Document
AIA A-XXXX-xxxx Control Reallocation Strategies for Damage Adaptation in Transport Class Aircraft Karen Gundy-Burlett, K. Krishnakumart, Greg Limes', Don Bryan?
Abstract technologies and mission needs drive the upgrade of specific components. These combined effects mean that if one is to field a neural flight control system on This paper examines the feasibility, a transport aircraft, one must consider how to retrofit potential benefits and implementation issues the technology to the existing aircraft designs, rather associated with retrofitting a neural-adaptive flight than rely on a completely new aircraft. While it is control system (NFCS) to existing transport aircraft, conceptually easy to think of software retrofits for including both cable/hydraulic and fly-by-wire configurations. NFCS uses a neural network based fly-by-wire planes, there will continue to be a large direct adaptive control approach for applying alternate number of cable-hydraulic aircraft in the fleet for the sources of control authority in the presence of damage next several decades. The natural question is whether i t would be useful, possible and/or feasible to or failures in order to achieve desired flight control incorporate adaptive flight controllers on performance. Neural networks are used to provide cable/hydraulic aircraft. Integration issues relative to consistent handling qualities across flight conditions, adapt to changes in aircraft dynamics and to make the both classes of aircraft will be discussed in this paper.
The neural network based approach controller easy to apply when implemented on different aircraft. Full-motion piloted simulation incorporates direct adaptive control with dynamic studies were performed on two different transport inversion's2 to provide consistent handling qualities models: the Boeing 747-400 and the Boeing C-17. without requiring extensive gain-scheduling or Subjects included NASA, Air Force and commercial explicit system identification. This particular architecture uses both pre-trained and on-line learning airline pilots. Results demonstrate the potential for neural networks, and reference models to spec@ improving handing qualities and significantly desired handling qualities. Several different control increased survivability rates under various simulated allocation techniques have been incorporated, failure conditions.
including a-priori daisy chain, a table-driven reallocation technique and a full simplex method Introduction linear programming theory reallocation technique.
Piloted simulation studies were w o r m e d at Design and production of entirely new NASA Ames Research Center on two Merent aircraft models have plummeted over the last few transport aircraft simulators, a Boeing 747-400 and a decades due to the high design cycle cost and long Boeing C-17. Subjects included NASA and Air Force development time. Instead, most "new" a i r c r a f t are test pilots and commercial airline pilots. This paper incremental improvements over existing models.
contains a brief overview of the system architecture While most current transport aircraft have been and presents simulation results comparing the neural- designed for a thirty-year life cycle, they are often adaptive controller performance to the aircraft's flown long beyond the design lifetime (i.e. the B-52 native control systems under nominal and simulated is now 50 and is anticipated to still fly for failure conditions.
approximately another 45 years). No new B-52 airframes are being produced. Instead, new Svstem Architecture Research Scientist, NASA Ames Research Center, The neural flight control architecture is based Moffett Field, CA upon the augmented model inversion controller, * Computer Engineer, QSS Inc. Ames R e s e a r c h developed by Rysdyk and Calise.' This direct adaptive Center, Moffeff Field, CA tracking controller integrates feedback linearization 'Pilot, Q S S Inc. Ames Research Center, Moffett theory with both pre-trained and on-line learning Field, CA neural networks. The Integrated Vehicle Modeling Environmen? is utilized to generate estimates of Copyright 0 2003 by the American Institute of stability and control derivatives. These derivativw are Aeronautics and Astronautics, Inc. No copyright is stored in pre-trained neural networks and are used to asserted in the United States under Title 17, U.S.
provide estimates of aerodynamic stability and control Code. The U.S. Government has a royalty-free license characteristics required for model inversion3. On-line to exercise all rights under the copyright claimed learning neural networks are used to generate herein for Governmental purposes. All other rights command augmentation signals to compensate for are reserved by the copyright owner.
