, as we know that discontinuities likely exist in the optimal control function, it may be more computationally efficient to have a supervisory logic that defines these boundaries (explicitly or implicitly) and subsequently use a continuous universal approximator on the compact set defined therein.Developing an effective Tail-Chase controller is further complicated by the fact that the control algorithm is contingent on the other vehicle's behavior.For example, if two vehicles have the same Tail-Chase controller, neither vehicle will succeed in reaching a desired region without a performance advantage; i.e., vehicles with the exact same performance (speed, load factor, etc.) will not be able to close the distance to an arbitrary desired region.Additionally, one potential way to make a controller more robust to unknown behaviors is to introduce uncertainty into the target vehicle's movement.This prevents overfitting a solution to particular conditions and therefore improves confidence in the system's ability to generalize.This paper proposes a solution that has several advantages over other approaches.First, the ability to identify decision boundaries implicitly in the input space avoids the need to define them a priori as would need to be done for an analytical solution.Secondly, the scenario can be modified to include any arbitrary desired penalty and reward regions relative to another vehicle.This allows the accommodation of much more complex systems and behaviors while retaining tractability.Lastly, this closed-loop control method is robust to unknown behaviors in the target UAS.The solution continuously reacts to current state information and is trained to account for a certain level of state uncertainty.The benefit of utilizing uncertainty during training is that it allows for a smaller set of training cases while improving the ability to generalize over the state-space.In Section III the specific problem being addressed is described.This is followed by a detailed explanation of the methodology of the proposed solution in Section IV.In Section V the results of the study are presented.Lastly, conclusions and future opportunities for expanding on this work are presented in Section VI. +III. Problem DescriptionThe problem being considered is a one-versus-one Tail-Chase scenario where one UAS, the pursuer, is attempting to reach a Reward Region (RR) relative to another UAS, the target, while avoiding a Penalty Region (PR).The PR and RR are defined by ranges of the relative states of the two vehicles.The relatives states considered are the distance, d, the relative angle, ψ R A P , and the relative heading, ψ RH P .The relative states d, ψ R A p , and ψ RH p are defined in Eqs. ( 8), (9), and (10), respectively.Specifically, RR = {R|d ≤ 50,-π 6 ≤ ψ R A P ≤ π 6 , -π 6 ≤ ψ RH P ≤ π 6 }. The PR is relative to the target vehicle such that PR = {R|d ≤ 50, -π 6 ≤ ψ R A T ≤ π 6 }.A visualization of this is shown in Fig. 1. +Fig. 1 Pursuer and Target UAS with Penalty and Reward RegionsThe vehicles are constrained to 2-D level flight with constant speed and maximum turn rates, ψ max j .The values were set such that the pursuer has an advantage in load factor and speed while keeping the same minimum turn radius, R min .The maximum turn rates and minimum turn radii were calculated from Eqs. ( 1)-( 2) with values of v P = 50 m s and v T = 20 m s for the pursuer and target, respectively.This turns the vehicles into Dubins vehicles [12] and the dynamic model in global coordinates is given by Eqs. ( 3)- (5).These can then be integrated to obtain the global states, x j , y j , and ψ j at each time step.ψ max j = g v j N j 2 -1 , j = P, T(1)R min = v j ψ max j(2)x j = v j cos(ψ j )(3)y j = v j sin(ψ j )(4)ψ j = u j , u j ∈ -ψ max j , ψ max j(5)In order to develop a controller with fewer required states, the relative states of the vehicles, d, ψ RH , and ψ R A , are calculated from the global states.These are shown in Eqs. ( 6)- (10).e x P = x T -x P(6)e y P = y T -y P(7)d = e 2 x P + e 2 y P(8)ψ R A P = tan -1 e y P e x P (9)ψ RH P = ψ T -ψ P(10)ψ RH P and ψ R A P are then corrected such that ψ RH P , ψ R A P ∈ {R| -π ≤ ψ RH P , ψ R A P ≤ π}.This information shall be used to direct the pursuer to the RR without entering into the PR.To achieve this goal, a controller must use the relative state information to determine the appropriate corresponding turn rate. +IV. Proposed SolutionThe proposed solution to this problem is to use a Neural Network to partition the state-space into discrete regions.Inside these regions, either a navigation controller or an avoidance controller will be employed to control the path of the vehicle.These two controllers direct the pursuer to either steer towards or away from the target.The controller for navigating towards the target is shown in Eq. (11).ψ nav P = ψ max P sgn(ψ R A P )(11)The avoidance controller is more complicated due to needing more information about the relative states of the two UAS and the possible modes for steering away.This controller and its corresponding logic is detailed in Section IV.A. +A. AvoidanceThe avoidance controller was developed with inspiration from the methods used in Ref. [13,14].It uses a similar logic based on the intersection point of the UAS.The intersection point is defined as the point of intersection of the projected current headings between the two UAS.The distance to the intersection point, d I j , is then calculated by Eq. ( 12)- (14).A visualization of the intersection point (x I , y I ) and the distances d I P and d I T are shown in Figure 2. +Fig. 2 Intersection point and distances from UASThis distance is used to determine if the pursuit UAS should go in front of or behind the target UAS using the relationships shown in Eq. ( 15).An intersection distance margin, denoted d M , is used in order to decide if there is sufficient distance to safely go in front of the target UAS.x I = y P -y T -x P tan (ψ P ) + x T tan (ψ T ) tan (ψ T ) -tan (ψ P )(12)y I = y P tan (ψ T ) -y T tan (ψ P ) + (x T -x P ) tan (ψ P ) tan (ψ T ) tan (ψ T ) -tan (ψ P )(13)d I j = (x I -x j ) 2 + (y I -y j ) 2(14)f ront = 1 if (d I P + d M ) < d I T 0 otherwise(15)After the UAS decides if it should go in front, it then selects an appropriate Fuzzy Logic Controller (FLC) that determines the actual continuous output, ψ.Note that FLCs are simply Fuzzy Inference Systems (FISs) that are specifically applied to control problems.The FLCs each utilize triangular membership functions, Ruspini partitioning, input normalization, the product method for rule association, and weighted average defuzzification, as desribed in Ref. [15].This results in an output that is of the form shown in Eqs. ( 16)- (18).A general FLC structure and internal processes is shown in Fig. 3.The input set, c, and output set, U, have the forms shown in Eq. ( 19) and (20), respectively.(Note: a bold variable indicates it is a matrix of values.)Due to the properties of the FLCs in question, these sets represent the center points of their triangular membership functions.The values of these parameters, as well as the normalization and scaling gains, k in and k out , were found using the methods described in Section IV.C.Note that, due to normalization, the values for c 1 and c 4 were set to -1 and 1, respectively.ψ avoid P = k out 3 i=1 µ i U i (16)µ i = ψ R A k i n -c i c i+1 -c i (17)µ i+1 = c i+1 -ψ R A k i n c i+1 -c i (18) c = [ c 1 c 2 c 3 c 4 ] (19) U = [U 1 U 2 U 3 U 4 ](20)Fig. 3 FLC structure and internal processes +B. Neural NetworkThe constructed Neural Network is a fully-connected, feed-forward network with two hidden layers.NNs with multiple hidden layers have been shown to be universal approximators for nonlinear classification decision boundaries [16].The hidden layers each contain five nodes, whereas the output layer has a single node.The activation functions within the hidden layers are hyperbolic tangent functions and the final output layer uses the sigmoid activation function.These functions are not typically used in deep-learning NN applications due to the problem of vanishing gradients.However, in this work a type of evolutionary reinforcement learning is used, and is therefore derivative-free.One benefit of using these atypical activation functions (such as a Rectified Linear Unit (ReLU) function) is that there have been recent efforts to formally verify NNs with ReLU activation [9].This could help increase the confidence that the system would behave as intended over the entire input domain.The inputs to the NN are d, ψ RH , and ψ R A .These inputs as well as a bias value are combined into vector form, denoted X.The output, ŷ, is converted into a binary value, ȳ, by thresholding the output at 0.5.This allows the inputs to be separated into the two classes: avoidance and navigation.If the inputs are classified as avoidance, the avoidance subcontroller is activated, otherwise, the navigation controller is activated.The operations that occur within the NN, along with the activation in each layer, are shown in Eq. ( 21)-( 24).(Note: s 3 is a scalar due to having a single output in the last neural network layer.)A visual representation of the NN architecture and pursuer logic with FLCs are shown in Fig. 4 and Fig. 5, respectively.Note that the bias is not depicted in Fig. 4. Now that the structure of the control system has been created, the various parameters need to be found.s l = w l T X if l = 1 w l T h l-1 if l = 2, 3(21)h l = tanh (s l ) , l = 1, 2(22) +C. Learning Control ParametersIn order to learn the parameters of this system, a Genetic Algorithm (GA) was utilized.This method was chosen because the optimal control is unknown and therefore actual error data are not available for supervised learning methods.Thus, a derivative-free search method with a critic, or cost, function is needed.The parameters to be learned are the membership function center points c and U, the gains for the FLCs k in and k out , and the NN weights w l .This constitutes an individual in the population that is being optimized by the GA.Better performing (i.e.lower cost) individuals are selected for mutation and recombination in order to iteratively refine the population towards a global optimum.The actual GA used was the default algorithm included in the Global Optimization Toolbox from Mathworks [17].To facilitate this, a simulation environment was created based on the vehicle properties described in Section III.The control system structure was defined and tested over a number of initial conditions.These conditions were such that the target vehicle started at fifty different values of ψ R A P in [-π, π], two different values for ψ P at (0, π), and an initial separation of 375 m.This gives one hundred trials of initial condition values for a single test run.Since the dynamics of the vehicle are functions of ψ (either directly or through coupling), these initial conditions were deemed sufficient for covering the state-space.A cost function is needed to assign a performance metric value to a particular test run.In terms of a GA, this is the fitness value for a particular individual in the population.The only difference here is that the value is being minimized, so it is conventionally labeled as a cost.The cost function is shown in Eqs. ( 25)-( 29).The value used for the collision distance, d C , was 10 meters.J = 1 n n i=1 τ i t=0 E x i (t) 2 + E y i (t) 2 + 100000 k C i (25) E x i (t) = k PR (t)e x i (t)(26)E y i (t) = k PR (t)e y i (t) (27) k PR (t) = 1000 if in PR 1 otherwise (28) k C i = 1 if d < d C 0 otherwise (29)This cost function therefore penalizes the pursuer for each time step that it exists within the PR.Although there is no explicit term that rewards it for reaching the RR, this happens implicitly by stopping the simulation once it is reached.Once the simulation terminates, the error terms in Eqs. ( 26)-( 27) will no longer be accumulating.This is related to the path length, and therefore the final time, due to having a constant speed.Note that the simulation does have a maximum time limit which will terminate that trial if reached.The final term in Eq. ( 25) is a penalty for collisions.If there is a collision then the pursuer is assigned an extremely heavy penalty.This value is arbitrary, but was selected after empirical examination of early runs of the optimization algorithm.The value chosen was deemed sufficient to dominate the cost if a collision were to occur.An occurrence of a collision also constitutes a terminating condition for the simulation.Overall, this cost function describes a controller that drives the pursuer to get to the RR as quickly as possible while avoiding the PR and collisions.The cost function value is stored for each trial in a particular test run and then averaged to get the final cost for that particular individual in the population. +D. Uncertainty in Target Vehicle MovementOne concern with learning models numerically is overfitting to the training data.Additionally, although the training data may offer relatively dense coverage of the state-space, the cost function may also inadvertently drive the system towards overfitting.To help combat this, uncertainty was added to the system in the form of noise to the target vehicle's input.Initially, the system was tested in a scenario where the target vehicle's input, u T , was set to be zero (i.e. it flies along a straight path).This ensured that the problem could be solved for a more simplified scenario.To help improve the robustness of the developed system, uncertainty was then added to the target's turn rate.This helped ensure better generalization in the learned model by accounting for possible states the vehicle can reach.This was achieved by augmenting Eq. ( 5) with a noise term, ρ, as shown in Eq. (30).ψ T = u T + ρ(30)The noise was added by sampling a truncated normal distribution, as described in Ref. [18], with zero mean and truncation points at ± ψ max T .The standard deviation of the distribution was chosen to be ψ ma x T +3. While this was used throughout the scope of this study, there are potential benefits to testing the system with larger standard deviation values.For example, an increased standard deviation would force the controller to be more conservative due to an increased probability of higher turn rates. +V. ResultsThe methodology described in Section IV was implemented for a number of additional test cases.An example of the cost from Eq. (25) during the learning portion of the development is shown in Fig. 6.As the GA modified the parameters of the system, it can be seen that the mean fitness value decreased.This shows the search honing in on more fit individuals as higher cost solutions are removed.The final values found for all weights, gains, and membership functions are not displayed, but can be provided by the authors upon request. +Fig. 6 Average and best individual cost vs. generation during GA learningFigure 7 shows the performance of the system, with no turn rate uncertainty, found through learning.The same initial conditions, as described in Section IV.C, are shown.It can be seen that in all cases, the pursuer successfully reaches the RR and avoids the PR.To further validate the system performance and generalization capabilities, many more initial conditions were tested that were not seen during training.By increasing the number of trials tested, the performance of the system with a higher granularity with respect to the state-space can be examined.A total of 10,000 trials were run with varying initial relative positions and headings between the target and pursuer.It was discovered that there were collisions in 1.48% of these cases.These collisions occurred in a few concentrated ranges of initial relative angles: {R| -179°≤ ψ R A P ≤ -172°, -143°≤ ψ R A P ≤ -137°, 171°≤ ψ R A P ≤ 179°}.These collisions could be due to the NN selecting the wrong subcontroller, the FLC subcontrollers learning incorrect behaviors, or some combination of both.To further improve the performance of the controller, additional training could be performed in order to expand coverage of the input space.Outside of the collision cases, the pursuer always successfully reached the RR. +Fig. 7 Simulation results for a test run of the best individual found during learning (no uncertainty)Once the system had been trained for the case with no uncertainty, uncertainty was added to the target UAS's turn rate, as described by Eq. (30).The same parameters were then again trained using a GA.For this second round of training, the final solution from the no uncertainty case served as an individual in the initial population for the uncertainty case.Figure 8 shows the system performance after being trained with uncertainty in the target UAS's turn rate.Note that although the system was trained such that the target UAS was performing a random walk, in order to compare these results to those shown in Fig. 7, the same test cases were used.That is, there is no uncertainty in the target's turn rate in the test cases shown.It can be seen from the results that the system is still able to reach the RR in all trials tested.The same 10,000 trials were also evaluated as in the no uncertainty case.No collisions occurred in these testing instances.Thus, the additional training with uncertainty in the target's turn rate lead to a more robust control system. +VI. ConclusionIn conclusion, a method was presented for obtaining near optimal solutions to a one-versus-one Tail-Chase UAS scenario by combining an NN supervisor with lower level FLC control.A GA was utilized in order to learn the parameters of both the FLCs and the NN with no precise knowledge about the target UAS or its PR.The methodology could be extended for any arbitrary one-versus-one Tail-Chase scenario by adjusting the RR, PR, and cost function.Overall, this represents a novel method for finding discontinuities in a complex input space and then enacting appropriate low-level continuous control.One possible application for this type of control includes leader-follower formation control in swarming agent systems.This approach could potentially be scaled up for one-on-multiple or even multiple-on-multiple UAS scenarios.This would require other supervisory and complementary functions in order to provide target assignments to the pursuing vehicles.Also, to improve scalability, another function is needed to ignore potential target vehicles outside of a sphere of influence.This could also be written into the target assignment protocol.Additionally, the cost function would need to be modified in order to account for multiple penalty regions.To improve the confidence in correctness, the Neural Network could use other activation functions that are more amenable towards formal verification, such as ReLU.Being able to definitively show that there are no collision cases and the pursuer always reaches the reward region, regardless of state value, is invaluable for ensuring system correctness and effectiveness.The approach could also be extended by considering other inner loop control methods for different scenarios.Furthermore, in cases where the inner loop control reduces to a multiple classification problem (discrete control), the NN output could instead be used directly for control using a Softmax function.Lastly, in an effort to validate the algorithms and methods presented within, the algorithms could be tested on actual flight systems.These flight systems would be small, fixed-wing RC aircraft controlled by sending waypoint updates from a ground station.The test results would then be collected, examined, and compared with the simulation results.µ= membership value ρ = turn rate noise τ = simulation termination time ψ = heading of vehicle ψ = turn rate of vehicle c = input membership function center points d = distance e = error E = scaled error g = gravitational acceleration h = Neural Network intermediate output (post-activation) J = cost function value k = scaling constant n = number of simulation runs N = load factor R = turn radius s = Neural Network layer aggregate t = simulation time u = control output U = output membership function center points v = UAS speed w = Neural Network weights x = Cartesian x location x = speed of vehicle along Cartesian x direction X = Neural Network input vector y = Cartesian y location y = speed of vehicle along Cartesian y direction ŷ = raw Neural Network output ȳ = piecewise decision boundary of Neural Network output Subscripts avoid = avoidance mode C +Fig. 4 Fig. 545Fig. 4 Hybrid NN and FIS structure for supervisory control +Fig. 88Fig. 8 Simulation results for a test run of the best individual found during learning (with uncertainty) + + + + + + + + +AcknowledgmentsThis research was conducted with Government support under and awarded by DoD, Air Force Office of Scientific Research, National Defense Science and Engineering Graduate (NDSEG) Fellowship, 32 CFR 168a. + + + + + + + + + A reachability-based strategy for the time-optimal control of autonomous pursuers + + CFChung + + + TFurukawa + + + + Engineering Optimization + + 40 + 1 + + 2008 + + + Chung, C. F., and Furukawa, T., "A reachability-based strategy for the time-optimal control of autonomous pursuers," Engineering Optimization, Vol. 40, No. 1, 2008, pp. 67-93. + + + + + Scalable value approximation for multiple target tail-chase with collision avoidance + + AHashemi + + + DWCasbeer + + + DMilutinović + + + + IEEE + + + 2016. 2016 + + + Decision and Control (CDC) + Hashemi, A., Casbeer, D. W., and Milutinović, D., "Scalable value approximation for multiple target tail-chase with collision avoidance," Decision and Control (CDC), 2016 IEEE 55th Conference on, IEEE, 2016, pp. 2543-2548. + + + + + Markov inequality rule for switching among time optimal controllers in a multiple vehicle intercept problem + + DMilutinović + + + DWCasbeer + + + MPachter + + + + Automatica + + 87 + + 2018 + + + Milutinović, D., Casbeer, D. W., and Pachter, M., "Markov inequality rule for switching among time optimal controllers in a multiple vehicle intercept problem," Automatica, Vol. 87, 2018, pp. 274-280. + + + + + Multilayer feedforward networks are universal approximators + + KHornik + + + MStinchcombe + + + HWhite + + + + Neural networks + + 2 + 5 + + 1989 + + + Hornik, K., Stinchcombe, M., and White, H., "Multilayer feedforward networks are universal approximators," Neural networks, Vol. 2, No. 5, 1989, pp. 359-366. + + + + + Approximation capabilities of multilayer feedforward networks + + KHornik + + + + Neural networks + + 4 + 2 + + 1991 + + + Hornik, K., "Approximation capabilities of multilayer feedforward networks," Neural networks, Vol. 4, No. 2, 1991, pp. 251-257. + + + + + Fuzzy logic, neural networks, and soft computing + + LAZadeh + + + + Communications of the ACM + + 37 + 3 + + 1994 + + + Zadeh, L. A., "Fuzzy logic, neural networks, and soft computing," Communications of the ACM, Vol. 37, No. 3, 1994, pp. 77-85. + + + + + Fuzzy systems are universal approximators + + L.-XWang + + + + IEEE International Conference on + + IEEE + 1992 + + + + Fuzzy Systems, 1992 + Wang, L.-X., "Fuzzy systems are universal approximators," Fuzzy Systems, 1992., IEEE International Conference on, IEEE, 1992, pp. 1163-1170. + + + + + Fuzzy systems with defuzzification are universal approximators + + JLCastro + + + MDelgado + + + + IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics) + + 26 + 1 + + 1996 + + + Castro, J. L., and Delgado, M., "Fuzzy systems with defuzzification are universal approximators," IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), Vol. 26, No. 1, 1996, pp. 149-152. + + + + + Reluplex: An efficient SMT solver for verifying deep neural networks + + GKatz + + + CBarrett + + + DLDill + + + KJulian + + + MJKochenderfer + + + + International Conference on Computer Aided Verification + + Springer + 2017 + + + + Katz, G., Barrett, C., Dill, D. L., Julian, K., and Kochenderfer, M. J., "Reluplex: An efficient SMT solver for verifying deep neural networks," International Conference on Computer Aided Verification, Springer, 2017, pp. 97-117. + + + + + Formal verification of piece-wise linear feed-forward neural networks + + REhlers + + + + International Symposium on Automated Technology for Verification and Analysis + + Springer + 2017 + + + + Ehlers, R., "Formal verification of piece-wise linear feed-forward neural networks," International Symposium on Automated Technology for Verification and Analysis, Springer, 2017, pp. 269-286. + + + + + Toward Scalable Verification for Safety-Critical Deep Networks + + LKuper + + + GKatz + + + JGottschlich + + + KJulian + + + CBarrett + + + MKochenderfer + + arXiv:1801.05950 + + 2018 + + + arXiv preprint + Kuper, L., Katz, G., Gottschlich, J., Julian, K., Barrett, C., and Kochenderfer, M., "Toward Scalable Verification for Safety-Critical Deep Networks," arXiv preprint arXiv:1801.05950, 2018. + + + + + On curves of minimal length with a constraint on average curvature, and with prescribed initial and terminal positions and tangents + + LEDubins + + + + American Journal of mathematics + + 79 + 3 + + 1957 + + + Dubins, L. E., "On curves of minimal length with a constraint on average curvature, and with prescribed initial and terminal positions and tangents," American Journal of mathematics, Vol. 79, No. 3, 1957, pp. 497-516. + + + + + UAS Collision Avoidance, Navigation, and Target Assignment in a Congested Airspace Using Fuzzy Logic + + BCook + + + TArnett + + + BRich + + + EKivelevitch + + + 2015 + AIAA Infotech@ Aerospace + 2031 + + + Cook, B., Arnett, T., Rich, B., and Kivelevitch, E., "UAS Collision Avoidance, Navigation, and Target Assignment in a Congested Airspace Using Fuzzy Logic," AIAA Infotech@ Aerospace, 2015, p. 2031. + + + + + A Fuzzy Logic Approach for Separation Assurance and Collision Avoidance for Unmanned Aerial Systems + + BCook + + + TArnett + + + KCohen + + + + Modern Fuzzy Control Systems and Its Applications + + InTech + 2017 + 12 + + + + Cook, B., Arnett, T., and Cohen, K., "A Fuzzy Logic Approach for Separation Assurance and Collision Avoidance for Unmanned Aerial Systems," Modern Fuzzy Control Systems and Its Applications, InTech, 2017, Chap. 12, pp. 225-256. + + + + + Verification of Genetic Fuzzy Systems + + TArnett + + + 2016 + + + University of Cincinnati + + + Master's thesis + Arnett, T., "Verification of Genetic Fuzzy Systems," Master's thesis, University of Cincinnati, 2016. + + + + + Approximation by superpositions of a sigmoidal function + + GCybenko + + + + Mathematics of control, signals and systems + + 1989 + 2 + + + + Cybenko, G., "Approximation by superpositions of a sigmoidal function," Mathematics of control, signals and systems, Vol. 2, No. 4, 1989, pp. 303-314. + + + + + Matlab and Global Optimization Toolbox R2016a, The MathWorks Inc + + + 2016 + + + Matlab and Global Optimization Toolbox R2016a, The MathWorks Inc, 2016. Https://www.mathworks.com/help/gads/genetic- algorithm.html. + + + + + Simulation from the Normal Distribution Truncated to an Interval in the Tail, GERAD + + ZIBotev + + + PEcuyer + + + 2016 + + + École des hautes études commerciales + Botev, Z. I., and L'Ecuyer, P., Simulation from the Normal Distribution Truncated to an Interval in the Tail, GERAD, École des hautes études commerciales, 2016. + + + + + + diff --git a/file23.txt b/file23.txt new file mode 100644 index 0000000000000000000000000000000000000000..ecff856176176750793d6ec7526a255cdea56d00 --- /dev/null +++ b/file23.txt @@ -0,0 +1,243 @@ + + + + +I. Introductionn recent years, new complex control methods have developed more quickly than the methods used to ensure their adherence to safety and performance requirements.Whereas many analytical methods have been used to prove that a classical controller can achieve stability, or other performance measures, at this time there are no standard +II. Problem DescriptionThe problem that this method is attempting to address is to analyze an FLC for the following dynamic system.The analysis will involve proving that the FLC adheres to specifications that it should always give positive feedback and be stable in the sense of Lyapunov. +A. Dynamic System DescriptionThe dynamic system that was analyzed in this study was a single degree-offreedom inverted pendulum.In this system, an input torque was supplied by a controller to keep the pendulum inverted throughout time.In this study, two controller types were developed and tested: a classical Proportional-Derivative (PD) controller and a non-linear Fuzzy Logic Controller (FLC).The configuration of the dynamic system setup can be seen in Figure 1.Whereas this system seems to be fairly simple in nature, the non-linear FLC controller imposed interesting characteristics not typically seen by a classical controller.Throughout simulation, the performance of the FLC was superior to the classical linear controller with respect to both rise and settling times.In each of these cases, experiments were conducted to show that the controller was able to meet the system requirements for positive feedback and Lyapunov stability. +B. Positive FeedbackThe first area of stability analysis conducted in this study was checking for positive feedback.In a dynamic system, a controller is created to provide negative feedback to the system.In essence, this means that the controller is not increasing the state error or error rate at any time.If the controller produces positive feedback, this is undesirable and will force the system to diverge from the desired set point.If positive feedback occurs, the system is not necessarily unstable, as it could push the system into a limit cycle; however, this is not a desired output of the system since it worsens performance. +C. Lyapunov StabilityThe second area of stability analysis performed was testing for Lyapunov stability.This testing criteria is achieved by selecting a candidate function that is positive semi-definite and often includes a notion of energy in the system.To check stability in the sense of Lyapunov, the time derivative of the candidate function is found.If this function can be proven to be negative-definite over all conditions of interest, the system will be asymptotically stable 2,3 .Therefore, over time the controller will force the system to converge to zero error and error rate.Similarly, if the function is negative semi-definite, the system is stable in the sense of Lyapunov and will remain in a set region around the desired position.Lastly, if the function is found to be positive for any conditions of interest, the stability of the system cannot be concluded using the candidate function chosen.For a given system, there is no guarantee that such a function that satisfies these conditions can be found, or even exists.This does not necessarily imply that the controller is unstable; this simply means that stability in the sense of Lyapunov cannot be proven. +D. SMT SolversThe properties of the controllers were tested for positive feedback and Lyapunov stability utilizing a Satisfiability Modulo Theories (SMT) solver.SMT solvers are tools that extend the ability to solve satisfiability problems to higher order logics.Satisfiability problems are problems that involve a search for variable assignments such that a formula of interest evaluates to true.When the variables have binary truth assignments, this search is known as the Boolean Satisfiability Problem.SMT solvers extend this to include First Order Logic (FOL) sentences.FOL extends propositional logic to include quantification of variables and predicates, and can include theories such as real and integer arithmetic, bitvectors, and arrays.Z3 8 is one such solver developed by Microsoft and was used throughout the scope of this work.It has the ability to solve SMT queries involving non-linear real arithmetic and was the top performing solver in competition-wide scoring for the 2016 SMT competition 11 . +III. Proposed SolutionPrior to developing the controllers for the inverted pendulum system, a model for the single degree-of-freedom inverted pendulum was developed using a Lagrangian approach.In this approach, energy methods are used to develop the equation of motion.This equation of motion was found to be:1 3 𝑚𝐿 2 θ ̈-𝑚𝑔 𝐿 2 sin θ = 𝑇 (1)where is the acceleration due to gravity, is the mass of the pendulum, is the length of the pendulum, and is the applied torque (i.e. the output from the controller).The physical parameters had values of 9.81 2 , 1 , and 1 , +A. PD Controller DevelopmentThe first controller that was created was a linear PD controller that has the form:𝑇 = 𝐾 𝑝 𝑒 𝜃 + 𝐾 𝑑 𝑒̇𝜃(2)where and are real-valued gains specified by the designer, is the error, ̇ is the error rate, and is the controller output.The set point used was the equilibrium point θ = 0 which gives the relationships = - and ̇ = - ̇.Eq. ( 1) was then linearized about θ = 0.The gains were selected using traditional tuning techniques to be: = 113, = 7.6. +B. Fuzzy Controller DevelopmentNext, a Mamdani-type Fuzzy Logic Controller, based on Fuzzy Set Theory introduced by Lofti Zadeh in the 1960's 9 , was developed.A Fuzzy controller uses input classification and rules associating different Fuzzy sets to produce complex and highly non-linear inputoutput relationships.Using a Fuzzy controller has many benefits including the ability to be developed by a designer that has some expert knowledge about how the system should desirably act.Fuzzy controllers are also universal approximators and can approximate any given function to an arbitrary degree of accuracy.This is useful when there is some unknown optimal control surface.Due in part to Fuzzy controllers' ability to approximate some (potentially non-linear) optimal control surface, the input-output relationship can be complicated, and their nonlinear nature precludes analysis using traditional design methods.Due to this, the authors aim to develop an approach for analyzing these inputoutput relationships.This is necessary for these types of systems to ensure they perform as expected prior to being implemented into safety-critical systems.In each Fuzzy system, the user must define a rule base to govern the action of the controller for a set of inputs.The Fuzzy system used in this study was trained using a combination of expert knowledge and Genetic Algorithms, both of which are beyond the scope of this paper.The rule base governing the FLC can be seen in +C. Conversion to Piecewise Polynomial RepresentationIn order to simplify the input-output relationship, the FLC was constrained and converted to a piecewise polynomial system using previously developed methods 10 .The constraints for the FLC are as follows:  Triangular membership functions: This constraint limits the membership function shape to be triangular.The result is a piecewise linear membership function of the input value for a particular membership function. Fuzzy partitioning: The membership functions are partitioned such that the end points for the triangle defining a membership function coincide with the center points of neighboring membership functions.A visualization of this is shown in Figures 3 and4 where the dashed lines represent the locations of the membership function center points.The center points are defined as the input value that gives the maximum membership value of one for that particular function. Normalization: The input and output sets were defined on a normalized range such that any input value is within positive or negative unity.Gains are used to reduce the inputs to this range and a gain is applied to the output as well.Also, if a value is outside of the universe of discourse, it is reduced to the boundary value. Product method: For rule associations, the product method was used to scale the contributions of each active membership function based on its membership value for a given input.An active membership function is one that has a non-zero membership value for a given input value.For example, if a rule in the FLC is described by If 1 is and 2 is then output is , , the product method would combine these to give = , .Where and are the membership values in the and membership functions and , is the center point of the output membership function associated with and . Weighted average defuzzification: In FLCs, there are usually multiple membership functions that are active for each input.This results in multiple rules that are active as well, and is referred to as the aggregate.To resolve a crisp output from this aggregate, a process called "defuzzification" is used.For example, a common method for defuzzification is the centroid method.This method takes the centroid of the aggregate to give a single, crisp output that is most representative of the components of the aggregate.The method used in this work for defuzzification was weighted average.This method is similar to centroid, although it only needs the membership values of the active membership functions along with the center points of the corresponding output membership functions, instead of the entire geometric representation of the aggregate.This reduces computational complexity and allows for easier translation to an explicit representation while still giving similar performance to the centroid method.Eq. ( 3) shows the output of a two-input, one-output FLC using this defuzzification method in combination with the product method for rule associations.𝑦 = ∑ ∑ 𝜇 𝑖 𝜇 𝑗 𝑈 𝑖,𝑗 𝑛 2 𝑗=1 𝑛 1 𝑖=1 ∑ ∑ 𝜇 𝑖 𝜇 𝑗 𝑛 2 𝑗=1 𝑛 1 𝑖=1 (3)Here, is the output of the FLC, and 1 and 2 are the number of membership functions for the first and second inputs, respectively.Note that the denominator in Eq. (3) will always be unity and there will only be two active membership functions per input due to the chosen partitioning constraint.Using these constraints, the FLC could be converted into a piecewise polynomial form.Through this conversion, the input-output relationships are simplified when compared to typical FLC implementations.The input-output relationship of the FLC has the following form:𝑇 = 𝐾 1 𝑒 𝜃 + 𝐾 2 𝑒̇𝜃 + 𝐾 12 𝑒 𝜃 𝑒̇𝜃 + 𝐾 𝑐(4)where the coefficients 1 , 2 , 12 , and are constant for a particular mode.The modes are defined as the regions between the center points of the input membership functions.The transitions occur such that the functions governing each mode are continuous and are inclusive on the lower boundary of the mode.The exception to this is the "upper" mode (i.e. when either input is in its "PM" and "PB" membership functions).In this mode, the domain is inclusive on both boundaries.As an example of the structure of the piecewise polynomial representation of the FLC, consider a similar FLC that only has three membership functions instead of the seven shown in Figures 3 and4.These membership functions, for both inputs, would have center points at {-1,0,1}, and the system would have four total modes.When either input crosses zero, the system would transition to a different mode that is governed by a continuous function in the form of Eq. (4).A visualization of this 4-mode system is shown in Figure 5.The structure is the same for the FLCs in this work, with the difference being that since seven membership functions are used for each input, the total number of modes is 36. +D. Formalizing Constraints into First Order LogicAfter the controllers were developed, the system equations and the constraints for each stability criteria for both controllers were formalized into FOL sentences.Using these FOL sentences, the equations need to be proven to hold over the entire bounded domain.This can be done by negating the entire statement and then proving that the statement and its constraints are unsatisfiable.As previously mentioned in the description of positive feedback requirements, the controller shall always drive the error and error rate towards zero.This expression can be formalized using the following FOL sentence:∀𝑒 𝜃 ∀𝑒̇𝜃 ( ((𝑠𝑔𝑛(𝑒 𝜃 )˄ 𝑠𝑔𝑛(𝑒̇𝜃)) → 𝑠𝑔𝑛(𝑇)) ˄ ((¬𝑠𝑔𝑛(𝑒 𝜃 )˄¬𝑠𝑔𝑛(𝑒̇𝜃)) → ¬𝑠𝑔𝑛(𝑇)))(5)where refers to a predicate that resolves signed real variables to Boolean values.In this case, is inclusive of zero on the positive side.This expression can be verified by examining the controller input-output equation(s).This approach was tested for both the PD and Fuzzy controllers.For the case of checking that the system is stable in the sense of Lyapunov, a candidate function was created to meet the set criteria.Our candidate function is as follows:𝑉(𝑥̅ ) = 1 2 𝑥 2 2 + 𝑎(1 -cos 𝑥 1 )(6)where 1 = and 2 = ̇.The coefficient had values of 334 for the PD case and 243 for the FLC case.This function meets the constraints for a candidate Lyapunov function that (̅ = 0) = 0 and (̅ ≠ 0) > 0 and was created with intuition and minor testing using simulation runs.Next, ̇(̅ ) can be found using the following expressions:𝑥̇1 = 𝑥 2(7)𝑥̇2 = 3(𝑇 + 4.905 sin 𝑥 1 )(8)𝑉 ̇(𝑥̅ ) = [ 𝜕𝑉 𝜕𝑥 1 𝜕𝑉 𝜕𝑥 2 ] [ 𝑥 1 ẋ2 ̇](9)By taking the partial derivatives of Eq. ( 6) and substituting into Eq.( 9), we get the following two equations for the PD and FLC cases, respectively. ̇(̅ ) = 3 2 ( + 116.238 sin 1 ) ̇(̅ ) = 3 2 ( + 85.905 sin 1 )We know that the candidate Lyapunov function, ̇(̅ ) must be negative semi-definite.Thus, the constraint equations for testing stability in the sense of Lyapunov becomes:∀𝑥 1 ∀𝑥 2 (𝑉 ̇(𝑥̅ ) ≤ 0)(12)Lastly, after the constraint equations had been written in FOL, the representations of each controlled system and constraints were implemented into an SMT Solver to verify that the system always meets the specification criteria. +IV. Test Methods +A. Positive FeedbackTo verify that the linear PD controller produces negative feedback for all possible inputs, a Z3 script was developed to verify this constraint satisfaction problem.Therefore, the inputs of error and error rate first had to be declared within the Z3 architecture.This was done by utilizing the knowledge that = - and ̇= - ̇, due to the set point being = 0.The assertions that the output torque from the controller was always in the desired direction were then described.That is, if both and ̇ were positive, the torque must always be positive.On the other hand, if both and ̇ were negative, the torque must necessarily be negative.In the cases where is positive and ̇ is negative and vice versa, the desired controller output is magnitude dependent.Thus, the positive feedback specification is not applicable.By setting hard constraints on the ranges of and ̇ for each case (i.e.strictly negative or strictly positive), the constraints on the controller output could be negated, resulting in the following expression:∃𝑒 𝜃 ∃𝑒̇𝜃 ¬ ( ((𝑠𝑔𝑛(𝑒 𝜃 )˄ 𝑠𝑔𝑛(𝑒̇𝜃)) → 𝑠𝑔𝑛(𝑇)) ˄ ((¬𝑠𝑔𝑛(𝑒 𝜃 )˄¬𝑠𝑔𝑛(𝑒̇𝜃)) → ¬𝑠𝑔𝑛(𝑇)))(13)Thus, if the SMT solver was used to check if the above system is satisfiable, and found it to be true, positive feedback can occur in the system.However, if it is found to be unsatisfiable, the system is guaranteed to have negative feedback for all inputs within the analyzed domain.This means that there are no possible values that violate the original requirement stated in Eq. ( 5).Therefore, it is proven to hold over the entire domain of the inputs.Similarly, the possibility of positive feedback occurring in the Fuzzy controller case was analyzed using the same approach.The only difference here is that whereas the PD case has only one mode, the Fuzzy controller has ( 1 -1)( 2 -1) possible modes that need to be analyzed, where 1 and 2 are the number of membership functions for each respective input. +B. Lyapunov StabilityFinally, both controllers were tested for Lyapunov stability.Using the equation previously defined for ̇(̅ ) and constraining the inputs to the controller, a constraint satisfaction problem can be developed to ensure that ̇(̅ ) is negative semi-definite.The constraints on the inputs are simply normalized bounds based on the gains found during training.To check the stability specification, it is again negated which gives the following expression:∃𝑥 1 ∃𝑥 2 ¬(𝑉 ̇(𝑥̅ ) ≤ 0) (14)By proving that the above expression is unsatisfiable, we can guarantee that the system is stable.For this case, the controller output, , was simply changed based on the type and mode of the controller.The PD controller has only one mode.However, the Fuzzy controller has 36 modes.These modes were checked along with the corresponding input bounds (defined by the membership function center points) for unsatisfiability.It should be noted that Z3 cannot handle transcendental functions and the sine functions in Eqs.(10) and (11) were approximated using the first three terms of a Maclaurin series expansion -sin() ≈ - .Although this approximation is not exact, the error at the bounds of 1 is acceptable as the maximum magnitude of 1 is approximately 0.1255 .The maximum error magnitude due to this approximation is 9.72 -11 and occurs at the outer boundaries of the input space. +V. Results +A. Positive FeedbackOnce each of the above cases was developed, they were analyzed using Z3.When checking for positive feedback in the linear controller, Z3 found that the model was unsatisfiable in all cases.Therefore, we have proven for the domain ranges tested we can ensure that the system will only produce negative feedback.When checking for positive feedback in the Fuzzy controller, we again found that the system was unsatisfiable in all modes analyzed.Thus, we again showed that the FLC would always produce negative feedback to the system.As a counterexample for the FLC positive feedback check, a bad rule was intentionally placed in the rule base, as seen in Figure 6.This rule now says that if the angle error and error rate are both large in the negative direction, the controller should output a positive torque.Due to the sign conventions of angle error, error rate, and torque, this means that the pendulum is far away from the set point, is moving further away, and the torque input is such that it is pushing it even further still.By creating this bad rule, Z3 was able to identify the mode that violated this specification.This was used to ensure that the problem was formulated correctly in Z3 and to check that Z3 could identify this potential hazard in the proposed Fuzzy controller.Although positive feedback is not in itself conclusive for guaranteeing safety or stability of a controller, it is easily verified using an SMT solver.This also represents a powerful tool for design.Incorporating specifications such as these during a training or optimization process will lead to more confidence in the controller performance. +B. Lyapunov StabilityPrior to evaluating both the PD and Fuzzy controllers using the Z3 formulation, a MATLAB simulation was developed to test a range of inputs for the system.The domain used in testing was the universe of discourse for each input.More formally, = {ℝ| -0.1255 ≤ 1 ≤ 0.1255, -1.628 ≤ 2 ≤ 1.628}.In these cases, we wanted to ensure that the derivative of our candidate Lyapunov function was negative semi-definite for all points tested.Upon evaluation over a range of values for 1 and 2 , it was concluded that both the PD and Fuzzy controllers satisfied this condition.A figure showing the results of the simulated candidate function for the FLC can be seen in Figure 7.Next, the PD controller was tested for stability in the sense of the Lyapunov using Z3.When the program ran the same range of inputs that were analyzed in the MATLAB simulation, Z3 found conditions that the simulation points missed.Whereas the simulation analyzed a large number of discrete input values, it did not analyze the entire continuous range of inputs.Therefore, the simulation missed points that make ̇(̅ ) positive (i.e.violate the system requirement for stability).Similarly, the Fuzzy implementation had the same outcome.In the MATLAB simulation, ̇(̅ ) was negative semidefinite over the range of inputs.In addition, dynamic simulations were run over a range of initial conditions for and ̇.These initial conditions were equally spaced over the domain .The time histories of both and ̇ for each of these runs for the FLC controlled system converge to zero, as shown in Figure 8.Additionally, a phase portrait is shown in Figure 9 for every tenth run.These results seem to indicate that the controller is stable within this region, but this method cannot be used to prove its stability within a finite amount of time.Although the simulation results are promising, when realized and tested using FOL in Z3, eight modes were found to be satisfiable.Meaning values for and ̇ were found that drove ̇(̅ ) positive and violated the specification.It is important to note that this does not mean that the controlled system is unstable; it simply means that no conclusions can be drawn about the system stability using our selected candidate Lyapunov function.If a candidate function could be found that does satisfy these requirements, we could then conclude that the system is stable (i.e.asymptotically or in the sense of Lyapunov).However, the power to identify these conditions that would otherwise be missed by the simulation results provides value when making claims about a system's adherence to its requirements.Because stability needs to be a safety property (a property that holds over the entire space), one can conclusively say that it holds, or identify cases where it does not.This is invaluable in the design phase of the controller development, as it can lead the designer to be able to correct those modes.The modes that were identified as violating the specification are shown in the Fuzzy rule base in Figure 10.The different shades for the membership functions are due to the fact that each mode contains four membership functions.This is due to the partitioning constraint detailed in Section III.C.Therefore, the darker the shade, the more modes that contain that membership function have values violating the specification.In addition to the information about which mode violated the specification, the values for and ̇ that satisfied the negated specification are known.Although the Lyapunov candidate function was not able to say that the controller was definitively stable in these modes, it may very well be.After identifying these modes, several simulations were run within these modes to gather more evidence that the controller is in fact stable.The phase portrait of the system response within these modes is shown in Figure 11.The initial conditions for and ̇ were set to be 25 equally spaced values from one boundary of the domain to the other within each mode for a total of 5,000 runs.Each simulated case ran for the same specified time as the responses shown in Figure 8.This was to show the overall stability, as opposed to ending the trace once it reached the boundary of the mode.These simulation traces similarly do not prove that the system is stable, but the simulation results do contribute to the overall body of evidence.Combined with the SMT methods mentioned, these inspire high confidence in the controller's ability to stabilize the system effectively.Although the chosen Lyapunov candidate function was not able to prove the Fuzzy controller was stable in all modes, being able to identify those inconclusive modes represents an important capability.The controller could be retrained using some sort of optimization algorithm and then reevaluated until the Lyapunov stability requirements are met.In addition, this training loop could be closed if the information about the violating modes could be utilized to guide the next optimization attempt. +VI. ConclusionUsing the approaches described in this study, we have shown that utilizing an SMT solver is a valid method for analyzing several desirable (and undesirable) characteristics of both a linear and non-linear Fuzzy Logic controller.The SMT solver offers conclusive proofs for whether a requirement holds, given that it can be properly represented in formal logic.Both controllers were proved to not produce positive feedback by showing that conditions that do give positive feedback do not exist (i.e. they are unsatisfiable).Lastly, although simulation runs provided negative semidefinite values for ̇(̅ ) and seemed to indicate stability from time histories and phase portraits, Z3 showed conditions exist where ̇(̅ ) is positive.Therefore, the Fuzzy controller cannot be proven stable using this candidate function.An important caveat is that this method can be very conservative, and the controller may very well be stable while not being able to satisfy the Lyapunov criteria for candidate functions that are selected.This represents the main difficulty with this method as selecting a candidate function is not trivial.Also, when determining performance and stability for dynamic systems, simulation runs do contribute towards the overall body of evidence that the system will perform as expected.The simulation runs that were performed indicate that it is likely that the controller is indeed stable within the boundaries evaluated.Overall, these techniques can aid with the design phase for FLCs of this type, since unstable modes can be identified and corrected using optimization protocols.There are several opportunities for extending this work.One of which is exploring different methods of candidate function selection.Although the results are inconclusive for proving Lyapunov stability given the prescribed candidate function, another method could be used to potentially find parameters for the candidate function that prove stability.In systems where certain unknown parameters affect the truth evaluation of a constraint satisfaction problem, it is sometimes possible to perform parameter synthesis to find satisfying assignments.Consider a candidate Lyapunov function of the following form: (̅ ) = 1 (1 -cos 1 ) + 1 2 2 2 2(15)The coefficients 1 and 2 are unknown but could potentially be found by searching with SMT.This could be done by converting this into an exists-forall formula.These formulas have the form ∃. ∀.(, ), meaning that there exists some values for that satisfy for all values of .This can then be rewritten as ∀.(, ) with constraints on in order to remove the existential quantifiers and search for satisfying values.The main issue with this approach is the computational expense of doing so.Others have done this for controllers with few modes, but it is difficult to find parameters that globally satisfy the Lyapunov criteria for all of the modes in a given FLC.Finding separate parameter values within each mode could also be accomplished, but could lead to a weaker stability proof.At this time, some of these techniques have been explored, but have yielded inconclusive results.Being able to find these parameters would be powerful, as they could then be used to conclusively prove stability for the FLC.Figure 1 .1Figure 1.Single degree-offreedom inverted pendulum +Figure 2 .2In this figure, the linguistic terms of each entry and their respective abbreviation are as follows: negative big [NB], negative medium [NM], negative small [NS], zero [ZE], positive small [PS], positive medium [PM], and positive big [PB].The inputs are the pendulum angle error, [e], and the angle error rate, ̇ [de/dt].The membership functions for the input sets are shown in Figures 3 and 4. +Figure 2 .Figure 3 .Figure 4 .234Figure 2. FLC rule base de/dt NB NM NS ZE PS PM PB NB NB NB NB NM NM NS ZE NM NB NB NM NM NS ZE PS NS NB NM NM NS ZE PS PM ZE NM NM NS ZE PS PM PM PS NM NS ZE PS PM PM PB PM NS ZE PS PM PM PB PB PB ZE PS PM PM PB PB PB e +Figure 5 .5Figure 5. Piecewise polynomial representation of 4-mode FLC with continuous modes and discrete transitions +Figure 6 .Figure 7 .67Figure 6.FLC rule base with bad (highlighted) rule de/dt NB NM NS ZE PS PM PB NB PB NB NB NM NM NS ZE NM NB NB NM NM NS ZE PS NS NB NM NM NS ZE PS PM ZE NM NM NS ZE PS PM PM PS NM NS ZE PS PM PM PB PM NS ZE PS PM PM PB PB PB ZE PS PM PM PB PB PB e +Figure 8 .Figure 10 .810Figure 8.Time histories for (left) and ̇ (right) over range of initial values for FLC controlled system +Figure 11 .11Figure 11.Phase portrait for 5,000 simulation runs within modes that violated Lyapunov specification + Downloaded by NASA AMES RESEARCH CENTER on January 9, 2017 | http://arc.aiaa.org| DOI: 10.2514/6.2017-1773 + + + + +AcknowledgmentsT. Arnett thanks Matthew Clark, Dr. Kuldip Rattan, and Jon Hoffmann of AFRL for their help and previous work in the area.T. Arnett and B. Cook also thank Dr. Kelly Cohen for his invaluable advice and support.This work was partially supported by the Dayton Area Graduate Studies Institute. + + + + + + + + + Design and analysis of a multistage Fuzzy PID controller + + KSRattan + + + MAClark + + + JAHoffman + + + + American Control Conference (ACC) + Chicago, IL + + 2015. 2015 + + + + K. S. Rattan, M. A. Clark and J. A. Hoffman, "Design and analysis of a multistage Fuzzy PID controller," American Control Conference (ACC), 2015, Chicago, IL, 2015, pp. 5726-5731. + + + + + The general problem of the stability of motion + + AMLyapunov + + + + International Journal of Control + + 55 + 3 + + 1992 + + + Lyapunov, A.M., The general problem of the stability of motion. International Journal of Control, 55(3), pp.531-534, 1992. + + + + + Dynamic Systems Recitation 6 + + MDahleh + + + 2003 + Cambridge, MA + + + Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology + + + Lyapunov Methods. unpublished + Dahleh, M., "Lyapunov Methods," Dynamic Systems Recitation 6, Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, 2003 (unpublished). + + + + + Stability analysis of discrete-time fuzzy dynamic systems based on piecewise Lyapunov functions + + GFeng + + + + IEEE Transactions on Fuzzy Systems + + 12 + 1 + + 2004 + + + Feng, G., Stability analysis of discrete-time fuzzy dynamic systems based on piecewise Lyapunov functions. IEEE Transactions on Fuzzy Systems, 12(1), pp.22-28, 2004. + + + + + Stability of a Fuzzy Logic Based Piecewise Linear Hybrid System + + AaronWSeyfried + + + 2013 + + + Wright State University + + + Thesis + Seyfried, Aaron W. "Stability of a Fuzzy Logic Based Piecewise Linear Hybrid System", Thesis, Wright State University, 2013. + + + + + A multiple Lyapunov function approach to stabilization of fuzzy control systems + + KTanaka + + + THori + + + HOWang + + + + IEEE Transactions on fuzzy systems + + 11 + 4 + + 2003 + + + Tanaka, K., Hori, T. and Wang, H.O., A multiple Lyapunov function approach to stabilization of fuzzy control systems. IEEE Transactions on fuzzy systems, 11(4), pp.582-589, 2003. + + + + + A new fuzzy Lyapunov function approach for a Takagi-Sugeno fuzzy control system design. Fuzzy sets and systems + + BJRhee + + + SWon + + + 2006 + 157 + + + + Rhee, B.J. and Won, S., A new fuzzy Lyapunov function approach for a Takagi-Sugeno fuzzy control system design. Fuzzy sets and systems, 157(9), pp.1211-1228, 2006. + + + + + + + SoftwareZ3 + + + Package + + + 2014 + Redmond, WA + + + Ver. 4.3.2, Microsoft + Z3, Software Package, Ver. 4.3.2, Microsoft, Redmond, WA, 2014. + + + + + Fuzzy Sets + + LAZadeh + + + + Information and Control + + 8 + + 1965 + + + 9 Zadeh, L. A., "Fuzzy Sets." Information and Control 8, 1965, pp. 338-353. + + + + + Piecewise Affine Hybrid Automata Representation of a Multistage Fuzzy PID Controller + + MAClark + + + KSRattan + + + + + 11th International Satisfiability Modulo Theories Competition SMT-COMP 2016 + 2014 AAAI Spring Symposium Series + Palo Alto, CA + + 2014 + + + Scoring results + Clark, M.A. and Rattan, K.S., "Piecewise Affine Hybrid Automata Representation of a Multistage Fuzzy PID Controller", 2014 AAAI Spring Symposium Series, Palo Alto, CA, 2014. 