American Institute of Aeronautics and Astronautics errors in the estimates and from the model inversion. function, over the available control surfaces. The The on-line learning neural networks also provide cost function biased the solution toward the additional potential for adapting to changes in aircraft minimum drag configuration and the smallest dynamics due to damage or failure. Aircraft refmnce possible surface deflections to achieve the desired models are used to filter command inputs in order to rates. Structural limitations for the subject aircraft are not known, and were not incorporated into the specify desired handling qualities. A Lyapunov stability proof guarantees boundedness of the tracking cost function, but the technique admits their potential error and network weights! For a detailed discussion inclusion in the future.
of the NFCS algorithm as applied in these studies, The final reallocation technique used please see Kaneshige and Gundy-Burlee involved a more complex, but fixed hierarchical schedule table. This table was initially derived by monitoring the solution space of the LP solver. It Control Reallocation StratePies Several different control reallocation techniques was hand-tuned to avoid parts of the solution space which we felt would cause structural degradation of were utilized in the course of these studies. In the the aircraft. It was also felt that the fixed nature of case of a potential retrofit to cable-hydraulic aircraft, the table would make this technique easier to certify it was assumed that significant system upgrades than the full LP solver.
would need to occur for NFCS to be enabled on the aircraft. In particular, the control surfaces on the Test Articles and Facilities aircraft would need to be upgraded with systems such as electric ailerons, power by wire systems or the The neural flight control system was evaluated utilizing two separate transport aircraft appropriate upgrading with full-authority hydraulic actuators which could use inputs from both the native types in two separate full-motion simulators located cable system and the overlaid neural flight control at the Crew Vehicle Systems Research Facility (CVSRF) at NASA-Ames Research Center. The h s t system. To reduce implementation issues, small, isolated integrated sensingkontrol systems could be type was a FAA Level-D certified Boeing 747-400 applied axis-by-axisin regions localized to the control simulator, shown in Figure 1. The second test bed system (to minimize re-wiring of the aircraft). This was the Advanced Concepts Flight Simulator concept was designated as NFCS (decoupled). This (ACFS)’, which has been modified to accommodate a system has the disadvantage that control requirements model of a Boeing C-17 aircraft. Both simulators are in one axis could not be reallocated to surfaces equipped with a six degree-of-freedom motion system, utilized to control of other axes. programmable flight displays, digital sound and aural cueing system, and a 180-degree field of view visual The second control reallocation scheme utilized a “daisy-chain” approach. In the longitudinal svstem.
axis, control is provided first by elevators, then !
symmetric ailerons and spoilers. Lateral control is primarily provided by asymmetric ailerons and spoilers with yaw-based roll control used as a secondary mechanism. Propulsion ~ o n t r o l ~ ~ ~ was not utilized in this experiment because (1) there was no access to the engine FADECs and (2) independently back-driven throttle levers are not currently available on the aircraft or simulators (and would be inordinately expensive to implement). In previous experiments, it was found that the pilot’s situational awareness of the control being utilized is critical to the strategic maneuvering of the aircraft, and that awareness cannot be adequately provided in propulsion control without back-driven throttles.
Figure 1. Boeing 747-400 flight simulator.
The next two control reallocation strategies were implemented only on the fly-by-wire aircraft.
The Boeing 747-400 is a cable-hydraulic One utilized a full simplex method l i n e a r actuated aircraft with stabilizer, two elevators, four programming (LP) theory technique. In this ailerons, twelve spoiler panels and a single rudder, for strategy, control derivatives for each axis were a total of 20 available control surfaces. For this estimated for every available control surface on the aircraft, the ailerons and spoilers on each side were aircraft and were provided to the LP solver. The ganged together in operation.
dynamic inversion solver was utilized to provide The Boeing C-17 is a fly-by-wire transport virtual roll, pitch, and yaw commands which were aircraft with a stabilizer, four elevators, two ailerons, then distributed optimally, according to a cost eight spoiler panels and two rudders, a total of 17 American Institute of Aeronautics and Astronautics surfaces used for active flight control. Slats, flaps the ground needed more study. In that experiment, and en-gines were not used by NFCS for cod-mation the neural adaptive flight control system transitioned control of either aircraft type. The pilots could to a simple gain scheduled system when the wheels manually utilize these surfaces for trim control of the touched the ground. This led to hard-to-control a i r c r a f t . transients when the damage adaptive control au_gnentation suddenly disappeared on touchdown.