11 "Scoring results", 11th International Satisfiability Modulo Theories Competition SMT-COMP 2016. [url: http://smtcomp.sourceforge.net/2016/SMT-COMP-2016.pdf]. + + + + + + diff --git a/file24.txt b/file24.txt new file mode 100644 index 0000000000000000000000000000000000000000..0e81e223d81c0e65afa67e2ef242f3df5aa97285 --- /dev/null +++ b/file24.txt @@ -0,0 +1,729 @@ + + + + +LIST OF FIGURES +INVESTIGATING THE NATURE OF AND METHODS FOR MANAGING METROPLEX OPERATIONSStephen Atkins, 1 Brian Capozzi, 1 Jim Hinkey, 1 Hunsi Idris, 2 and Kent Kaiser 1 Ames Research Center +INTRODUCTION +Motivation and ObjectivesA combination of traffic-demand growth, Next-Generation Air Transportation System (NextGen) technologies and operational concepts, and increased utilization of regional airports is expected to increase the occurrence and severity of coupling between operations at proximate airports.These metroplex phenomena constrain the efficiency and/or capacity of airport operations and, in NextGen, have the potential to reduce safety and prevent environmental benefits.Without understanding the nature of metroplexes and developing solutions that provide efficient coordination of operations between closely spaced airports, the use of NextGen technologies and distribution of demand to regional airports may provide little increase in the overall metroplex capacity.However, the characteristics and control of metroplex operations have not received significant study.This project advanced the state of knowledge about metroplexes by completing three objectives:• We developed a foundational understanding of the nature of metroplexes.• We provided a framework for discussing metroplexes.• We suggested and studied an approach for optimally managing metroplexes that is consistent with other NextGen concepts. +Scope and Report OrganizationThe goal of this report is to summarize all of the project work, including new work accomplished since the last project report in May 2009.Following this introduction section, the report is organized into the following sections:Sections 6 -0 Describes recent research related to the Metroplex Planner, mostly not presented in prior reports Sections 0-5 Briefly summarizes earlier project work that has been presented in detail in prior reports Sections 6 -0 Describes recent research related to the Metroplex Planner, mostly not presented in prior reports Section 0 Describes research on segregation of metroplex flows by aircraft speed rather than airport, research that was not included in prior reports Section 14 Presents valuable information-describes the aircraft model-developed within the project Section 0Summarizes the project conclusions and suggests important future work +PublicationsThe project produced numerous publications [1][2][3][4], as well as many planned publications based on work conducted during the project:• Brian Capozzi, Stephen Atkins, and Jim Hinkey: Comparison of MILP vs. greedy FCFS scheduler performance on a select set of metroplex interactions.ATIO, 2010: This paper explores the conditions under which the "optimal" solution provides significant benefits relative to a greedy search algorithm and the computational performance of different linearization techniques.• Brian Capozzi, Stephen Atkins, and Jim Hinkey: Hybrid Genetic Algorithm/Mixed Integer Linear Program Planner.AIAA Modeling and Simulation Conference, 2010: This paper describes a hybrid architecture combining a genetic algorithm that stochastically searches over the binary variable space with a pure linear programming kernel.By exercising ] this planner on a set of representative problems, we compare computational time and objective function values obtained against those obtained with the full mixed integer linear program (MILP) planner and a greedy first-come, firstserved (FCFS) baseline.• Husni Idris: Improving Metroplex Operations Efficiency Using Speed Segregation and Trajectory Flexibility.ICAS (presentation only), 2010.• Seongim Choi and Brian Capozzi: Application of a mixed integer linear program planner to understand scheduling requirements associated with different super-density operations (SDO) route topologies.Unknown conference, 2010.This paper discusses different route topologies under different uncertainty models to try to abstract out the "optimal" schedule behavior to set requirements for future SDO scheduling algorithms.Using the Los Angeles International Airport, Los Angeles, California (LAX) as the focus area, the paper contrasts 1-, 2-, and N-point schedulers. +METROPLEX OBSERVATIONS +Literature SurveyTo provide a foundation for conducting original, foundational research, the project started with a significant literature review activity that produced a report summarizing the survived documents [5].Overall, metroplex issues and traffic management have not received significant study.The Federal Aviation Administration (FAA) has produced numerous traffic forecasts that predict growth in the number of metropolitan regions with multiple busy airports.The use of regional airports, including the impact of very light jets and affordable general aviation avionics, is a significant uncertainty in the metroplex forecasts.At the time of the report, air taxi, fractional ownership, very light jets, and general aviation growth were important topics.Economic and other changes have since made these topics less critical to metroplex research.Literature on metroplex definitions, which is limited, and emergence of regional airports were reviewed.Literature on airspace redesign, such as the FAA's New York Airspace Redesign Project, departure management from airports that share departure routes/fixes, and managing arrival/departure trade-offs was reviewed for applicability to metroplexes.Additional topics included metroplex simulation capabilities, Next-Generation Air Transportation System (NextGen) operational changes, and the impact of precision trajectories. +Site VisitsTo ensure research was relevant to the real-world metroplex issues, the project team visited a significant number of air-traffic-control (ATC) facilities in four different potential metroplexes to collect current operational information about metroplex phenomena and current management techniques: The information obtained during these site visits was documented in a contract report [6] and site visit notes.Notes from all of our site visits were delivered to the National Aeronautics and Space Administration (NASA) informally (i.e., not as an official contract deliverable).Moreover, the information gathered was used to formulate our generalization of metroplex phenomena and management approaches. +San Francisco Bay Area MetroplexThe project focused on the San Francisco Bay Area metroplex (Figure 1), which contains three large airports-San Francisco International Airport (SFO), Oakland International Airport (OAK), and San Jose International Airport (SJC); four regional airports-Hayward (HWD), San Carlos (SQL), Palo Alto Airport (PAO), Reid-Hillview Airport (RHV); and Moffett Federal Airfield, Moffett Field, California (NUQ).Half Moon Bay Airport (HAF) and Livermore Airport (LVK) do not appear to significantly interact with the other airports, possibly because they are separated from the other airports by terrain (hills).SFO and OAK are separated by 10 miles, while SJC is 30 miles to the south.A conference paper [1] and a NASA Airspace Systems Program Technical Interchange Meeting (TIM) presentation [7] contain considerable detail about the phenomena observed at this metroplex.Figure 2 shows radar track data of arrival and departure flows for the three largest of these airports during a single day, while Figure 3 shows a typical pattern during South-East Plan operations. +Generalized Metroplex ObservationsAll of the metroplex phenomena observed to date result from a limited resource being shared by operations at different airports.Sharing of several types of resources was observed.Specific points in the local airspace, such as arrival fixes and departure fixes, may be shared.Downstream resources may be shared, resulting in departure operations being subject to combined Traffic Management Initiatives (TMIs) such as miles in trail (MIT).Specific local trajectories may be shared, such as standard terminal arrival routes (STARs) or departure procedures (DPs, also known as SIDs).Airspace where trajectories to or from different airports cross, or would cross if not procedurally separated (e.g., using altitude restrictions), may be shared.An ATC sector is frequently responsible for a flow to or from a large airport and also traffic to or from a regional airport.Although the regional airport traffic may not otherwise interact with the traffic at the large airport, the sharing of the controller's attention may result in delay or inefficiency at one or both airports.Although not explicitly observed, other possible future shared resources include noise or environmental quotas.No interaction between visual-flight-rules (VFR) and instrument-flight-rules (IFR) traffic at different airports was observed.No metroplex impacts on safety were observed.Metroplex phenomena are site-specific and detailed.Since the phenomena depend on the airport configurations and procedures in use-and there are a large number of combinations of these configurations and procedures-there are many metroplex configurations, each with distinct phenomena.Several properties were observed to contribute to the presence of metroplex phenomena:• The geometry of the airports-the distance between airports and relative runway orientation-can affect metroplex phenomena.• Terrain and noise restrictions contribute to the presence of metroplex phenomena by reducing the available airspace near the airports.• Wind direction and speed (and also noise considerations) can affect the runway configurations in use at each airport.• Visibility (i.e., whether or not visual approaches are allowed) determines procedures in use at each airport.Visibility also affects the volume of IFR traffic.For example, HWD, OAK, and SJC serve numerous flights that typically operate VFR but, when the weather necessitates, operate IFR.• Time of day can affect the procedures in use, for example if noise restrictions are different between day and night.• Limited procedures, such as overnight when more strict noise restrictions are in effect, result in metroplex phenomena that do not occur when other procedures are available. +Generalized Metroplex Management ApproachesMetroplex phenomena are visible in current operations, but they result in only small delays (except possibly at New York); in many cases, however, the trajectories are less efficient.Current methods for managing metroplex phenomena can be organized into approaches that 1) spatially decouple the airports, 2) temporally coordinate the use of shared resources, or 3) modify the airspace to minimize the metroplex phenomena.Spatial decoupling uses predefined, conflict-free routes.Because of the limited horizontal extent of the metroplex area, minimum and maximum altitudes are required to allow flows to cross without flight-by-flight coordination.Based on observations to date, spatial separation is the preferred approach to manage interactions where the number of flights that exhibit the interaction is large (i.e., between large airports).This approach makes intuitive sense because spatial decoupling minimizes controller workload.The primary airport generally is given routes that exhibit minimal impact due to the presence of the other airports.Routes to and from the other large airports are then designed around, over, and under the routes of the primary airport, producing decoupled traffic but less-efficient trajectories.The Federal Aviation Administration (FAA) has built large consolidated terminal radar approach control facilities (TRACONs) in several of the regions that currently exhibit metroplex issues, including New York, Los Angeles basin, and the San Francisco Bay Area, suggesting that the FAA has already reacted to the need to manage metroplex phenomena.The creation of these large TRACONs facilitates airspace design that decouples the major airports in the region.Temporal deconfliction (i.e., coordinating the times at which individual flights may use a shared resource) appears to be the preferred approach when the number of flights that exhibit the interaction is small.This may be the preferred approach due to not wanting to over-constrain the predefined routes used for the large airports; spatially separating the routes to and from large airports and small airports with little daily traffic may require too many routes, forcing each to be contained to a smaller area.Without automation, temporal deconfliction requires significant workload to select when each flight may use the shared resource and then control each flight to comply.However, temporal deconfliction provides a solution in situations where the proximity and geometry of the airports preclude the use of static, spatial deconfliction.Temporal deconfliction also is (currently) inefficient because of the uncertainty in when each flight will actually use the shared resource, requiring some slack between operations.NextGen technologies could make temporal deconfliction much more efficient.Both spatial and temporal approaches are predefined through published procedures.These procedures are typically described in STARs, DPs (or SIDS), local standard operating procedures (SOPs), and letters of agreement (LOAs).Airport runway configurations are varied because of weather and other factors, such as balancing noise exposure.Separate procedures are defined for each combination of airport configuration, and some combinations are more efficient than others.Therefore, the coordination of runway configurations used at each airport to create the most efficient, feasible metroplex configuration adds a time-varying element to metroplex management and is the third class of current techniques used to manage metroplex phenomena.New York's use of this class of approach includes reallocating a piece of airspace between LaGuardia International Airport (LGA) and John F. Kennedy International Airport (JFK).Access to the airspace enables either airport to use an additional procedure that increases the capacity of that airport.Note that significant overlap exists between traffic management at a metroplex and at a busy single airport.The types of dependencies between two closely spaced airports, each with one runway, are very similar to those observed between runways at a single airport with multiple runways.Moreover, the approaches used to manage metroplex phenomena and busy single airports are similar.For example, managing release times to coordinate departures from metroplex airports on common or crossing trajectories is similar to using release times to achieve MIT spacing for consecutive departures to a common departure fix at a busy single airport.Metroplex management also exhibits the common phenomena that local procedures have evolved to meet specific local needs, and have been created by local individuals using their experience and preference, but have not been optimized.In summary, spatial deconfliction decouples airports, results in little or no capacity loss due to the metroplex phenomena, requires some less-efficient trajectories, and gives preference to the larger airports.Temporal coordination is currently inefficient because the point at which aircraft are controlled is often separate from the shared resource, and uncertainties (both in compliance with the controlled time and between the control point and the shared resource) require significant safety buffers.Temporal coordination permits efficient trajectories but creates a direct capacity trade-off between airports, as well as higher controller workload currently.Dynamic airspace and temporal coordination are currently underutilized.Dynamic airspace has the potential to match resource allocation to demand, where the demand characteristics vary in time. +METROPLEX DEFINITIONThe Joint Planning and Development Office (JPDO) Concept of Operations for the Next-Generation Air Transportation System (NGATS) [8] has defined a metroplex as "a group of two or more adjacent airports whose arrival and departure operations are highly interdependent".Although easy to understand, this definition does little to describe metroplex characteristics or suggest methods for managing a metroplex.Precisely drawing a line around a metroplex is not likely to be a tremendously useful way to define the metroplex.For example, operations at one airport within the physical boundary of a metroplex might be decoupled, for some reason, from operations at the other airports.Two metroplexes might overlap, sharing some airports but not others.A good definition must address the need to which it is applied.In this project, the definition should help guide further research and coordination with other NGATS research.Consequently, we focused on definitions that captured the type and extent of the interdependencies rather than the metroplex geometry.A metroplex definition should provide a common foundation for discussing metroplexes.Furthermore, the definition should be useful to locate, quantify, and compare metroplexes by measuring their type and magnitude.Lastly, the definition should provide insight into methods for managing metroplex traffic. +Qualitative Metroplex DefinitionTo facilitate modeling the effect of metroplexes in Next-Generation Air Transportation System (NextGen) and studying concepts for managing metroplex phenomena, a metroplex definition was sought that captures the type, magnitude, and impact of interdependencies rather than a geographic boundary.A qualitative metroplex definition was offered first [9].A metroplex is a set of airports that exhibit metroplex phenomena.Metroplex phenomena, we defined, as interdependencies between operations at two or more closely spaced airports that cause a reduction in capacity, efficiency, or safety (or an increase in environmental impact) at one or more of the airports, relative to what the operations at those airports would be if each airport were the only airport.The interdependencies are considered to be fundamental to the metroplex; they result from the geometry, aircraft separation standards, aircraft performance characteristics, and the aircraft types that use the airports.The types of impacts that result from metroplex interdependencies depend on the coordination used to manage the interdependencies.For example, spatial deconfliction may result in no capacity trade-off between the airports but less-efficient trajectories for one or both airports, while temporal deconfliction may result in optimal trajectories for each airport but a capacity trade-off between the airports.Fundamentally, metroplex management includes an element of balancing competing objectives-efficiency and capacity.The magnitude of the impacts further depends on the amount of traffic, and requires a baseline to measure, chosen to be a hypothetical situation in which the other nearby airports do not exist.Therefore, a set of airports may have the potential to be a metroplex because of interdependencies but may not exhibit metroplex phenomena because either traffic volume is too low or coordination methods in use mitigate the impact of the interdependencies.Note that temporal deconfliction would produce no delays when there is insufficient traffic at both airports, while spatial deconfliction may still result in measurable increase in distance flown.Some metroplex impacts are difficult to measure in current operations because of the management approaches already being applied.For example, measuring the impact of metroplex phenomena that are managed through procedures that spatially decouple operations to and from different airports is challenging since the baseline-what the trajectories would be in the absence of other airports-requires judgment.Other factors, such as terrain and noise restrictions, can also influence trajectories close to the airport. +Types of Quantitative DefinitionsOur original goal was to introduce a unified metroplex definition as well as a set of dimensions along which metroplexes may be measured.However, as we studied a variety of methods for measuring metroplex phenomena, we demonstrated that different analytic definitions are useful for describing different aspects of metroplex phenomena, but no one definition provides a complete picture.We classified definitions into three types [10]-observation-based or "black box," model-based or microscopic, and simulation-based.Observation-based definitions treat the metroplex as a black box.These definitions attempt to quantify metroplex characteristics from the relationship between observable input and output data.Therefore, these definitions attempt to avoid the need for detailed knowledge of local procedures.As a result, this approach can readily be applied to a large number of potential metroplexes.However, separating metroplex interactions from other effects such as noise restrictions can be infeasible without applying local knowledge.This type of definition may be applied to real-world data or simulation outputs.However, the baseline against which to measure the impact of the metroplex phenomena is not well defined, often requiring comparisons against other airports rather than on an absolute basis.In fact, the metroplex phenomena and result of current management approaches are combined in the result.Lastly, this type of definition does not directly support identifying viable metroplex management approaches.We studied a variety of observation-based metrics [10] such as excess distance flown and several airspace complexity metrics.Husni Idris studied the New York metroplex using a queuing model that is an example of observation-based metrics [3], [4], and [11].Model-based (or microscopic) definitions require detailed knowledge of local procedures, making them harder to apply to future environments.However, definitions in this class excel at visualizing metroplex phenomena and identifying effective traffic management approaches.Our approach to this type of definition was to extend the classic arrival-departure capacity curve by adding additional constraints (in various dimensions) to model the metroplex phenomena, such as shared use of a departure fix or piece of airspace [9].The approach, like many of the potential definitions, was useful for some reasons but awkward for describing some impacts of interdependencies such as less-efficient trajectories or capturing restrictions on timing of events, which can result in delays without affecting capacity.Simulation-based definitions are ideal for comparing alternative management concepts and studying future environments or traffic levels.We employed a simulation-based definition throughout our study of metroplex management concepts. +METROPLEX MODELINGMetroplex modeling activities during the project may be grouped into three areas: assessing the current cost of metroplex phenomena, developing part-metroplex models of specific types of metroplex phenomena, and developing a simulation environment for studying metroplex management concepts. +Modeling Current Metroplex CostsThe purpose of modeling current metroplex operations was to better understand the existing magnitude of the lost capacity or inefficiencies due to metroplex phenomena.We focused on two interactions in the San Francisco Bay Area: coordination of San Francisco (SFO) and Oakland (OAK) departures during quiet and silent operations and Hayward (HWD) instrumentflight-rules (IFR) departures blocking Oakland (OAK) arrivals.Using actual traffic data from a single day in West Plan (runway use plan), we modeled the operations independently for each airport and compared those results to a model of both airports operating with the metroplex interaction.Table 1 shows the results for the SFO-OAK interaction.Overnight, to reduce the residential noise impact, both SFO and OAK departures stay over the center of the bay, requiring departures from the two airports to be merged into a common stream immediately upon takeoff.SFO and OAK share the capacity of this one stream, and departures must be coordinated in time to enable the merge.Similarly, the metroplex interaction added approximately 2 minutes of delay per HWD departure, and 19 HWD departures caused 93 minutes of arrival delay at OAK.Additional information about our modeling of current metroplex operations appears in reference [12]. +Part-Metroplex ModelsTo focus on specific types of metroplex phenomena and the performance of different management strategies under different assumptions about uncertainty (due to aircraft equipage or Next-Generation Air Transportation System (NextGen) technologies), we modeled isolated metroplex interactions in MATLAB and explored managing the interaction with significant parametric flexibility.In one example, we studied the problem where multiple airports are delivering departures to a shared departure fix.We studied the effectiveness of applying control on the airport surface versus in the air and, in the former case, both uncoordinated control (e.g., miles in trail (MIT) between departures from the same airport) and control coordinated between the airports.Figure 4 illustrates the model.We considered cases where the flight time from the airport to the fix was predictable and uncertain (Figure 5).The various scenarios highlighted the need to know where uncertainty enters the system and the cost of uncertainty after control is applied.In a highly unpredictable environment, control does not need to be coordinated.As individual flight operations become more predictable, coordination in the control provides increasing benefit.Also, as the number of airports increases, coordination becomes increasingly important.We also studied the importance of vertical restrictions on metroplex capacity.Using actual track data, we modeled how many conflicts could have occurred if flights were allowed to select their own vertical profile within vertical windows of various sizes.We counted the potential conflicts as a function of the flows they were between and plotted the location of the conflict.For example (Figure 6), OAK arrivals and SFO arrivals would have approximately 160 conflicts per day if each flight were allowed a 6000-vertical-foot window around the current trajectory.We also estimated the amount of temporal control that would have been required to avoid each conflict.Additional information about our part-metroplex modeling appears in [11], [ 12], and [13]. +Metroplex Simulation EnvironmentTo support research on various metroplex management concepts, we developed a focused but flexible metroplex simulation capability.Figure 7 presents a notional design of the metroplex simulation environment.The Metroplex Planner block receives information about the metroplex and traffic demand and produces a traffic management plan that assigns flights to {runways, fixes, routes} and schedules those flights at scheduling points in the metroplex and at its boundaries.The Metroplex Planner block requires information about feasible minimum and maximum flight times along route segments; this information is provided by the Aircraft Models block.The Simulation block receives the traffic management plan from the Metroplex Planner block and simulates the actual movement of aircraft within the metroplex.The Simulation block uses parameters specified in the Airspace Structure block to define how precisely aircraft are expected to conform to the defined routes and planned times.The Simulation block also calls the Aircraft Models block to produce the actual aircraft movement.O A K ( D ) / S F O ( D ) O A K ( A ) / S F O ( A ) O A K ( A ) / H W D ( D ) S F O ( A ) / S J C ( D ) O A K ( A ) / P A O ( A ) O A K ( A ) / S F O ( D ) O A K ( D ) / S F O ( A ) O A K ( A ) / H W D ( A ) S F O ( A ) / P A O ( D ) O A K ( A ) / S J C ( D ) 4nmi/2000ft 4nmi/1000ft 4nmi/500ft 5nmi/2000ft 5nmi/4000ft 5nmi/6000ftIn the "static" mode, the Metroplex Planner uses the flight-plan data (which includes both earliest/requested times for departures to take off and arrivals to cross arrival fixes into the metroplex airspace) and produces a single plan.The output of the Simulation block simply goes to the Data Collection block to evaluate how well the plan of the Metroplex Planner worked after being exposed to the various modeled uncertainties.In the "dynamic" mode, the Simulation block also outputs new estimates of where flights are along their routes in the metroplex, when departures want to take off, and when arrivals want to cross the arrival fixes.The Metroplex Planner receives updated status of aircraft in the metroplex and demand for future traffic and produces a new plan.We have delivered the software to NASA and supported NASA's use of it for internal NASA research.Discussion of work on and application of the metroplex simulation environment is available in references [2], [11], [12], [14], and [15]. +TRACWe used NASA's TCSim Route Analyzer/Constructor (TRAC) software to generate the metroplex airspace structure by defining waypoints, routes, and altitude requirements.We developed software to convert the output format produced by TRAC into the Extensible Markup Language (XML) format used by our metroplex planner/simulation software.Figure 8 and Figure 9 are examples of the Bay Area metroplex structure defined using TRAC.Additional information about our use of TRAC is described in [12] and [14]. +METROPLEX TRAFFIC MANAGEMENT DECOMPOSITIONAs previously reported, our approach to metroplex traffic management in this project focused on optimally combining spatial and temporal techniques. +Space of NextGen ApproachesUsing our metroplex observations and initial modeling activities, we developed a few ways to organize possible metroplex management approaches.Table 2 presents a three-dimensional (3-D) space of approaches: 1) spatial vs. temporal, 2) static vs. dynamic airspace, and 3) individual flight-based vs. flow-based.We focused our work on optimally combining spatial and temporal techniques for individual aircraft within a static airspace structure.We carefully designed our approach to be consistent with other Next-Generation Air Transportation System (NextGen) concepts, especially NASA's concept for super-dense operations. +Spatial and Temporal Metroplex Management ConceptsMetroplex phenomena occur when two or more airports want to use the same limited resource.The most common shared resource is a portion of airspace.Approaches to managing metroplex phenomena may be classified as time-based, spatial, and mixed.Time-based approaches coordinate the use of the shared resource so that each airport may use it at different times.Spatial approaches move the demand of one or more airports to a different resource (e.g., a different piece of airspace) to remove the original contention.Mixed approaches apply both time-based and spatial approaches.Spatial approaches may be predefined routes or airspace corridors that are spatially decoupled and, therefore, the position of aircraft along these routes does not need to be controlled (except to prevent overtakes).The predefined routes may be static or may be changed periodically but held constant between these occasional changes and always spatially deconflicted.Temporal (or coordinated) approaches require the timing of aircraft movements along routes or through airspace regions to be controlled to prevent conflicts.The coordination may be applied at the entry to the metroplex (i.e., runway for departures and fix for arrivals) and the aircraft fly open-loop through the metroplex, or applied at various points or continuously through the metroplex airspace.Four-dimensional trajectories-defining unique trajectories for each flight that are decoupled where a piece of airspace may be used by different airports at different times-is an approach that combines spatial and time-based approaches.Examples of other shared resources are controller attention and environmental quotas.Temporal approaches applied to contention for these resources similarly deconflict the use of the resource in time.Spatial approaches involve modifying the demand to use a different resource; this approach may not be available without changing the resources, such as what controllers work what airspace.In this report, we primarily consider contention for airspace resources between metroplex airports.There is a trade-off between capacity and efficiency.The San Jose (SJC) loop departure procedures are a good example.The SJC departures and San Francisco (SFO) arrivals could be coordinated in time to avoid the need for the longer path length flown by the SJC departures.However, during periods of simultaneous demand, some flights would be delayed by this coordination, reducing the total number of operations in the metroplex during those periods of time.Historically, maximizing capacity has been the principal objective in air traffic management research.Recently, fuel-efficiency considerations have moved to the forefront as traffic has decreased, fuel costs have increased, and environmental considerations have become more important.We can expect over the lifetime of any metroplex management concept that the relative importance of different objectives will vary.In fact, the blending of objectives may vary as a function of time of day in some situations.Therefore, we strive to propose approaches that are flexible to relatively weighting different objectives.The optimal trade-off between spatial and temporal approaches to managing metroplex operations depends on the technologies available for time-based coordination.Historically, spatial approaches have been applied when possible to decouple operations at metroplex airports.Spatial approaches have been more efficient than temporal approaches because a lack of good coordination automation has made temporal coordination require higher workload, and large uncertainty requires large safety buffers between consecutive operations.However, one result of dedicating different corridors of airspace to each airport is that each airport has fewer departure corridors.As a result, additional spacing is often required between consecutive departures from the airport, since diverging headings are not available.Moreover, during periods of high demand at one airport but low demand at other metroplex airports, the static airspace structure does not allow the busy airport to take advantage of underutilized airspace allocated to the other airports.Figure 10 illustrates the difference between static allocation of spatially separated routes and shared routes on which operations must be coordinated in time, using an example in which there are two airports and two resources (e.g., routes or departure fixes).In a NextGen environment, precise temporal coordination and precise compliance with planned resource usage times could allow temporal coordination to be much more efficient than in current operations and provide more different departure corridors to each airport.An interim approach may be to dynamically reallocate resources to each airport, where at any point in time the allocations are spatially decoupled. +Spatially Separated Temporally CoordinatedFigure 10.Spatial versus temporal approaches. +Problem DecompositionThe metroplex management problem may be decomposed into four sub-problems that may operate at different time horizons:1. Select airport configurations.2. Select airspace configuration.3. Assign flights to routes and schedule at shared points.4. Apply tactical control in response to uncertainty.This decomposition assumes that the airspace structure consists of defined routes.This assumption is mostly consistent with NASA's current concept for airspace super-density operations (SDO) in which each flight will be on a known route or trajectory at all times.The applicability of the results presented in this report to other concepts for airspace structure, including the version of the NASA SDO concept where the routes are not predefined but calculated specifically for each flight, are discussed in the following sections. +Problem 1: Airport ConfigurationsThe first sub-problem is to select the configuration at each airport (or schedule of airport configurations).For this work, we assume the airport configurations are provided by other automation or human decision.Our current research for the NASA System-Oriented Runway Managament (SORM) effort is addressing this problem. +Problem 2: Airspace ConfigurationFor the given set of airport configurations (or schedule of configurations), we must create the route structure.If a unique 4-D trajectory will be defined for each flight, this step may still define some airspace allocation, such as what fixes are being used or what regions of airspace will be used for only specific traffic flows.If a playbook of predefined sets of routes is used, then this step selects the set of routes (or schedule of sets).The simplest case is that a single "play" exists for each set of airport configurations.Between these extremes, routes are defined (as opposed to unique 4-D trajectories) but are not predefined; a set of routes is defined to be optimal for current conditions.This option likely does not offer significant advantage over the "playbook" approach. +Problem 3: Flight PlanningIf the airspace configuration uses defined, known routes, then the metroplex flight planning problem, which was the focus of our work, is to assign each flight to a route through the metroplex (which may include assigning the transition fix and runway) and scheduling the flight at certain points along that route to avoid conflicts at intersections (which include runways and transition fixes).Vertical restrictions could be applied as an additional dimension of control or could be applied by predefining multiple routes with different vertical constraints (e.g., a route with vertical flexibility and a separate route with vertical restrictions to spatially eliminate crossing points).If the airspace configuration concept defines only regions of airspace available for each flow and allows flight-unique trajectories, then the flight planning problem takes a different form, which we did not study in this project. +Problem 4: Tactical CorrectionsOnce the aircraft is flying along the route, airborne control may be required to react to compliance errors or other uncertainties.One modeling option is to set the uncertainties (e.g., compliance with a route and error in the predicted speed) to equal values that represent the motion of the aircraft after airborne control is applied.Another research option is to study the prior steps in an environment where airborne control is not available. +Relevance to ResearchThis decomposition of the metroplex management problem, we believe, handles most concepts for airspace structure within which the metroplex management problem must operate.The concept in which each flight receives unique 4-D trajectories is distinct.Such a concept requires a slightly different decomposition-effectively dividing the airspace configuration problem (problem 2 described previously) into two parts.The first part defines the regions of airspace available for the 4-D trajectories.The remainder of the airspace configuration problemdefining the "routes"-is combined with the flight planning problem (problem 3 described previously) since defining the "routes" and assigning flights to routes (and scheduling them along the routes) are actually the same problem when each flight is assigned a unique trajectory.Problems 1 and 2 could likely be combined.However, problem 1, the airport configuration selection, is the focus of other work and is not studied here, motivating the separation.Problem 2 was studied briefly by solving problem 3 for different airspace structures and comparing, rather than building an optimization algorithm to solve problem 2 in a real-time automation system.The latter is proposed for future work.Problem 4, which is not unique to metroplexes, could be studied by considering whether or not speed control is sufficient to maintain separation at intersection points, given an assumption about aircraft equipage/technology and the resulting compliance errors and prediction uncertainty.Problem 3 was our focus.We developed a "static" metroplex planner that produces a set of assignments and scheduled times on a single set of input data.We further applied this static planner within a dynamic planner-one that uses time windows and freeze horizons to incrementally advance the solved time window while using new information about flights further away in time.That work is described at the beginning of this report as well as in references [2], [11], [12], [14], and [15]. +METROPLEX PLANNER INTRODUCTIONThis section describes the evolution of our thinking and approach to solving the problem of managing metroplex operations.Atkins [1] summarized numerous metroplex phenomena, particularly in the San Francisco Bay Area, illustrating the current use of both spatial deconfliction (i.e., use alternate routes or resources) and temporal coordination (i.e., interleave utilization of shared resources) to manage different metroplex interactions.Enabling Nest-Generation Air Transportation System (NextGen) technologies will allow more dynamic selection of a spatial or temporal solution to a metroplex interaction, either on a flight-by-flight basis or for flows during a period of time.Our goal, therefore, became to develop an algorithm to optimally choose between spatial and temporal approaches to manage metroplex interactions.The point of departure for this research was the realization that a predominant factor contributing to metroplex operations is the need for sharing of critical resources across multiple airports within the metroplex-whether those resources be airspace, information, or controller attention.If we consider existing techniques that have been put in place to manage metroplex interactions at various locales throughout the National Airspace System (NAS), we can broadly place these approaches into two major categories: spatial strategies and temporal strategies.Spatial strategies handle potential conflicts by defining completely separate physical "corridors" for aircraft to travel within, where the corridors themselves have been procedurally deconflicted.An example is the use of altitude restrictions on departures climbing out of a given airport such that the departures will stay below the arrival streams of another airport.Temporal strategies, on the other hand, use explicit timing of individual flight operations to interleave aircraft through the shared resource.As it turns out, however, the "best" strategy to apply in a given situation depends on many factors, and it generally can change over time.Given this classification, our approach was to pose the metroplex traffic management problem in the context of an optimization problem.Rather than solve the route assignment, sequencing, and scheduling problems as a series of subproblems, we instead attempted to solve these problems simultaneously.We desired to see whether the appropriate management strategy (spatial vs. temporal) in a given situation could fall out "naturally" from the optimization-as opposed to having to be defined a priori as in currentday operations.In this sense, the approach applied could change minute-by-minute or flight-byflight.We thus set out to formulate the metroplex problem in an optimization context.Inspirations came from previous work on surface routing and scheduling-where considerable research has been conducted on optimization algorithms related to conflict-free scheduling of surface movements.Table 3 summarizes the key references that influenced our thinking and highlights how each reference modeled the various functions: route assignment, sequencing, and scheduling.In particular, we began by studying the taxi scheduling work of Rathinam, Montoya, and Jung [13], which, although not handling the route assignment problem (taxi routes were assumed to be defined a priori), developed a mixed integer sequencing and scheduling problem similar to that which we were seeking.It should be noted, however, that the results reported in their paper correspond to a pure linear program; the binary sequencing variable values were set through a preprocessing step based on "first come, first-served" (FCFS) ordering.Nonetheless, the ideas expressed in the paper significantly influenced our thinking.We studied other surface optimization formulations that included both routing and scheduling aspects-but modeling time as discrete with discrete routing decisions at each node (Marin [16]; Roling and Visser [17]; and Smeltink, Soomer, DeWaal, and Van Der Mei [18]).We developed initial prototype implementations of several of these formulations to assess their numerical performance.At least given the solvers we had readily available, however, the realistic size of the problems that could be solved in any sort of reasonable time was limited.The formulation presented in Keith, Richards, and Sharma [19] was the closest to what we were looking for, namely modeling discrete routing and sequencing decisions and continuous time scheduling.However, their approach treated each node within the network as a discrete decision point, resulting in a discrete decision variable space significantly larger than what we thought was necessary.The formulation presented in the following sections represents our fusion of the inspiration provided by these previous efforts.The basic formulation, presented in Capozzi, Atkins, and Choi [2], is similar to that of Keith, Richards, and Sharma [19], but rather than treating each turn as a discrete decision, treats routing options in the form of complete routes.For our purposes, rather than scheduling, sequencing, and deconflicting at every point along a route, it is only strictly necessary to make decisions at key points that "matter"-particularly where flights can potentially conflict given underlying route geometry.As described in the following sections, however, the dynamic form of the problem (e.g., planning "in front of the aircraft" as it moves along its most current plan) begins to resemble that of Keith, et.al.In the dynamic planning application, we solve a sequence of deterministic problems.Each problem comprises a "demand snapshot", a "route network", and a function that maps route options onto each flight in the demand set.The general concept is to over-specify the route option set to include options representative of both spatial and temporal strategies.In this fashion, the metroplex management strategy to use can naturally fall out of the optimization process.As stated previously, our objective in this facet of the research conducted under this NASA research announcement (NRA) effort was to develop an approach to simultaneously and optimally solve the route assignment, sequencing, and scheduling problems in the context of metroplex operations.From an algorithm perspective, this work extends NASA's prior scheduling-only work to include route assignment and could be applied to ongoing Safe and Efficient Surface Operations (SESO) research.This problem is a challenging one-primarily because of the dimensionality of the underlying discrete state space.However, it should be noted that it was not our intention to leverage expensive, state-of-the-art solvers and to be content in solving the problem at all, regardless of how long it took.Rather, we focused on the development of an optimization-based formulation that could be solved with open-source solvers in the context of a real-time automation system-practical optimization. +FORMULATION AS A MIXED INTEGER LINEAR PROGRAMThis section describes the approach taken to achieve the capability of simultaneously and optimally solving the route assignment, sequencing, and scheduling problem-namely, casting the problem in the form of a mixed integer linear program (MILP).This choice was a natural one given the discrete nature of the routing and sequencing decisions and the continuous nature of the scheduling aspects of the problem.This section describes the MILP "kernel" that is used to solve a given deterministic problem instance within the overall dynamic planning framework. +Preliminaries and NotationThe geographical scope of the problem to be solved is taken to be the combined terminal maneuvering area for the set of airports included in a given metroplex.From a modeling perspective, we define a point at a virtual metroplex "center" and draw an arc at a distance of approximately 200 nautical miles (nmi) from this center point.This arc defines the boundary at which arriving flights are assumed to "enter" the metroplex and the point at which departures are assumed to "exit" the metroplex.The demand set to be modeled includes arrivals and departures into and/or out of any or all of the metroplex airports (see Figure 11), in one of various operational "phases":• Arrival flights outside the metering fixes, en route to one of the metroplex airports• Arrival flights inside the metering fixes, en route to one of the metroplex airports• Departing flights waiting to push from the gate• Departing flights taxiing to the runways• Airborne departures en route, exiting the metroplexThe following assumptions are utilized in establishing a given problem instance:• The set of flights to be scheduled is .An individual flight is denoted by .• We assume the existence of a globally defined set of complete routes, .Each route uniquely connects a metroplex "entry" point to a metroplex "exit" point.• Each flight can be feasibly assigned to one or more routes, .We assume the existence of some function that prunes the set of available routes and identifies and populates the subset of route options to be considered for each flight in the demand set.• Each route contains an ordered set of scheduling points, .• The initial scheduling point on any given route is denoted by .• The The final scheduling point 4 on any given route is denoted by .Note that no a priori assumptions are made regarding the routing structure that is defined for a given planning cycle; the route structure is a function solely of the airspace design concept being evaluated.As such, the routing structure may include only a single route from a given arrival fix to a particular arrival runway (thus implicitly defining the airspace and runway configurations).Or, at the other extreme, there may be a route from every arrival fix to every runway in both directions-thus allowing the runway usage to be determined as part of the overall optimization process.Figure 12 shows an example routing structure and highlights several aspects of the associated data, including several "entry" and "exit" points.Note that a formal distinction is made between schedule points-where separation is actively controlled-and waypoints, at which separation is not controlled.The specification of which points in the routing structure "matter" and should be defined as schedule points is left to the user.Similarly, the formulation makes no assumptions regarding the definition of feasible route options for each flight.Rather, the user is provided the flexibility to define the mapping function of choice.Typical mapping functions that might be considered include the following (generally ordered from more to less restrictive):• Assign a single-route option a priori to each flight, eliminating the route assignment subproblem and leaving only sequencing and scheduling as free variables.• Include routes that connect from a given "fix" to a given "runway".Such a mapping implicitly defines the runway usage/configuration.• Include routes that connect from a given "fix" to all usable runways at a given "airport", allowing investigation of the value of relaxing the notion of runway configuration to allow flights to be routed to any usable runway.• Include routes that connect from a given "entry direction" (e.g., the northeast) to a given airport runway.Such a mapping could be used to explore the value of sharing multiple airspace fixes across several metroplex airports.• Include routes that connect from a given "entry direction" (e.g., the northeast) to all usable runways at a given airport.Such a mapping could be used to explore the value of sharing multiple airspace fixes across several metroplex airports.• Include routes that connect from a given "entry direction" to all runways at all airports.Such a mapping could be used to explore the value of distributing traffic across multiple metroplex airports within a geographic region.As should be clear from these examples, almost any mapping function could be implemented.It should also be noted that the effective routing structure-particularly in the dynamic planning context-changes as flights move through the metroplex (see Figure 13) to include only those portions of route options "in front" of each aircraft. +Decision VariablesThe primary decision variables are defined as:• : Continuous variable representing the scheduled time for flight at point on route option +•: Binary decision variable that takes on the value of 1 if flight is assigned to route option and zero otherwise +•: Takes on the value of 1 if flight (assigned to route option ) is scheduled before flight (assigned to route option ) at a scheduling point (that is common between routes and ), and zero otherwise +Notions of TimeSeveral notions of time are used within the planning framework:• Scheduled time of arrival (STA): This value is output by the planner each cycle, , and represents the target time at which the flight should cross a given point on route option .• Actual time of arrival (ATA): ATA is the time at which a given flight actually crosses a particular point on its assigned route.The degree to which the ATA matches the STA for a given flight is a function of its navigation performance model.• Earliest time of arrival: This value defines the earliest feasible time at which a flight can physically reach a given point along a potential route option , denoted .• Aircraft-optimal time of arrival: This value, denoted , represents the time at which a flight would arrive at a given point along route option if allowed to fly completely uninfluenced by the presence of any other flights.This value generally would correspond to the most efficient trajectory for that individual aircraft.• Estimated time of arrival (ETA): ETA is the best estimate of the time at which a given flight will reach a given point on route option , denoted .Associated with the ETA to a given point is an assumed distribution, parameterized by an equivalent standard deviation, . +Objective FunctionThe primary objective function studied was of the form:(which minimizes the sum of the "exit" time out of the metroplex over all flights.For arrivals, the "exit" time is typically taken to be the point at which airplanes touch down on the runway, whereas for departures, the exit time is taken to be the time that a flight crosses its departure fix.An alternative form of objective function that included a penalty on initial time was also studied over the course of this research, namely:(This alternative form had the effect of resulting in similar exit times to those obtained with the previous objective function but with smaller "entry" time.In effect, the addition of the penalty on initial time shifted the distribution of required delay absorption from outside to inside the metroplex. +ConstraintsThis section describes the set of constraints that are required such that the solutions obtained via the optimization process are conflict-free and physically feasible given assumptions regarding the motion of each aircraft. +Crossing Time Constraints at Initial Scheduling PointThis constraint bounds from below the time at which a flight can reach the initial scheduling point that falls within the planner scope (see Figure 14).This constraint forms the basis of all additional timing constraints related to each flight.Mathematically, this can be expressed as:(where is the earliest feasible time at which the flight can be scheduled at its initial scheduling point.In general this bound corresponds to a flight taking the most-direct routing and fastest feasible speed to the initial point on the route.Note the dependence on the route assignment variable in eq. ( 3).If desired, an upper bound on the scheduled time can be included at the initial point:Application of both constraints is equivalent to defining a feasible scheduling window at the initial scheduling point.This window could correspond, for example, to an expected-departureclearance-time (EDCT) window at the runway for a particular departing flight due to the action of some traffic-flow-management (TFM) initiative.Note that for most of the research conducted to date, no "latest" time constraint was enforced for the initial scheduling point.Rather, flights were allowed to be delayed indefinitely prior to entering the metroplex-representing the effect of airborne holding or ground holding. +Crossing Time Constraints at Final Scheduling PointThe formulation includes the ability to bound from above and below the time at which a flight must reach its final scheduling point within the planner scope, namely:(Such a constraint can be used to model a particular runway window for an arrival flight or, alternatively, a fix crossing time window for a departing flight. +Transit Time Constraints between Successive Scheduling PointsIn order for the schedule obtained to be consistent with physical limitations of aircraft performance and air traffic operating procedures, constraints are introduced to bound from above and below the time for the aircraft to transition between two successive scheduling points on each of its potential route options: +Ordering Constraint at Potentially Shared Scheduling PointsFor each pair of flights that can be assigned to routes that have at least one common point, the order at which the flights are to visit each of these shared points must be uniquely defined. +Expressed in words:If:• Flight is assigned to route : and• Flight is assigned to route : , Then: for each scheduling point that is shared by routes and , either flight must be scheduled ahead of flight or vice versa, and can be expressed mathematically via the constraint: (9) Thus, the binary sequence indicator variables have only nonzero values when both of the corresponding pairs of route assignment variables are nonzero. +Separation Constraint at Potentially Shared Scheduling PointsSuccessive crossings of a given point by any pair of flights must often satisfy some separation constraint, which is often a function of some characteristic of each aircraft pair.Runway separation requirements due to wake vortex are an example of such a constraint.Expressed in words, this constraint is equivalent to the following set of conditions:If:• Flight is assigned to route : and• Flight is assigned to route : and• Flight is sequenced ahead of at scheduling point , which is shared by routes and , Then: the scheduled crossing time of at point must be at least the minimum required separation seconds later than that of flight .Mathematically, this condition can be expressed by the constraint equation: (10) where one such inequality must be written for each potential ordering, for each pair of flights, at each potentially shared point. +Transformation to Linear FormBecause of the dependence on route assignment, the objective function and the constraints defined in the previous sections involve a product of logical (binary) variables and/or a product of logical and continuous variables.In order to leverage open-source, self-contained mixedinteger-linear-program (MILP) solvers, it is necessary to transform the nonlinear constraints developed in the previous section into a set of equivalent linear constraints. +Product of Logical VariablesEach occurrence of a product of logical variables of the form can be replaced through the introduction of an auxiliary binary variable (Bemporad and Morari [20]): (11) Then, we append a set of auxiliary constraints: (12) +Product of Logical and Continuous VariablesWe can replace the occurrence of a product of logical and continuous variables of the form appearing in a constraint equation by replacing it with the auxiliary continuous variable (Bemporad and Morari [20]): (13) And appending the set of additional auxiliary constraints:(14) +Summary of Linearized Mixed Integer Optimization ProblemApplying the linearization technique described previously results in the following MILP:Objective:Initial condition:Minimum Transit: (17) Maximum Transit: (18) Ordering:Separation: (20) Where the following variable substitutions have been made: +Alternate Linearized FormIt is possible to construct linear constraints using an alternative approach that reduces the number of binary variables required. +Objective FunctionIf we define a set of additional constraint equations: (21) where is a large positive constant value, then if is zero (e.g., flight is not assigned to route option ), then the scheduled time at all points on that route option will be forced to be zero, allowing us to rewrite the original nonlinear objective function expressed by eq. ( 1) in the linearized form: (22) +Initial Scheduling Time ConstraintConsider the initial scheduling constraint given by eq. ( 3) and repeated here for convenience:Effectively, this constraint expresses the statement: "If flight is assigned to route option , then the scheduled time at the first point on that route can be no earlier than ."An equivalent way to express this mathematically is: (23) If the value of is zero, the constraint is trivially satisfied given a sufficiently large positive constant, .Unlike the linearization technique given by eq. ( 13), the approach to linearization used in eq. ( 23) does not require the introduction of additional auxiliary constraint equations. +Transit Time ConstraintsThe transit-time constraints, repeated here: can similarly be linearized through the introduction of a large positive constant: (24) (25) +Sequencing ConstraintThe sequencing constraint at each potentially shared point (refer to eq. ... ( 9)) can equivalently be expressed via the pair of linear constraint equations:(26) (27) Note the effect of the term scaling the difference in scheduled time.Consider the case where flight is assigned to route and flight is assigned to route , respectively.If the quantity in parenthesis is some positive value (thus implying that flight is scheduled before flight at shared point ), then eqs.