During the course of this study, it was found that Implementation Considerations acceptable performance was obtained if: B747-400 Integration of a fly-by-wire oriented flight NFCS was engaged at rotate speed on control system with a cable-hydraulic airplane such as takeoff the Boeing 747 presents difficult issues. Two NFCS was disengaged when the wheels different implementation concepts were considered.
were on the ground and the plane had The first would be to independently control the pitch, decelerated to GO speed.
roll and yaw axis (decoupled option). This could potentially allow distributed processors and sensors Test Obiectives located near their primary control effectors (minimizing rewiring of the aircraft). For the The purpose of these studies was to evaluate different flight control reallocation techniques and decoupled option, it became apparent that if one could retrofit just one axis due to cost or other constraints, their affect on the handling qualities of the test aircraft relative to their native flight control systems. In the retrofitting just the pitch axis offered the greatest case of the Boeing 747, the aircraft’s normal handling single improvement in handling qualities on the characteristics are shaped by the cablehydraulic plane. In order to allow primary control reallocation to move to unconventional surfaces, the system system with yaw dampener (YD) that is intrinsic to the Boeing 747-400 aircraft. The Boeing C-17 is would need to be fully coupled and would likely equipped with a reference-model following stability require a full fly-by-wire overlay on top of the control and augmentation system (SCAS).
mechanical system. In either case, the hydraulic actuation systems on the airwaft would need to be B747-400 Test Results and Discussion retrofitted to full authority systems (for redundancy in case of damage). It was also necessary to postulate The flight control systems were evaluated by hydraulic actuation systems which would accept both a total of 4 pilots (3 NASA test pilots and one mechanical and electronic inputs (or a concept such as commercial a i r l i n e pilot) on the full-mission motion- electric ailerons). based simulator. Pilots evaluated the handling characteristics using the Cooper-Harper (CH) rating C-17 scale for maneuvers in high-altitude fight, approach The C- 17 has a complex spoiler/flap/throttle and landing, and take-off scenarios under a range of interconnect system which provides anti-ballooning failure conditions. The test pilots evaluated in-flight during configuration changes, speed brake modes handling qualities through a series of pitch and bank (spoilers extend and stay out until commanded to maneuvers in nominal conditions and with failures.
retract) anddirect lift control modes (spoilers pop out For the Boeing 747400, failure scenarios included a and back in to finely manage sink rate). These are full tail failure (aU surfaces frozen at trim), a coupled essential functions to the C-17 control system and the stabilizer/rudder failure (stabilizer nose down 3 NFCS command signal had to be overlaid on these degrees, rudder 5 degrees offset) and two engines out signals after the C17 control mixer had operated. on one side. Pilot workloads for some cases were The C-17 has a mechanical backup to the increased through addition of moderate turbulence and fly-by-wire control system. For transient-k crosswinds, low visibility conditions with an switching, the C-17 SCAS runs continuously (even obstructed runway.
in mechanical mode), but certain integrators are E- Table 1 provides information on the landing initialized if the flight controller commands are not scenarios discussed here. Figure 2 utilizes control being utilized. That philosophy was extended for column movement as an indication of pilot workload transient-fleemode switching between the NFCS and The figure displays average for landing scenarios.