( 26) and ( 27) reduce to:Since is a binary variable, then the only value that satisfies both constraints is unity. +Separation Constraint at Potentially Shared NodesThe nonlinear separation constraint expressed in eq. ( 10) can alternatively be written as:(28)(29)This pair of equations can be interpreted as follows:Consider first the case: .Then, if is sequenced ahead of (e.g., ), the following must hold: while eq.( 29) is trivially satisfied.However, what if is sequenced ahead of ?In this case, eq. ( 28) is trivially satisfied and eq.( 29) must hold:If either or is zero, then both eqs.(28) and (29) are trivially satisfied as long as is larger than the separation value. +Heuristics to Reduce Computation TimeAs formulated in the previous section, the resulting MILP consists of a considerable number of binary variables.Since the time required for the solver to perform the branch and bound iteration scales strongly with the number of integer (binary) variables, several heuristics have been defined to provide the option of reducing the problem size in certain situations. +No-Passing Constraint within a RouteIf flights cannot pass one another within a given segment, then a constraint that preserves the ordering of any pair of flights can be included in the problem formulation:(30)This constraint enforces the condition that the ordering at any point along the assigned route is the same; flights are not allowed to "pass" one another in the optimal solution.In general, this constraint is applied only when the pair of schedule points ( ) appears strictly consecutively in each route option.Consider the routing network depicted in Figure 15, which shows two routes: and .The subsequences of points common between these two routes-and -are shown in purple.Thus, the process of writing out the constraints related to a pair of flights ( ) in which one flight includes the route in its route option set, and the other includes the route associated with the no-passing constraint, is depicted in Figure 16.As can be seen, this heuristic is applicable to cases where routes converge and/or diverge.For example, in this case, although the order of flights between point B and point D is maintained, a flight F1 routed on R1 that entered B behind a flight F2 routed on R2 could in fact be ordered in front of F2 at points I and J.By comparison, for each pair of flights that include in their route option set, the set of variable substitutions defined by the no-passing constraint on are shown in Figure 17.For a pair of flights routed on R1, the order that the flights are sequenced at A is thus preserved over the entire length of the route. +Natural Ordering ConstraintIn some cases, it may be desired to limit the degree to which the solver searches for sequences of pairs of flights at common points.For example, if flight B would naturally reach a common point "significantly later" than flight A, then it may be considered "unlikely" that the optimal solution would involve swapping this order.The degree to which this assumption is true is related to the maximum "speed up" and "slow down" possible for each flight along its respective route prior to reaching the common point.In some cases, the application of this heuristic can result in infeasibility.Such is the case if there is not sufficient controllability to keep a faster flight "behind" a slower flight that happens to reach the initial common point earlier in time.Nonetheless, we provide this option to the user and allow specification of the "first-come, first-served threshold" beyond which flights will be assumed to be fixed at their natural ordering based on their estimated time of arrival (ETA) at a given common point. +Application of WindowingIn order to optimize the scheduling and assignment of larger demand sets, we explored the use of a windowing technique to carve up the demand into a sequence of smaller demand sets.The windowing approach taken was to solve the first window, then solve the second given the solution of the first window.In general, the future window is solved given the solution obtained for previous windows.. Previous window solutions effectively serve as additional constraints on the solution of a future window.This process is repeated until all flights in the demand set are scheduled and assigned.The optimal window size is generally determined iteratively based on the sensitivity of the solution time required to the size of the window.The largest window that demonstrates acceptable run-time performance for a given problem provides the appropriate compromise between solution quality and speed.Note: In the research conducted to date, overlap between successive demand windows was not considered.This overlap was observed, in some cases, to result in sub-optimal solutionsparticularly for flights near the edges of neighboring windows. +MODELING CONSTRAINT PARAMETERSSection 7 defined the structure of the constraint equations.However, to construct a real problem instance, the value of the parameters that appear within the structure of these constraint equations must be defined, in particular:• Earliest time for each flight at the first point on each potential route option,• Transit time bounds between any two points on a potential route option, defined by and• Separation required between two flights at a potentially shared point, defined by This section identifies issues associated with each of these values and describes the approach taken in this research to define these values. +Computing Earliest Time at Initial Scheduling PointConsider an aircraft traveling at speed a given distance from the initial scheduling point on a potential route option.The first challenge is to identify the earliest time at which the aircraft could reach the point .We start by assuming the following:• There is a crossing restriction at point that specifies an upper and lower bound on the speed at which the aircraft must cross the point, defined by the closed range .• There is some realistic physical limit on the deceleration or acceleration that the aircraft can achieve that constrains its ability to attain a particular speed.• There is some realistic physical limit on the minimum and maximum airspeeds that the aircraft can fly in a particular aerodynamic configuration.We then ask the question: How could the aircraft change speed from to some speed so as to reach the point the fastest?From the perspective of estimating the earliest time at point , consider first the case where the current speed is below the maximum crossing speed.Then the minimum time policy (see green line in Figure 18(a)) would be to accelerate at until reaching and then holding a constant speed until reaching point .Note that this approach would be a "best-effort" approach.If the aircraft could not actually achieve the target speed prior to the point , then some residual speed error would exist at the crossing time.Conversely, as depicted in Figure 18(b), if the aircraft were currently flying faster than the crossing restriction, then the earliest time would be attained by continuing to fly at the current speed and then decelerating as quickly and as late as possible to achieve the crossing restriction just as the aircraft is crossing point .An alternative approach is to assume that the aircraft will linearly accelerate or decelerate from its current speed as necessary to achieve the maximum crossing speed restriction just as the aircraft reaches the crossing point .Under such an assumption, the following logic is defined for determining the effective maximum speed:Then the earliest time is computed based on: where represents the current time and is the current distance to the point . +Estimating Fast and Slow Transit Times between Schedule PointsThis section describes different mechanisms explored for use within the planning framework for purposes of estimating upper and lower bounds on transit times between schedule points, as required by the constraints defining the optimization problem. +Controlling Transit TimesIn general, the maximum variation in the time at which a flight can reach a particular point is a function of the path taken to that point as well as the speed of the aircraft along that path.If the aircraft is already at its maximum speed, then the time to reach a point can be reduced only by shortening the distance to that point (e.g., by shortcutting, as depicted in Figure 19(a)).Likewise, the ability for the crossing time at a given point to be delayed is a function of the degree to which the aircraft can be slowed or its path can be "stretched", as depicted in Figure 19(b).The limiting case, of course is the use of explicit holding-whether via a fixed pattern for airborne aircraft or on the ground-in which case flights may be delayed arbitrarily prior to crossing a particular point.In the abstract, the formulation described in section 7 does not consider the nature of what path or speed modifications are necessary to achieve a given transit time between any two points.Rather, it takes advantage of whatever range of transit times is available to produce the most optimal solutions it can.However, if simulation of the actual motion of flights is desired, the options are either (a) ignoring implied path deviations (typically resulting in unfeasibly slow or fast speeds) or (b) actually modeling the path deviations required to achieve a given speed-up or slow-down of a given flight.Choosing the latter approach requires modeling the effect of path variations during the planning process and then within the simulation such that the actual flown duration on the path is consistent with that assumed by the planner in generating the time constraints.Otherwise, the time-constrained plans output by the planner will be meaningless.Early in the research conducted in this effort, we chose the simplified approach in which time was dealt with in the abstract.We thus limited our attention to temporal changes only, without concerning ourselves with the details of how one might actually achieve such changes 5 .In later research, we have begun to address the issue of how such variations in transit time might be physically realized, limiting our attention to speed control as the sole mechanism for adjusting transit times. +Controlling Transit Time via SpeedThe question that the planner is trying to answer is: For a given speed profile, what is the range of transit times that could be achieved between each pair of points on a route?This section examines this question in the form of a representative example.Consider a route consisting of three schedule points (A,B,C) as depicted in Figure 20.At each point, we assume the specification of a speed crossing restriction, expressed as a fastest and slowest speed at which aircraft can cross the point.In current operations, these speed restrictions might model the action of controllers to manage speeds within the terminal area.It is further assumed that these speed restrictions are "within" the physical speed limitations of aircraft maneuvering along a given route-and thus can feasibly be obtained by the aircraft.Given this route and assumed speed profile, the objective in this section is then to explore the range of upper and lower bounds on transit time achievable between successive schedule points.Mathematically, if we assume a constant acceleration (which may be zero), then the transit time between any two points is defined by the ratio of the distance between the points and the average speed:(31)The first question to answer is the nature of the aircraft speed along a given segment.Should it be modeled as a constant speed or involve some acceleration or deceleration?Consider first a constant-speed strategy in which we set the values of both and in eq.(31) to the maximum or minimum speed of the point immediately in front of the aircraft.In this example, setting the speeds based on the restriction bounds defined at point B results in minimum and maximum transit times between points A and B of 491 and 568 seconds, respectively.Another strategy would be to select the maximum and minimum speeds at either endpoint, referred to as the "Max at Either End" and "Min at Either End" strategies in Table 4, resulting in fast and slow transit times of 432 seconds and 568 seconds, respectively.Of course, such strategies are not strictly feasible in the sense that they, by definition, violate the crossing speed restriction at one or the other end of the segment.The next class of strategies to consider is constant, nonzero acceleration strategies.Under such an assumed motion, the maximum transit time would be for the aircraft to cross point A at the minimum crossing speed at point A and then cross point B at its corresponding minimum speed.This strategy, referred to as the "Min Speed at Each End" in Table 4, results in a maximum transit time between points A and B of 514 seconds.The minimum transit time on the segment is obtained in an analogous fashion under such a motion model, with the speeds set to the fastest values at each endpoint, resulting in a fastest transit time of 460 seconds.Comparing the transit times of the constant acceleration strategies to those obtained through the constant-speed motion shows that while the range of times achievable under the constant accelerations is somewhat smaller, these strategies have the advantage of being consistent with the speed restrictions defined at each endpoint.Motivated by the earlier discussions of the estimation of earliest time at the initial point and the maximum and minimum time speed profiles described in Figure 18, let us consider one final class of strategies that combines constant-speed and constant-acceleration segments.Let the maximum aircraft acceleration be defined by and the maximum aircraft deceleration be defined by. For purposes of this example, assume that the values of these parameters are and = -2, respectively, in units of knots per second.Assume that the current aircraft speed as it crosses point A is 250 knots.Then, the minimum transit time is obtained by maintaining a speed of 250 knots until the time at which the maximum deceleration can be applied to drop the aircraft speed to the maximum crossing speed restriction at point B (220 knots).The kinematic relationship implies a deceleration time of fifteen seconds for a final speed: then for a final speed of 220 knots and an initial speed of 250 knots, this relationship implies a time required to decelerate of 15 seconds.The distance traveled during these 15 seconds is equal to: which, in this example, takes on a value of 0.979 nautical miles (nmi).The total distance between the two points can be computed based on the distance covered at constant speed and that covered during the constant acceleration phase of motion:The time the aircraft spends during the constant-acceleration portion of the segment is equal to: Substituting this value into eq.(32), we obtain: which can be simplified to:(33) Equation (33) can then be solved for the time that the aircraft spends on the constant-speed portion of the segment.For the values considered, this fix equates to covering a distance of just over 29 nmi at a speed of 250 knots spanning of 418 seconds.Combining this time with the time on the deceleration segment gives a total transit time of = 433 seconds.One can perform the calculations for a similar strategy to determine slowest transit time consistent with the constraints.In this case, we assume the aircraft crosses point A at a speed of 230 knots (the minimum at point A) and reaches a speed of 190 knots at point B. In this case, the longest transit time results from achieving that speed as soon as possible and then flying at 190 knots over the remainder of the segment.In this case, it takes 20 seconds for the aircraft to decelerate from 230 to 190 knots, covering a distance of 1.16 nmi.The aircraft can then fly the remaining distance at a constant speed of 190 knots in 546 seconds, resulting in a total transit time of 566 seconds.It can be concluded that the two-stage (constant-speed/constant-acceleration) strategy results in transit-time bounds that (a) are consistent with the speed restrictions specified at either end of the segment and (b) are nearly equal to the fastest and slowest values that can be obtained-even by strategies that violate the speed restrictions at either endpoint.So, given the previous analysis, the question remains: In the context of the planner, which must estimate the fastest and slowest transit times, what approach should be used?Based on the analysis, the two-stage strategy would result in reasonable bounds.However, a secondary issue needs to be considered related to the typical behavior of solutions to constrained optimization problems.That is, the solution obtained often lies on the constraints.What this reality means in the context of scheduling aircraft along routes subject to minimum and maximum transit-time constraints is that the optimal solution often entails aircraft flying at the fastest speed-raising the issue of the effect of aircraft speed on the efficiency with which it is able to fly along a given route.It is likely that flying as fast as possible (given the speed profile restrictions), while resulting in an optimal "system" solution as defined by the objective function in eq. ( 1), is not optimal from the perspective of individual aircraft.Further, such a solution may not be desirable from an environmental/noise perspective.This is not to say that these bounds are incorrect in and of themselves.Rather that they need to be used in the context of an overall optimization problem that includes these other factors in some fashion such that these transit-time bounds are not blindly adhered to but rather are used only when they provide an appropriate "net" advantage over all considerations-both system and aircraft.Because the formulation as described in section 0does not include such considerations, we chose to estimate transit bounds using the constant-acceleration motion described previously, which is captured mathematically by eq. ( 31).The initial and final speed values used in this equation are derived from the crossing restrictions at the "from" and "to" ends of the segment, respectively.We compute an "effective" minimum speed between the "from" point ( ) and "to" point ( ) as the average of the minimum speeds at either end of the segment:Then the maximum transit time between two successive points is computed via:where is the distance between points and .We likewise compute the effective maximum speed on the segment as the average of the maximum speed values defined at each endpoint: with the corresponding minimum transit time computed via:The values shown in Table 4 for the constant-acceleration strategy were obtained in this fashion.As discussed previously, in this example the bounds obtained using this technique span a smaller time range than those obtained using the more aggressive strategies.But given our objective of limiting the speed variations somewhat, this implementation is a suitable initial one.Note that the relationship between transit times achieved via this "average-speed" method and those obtained via the minimum-time or maximum-time policies depends, in general, on whether the effective changes in speed implied by the user-supplied speed restrictions are more or less aggressive than the physical acceleration or deceleration limits assumed for the aircraft.Note: Currently, no physical acceleration limits are being imposed within the planner or the simulation.Thus it is possible that the speed changes required are infeasible for the aircraft to actually perform.This area is one in which increased realism is desired, and it will be the subject of future research in applying this formulation.Note: Currently, no minimum or maximum airspeed limits are imposed within the planner or the simulation.Thus it is possible that the speeds required are inconsistent with aircraft aerodynamic or structural limitations.This area is another in which increased fidelity is desired, and it will be pursued in later applications of this formulation in more realistic contexts. +Establishing Required Separation between Flights at Schedule PointsThis section addresses the issue of establishing the minimum time required between successive flights crossing a potential common point.To begin, we consider the time at which a given flight is expected to cross the "next" point ahead of it on one of its route options, as depicted in Figure 21.Here we assume that the time of arrival at the point can be described by some distribution.We assume that the "most likely" time of arrival is defined in some manner based on this distribution; for example, this value could be taken as the mean, mode, or some other value.The middle bar denotes the extent of the modeled (and thus expected) variation in the crossing time for this flight at this point.In other words, based on the current plan, and given the understanding of the way the aircraft is likely to navigate along that plan, this range is the range of crossing times that could be attained-and our best guess as to the most likely crossing time.To the left of the distribution is an additional green shaded bar extending to the left.A similar red bar extends to the right.These two bars depict the degree to which this flight could be sped up or delayed relative to its current most likely crossing time.Now consider the task of scheduling two flights at a common point.As shown in Figure 22, the minimum required separation between these two flights requires that the trailing flight (B) be delayed somewhat.The question is, by how much?One option is depicted in Figure 23, in which the minimum separation required is applied with respect to their expected times of arrival (ETAs) or most likely arrival times.In principle, if each flight were to arrive precisely at its "most likely" time, then this delay would be sufficient.However, given the expected variation in crossing time for both flight A and flight B, there is still a nonzero chance that the two flights would conflict (because of the overlap in the "tails" of their respective time-of-arrival (distributions).Thus, a buffer is added to account for the degree to which flight A could be "late" and flight B could be "early", as shown in Figure 24.Note: The "double buffering" depicted in Figure 24 has been the approach taken in the research conducted to date.The uncertainty bounds defining the expected variation in time of arrival for each flight at a given point are assumed to be provided by a user-defined function that implements a particular software interface.No constraints on the functional form of the uncertainty that is returned by this implementation are enforced.Rather, the variability in uncertainty can be related to the aircraft type (or sophistication of available automation), the particular route, the particular point, and/or the current aircraft location.Current-day operations might model the uncertainty as varying somewhat monotonically with increasing distance from the point, with the uncertainty decreasing in some fashion as the aircraft approaches the point.Future Next-Generation Air Transportation System (NextGen) operations might be modeled differently-with the uncertainty defined by values significantly smaller than those representing current-day navigation capabilities and with different-shaped characteristic curves governing the variation of uncertainty with distance.Figure 24.Buffering applied to nominal separation to account for "late" leader and "early" follower. +DYNAMIC PLANNING APPLICATIONSThis section describes the dynamic planning framework that has been developed over the course of this effort to allow us to incorporate the optimization problem "kernel" described in the previous sections within the context of a real-time automation system.Using this framework we are able to dynamically "construct", "solve", and then "implement" the solution to a sequence of deterministic problems over time.With this framework in place, we were able to demonstrate the benefit obtained from taking advantage of improved knowledge of the precise nature of the demand that will flow through the metroplex as it becomes available.We also studied the effect of key parameters that define the scheduling behavior-namely: "how far" into the future we plan, "how often" we plan, and "how long it takes" to plan-on the nature of the solutions obtained. +ArchitectureThe key components that comprise the dynamic planning framework are depicted in Figure 25.The most significant behavioral components are the Simulation, currently implemented within MATLAB, and the Planner, currently implemented as a stand-alone Java application.• Simulation: Responsible for advancing time.Manages the creation of targets at specified location and time.Constructs the "demand" snapshot at a given instant of time.Delegates to a Trajectory Model implementation that handles the "actual" movement of flights along their most recent plan and "blends" motion between successive plans.• Planner: Responsible for constructing conflict-free motion plans for each aircraft in a given demand set.Each motion plan consists of a sequence of waypoints with an associated scheduled time of arrival (STA).The interprocess communication used to synchronize these two processes is the local file system.The Planner application includes a "Demand Listener", which listens for updates to a specific file (LatestDemandSet.xml).Upon detecting an update, this listener sends an event to the Planner triggering a planning cycle.The Simulation includes an analogous "Plan Listener" that detects when plans are available to be "implemented" within the simulation. +Greedy First-Come, First-Served PlannerAs an alternative to the mixed-integer-linear-program (MILP) planner-both to provide an analytical reference and to demonstrate the plug-and-play dynamic planning framework developed under this effort-we have implemented a "greedy" first-come, first-served (FCFS) planner.The greedy FCFS planner sorts the flights within a given demand set in the order of their earliest time at their initial scheduling point.It then processes each flight in turn, searching over all route options to find the route at which the flight can be scheduled that minimizes its scheduled time at its final scheduling point.As such, the objective function used within the greedy FCFS planner is analogous to the "final time" objective defined for the MILP planner.As the greedy FCFS planner attempts to schedule a given flight on a potential resource, it accounts for the resource usage associated with any previously scheduled flights on that resource, complying with the separation requirement.It should be noted that the greedy FCFS planner utilizes the same uncertainty model with respect to time-of-arrival behavior.Thus, the interflight spacing defined within the greedy FCFS planner is consistent with that applied by the MILP planner for the same pair of flights.The greedy FCFS planner also leverages the same "earliest" and "latest" time calculations such that transit times are consistent with the MILP planner.As such, plans generated by the greedy FCFS planner can be directly compared with those output by the MILP planner for a given demand set. +Interfaces between Simulation and PlannerThe inputs required by the Planner to construct a plan are: (a) the demand set, (b) the set of routes defined for the planning cycle with associated speed profiles, (c) the set of schedule points defined for the planning cycle with associated separation rules, and (d) the expected variation in time-of-arrival behavior. +Demand-Set DescriptionEach demand set, which is tagged with the simulation time at which the "snapshot" was taken, consists of the following information for each flight:• Unique flight identifier• Current position (expressed as nautical miles east and north of metroplex "center")• Current altitude (feet mean sea level (MSL))• Current ground course (degrees relative to true north)• Current ground speed (knots)• Origin airport• Destination airport• Type (used for mapping to separation rules and time-of-arrival behavior)• Category (arrival or departure)• Cardinal direction from which the aircraft is approaching the metroplex (e.g., E, NE, S)• Earliest gate departure pushback time (secs) +Route DescriptionThe routing network that exists at the start of each planning cycle must define:• The set of airports within the metroplex• The set of runways available at each airport• The set of fixes• Any waypoints, not including runway endpoints or fixes• The set of routes, each of which must include a fix and runway +Scheduling-Point DescriptionThe input data for the planner must also include a specification of the points at which separation rules are to be enforced by the schedule.Each scheduling point includes:• The name of the point, which must be the name of a fix, runway, or waypoint defined as part of the route description• The minimum required separation at that point, specified as a number of seconds by which the "trailer" must follow the "leader" at that point• Optionally, defining of aircraft type-specific rules that provide a separation that is a function of the "leader" and "trailer" type (see section 0) +Expected Time-of-Arrival BehaviorAs described in section 0, time-of-arrival behavior is parameterized in the form of a standard deviation about the nominal time of an aircraft.This value is obtained by calling out on a well-defined software interface to a user-specified implementation.The current default implementation provides an uncertainty value (which is assumed to represent the 1of the underlying distribution) that increases with increasing distance from the point, as depicted in Figure 26.This model outputs an uncertainty value that decreases linearly to zero as the distance to the point goes to zero.The uncertainty value maxes out at a value of 300 seconds for aircraft more than 100 nmi from a given point.It should be noted that this model is not intended to be realistic-rather it is solely a starting point for analysis.By implementing the required software interface, users can provide an uncertainty model that is best suited for a given scenario. +Planner OutputThe planner output is a set of plans, one per flight in the demand set, each consisting of the name of the assigned route and a sequence of timed points.In addition to the time-of-arrival (STA) value at each point, the earliest and latest times as well as the "unimpeded" or "aircraft-optimal" times at that point as computed by the Planner are specified.Also provided is an indication as to whether the route assignment has been frozen and whether or not the Planner has frozen the timing of each individual scheduled point. +Overview of Processing FlowEach planning cycle consists of the following steps:1. Determine the most recent demand file, as described in the LatestDemandSet.xmlfile.2. Read the most recent demand set.3. Identify the set of routes that are feasible for each flight in the demand set.4. Prune the route options for each flight such that the remaining schedule points consist only of those points "in front" of the current aircraft location.5. Compute the earliest and latest times at which the aircraft can reach each point along each route option.6. Construct a conflict-free motion plan that satisfies all constraints.7. Write out the plan to a file.8. Update the LatestPlan.xmlfile to indicate that the plan is ready to be "implemented".The Simulation, which is itself blocked while the Planner is computing a given plan, incorporates a PlanListener that then detects the update of the LatestPlan.xmlfile.In response, the Simulation then reads in the updated plan and distributes the individual time-based routes to each aircraft modeled within the simulation.The Trajectory Model for each aircraft, upon receipt of a new route, prunes the route to keep only those points that are physically and temporally "in front of" the aircraft at the time the plan is received.The Trajectory Model then performs a calculation to update its target speed to comply with the STA at the point immediately in front of it.Specifically it computes the distance between the current location of the aircraft and its next schedule point, , as well as the time available to reach that point:(34) Within the Trajectory Model, a low-pass filter with a 10-second time constant is utilized to generate the actual aircraft speed in tracking to the commanded speed.Currently, no "error" is introduced within the "actual" trajectory model-other than the time response implied by the low-pass filter.Rather, each aircraft generally crosses each schedule point at the most recent STA defined at that point.This modeling was done so that we could separately study the effect of demand uncertainty from execution uncertainty. +Control of TimeThe interaction between the Simulation and Planner occurs in a lock-step fashion, with the Simulation responsible for advancing time forward.Controlling time in this manner allows us to model and study the effect of different computational times without actually having to achieve that computation time.At a high level, the flow of execution is described by the following sequence:1. Simulation advances time to the start of the next planning cycle (Plan ).Trajectory Model moves aircraft according to Plan .2. Simulation advances time to the point at which Plan is available for implementation, modeled via the parameter PlanComputationSecs, allowing us to study the effect of different computational times required to solve a given problem instance.3. Simulation constructs the demand set based on the current location of all aircraft, and including flights that will be active between the current time and PlanHorizonSecs into the future.Simulation writes out the demand set and updates the content of the LatestDemandSet.xmlfile to identify the file containing the data. +Simulation blocks [[blocks what? flights?]], awaiting for event indicating that Plan is ready for implementation.5. Planner detects the update of the demand set and generates Plan .6. Planner writes out Plan and updates LatestPlan.xml.7. Simulation detects the update and reads in Plan and distributes it to individual aircraft.8. Go back to step 1 until all flights have reached their final scheduled point.Figure 27 graphically depicts a series of planning cycles.Note that the planning horizons are generally intentionally defined such that they have significant overlap.Only a certain portion of any given plan is actually intended to be "implemented" per se.Overlapping successive horizons allows flights to be factored into multiple plans.Key parameters available within the framework for study include:• PlanHorizonSecs: Controls how far into the future to plan by limiting demand only to those flights that fall within the range [now, now + PlanHorizonSecs] in each cycle• PlanComputationSecs: Models how long it takes for the plan to complete in simulation time, independent of wall-clock time• PlanUpdateSecs: Defines how often a plan is computed Note that these variables cannot necessarily be varied independently.Certain relationships must be maintained between these parameters to ensure they remain meaningful.For example, the time required to compute a plan must be shorter than the update period. +Plan Stability: Freeze HorizonsAs time evolves and aircraft move along a route, there comes a time at which at least part of the plan "ahead" of each aircraft must be "frozen" so that it can be executed.Figure 28 shows an abstract view of the notion of a freeze horizon.Extending forward from the left-most edge of the planning horizon, the freeze horizon is used to select the subset of flights whose existing motion plans should be frozen.Thus, the demand set within a given planning cycle actually consists of the set of flights that have not yet been planned plus the set of flights that have previously been included in a plan but have some future elements of the plan that can still be changed because their timing falls "to the right" of the freeze horizon.The amount of "implementation" lead time required depends, of course, on the operational concept in place for communicating plans as well as the automation available for assisting pilots and controllers in verifying and executing a given motion plan.For example, the Traffic Management Advisor (TMA) would freeze scheduled times for flights once within so many minutes of the metering fix to which it was scheduling-so as to provide controllers with adequate time to issue clearances via speeds and/or vectoring-to actually build the sequence and/or schedule.However, under future Next Generation Air Transportation System (NextGen) operations, leveraging advances in datalink communications and onboard flight-deck separation and required-time-of-arrival (RTA) navigation capabilities, it may be feasible to defer "freezing" either route assignments or timing until significantly later to maximize efficiency.In this research, we treat the notion of freeze horizon somewhat generically, allowing it to be applied separately to the assignment, sequencing, and scheduling decisions included in a given plan.Thus, one could specify that assignments are frozen at a certain point, sequences are frozen at another, and scheduling is frozen at yet another point.In this manner, we can study the effect of different freeze strategies (or requirements) on the nature of the solutions obtained.Figure 29 shows two examples of such strategies.In figure 29(a), the freeze horizon is defined relative to the runway, whereas in figure 29(b), the horizon is defined at the arrival fix. +Simple Dynamic Planning ExamplesThis section provides two simple examples to explain the behavior of the dynamic planning framework and to verify proper functioning of the components. +Arrival ExampleWe first consider two arrival flights heading into San Francisco (SFO) along a single route consisting of three schedule points: the arrival fix (E1), the intersection (P1), and the runway (SFO28R), as depicted in Figure 30.Aircraft appear 200 nmi from the metroplex "center" along a given radial (e.g., "E") with an initial speed of 200 knots.The first aircraft target appears at 1000 seconds and the second target appears at the 200-nmi boundary 100 seconds later.The nominal spacing between the aircraft at the 200-nmi boundary is approximately 6 nmi.The minimum required separation at the points along the route is defined to be 60 seconds at the runway (SFO28R) and 90 seconds at the intersection P1 and fix E1.The model governing the time-of-arrival uncertainty is that depicted in Figure 26.Running the framework on this example results in the "time-of-arrival" graphic depicted in Figure 31 at the arrival fix E1.In this graphic simulation time is plotted on the x-axis while various values of the crossing time at the point E1 are plotted on the y-axis.Notionally, this plot allows one to analyze the evolution in various aspects of the crossing time for each flight in time as influenced by the periodic plan updates.For each flight four "lines" are included:• The red lines (with diamonds) depict the latest crossing time for each flight as computed at the beginning of each planning cycle.• The green lines (with asterisks) depict the earliest crossing time for each flight as computed at the beginning of each planning cycle.• The blue line (with circles) depicts the scheduled crossing time for each flight as computed by the Planner and output at the end of each planning cycle.• The black dashed lines depict the actual crossing time for each flight as generated by the Trajectory Model within the Simulation.Several observations regarding this graphic can be made.First, notice that the sequence of STA values for the "leader" flight, FA1, ride along the earliest feasible time line at each planning cycle.Second, notice that the "follower" STA values lie above the "latest time" line for FA2, a consequence of the fact that the "latest time" computation accounts only for delays related to speed control.As described in section 0, in this case no upper bound was placed on the scheduled time at the initial schedule point (E1).Thus, some other form of control-through path stretching or holding-would be necessary to absorb the delay implied by Figure 31.A third observation can be made regarding the vertical distance between the STA lines for FA1 and FA2.At the end of the first planning cycle, when both aircraft are on the order of 200 nmi away from point E1, the separation enforced by the plan is approximately 700 seconds.Recall that the minimum required separation is 90 seconds.The excess padding is a result of the modeled uncertainty, which, based on Figure 26, is on the order of 300 seconds per flight.However, as the flights progress toward the arrival fix, note how the distance between the STA lines gets smaller, reaching a minimum value of 500 seconds as flight FA1 crosses point E1.Note that by the start of the planning cycle at 3100 seconds, flight FA2 is flying along the fastest speed bound.At this point, as depicted in Figure 32, the flight FA2 is ~40 nmi from E1.As such, the required separation between the flights is 90 seconds plus that due to uncertainty, which is on the order of 100 seconds.Thus, only approximately 200 seconds is strictly required.However, since by this point FA2 is flying almost as fast as it can on this segment (as limited by the crossing speed restrictions at point E1), it is unable to make up any further distance prior to reaching E1.Figure 33 shows the variation of the actual speed as determined by the Trajectory Model for each aircraft during this simulation.Here the convergence of the leader FA1 to its fastest speed is apparent as it approaches E1.By comparison, the speed of the trailer FA2 is held low early on to satisfy the separation requirements dominated by the arrival time uncertainty at E1.Then, as the aircraft approach E1, we see the speed of FA2 begin to increase until reaching its maximum speed-at which point it runs out of time to "close the gap" created in the initial planning cycles.Note that the speed at which the aircraft cross E1 is within the speed restriction range-with FA1 at the fast boundary and FA2 at the slow edge of the range.However, note that both aircraft assume a speed of 235 knots after crossing E1-a speed that is outside of the restriction range defined at P1 [200,220].This situation corresponds to each flight riding along the minimum transit-time boundary between E1 and P1, which, as described in section 0, is computed by taking the average of the maximum speeds at either end of the segment ( , ).Such mismatches between the actual speed and the crossing restriction ranges are not unexpected.Currently, the Trajectory Model within the Simulation is not aware of the crossing restriction.Rather, it simply adjusts speed on each plan update so that its actual crossing time matches the latest STA at any given point along its assigned route.Thus, in the current implementation, the Planner is solely responsible for constructing transit-time bounds that are physically realizable and consistent with crossing-point speed restrictions.Clearly the simple "average-speed" model used in this example and described in more detail in section 0 does not satisfy these requirements.Utilization of a more sophisticated speed profile-both within the Planner for estimating transit times and within the Simulation for generating actual transit times-would be desirable to enhance the realism of future studies. +Departure ExampleA second example is presented to illustrate application of the dynamic planning framework to a departure scenario.In this case, the demand set consists of two departures out of San Jose (SJC): FD1 with an earliest gate pushback time of 2600 seconds and FD2 with an earliest gate pushback time of 2722 seconds.Prior to pushback the aircraft are assumed to be located 0.2 nmi away from the departure runway threshold.A constant, default taxi speed of 10 knots is assumed for both flights.The Trajectory Model for departures is designed such that flights will wait at their gate and push only when they can taxi continuously at 10 knots to meet their STA at the runway threshold.Any subsequent plan updates that occur while the aircraft is taxiing to the runway will update the taxi speed according to eq. (34).In this example, the routing structure and speed profile assumed are defined in Table 5, which also includes the minimum and maximum transit time between each successive pair of schedule points.The "earliest time" for each aircraft at the runway (prior to push) is estimated using:These estimates result in earliest feasible runway times of 2672 and 2794 seconds, respectively, for FD1 and FD2.After pushing from the gate, the earliest time at the first schedule point "ahead" of each aircraft is then estimated based on the current time ( ), the current distance to the point, and the current speed obtained from the latest target update.: Table 6 summarizes the planning cycles that were executed in this example, including the time at which the computation for each planning cycle was initiated, the time each plan was implemented, and the STA that was defined in each plan for flight FD2 (the trailer flight) at each of the schedule points on its route.Note first that the plan for FD2 does not change during the first two planning cycles because the earliest gate-pushback time of the flight is 2600 seconds.But since it takes only 72 seconds in this example for the flight to cover the distance from gate to runway, the effective pushback time for the flight is shifted to 2911 -72 = 2839 seconds.Thus, it is not until Plan 2 that the aircraft is actually "active".Secondly, note that, similar to the behavior of the "trailer" flight STAs in the previous arrival example, the STA at each point along its route moves earlier in time as the flight progresses along the route.This reality is, again, indicative of the "excess" padding due to uncertainty that is included in the early planning cycles being removed from the later planning cycles-allowing the flights to be scheduled closer together as time elapses.Figure 34 depicts the variation in STA as well as the earliest and latest bounds for flights FD1 and FD2 as they approach fix N1.The reduction in schedule padding in each successive plan once flight FD2 becomes active is clearly evident in this graphic.Note that the STA at N1 for flight FD2 prior to the flight actually reaching the runway lies above the latest time boundagain because there is no actual latest time bound enforced at the initial scheduling point in this example.The latest bound for FD2 at point N1 in this time range [prior to 2911 seconds] is drawn assuming that the flight will push from the gate at its earliest time.Figure 35 shows the variation in speed required by the FD2 Trajectory Model within the simulation to comply with the sequence of STAs output by the planner in this example.Note that in this example, like in the previous arrival example, the actual speed of FD2 fails to fully comply with the speed restrictions defined at the schedule points. +SUMMARY AND LESSONS LEARNEDThis chapter summarizes what we have achieved over the course of this project and the ways in which we are positioned for future work as a result. +Summary of AccomplishmentsThrough the research conducted under this effort, we have established the feasibility of applying an optimization-based approach to the management of metroplex traffic within a real-time framework.Specifically, we have succeeded in:• Extending previous mixed-integer-linear-program (MILP) formulations applied to network structures to allow for continuous scheduling combined with discrete route assignment and sequencing• Developing and demonstrating a framework for dynamic planning• Identifying and understanding issues associated with trajectory prediction and uncertainty• Applying the MILP formulation-in both static and dynamic contexts-to metroplex interactions, demonstrating the ability for the management strategy to naturally fall out of the optimization process• Supporting NASA Airspace Super Density Operations (SDO) researchers by providing a capability to analyze route topologies and scheduling strategies-specifically providing sequencing and route assignment search degrees of freedom (DOF) not available in their previously existing toolset• Developing an alternate greedy first-come, first-served (FCFS) planner as a basis of comparison for evaluating the benefits of additional sequencing and route assignment options as well as proving the plug-and-play architecture that was implemented as part of this effort• Adapting the basic MILP formulation for performance, including definition of heuristics to reduce or eliminate binary variables in some cases and utilization of alternate linearization techniques to reduce the number of auxiliary variables and constraints required to model a problem instanceFigure 36 shows a timeline view of key milestones related to the development of an optimization-based planner for metroplex operations under this NASA research announcement (NRA) effort. +Lessons Learned and Next StepsSpecific lessons learned over the course of this effort are:• There are challenges in applying an optimization-based approach because optimal solutions tend to lie on constraint boundaries.These challenges are specifically related to balancing two objectives: the desire to find efficient system-wide solutions and the desire to find efficient individual aircraft solutions.Constraint bounds must be realistic and not wholly inefficient from the individual aircraft perspective.• The objective function can be leveraged to shape the behavior of the optimization.Specifically, we have experimented with placing a penalty on both initial and final time, and we have seen that it results in a redistribution of delay as compared to when only the final time is penalized in the objective function.Further, we have realized the value of the addition of a weighted convex function penalizing deviations from aircraft-optimal arrival time in mitigating the tendency for the optimal solution to force aircraft to fly at the fastest speeds defined by the speed profiles on their assigned routes.Going forward, we will consider varying the structure of the cost function contribution for individual flights based on uncertainty and distance to go.• Results obtained to date, specifically within the dynamic planning context, point to the need for consistency between planner and actual aircraft trajectories.• A substantial reduction in computational cost is achieved by applying the effect of heuristics within the base constraints themselves (e.g., by substituting known values or eliminating decision variables from a given constraint) rather than appending them as additional constraints.• The same basic constraints can be converted to linear form in different ways-some more and some less efficient.Recent work has identified alternative formulations that reduce the number of binary variables.Initial tests with these alternative constraint representations look promising with respect to reducing the computation time required to solve a given problem instance.Next steps to be conducted as part of ongoing research in this area include:• Continued prototyping of a hybrid scheme combining an outer stochastic search over the binary variables with an inner pure linear programming scheduling core.Initial testing of this concept shows promise.Such a hybrid scheme is pursued as an alternative to waiting for the pure MILP solve to execute its branch and bound search over the massively large space defined by the combination of binary variables in a typical metroplex problem instance.Through proper design of the individuals within the population, we can (a) seed the solution with a good initial guess, (b) always have a feasible solution on hand, and (c) have a solution that can continue to improve with time.• Continued study and comparison of the MILP-based planner with the greedy FCFS planner in both static and dynamic contexts.The intent of these studies will be to identify the types of problems for which the MILP formulation consistently outperforms the greedy FCFS planner.As part of this effort, the realism of the metroplex problem will be increased-with respect to both shear problem size and trajectory models used in the Planner and Simulation contexts. +NUMERICAL RESULTSThis chapter briefly summarizes the types of numerical results obtained over the course of this NRA effort.In presenting these results, we attempt to provide insight into the following questions:• How does the computational performance of the mixed-integer-linear-program (MILP) formulation(s) scale as a function of the complexity of the problem?• What is the effect of different linearization techniques on solution time?• What is the effect of different cost functions on solution time?• How do solutions obtained with simple, heuristic-based planners (e.g., the greedy firstcome, first-served (FCFS) planner) compare with the MILP solutions? +Analysis of ScalabilityThe time required to solve a given MILP instance increases as a strong function of the number of binary variables in the problem.Given the nature of the constraints defined in the MILP formulation, the number of binary variables scales with:• The number of potentially shared points• The number of flights in the demand set• The number of route options for each flightThis section examines the nature of the computational time required to solve several simple problem instances as we increase the number of flights included in the demand set.The problem instances themselves are selected to pose different types of routing networks and thus underlying combinations of scheduling points and routing options. +Single-Route CaseWe first consider a single route consisting of three schedule points, as depicted in Figure 37: We consider numerous flights in the demand set ranging from N = 4, 5, 6, 7, 8, 9, and 10.The parameters varied in each experimental condition are:• Is the no-passing constraint applied?