C-17 SCAS controllers. The C-17 SCAS ran in absolute deflection and average absolute rate of deflection of the control column in both the roll and shadow mode even when NFCS was providing commands to the system, but integrators in the pitch axes’ for a variety of scenarios. The values were SCAS were suppressed o provide transient-free then normalized by those obtained in Scenario 1.
switching between control systems. Three flights were conducted without failures to fom One recommendation from the second- a baseline for the rest of the experiment. The pitch generation flight control system study was that the deflections for the NFCS scenarios are noticeably transition fiom flight to landing and rolling out on reducedrelativeto the native control system. NFCS American Institute of Aeronautics and Astronautics provides rate-command, attitude hold (RCAH) the 747-400 were insufficient to augment pitch capability while the normal cable-hydraulic system control, so both sets of ailerons were utilized here.
must be manually trimmed during flight. The pilots Scenario 5 included a low visibility condition in commented that this change provided significant which an aircraft became visible on the runway. - At improvement in handling qualities. an altitude of SOO’AGL, the pilots were ordered to sidestep to the adjacent m w a y . Despite the pitch- axis failure, the pitch column deflection was mluced relative to the un-failed cable-hydraulic controller in landing scenario 1. The pitch column rate w a s increased over all the un-failed cases, and probably represented Number additional effort associated with control dead-bands present in the daisy-chain scheme. The pilots all Failures. light turbulence.
-
landedsafely with the NFCS controller, even when NFCSDEC additionally challenged with a runway incursion a d 3 No Failures, moderate turbulence, sidestep (case 5). It should be noted that without NFCSDEC control reallocation of the pitch authority, the NFCS 4 Full Tail Failure, light turbulence, YD scheme would have provided no significant benefit 5 Full Tail Failure, light turbulence, over the cable-hydraulic system.
I runway incursion, NFCS-
In scenarios 7-9, the stabilizer was frozen in
1 Full Tail Failure, moderate turbulence, 6
I
a nose-down 3 degree position and the rudder was
I NFCS
frozen at a 5 degree offset. For the cable-hydraulic
7 I Stab/Rud, light turbulence, Yaw
I
system, the pilots all chose to have the copilot
I Dampener
manually provide the t r i m force needed. This is
I Stabmud, light turbulence, NFCSDEC - 8
reflected in the pitch column absolute deflection, I 9 I Stabmud. liphtturbulence. NFCSDEC I where on average, the pitch column was deflected about 4 times that of the undamaged cable -hydraulic Table 1. Landing scenarios €or the Boeing plane. The pitch column rate was also significantly 747-400 Experiment. increased. This represented a significant workload increase and required close coordination between the Average Control Input pilot and copilot. The pilots generally chose to modulate the roll t r i m to try to minimize wheel trim
1 4.5
force, however, it still was significantly greater than er for the un-failed case. For cases 8 and 9, NFCS was 2 3.5 BPitch Column Rate operated in a de-coupled mode. The pitch and roll I C 3 yoke deflections were significantly reduced over the Roll Yolk Deflection cable hydraulic system.
1.5 C-17 Test Results and Discussion
I 0.5
The goal of the C-17 experiment was to l o evaluate the performance of the third generation 1 2 3 4 5 6 7 8 9 1 control neural flight control systems (LP and table Scenario Number I driven reallocation with adaptive critic) relative to that of the second-generation system (daisy chain Figure 2. Control Column deflection and rate used to reallocation). Five C-17 pilots from the Air Force, I indicate pilot workload.
andNASA were used to compare the second and third generation flight control systems. The native C-17 ~ I In scenarios 4-6, the tail was frozen in the flight control system was not included in the trim position. Only 1 pilot, who had extensive evaluation because an early version of the flight experience with propulsion control, was able to land control system is incorporated in to the simulation, the un-augmented Boeing 747-400 system. That and it was felt that it does not adequately represent the pilot utilized symmetric and differential engines to current C-17 SCAS. The evaluation criteria for all control the aircraft, so no stick deflection data is the pilots included Cooper-Harper (CH) ratings, reported here. For cases 5 and 6, NFCS utilized the approach performance time history data, touchdown daisy chain control allocation system, wherein pitch snapshot data and pilot comments.
control w a s obtained through use of both ailerons The failure scenarios for the C-17 test are symetrically when elevators fail to provide an outlined in table 2. The scenarios were designed to adequate pitch response. It was found during the tests performance of the controllers relative to course of this experiment, that the inboard ailerons of primary failures in all 3 axes’ as well as a failure American Institute of Aeronautics and Astronautics .