• Is the FCFS natural-ordering constraint applied?If so, what is the threshold value?Two threshold values are considered, namely zero seconds (FCFS0) and 300 seconds (FCFS300).The variation in solution time for the single-route case is depicted in Figure 38.Note the rather steep increase in solution time as the number of flights increases above N = 8 when the nopassing constraint is not applied.The rate of increase in solution time generally lessens with increasing heuristic strength-becoming nearly linear for the FCFS0 case-in which case all the sequencing variables have been set based on the natural ordering of the flights at their initial scheduling point, leaving only the assignment binary variables to be solved.The reduction in the number of binary variables remaining as a result of applying the various heuristics is summarized graphically in Figure 39.Indeed, the application of the FCFS natural ordering heuristic results in a dramatic reduction in the number of binary variables remaining in the problem instance for a given number of flights, changing the curve from a sharp nonlinear growth to a more gradual linear growth rate.To better illustrate the effect of the size of the demand set on the solution time, Figure 40 shows the same data as in Figure 38, but plots the log (10) of the solver time on the y-axis.Here we can see that the effect of the no-passing constraint is a downward shift of this curve.Addition of the FCFS heuristic results in a further downward shift as well as a reduction in the slope of the curve.Given the curves in Figure 40-obtained using the MILP formulation resulting from the "alternate" linearization technique-the question is: What is the performance of the other linearization (that described by Bemporad et al. [20]) on this same problem set?Table 7 provides some initial insight into the answer to this question.We compare the Bemporad linearization (without NoPass Constraint) with the alternate linearization-both without and with the NoPass Constraint applied.We also compare the solution obtained with the greedy FCFS planner.As can be seen in Table 7, the alternate linearization generally results in a significant reduction in solution time-even without the no-passing constraint applied.The reduction in solution time is even more pronounced when this constraint is added. +Two-Independent-Routes CaseIn this case the demand set is identical in structure to that for the single-route case-namely a "line" of flights stretching to the east, starting at 200 nmi from the runway and initially spaced by 5 nmi.The routing structure in this scenario, however, is as depicted in Figure 41, with completely independent routes going to two separate runways.The separation requirements and speed profiles at each point are analogous to those applied in the single-route case.Each flight thus can potentially be routed on either of the route options.The solution time for this scenario for a number of flights ranging from 4 to 10 is shown in Figure 42, again illustrating the effect of various heuristics on the required solution time for a given demand size.Figure 43 shows the effect of these heuristics on the number of binary variables, while Figure 44 shows the variation in solution time on a logarithmic scale.Figure 42.Variation of solution time with demand size for two-independent-routes case.Figure 43.Effect of various heuristics in reducing the number of binary variables. +Two-Merging-Routes CaseWe now explore how solution-time scales with number of flights in a simple route network consisting of two routes that merge into a common route prior to the runway, shown in Figure 45.The solution time for this scenario for a number of flights ranging from 4 to 7 are shown in Figures 46 and47, indicating the effect of various heuristics on the required solution time for a given demand size.Figure 48 shows the effect of these heuristics on the number of binary variables, while Figure 49 shows the variation in solution time on a logarithmic scale.Figure 47.Zooming in on variation of solver time with size of demand set for two merging routes. +Effect of Penalty on Initial TimeWe also briefly explored the effect of adding a penalty on initial time to the objective function.It results in an objective function of the form: +Single-Route CaseFor the single-route case, Table8 illustrates the impact of adding the cost component related to the time at the initial scheduling point.Note that the objective function value-as defined as the sum of only the times at the final scheduling point-remains unchanged using either cost function.However, there is an impact on the solution time required; the time required to obtain the solution was reduced by approximately 40% for the N = 10 flight case.In addition, although the total delay is the same for both cost functions, the distribution of delay in the two cases is different.For the case where the time at the initial scheduling point is included in the cost function, the delay tends to be incurred within the route (e.g., via speed reduction), whereas with the original cost function the delay is incurred prior to the initial scheduling point. +Two-Independent-Routes CaseA similar trend to that observed previously in the single-route case is observed in Table 9 for the two-independent-routes case.However, the impact of adding the penalty on time at the initial scheduling point is not quite as dramatic in this scenario, but still reduces the solution time required by approximately 30% for the N = 9 flights case. +Two-Merging-Routes CaseAs a final example, we compare the solution times required for the two-merging-routes case when a penalty on time at the initial point is added to the cost function.As summarized in Table 10, again the effect of adding this penalty is to reduce the solution time (by ~30% for the N = 7 case) with an identical solution from the perspective of "final time" and total delay. +Comparison against Greedy FCFS PlannerThis section compares the computational time required to solve a given problem instance with the MILP planner (alternate linearization) with that required by the greedy FCFS planner.We also compare the nature of the solutions obtained in terms of the equivalent objective function value obtained, and the nature of the delays found in the solution.The time required for the greedy FCFS planner to solve problems involving hundreds of flights is inconsequential-on the order of less than 1 second of CPU time.Thus, from a pure speedof-performance perspective, the greedy FCFS planner significantly outperforms the MILP implementation using lpsolve.So the question really becomes: Under what conditions does the MILP solution provide a "better" answer?And what is "better"?It should be noted that the greedy FCFS planner objective function implemented in the initial set of comparisons presented in this section is to select the minimum delay route.This planner is slightly different from the MILP planner, which tries to minimize the total exit time out of the metroplex over all flights in the demand set. +Single-Route Case with Alternating Aircraft TypesWe first investigate a case in which the MILP planner should outperform the greedy FCFS planner from a runway-throughput perspective.We consider a set of eight departures from a single airport on a single runway.All aircraft originate at same location (200,0).Four of the aircraft are "heavies" and four of the aircraft are "smalls".The earliest gate-pushback time for this set of flights is defined such that their natural FCFS ordering at the runway is H, S, H, S, H, S, H, S. The separation matrix used at the runway in this scenario requires 180 seconds for an "S" following an "H" and 90 seconds otherwise.In other words, the natural ordering corresponds to an inefficient use of the runway as the required gaps are larger than if the flights were grouped according to size.Uncertainty in arrival time is assumed to be zero at all points (no buffering).The solution time required by the MILP planner on this problem was ~30 seconds, while the greedy FCFS planner computed its solution in less than 1 second.The solution obtained by the MILP planner was HHHHSSSS with a sum total of gaps of 720 seconds, a total objective function value of 31,165 seconds, and a total delay of 3,880 seconds, corresponding to the minimum delay/maximum throughput solution possible in this scenario.By comparison, the solution obtained by the greedy FCFS planner was HSHSHSHS, with a sum total of gaps of 990 seconds, a total objective function value of 32,244.8seconds, and a total delay of 3,960 seconds.Of course this solution was expected, given that the greedy FCFS planner schedules flights based on their earliest time at their initial scheduling point.Figure 50 depicts the sequence and time of arrival (STA) at the runway for both the greedy FCFS (blue circles) and MILP (red asterisks) solutions.It is clear that relative to an objective function of minimizing the sum of STAs at the runway for all flights, the MILP solution is superior.Figure 51 shows the distribution of delay incurred by each flight as a function of their order in the runway sequence.Again, the MILP solution outperforms the greedy FCFS solution.To summarize: The MILP solution is clearly better in this example.The question to be answered is whether the improvement in solution quality is worth the increased time required to compute it, particularly given the uncertainty present in real-world situations. +Two-Independent-Routes CaseWe now consider the application of the greedy FCFS planner to the two independent routes scenario described previously in section 0, for the case of N = 8 flights in the demand set.The earliest time for each flight at the runway on route 1 is 3,628.4seconds, while that for flights taking route 2 is 3,645.1 seconds.The separation requirements are identical on either route option: 120 seconds at the runway and 100 seconds at all other points.Solution time for the greedy FCFS planner for this problem was less than 1 second.The nature of the solution was to assign successive flights alternately to the two routes, with FA1, FA3, FA5, and FA7 assigned to route 2 and FA2, FA4, FA6, and FA8 assigned to route 1.The equivalent objective function value, the sum of STAs at the runway, was 30,534 seconds, with a total delay of 1,440 seconds.By comparison, the solution time for the MILP planner was approximately 500 seconds.The solution obtained was qualitatively different in that it routed the first four flights (FA1-FA4) on route 1 and the next four flights (FA5-FA8) on route 2. But the objective function value and total delay obtained in the MILP solution was identical to that of the greedy FCFS planner.Thus, equivalent solutions were obtained by the two planners in this case. +Shared-Departure-Fix CaseWe now consider the application of both the MILP and greedy FCFS planners to a problem involving departures from two airports.As depicted in Figure 52, each airport is modeled as having a single runway.The routing structure consists of two fixes, E1 and E2, with a route connecting each airport runway to each fix.Thus, departures from either airport can be assigned to a route going to either departure fix.The demand set for this example consists of four departures from San Jose (SJC) and four departures from San Francisco (SFO), with 30 seconds between successive earliest gatepushback times at each airport.Zero padding is applied associated with uncertainty in time of arrival at any point along each potential route option.The separation requirements at each fix are defined to be a function of the origin airport.Successive flights from the same airport are required to be separated by 60 seconds at the fix, while successive flights from different airports must be separated by 180 seconds.This increased separation is roughly representative of current-day operations in which successive flights from different airports generally have reduced predictability and control relative to successive flights from the same airport.The solution obtained by the greedy FCFS planner resulted in an equivalent objective function value of 15,879 seconds, while the MILP solution had an objective function value of 15,186 seconds.The total delay was 1,080 seconds in both solutions.The nature of the solution obtained by both planners can be observed by looking at the distribution of STAs, as depicted in Figure 53.In this example, the MILP solution, shown in red, consistently results in earlier STAs at the fix than the greedy FCFS solution.Further insight can be obtained by referring to Figure 54, which shows the time at which each flight reached a given scheduling point for both the greedy FCFS and MILP solutions.First, notice that the MILP achieves earlier STAs than the greedy FCFS planner-consistent with the reduction in objective function value.Second, notice that in each case, with a 3x multiplier on required separation for sequential flights crossing the fix from dissimilar airports, a "segregated"-like behavior is evident in which a single fix is predominantly displayed for each airport.Then, in order to investigate the effect of the relative proximity of the two airport runways to the two departure fixes, we consider the solution to the same basic scenario but now with the airport runways equidistant from the fixes.Thus, the distance from SFO to E2 is the same as the distance from SJC to E1.Likewise, the distance from SFO to E1 is the same as the distance from SJC to E2.In this case, the objective function value obtained by the MILP planner is 15,840 seconds, while that obtained by the greedy FCFS planner is 16,497 seconds.As illustrated in Figure 55, the MILP solution again results in STAs that are generally earlier than those found in the greedy FCFS solution.Of course, these results were for a greedy FCFS planner that was attempting to minimize the delay for a given flight.Thus, presented with two route options-each with equal delay but with potentially different STAs at the latest scheduling point-the greedy FCFS planner breaks the "tie" arbitrarily.What would happen if the greedy FCFS planner instead greedily selected from the available route options on the basis of the earliest STA at the final scheduling point?Then it and the MILP planner would be basing decisions on the same objective.In this case, as depicted in Figure 56, aligning the objective functions results in a greedy FCFS solution that matches the MILP solution. +DYNAMIC PLANNING DEPLOYMENTThis section presents information on running the dynamic metroplex planning software. +System Requirements and Configuration• Java Runtime Environment: Install either Java Runtime Environment (JRE) version 6.0 or higher.Be sure that JRE bin directory is in the system path.• MATLAB: Version 7.9.0 was used in testing of this software.• LPSolve : This linear programming library is required to run the suggested planner implementation (i.e., the Mixed Planner).Library files applicable to the operating system must be placed in the metroplex root directory (e.g., C:/metroplex/).o On UNIX or Linux: Copy the .aand .solibrary files appropriate to the bit type of the operating system to the metroplex directory (see sub-bullet). 32-bit original library files: Copy liblpsolve55.32bit.a,liblpsolve55.32bit.so,and liblpsolve55j.32bit.so to liblpsolve55.a,liblpsolve55.so,and liblpsolve55j.soappropriately. 64-bit original library files: Copy liblpsolve55.32bit.a,liblpsolve55.32bit.so,and liblpsolve55j.32bit.so to liblpsolve55.a,liblpsolve55.so,and liblpsolve55j.soappropriately.o On Windows: Copy appropriate operating system bit type .dllfiles to lpsolve55.dlland lpsolve55j.dll in the metroplex directory. +Application ConfigurationConfiguration options are available to the dynamic planning application via an application context file (see /config/app/dynPlanAppContext.xml).The application context consists of various application components that are specified as Java beans using Extensible Markup Language (XML).These planning application components are described in this section. +Planning ControllerThe planning controller references and controls access to all of the components within the application context. +Plan Constraint ManagerThe plan constraint manager holds various constraints to be considered by the Planner.Table 11 describes the constraints.In order for flights to be considered for FCFS ordering, the difference between their times being compared must be greater than this threshold.For example, if useFirstCommonPointFcfs is true, the difference between the unimpeded times at the first common point for two flights must be greater than the minThreasholdSecondsFcfs value.[doWindowing Break down demand into a sequential set of smaller demand "windows".Set property windowSeconds for the time size of each "window".true | false windowSeconds Size of each time "window" in seconds.Property doWindowing must be true for this to be applicable.[0, 1, 2, … n] timeoutSeconds Time at which the lpsolver will terminate and provide its current solution.If set to zero, no timeout will be applied. +PlannerImplements the logic for planning flights on resources, given constraints within the metroplex.The Planner must implement the IPlanSolutionCalculator interface.The following planner implementations are available:• Mixed Planner: Uses heuristics and linear programming in calculating plans.• FCFS "greedy" planner: Assigns flights to least-expensive route options in terms of the latest scheduled times for successive flights; assigns common flight resources on a firstcome, first-served (FCFS) basis. +Scenario File ManagerThe scenario file manager specifies the local file system directory containing the scenario files to be used in dynamic planning.This directory is the working directory for all input and output files used in the current dynamic planning application instance.The following files should be within the scenario directory:• Adaptation File: This file defines the metroplex airspace, including the routes, runways, fixes, and waypoints.Separation requirement elements are used to specify time separations between flights at the schedule points.Speed profile elements are used to specify minimum and maximum crossing speeds at the schedule points.• Navigation/Performance Model File: This file defines aircraft type-specific speed and fuel usage in the overall metroplex, at fixes, and/or at runways.The speed definitions may be used by the Planner.For example, the Mixed Planner uses speed definitions in estimating time of arrival at schedule points for enforcing FCFS constraints.• Demand File Directory: This file provides the location for the simulator to output demand set files.The dynamic planner uses these files as input for scheduling metroplex resources.• Plan File Directory This file provides a location for the planner to output plan files.The simulator processes the context of these files to simulate implementation of the flight plans. +Dynamic Plan ManagerThe Dynamic Plan Manager manages parameters specific to dynamic planning.The parameters include the following:• Solution coverage seconds: Time horizon covered by a plan.• Plan computation time seconds: Time estimated for the planner to compute a plan on a demand set.• Iteration interval seconds: Frequency at which the planner is to be invoked.• Raw demand file: The "target report" input to the simulator; it specifies the time at which each "target" appears, trajectory information, and basic flight plans.• Latest demand set reference file: Holds reference to the latest demand set.• Latest plan reference file: Holds reference to the latest plan.• Freeze strategy: Refers to the application context freeze strategy.• Apply freeze horizon: Specifies whether to apply the freeze strategy in planning.• Freeze horizon: Specifies the freeze horizon to be used by the freeze strategy. +Demand ManagerLoads the demand sets from the specified data source.Currently the demand manager implementation loads the demand sets from a file.Note: Future demand managers may load demand sets programmatically. +Navigation Performance ManagerThe navigation performance manager loads the navigation/performance model file to instantiate navigation/performance models. +Uncertainty CalculatorThe uncertainty calculator computes expected uncertainty of the time of arrival of the flights at schedule points along prospective routes.Uncertainty calculators must implement the IUncertaintyCalculator interface.Uncertainty calculator implementations available are the following:• UncertaintyCalculator: Applies scaled uncertainty buffer times to flights based on their distance to respective scheduled points.• UncertaintyCalculatorUnScaled: Applies rigid unscaled uncertainty buffer times to flights based on their distance to respective scheduled points +Simulator StartupThe simulator requires "target report" information.A raw demand file (as specified in the application context file) is used to provide this information as input to the simulator.The simulator must be launched from MATLAB.The following steps describe how to launch the simulator:1. Start MATLAB.2. Set the MATLAB Current Folder to metroplex root directory.3. Show the Current Folder panel on the MATLAB Desktop by selecting Desktop -Current Folder.4. Within the Current Folder panel, navigate to /Optimization/Simulation/ MatLab/.5. Select file DynamicPlanningSimMain.m, right mouse click it, and select Run File.The simulator reads in the raw demand set file, produces a demand set file to /DemandFiles/, and updates the latest demand set reference file.Then a message stating that the simulator is waiting while the plan iteration completes should show to the MATLAB command window:Waiting for plan for iteration 0 to completeThe simulator shuts down if an update to the latest plan reference file is not detected within the default timeout time of 120 seconds. +Dynamic Planning VerificationWhen the simulator completes implementation of a plan, it produces a demand set file and writes it to /DemandFiles/.Likewise, the planner produces plans and writes them to /PlanFiles/.Until either the simulator or planner is shut down, demand set files and plan files are written to the appropriate sub-directories under the scenario directory.In addition, the simulated flight information is available in MATLAB workspace objects for verification.This section describes how to verify dynamic planning via these outputs. +Demand Set File VerificationWhen the time stamp of a flight "target" (as found in the scenario raw demand file) is found within the planning horizon, an entry for that flight is written to the demand set file produced by the simulator.The X/Y coordinates of a flight previously "scheduled" by the planner should approach the next schedule point of the route currently assigned to the flight.The speed of a flight should ideally fall within the maximum and minimum required speeds of the next schedule point and previous schedule point (if any). +Demand Set File ContentsDemand set files produced by the simulator contain the following data:• Elapsed seconds: The time that has elapsed since the start of the startup of dynamic planning.This parameter may commonly be referred to as the "plan time".• Flights: Data for each flight.o ID: Flight ID.o Type: Aircraft type corresponding to the "type" attributes of the elements specified in the Navigation/Performance scenario file.o Category: "A" for arrival to a metroplex airport or "D" for departure from a metroplex airport.o Origin airport: Origin airport code (significant for metroplex departures).o Destination airport: Destination airport code (significant for metroplex arrivals).o Direction: Directions are with respect to the "center" of the metroplex.(Important: For a flight to be scheduled on a metroplex route, the direction value must correspond with the direction attribute value for one or more fixes as defined in the scenario adaptation XML file). For arrival flight: Direction from which a flight is entering the metroplex at its arrival fix. For departure flight: Direction to where a flight is exiting the metroplex at its departure fix.o Current X nmi: X coordinate in nautical miles with respect to the "center" of the metroplex.o Current Y nmi: Y coordinate in nautical miles with respect to the "center" of the metroplex.o Current altitude feet: Altitude in feet.o Current heading degrees: Heading in degrees with respect to the "center" of the metroplex.o Current speed knots: Ground speed in knots.o Route (optional): Assigned route for the flight.o Fix (optional): Assigned fix for a flight. +Plan File VerificationAs a demand set is processed by the planner, each flight is assigned a route.For each schedule point ahead of the flight along the route, crossing times are estimated based on resource constraints and aircraft performance limitations.If a freeze strategy is configured for the dynamic planning session, a flight found within the freeze horizon of an applicable schedule point will have its route assignment frozen and possibly its scheduled times of arrival (STAs) on one or more schedule points remaining along the assigned route. +Plan File ContentsPlan files produced by the planner consist of the following data:• Elapsed seconds: The time that has elapsed since the start of the startup of dynamic planning.This may commonly be referred to as the "plan time".• Flights: Data for each flight.o ID: Flight ID.o Route: Currently assigned route.o Route assignment frozen: Boolean flag indicating whether the route assignment is "frozen".o Schedule points: Fix, waypoint(s), and runway found ahead of the flight on the currently assigned route. Reference ID: Schedule point identifier corresponding to the ID used for the schedule point in the scenario adaptation file. Scheduled crossing time seconds (STA): Crossing time as scheduled by the planner. Schedule cross time frozen: Boolean indicator as to whether the scheduled crossing time is frozen. Earliest time seconds: Earliest time physically feasible for the flight to cross the scheduled point. Latest time seconds: Latest time for the flight to physically cross the scheduled point. Unimpeded crossing time seconds: The aircraft-optimal time at which the flight would cross the schedule point if allowed to fly completely uninfluenced by the presence of any other flights. +Verification in MATLABThe simulator in MATLAB provides a means for animating the implementation of the flights based on their schedules.Plots may be generated based on MATLAB data objects resulting from the dynamic planning simulation. +Simulation AnimationUpon simulator startup, a MATLAB figure window displays the metroplex adaptation and the active flight "targets".As the simulator detects and implements newly received plans, the flights advance to their next schedule points to meet the currently scheduled STAs.Flights should advance along their assigned or frozen routes.Flights should maintain required separation at schedule points both in cases of leader/follower instances along the same route and in cases where assigned routes intersect.Figure 57 shows an animation of arrival flight FA1 approaching runway SFO28R and departure flight FD1 approaching schedule point DP1 after departing from runway SJC30L.The adaptation demonstrates the use of routes intersecting at a common schedule point P1. +Plot VerificationUpon completion of a dynamic planning and simulation session, plots may be generated on the resulting MATLAB data objects.Plots such as the ones in Figure 58 may be used to compare various times of arrival.These plots were generated using script /Optimization/ Simulation/MatLab/createTimeOfArrivalGraphic.m.The lines are explained as follows:• Dashed line: Actual time of arrival (ATA) at schedule point.• Red circles (small) and red lines: Scheduled time of arrival (STA) at schedule point.• Red circles (large): Frozen STAs on schedule point.• Blue X symbols and lines: Aircraft-optimal time of arrival at schedule point.• Green asterisk symbols and lines: Earliest physically feasible time of arrival at schedule point.• Magenta diamonds and lines: Latest physically feasible time of arrival at schedule point.A few things worth noting regarding these two plots follow:• As the flights approach the schedule point, their STAs converge on the ATAs.• STAs do not change once frozen for a flight on the schedule point.(See the large red circles in the figures.)• STAs may be found outside of the estimated earliest or latest times of arrival if the aircraft is still "outside" of the metroplex.For example, at 1300 seconds the STA for FA2 exceeds the estimated latest time of arrival.This "spike" in the STA for FA2 is due to the planner scheduling departure flight FD1 ahead of FA2 at common schedule point P1.This STA value is reasonable since FA2 is over 180 nmi from the radial boundary of metroplex.• Earliest and latest times of arrival typically approach but often do not intersect with the ATA.• The time gap between the earliest and latest times of arrival for a flight will decrease as the flight approaches its ATA.To produce the time-of-arrival plot, execute the script in the following manner:• createTimeOfArrivalGraphic(2,flights, 'P1')Parameter description:• 2: Arbitrary number for the MATLAB figure.• flights: Data collection of flight objects that hold the time-of-arrival data.