sequence which would couple over all the axes. The the Gen-2 controller (which resulted in similar ratings roll and pitch axis scenarios were utilized during comments and ratings). The table-driven allocator normal landing operations, the yaw axis failure on a was designed to minimize control dead-bands. It takeoff sequence, and the coupled failure during a blends control across surfaces, so that ailerons is tactical descent scenario. The pilots were asked to utilized for augmenting pitch control much sooner perform handling qualities tests roll, pitch and yaw than the other two allocators, This leads to smaller dead-bands and improved CH ratings.
doublets) during the tests and provide individual CH ratings for each of the three axes. The second scenario involved failure of the The lateral and longitudinal CH ratings for primary wing roll control surface (spoilers and each of the four scenarios are shown in Figure 3. ailerons). The daisy chain architecture was designed Heavy horizontal lines indicate the specific pilot to utilize yaw-based roll control in response to primary roll control failure. The controller ratings. If more than one pilot gave the same rating, the number of pilots giving that rating is listed in commands the primary roll control surfaces to their Roman Numerals. The average CH rating is listed maximum extent and then passes excess command to the rudders. The resulting aircraft performance above each of the bars on the chart.
contains a large dead-band, and then utilizes slow aircraft dynamics to achieve roll control. The pilots C17 Scenario Characteristics: commented that t h i s made it difficult to make fine Scenario Winds 190 @ 10, light turbulence tracking adjustments near the ground and gave the Pitch Full tail failure. Stabilizer failed at trim.
controller poor CH ratings in the lateral direction axis 2 rudders, 4 elevators failed at 0 deg.
(average 6.6). The LP allocator had similar de3b Roll 2 ailerons and 8 spoiler panels failed at 0 bands to the daisy chain, but actively used differential axis I & . E .
elevator and rudder to provide coordinated turns. This Yaw I Two en.gines out on one side on takeoff.
gave an improved CH rating over the daisy chain axis I minimum climb speed + 10Kts.
allocation‘scheme. The tabledriven scheme utilized
Coupled 1 During tactical descent (failures on one
control blending in order to transition roll control to
failure I side)
the tail much faster than either of the other schemes.
23,000’ : Stab frozen at t r i m The table was also massaged to limit the severe rudder 20,000’ : 2 Elevators frozen at 0 deg.
and elevator deflections that the LP allocator was l7,OOO’ : Upper rudder hard over inducing. l l i s led to smaller dead-bands, but lower 15,000’ : Outboard flap fails retracted control authority for the table-driven allocation l4,OOO’ : Aileron frozen at 0 deg.
scheme and the pilots gave it lower CH ratings 13,000’ : Two outboard spoilers (average 5.2) than the other two schemes.
frozen at 0 deg.
The third scenario involved two engines on When engines come out of reverse: one side failing just after take-off at 10,OOOlb over the Outboard engine seizes.
theoretical two-engine-out take-off weight. The pilots slightly preferred the table-driven allocation scheme Table 2. Failure scenarios for the Boeing over the other two schemes. The pilots a l l C-I7 Experiment.
commented that the workload for all three controllers was distinctly reduced over the real aircraft (which requires substantial input into the rudder to control CH ratings for the full tail failure scenario the aircraft).
for each of the 3 controllers are shown on the far left The frnal scenario was that of a tactical columns of the figure. The three controllers are descent while under assault from multiple surface to comparable for the longitudinal control, but the table- air missiles. The scenario, described in Table 2, was driven allocator gave more consistent and lower CH designed to induce trim offsets and control failures in ratings than the other two. The Gen-2 controller was every axis under high workload conditions. No optimized for failure of the primary control effectors control dead-bands were induced i n this scenario. The on the pitch axis. The main complaint about the table-driven scheme ran out of nose-up control performance of the Gen-2 controller related to control authority during part of the descent requiring a rapid dead-bands induced by the strict hierarchical daisy re-configuration of the airaaft to maintain safe flight, chain employed for reallocation. Since there is no which reduced its CH