• 'P1': Name of the schedule point of interest. +ORGANIZING METROPLEX TRAFFIC BASED ON SPEED SEGREGATION AND TRAJECTORY FLEXIBILITYThe growth in air traffic demand and volume is increasing the number of operations at major and secondary airports.As a result, the interdependencies between nearby airports are also increasing, leading to the emergence of multiairport systems (metroplexes).Often, these metroplexes constitute the bottleneck capacities of the national air transportation network and a main cause of delay.The capacity limitations are caused by several inefficiencies, among which are the delegation of airspace among competing airports based on procedures that segregate traffic and assign airspace by destination airport and the mixing of slow and fast aircraft in single flows.This report proposes a new approach for delegating airspace and segregating traffic according to aircraft speed as opposed to the destination airport, with potential throughput gains.The proposed approach is analyzed, using a simulation of hypothetical scenarios, in terms of throughput and trajectory flexibility. +IntroductionExpectations for the Next Generation Air Transportation System (NextGen) environment include up to three times the current traffic demand by the year 2025 [21].In order to handle this traffic volume, it is essential to increase the capacity at the bottlenecks of the air transportation network, which are typically the airports.One key capacity limitation at airport systems is the increasing interdependence between nearby airports.This interdependence leads to the emergence of multiairport systems where one airport has to limit its operations in order to accommodate the operational needs of a neighboring airport.Therefore, the operation of these airports requires coordination between the traffic managers to maximize the overall throughput.A typical example of a metroplex is the New York metropolitan airport system, which includes four major airports: John F. Kennedy (JFK), LaGuardia (LGA), Newark Liberty International Airport, Newark, New Jersey (EWR), and Teterboro Airport, Teterboro, New Jersey (TEB), within 20 miles of each other, in addition to numerous secondary surrounding airports.The capacity limitations at a metroplex are caused by several inefficiencies, among which is delegation of airspace among competing airports based on procedures that predate the highdensity environment and hence are not optimized for current high-density operations.In particular, these procedures segregate traffic and assign airspace by destination airport, a situation that may not be optimal from a throughput perspective.Another major inefficiency results from mixing slow and fast aircraft in single flows.This report proposes a new approach for delegating airspace and segregating traffic according to aircraft speed as opposed to the destination airport, with potential gains in terms of throughput.The proposed approach is analyzed, using a simulation of hypothetical scenarios, in terms of throughput and trajectory flexibility.First, in order to provide context, the specific types of metroplex interdependencies and inefficiencies that are addressed in this report are presented in section 0 along with proposed solutions.Then the analysis approach and design are presented in section 0, including the simulation tool that was used.Preliminary results and insights gained from early analysis are discussed in section 0. Conclusions and future extensions are then described in section 0. +Metroplex Inefficiencies and Proposed SolutionsThe airspace and traffic in terminal areas (and sometimes in the Air Route Traffic Control Centers (ARTCC)) are segregated largely based on destination airport.This destination-based segregation often causes inefficiencies in high-density conditions, for example:1. Loss of capacity due to early mixing of aircraft with different speeds in a single airport/runway flow.Sequencing a slow aircraft behind a fast aircraft requires a large spacing between the two aircraft (due to wake vortex separation requirements).This separation opens up with time, because of the speed difference, if the merging is established early.Sequencing a fast aircraft behind a slow aircraft causes slowdown of the fast aircraft to match the speed of the slower aircraft.It also causes excessive spacing if the merging is established early with different speeds to avoid closing the spacing to a value below the separation requirement.2. Excessive travel distance due to procedural separation of the routes used by the different airports, as explained in the example that follows.3. Loss of capacity due to procedural delegation of shared airspace and routes to flows of different airports.For example, an airport may lose the usage of certain runways when a shared airspace is delegated to another airport in the metroplex.This phenomenon is also explained in the example that follows.One example of cases ( 2) and ( 3) is the interdependency between LGA and JFK, based on field observations at the New York metroplex [22].Specifically, when JFK lands instrument landing system (ILS) on runway 13L, LGA is obligated procedurally to land only ILS on runway 13.In addition to limiting the available arrival runways to only one, this procedure often reduces LGA to single runway operations when aricraft can depart only on runway 13.While this situation occurs only a few times a year, it is known to be the most limiting situation at LGA.The reason for this procedure, as clarified in the LGA standard-operating-procedure (SOP) excerpt in Figure 60, is sharing airspace areas 15 and 19.Typically, LGA owns area 15 from 10,000 feet and below and area 19 from 12,000 feet and below, except when JFK lands ILS on runway 13L.Under this condition, LGA has to give up altitudes 4,000 feet and below to JFK, to be used by the JFK arrivals.As shown in the diagram, LGA flights in this condition have to fly higher above the JFK arrivals and perform an additional loop, during which they also stay above the EWR and TEB traffic, and then descend and approach runway 13.Therefore, LGA flights fly a longer distance at higher altitude, inefficiently.Two potential methods to help mitigate these types of inefficiencies are:Reducing the effect of speed mixing by segregating the traffic by speed where possible.Initial ideas of segregation by speed for a single airport were published in reference [23] Generalizing this approach to a metroplex environment, traffic may be segregated by speed as opposed to, or in addition to, by the destination airport.This setup is shown in Figure 61 notionally.For example, by using multiple parallel downwind and base leg segments, the fast aircraft (larger triangles) are assigned on outer downwind/base legs relative to the slower aircraft (smaller triangles), which are assigned on inner downwind/base legs.Airports are shown as crosses in the figure.Allowing sharing of approach segments (such as downwind and base legs) by arrivals to neighboring airports.This method can be combined with speed segregation where, for example, aircraft of the same speed category but heading to different airports are assigned on the same approach segments.An example is shown in Figure 61, where the fast aircraft heading to two different airports (blue and red large triangles) share a downwind leg.The hypothesized benefits of this approach include:1. Increasing throughput by delaying the merging between slow and fast aircraft.3. Increasing throughput and reducing travel distance by sharing of airspace and routes currently delegated procedurally to different airports.In the example, LGA flights landing on runway 13 could share airspace areas 19 and 15 below 4000 feet with the JFK arrivals to runway 13L.Then each could perform a late exit to the respective airport.If prioritization is needed, this sharing can be allowed for some of the LGA flights only when there is a lull in the JFK flow to runway 13L, thus reducing the longer and higher travel for at least some of the LGA flights.It must be noted that such sharing methods require new controller procedures and possibly automation support.These requirements are not addressed in this report, which focuses mainly on making the benefit case. +Analysis Approach and DesignThis report presents preliminary experiments for proof of concept of the traffic organization approach described in the previous section. +Analysis PlanThe analysis compares the following scenarios: +Single airportOne approach path with mixed aircraft speeds (baseline)Multiple approach paths with different speed category per path Two airports One approach path for each airport with mixed aircraft speeds (baseline)Sharing speed-based approach segments among the airportsThe analysis compares the scenarios given in terms of the following metrics:1. Throughput 2. Trajectory flexibility (explained in section 13.3.2) +Simulation ToolThe tool used for the analysis is a MATLAB simulation environment that was developed for trajectory flexibility planning in the presence of controlled arrival time and hazard/traffic avoidance constraints [24].Trajectory flexibility is defined as the ability of a trajectory to accommodate disturbances while meeting constraints such as controlled time-of-arrival constraints and traffic-/hazard-avoidance constraints.Disturbances are events that pose risk of constraint violation, such as the uncertainty in the traffic/hazard dynamics.Relevant trajectory characteristics to measuring flexibility were identified; they included robustness and adaptability.Robustness is defined as the ability of a trajectory to remain feasible given disturbances, while adaptability is defined as the ability to regain feasibility if feasibility is lost because of disturbances.Details of formal mathematical definitions of these metrics and estimation techniques are given in reference [24].The tool implements a dynamic programming approach to estimate the trajectory flexibility metrics and to use them (along with other objectives) for trajectory planning.These methods are described in reference [24].Briefly, the method is based on discretizing space (currently only two dimensions) into square cells and time into steps.Then a solution space of all trajectories is built as a reachability tree connecting the resulting discrete nodes (each node represents a location as the center of a cell and time).The tree is based on reachability given discrete degrees of freedom (DOF), namely allowable speed and heading changes with discrete increments and within given ranges.At each node of the tree, adaptability is measured by the number of feasible trajectories that reach from that node to the destination (defined as a location with a time-ofarrival constraint).To estimate this number of feasible trajectories, a convolution process is used that starts at the destination and proceeds backwards, adding up feasible trajectories within the reachability bounds from each node.The convolution process is preceded at each time step by a filtering process that zeros out the number of trajectories at nodes that violate any constraints.The constraints include violation of separation from hazards and from other traffic or violation of time-of-arrival constraints at a node.After building the solution space tree with the metrics at each node, a trajectory is computed using a dynamic program that optimizes an objective function.For more details, see reference [24].The objective function used in this preliminary analysis maximized adaptability at each node along the trajectory.The tool has been developed initially for an en-route environment but has been partially adapted for simulating a terminal/metroplex environment.In particular the aircraft were forced to follow typical trombone approach patterns.The trombone structure was assumed in order to isolate the effect of the speed segregation only.In other words, the terminal route structure was maintained and the aircraft were assumed to follow it, because changing or relaxing this structure introduced other interactions that are out of the scope of the concept of speed segregation and resource sharing among airports.It is expected that if this structure were relaxed, there would be more interaction between aircraft because, for example, they would path stretch along the downwind as well as the base leg.This added interaction might create another bottleneck upstream of the runway and reduce throughput because of a factor other than speed mixing.Therefore, to isolate the speed factor and not deal with the effect of increasing interaction between trajectories, the aircraft were assumed to follow the route structure strictly (in this case the trombone approach).Polyhedral hazards were introduced in the simulation tool in order to design the terminal airspace and approach patterns by blocking out certain areas.Figure 62 gives an example screen capture for the two airport scenarios described in Figure 61.Trombone approach patterns were forced by blocking out the ability of the aircraft to path stretch along the downwind and runway centerline, allowing path stretching only along a base leg.Polygons were also introduced to limit the extremities of the airspace to about 50 nautical miles representing typical terminal airspace size.The hazards were also made dependent on the aircraft, such that each aircraft may be forced to a particular approach pattern by applying specific hazards to it.This scenario enabled, for example, giving slow aircraft different patterns than fast aircraft in some experiments.Other elements of the trombone were established by manipulating the speed and heading limits of the aircraft.While in the en-route environment, aircraft often speed up as well as slow down; in the terminal area they mostly only reduce their speed under strict control and in gradual steps.To model this behavior, speed and heading limits were made a function of time, a situation that enabled forcing the speed to monotonically decrease along the approach and within limits representative of the speed step-downs from the terminal entry speed towards the landing speed.Constraints were also added to the dynamic program to force either clockwise or counterclockwise heading changes along the trajectory, depending on the geometry of the source-airport pair.If the airport is to the left of the downwind segment, the aircraft are allowed to turn only counterclockwise.If the airport is to the right of the downwind segment, then the aircraft are allowed to turn only clockwise.However, the analysis presented currently is performed under simplifications that were not removed.For example, the analysis is performed in two dimensions only, while extending the tool to incorporate altitude is underway.The separation requirement is set to 3 nautical miles between all aircraft independent of weight category and wake vortex requirements. +Simulation Results and ObservationsThis section presents a preliminary analysis.Two scenarios are analyzed: a single-airport scenario and a two-airport scenario as mentioned in the previous section. +Single-Airport ScenarioThe scenario analyzed includes a single airport with speed segregation over multiple approach paths.Two cases are compared, as depicted in Figure 63: The scenario includes the following parameters:• Two landing-speed categories are used: Fast with 120 knots and slow with 80 knots.• Landing speed is met with 30-knot tolerance.• Maximum demand is assumed available at all time (to determine maximum throughput).• Aircraft speed categories are alternated in the demand stream to maximize the speed mix effect.• To maximize throughput, the time of arrival (STA) is minimized for each aircraft using the following algorithm:o For each aircraft i  Assign STA(i) = STA(i-  Separation requirement is set to 3 nautical miles, independent of the aircraft type. Path stretching is allowed only using trombone of the base leg. Time is discretized using 2-minute time increments, and space is discretized using 1 x 1 nautical mile square cells. The spatial span of the scenario is 50 nautical miles.Throughput and trajectory flexibility (adaptability) were compared between cases (a) and (b).As shown in Figure 64, throughput was higher in case (b), segregating speed categories onto two trombones, than in case (a), which combined the two speed categories in one trombone.In this example the throughput was higher by about 20%.The 10 aircraft in the scenario landed in 1020 seconds in case (b) compared to 1320 in case (a).However, it should be noted that this analysis is a theoretical simulation that is not validated against current operations.Therefore, the increase in throughput should not be interpreted as a potential increase if such a procedure is adopted relative to current operations, since the baseline may be different and may include some degree of segregation by speed and delay of merging, to the extent practiced by controllers.In terms of trajectory flexibility, it was observed that the faster aircraft have lower adaptability relative to slower aircraft in both cases (a) and (b), as shown in Figure 65.This difference is mainly due to the smaller speed range that is available to the faster aircraft (between the maximum speed and the landing speed).Some of the difference may be due to impedance by slower aircraft.It is clear in Figure 65 that the first aircraft has higher adaptability, because it does not encounter any aircraft ahead of it.Early signs were also observed that adaptability of the fast aircraft may increase (as shown in case (b) relative to case (a) of Figure 65) when they are separated from the slower aircraft and placed on an outer trombone.This increase in adaptability is due to at least two aspects:(1) providing more airspace for maneuvering to the aircraft that are placed on the outer trombone and (2) providing more speed for maneuverability to the faster aircraft, which can stay at higher speed longer in case (b) than in case (a) where they are mixed with the slower aircraft on single trombone. +Two-Airport ScenarioThe scenario analyzed in this section includes two airports with speed segregation over multiple approach paths.Two cases are compared in Figure 66:Case (a): Each airport has a single downwind leg, shared by all speed categories heading to the airport.In this case the two airports operate independently with separate approach paths.Case (b): The two airports use three downwind legs, two of which are dedicated to the slow traffic of each of the airports, with each airport using the leg closer to it.The third central downwind leg is shared by the fast traffic heading to either airport.The scenario includes the same parameters as in the single-airport scenario, where the aircraft speed categories are alternated in the demand stream for each airport to maximize the speed mix effect.They are also alternated between airports to maintain equal loads on the two airports.The STA is minimized for each aircraft (to maximize throughput) using the same algorithm.Time is discretized using 1-minute time increments and space is discretized using 1 x 1 nautical mile cells.Throughput and trajectory flexibility (adaptability) were compared between cases (a) and (b) in Figure 67 and Figure 68, respectively.As shown in Figure 67, throughput was higher in case (b), segregating speed categories onto two trombones with a shared central downwind for the fast aircraft, than in case (a), which combined the two speed categories in one trombone for each airport.In this example the throughput was higher by about 13% for the first airport and by about 6% for the second airport.The 10 aircraft landed in 480 seconds less for airport 1 and in 240 seconds less for airport 2, in case (b) compared to case (a).The difference between two airports is attributed mainly to the order in which the aircraft were introduced, which caused the second airport to lose some benefits because its first fast aircraft has to platoon behind the first fast aircraft heading to the first airport along the shared centerline.The gain is not as high as 20 percent, which was the gain of both airports if each had two independent trombones for its slow and fast aircraft without sharing resources with the other airport.The reduction in the gain is because while speed segregation increases throughput, the central downwind leg is shared between the two airports.Despite sharing, the speed segregation resulted in a throughput benefit in the range between 6 and 13%.Again, it should be noted that this analysis is a theoretical simulation that is not validated against current operations.Therefore, the increase in throughput should not be interpreted as a potential increase if such a procedure is adopted relative to current operations, since the baseline may be different and may include some degree of segregation by speed and delay of merging, to the extent practiced by controllers.It should also be noted that sharing the downwind segment enables other benefits that are not measured in this analysis-for example, enabling the airports to use additional runways or travel less distance, as described in section 11.In terms of trajectory flexibility, signs were also observed that adaptability of the fast aircraft may increase (as shown in Figure 68 in case (b) relative to case (a) for airport 1) when they are separated from the slower aircraft and placed on a separate downwind leg.The logarithm of adaptability (log of the number of feasible trajectories at each point) is plotted because adaptability grows exponentially with negative time.This increase in adaptability is due to at least two aspects: (1) providing more airspace for maneuvering to the aircraft that are placed on the outer trombone for each airport and (2) providing more speed for maneuverability to the faster aircraft, which can stay at higher speed longer in case (b) than in case (a) when they are mixed with the slower aircraft on single trombone.The gain in adaptability for the second airport is not as noticeable as for the first airport.One reason may be because of the order in which the aircraft are introduced.The fast aircraft of the second airport are introduced behind those of airport 1 on the shared downwind leg.This situation may have impeded the aircraft of airport 2 and resulted in less adaptability in addition to less throughput for airport 2. However, more analysis is needed to investigate the difference. +ConclusionsA preliminary analysis was described in this report as a proof of concept for organizing traffic in a metroplex based on speed segregation as opposed to (or in addition to) destination airport.This organization involved two techniques: (1) Sharing airspace and route resources among airports, thus opening up runway capacity and shortening travel distance and (2) segregating aircraft by speed, thus reducing the mixing between fast and slow aircraft in the same flow.This preliminary analysis demonstrated encouraging signs of benefiting from the solutions proposed.The benefits were demonstrated in terms of increasing throughput and trajectory flexibility, which allows better mitigation of the risk of constraint violation.For a single-airport scenario the theoretical increase in throughput was up to 20% because of speed segregation.For a two-airport scenario the increase in throughput was less, ranging between 6 and 13%, because of sharing resources between the two airports in addition to the speed segregation.The analysis presented in this report is preliminary, and further analysis of additional scenarios is needed.Future extensions of this research include validating the benefits by analyzing historic data and using a baseline that represents current operations in terms of the level of speed segregation that is practiced.Future research may also investigate the added benefits of increasing runway capacity and reducing travel distance because of sharing resources among airports more effectively than in current procedures. +AIRCRAFT MODELSThis section describes the aircraft models required as a part of the metroplex management solution approach.The aircraft models will be used in two ways.In the planning phase, the models will be used to provide feasible ranges of flight time along route segments and the associated cost function contributions.In the simulation phase, the models will introduce some error and possibly be given additional constraints along a route, such as required times at intersection points, and will be used to provide the actual aircraft motion.The predefined routes should be realistic.Therefore, we need the ability to define detailed routes between runways and fixes that aircraft are capable of following.We may use the aircraft models in advance to demonstrate that the routes are feasible.Note that the "aircraft models" from our perspective include not only the vehicle dynamics but also the vehicle control and guidance to follow defined routes.To support the planner, we need an aircraft model for an aircraft following a route where the route does not specify "longitudinal" motion (i.e., speed along the route).The aircraft model must include a model for the speed along the route.The aircraft model should represent aircraft turns as well as the vertical profile climbing and descending of the aircraft.When the route does not include vertical restrictions, or when it includes very loose restrictions, the aircraft model should represent an efficient/optimal climb/descent profile.The output of the model will be information such as the fastest the aircraft can traverse the route (and the associated cost) and the slowest the aircraft can traverse the route (and the associated cost), so that the planner may plan separation at intersection points.In addition to a route, the planner may assign the aircraft required times (RTs), which may be a point in time at which the aircraft should cross the specified position or a time window during which the aircraft should cross the specified position at various points along the route, including the runway, fix, and points where routes intersect.The aircraft model should be able to comply with these RTs or provide an error if an RT is not achievable.Eventually, the aircraft model should have different models/parameters to represent different aircraft types.In this project, we used a single aircraft model.To support modeling what "actually" happens when aircraft try to follow the plan produced by the metroplex planner, the aircraft model should be able to vary the aircraft performance in complying with the route (in both lateral and vertical dimensions) and RTs.This capability will also be used to study the impact of aircraft navigation performance on metroplex management efficiency and airspace structure design.Simulating the "actual" result of the planner output also allows us to study a dynamic situation where the planner is confronted with the actual results of the prior plan and must produce a revised plan. +Aircraft Dynamics ModelThe three-dimensional point-mass equations of motion for an aircraft are considered adequate for route feasibility in terms of maintaining realistic aircraft physics, as well as providing state information for optimization purposes.The point mass equations of motion predominantly given in aircraft performance texts do not include wind effects, and also usually consider the thrust to be aligned with the velocity vector.While these assumptions are valid for preliminary analysis, as fuel-burn modeling accuracy becomes more crucial, the effects of winds and thrust direction with respect to the velocity vector also become more critical.With that said, the equations of motion and fuel burn, as given in reference [25], are slightly more detailed than typically found in undergraduate-level texts and are expressed as: It should be pointed out that since no rigid-body dynamics are involved here, commanded rotations such as t α and a φ can be considered as instantaneous.If the effect of vehicle transient times to achieve these input values were ever deemed to be important, rate limiting the control inputs according to ( )( )( ) ( )( ) ( )( ) ( ) ( )V m T D g W W W V m L T g W W W V L m D mmin max t δ δ δ ≤ ≤   , where δ represents the control input, could be incorporated.Alternatively, the input could be passed through transfer functions, representing actual vehicle lag times expected to be seen, to achieve the transient effects without the need to resort to constructing full six-DOF equations of motion model.Besides initializing the vehicle simulation with initial state values, additional values will be used to determine a "nominal" flying condition, based upon a straight and level cruise condition; subsequent control variations will be made for maneuvering beyond this particular operating point.These nominal values are defined accordingly: Constructing a simple autopilot would be the easiest way to get the vehicle to track commands.Simulink provides a framework for rapid prototyping such a control system.While the aircraft model is only a point mass model, some of the conventional linearized 6-DOF autopilot design techniques can still be used.For example, the point mass model still maintains a phugoid-type motion that requires damping.The basis of the autopilot design will account for the guidance methodology and what corresponding vehicle states or combination of states the guidance will request for the vehicle to travel along a specified trajectory.In short, the autopilot must be designed with regard to the commands coming from the guidance routine.A model of the autopilot is illustrated in Figure 69.The inputs into this model represent commands coming from the guidance system and are depicted in red, while the outputs, in turn used by the guidance system, are shown in green.The input commands to the autopilot are: (1) bank angle needed to generate the turn rate requested by the guidance algorithm, (2) altitude associated with a given point in time along a specified route, and (3) velocity needed to satisfy the specified-time-ofarrival (TOA) or required-time-of-arrival (RTA) values associated with the waypoints.Some notable features of this model are that it uses proportional-integral control for achieving zero steady-state error in velocity tracking as well as using damping on the velocity and altitude feedback.Limiters are also used on the angle-of-attack (AOA), thrust, and bank-angle inputs going into the vehicle equations of motion.These limiters play a crucial role in determining whether or not a proposed route is achievable.In conjunction with the use of limiters, the load factor is also monitored to ensure the vehicle remains within the load-limited envelope (e.g., within the envelope described by the V-n diagram of the vehicle).It should also be noted that while the equations of motion are modeled about the true velocity vector, which is not the same as the inertial velocity vector, the actual control law uses inertial values for the feedback law, ensuring that the vehicle passes through the waypoints at the specified times regardless of the wind conditions.There are drawbacks in this method, and therefore it is subject to change in future analysis. +Guidance ModelA guidance law is required to direct the aircraft dynamics model to follow the route.The route provided to the guidance is in the form of an array, where each row of the array contains: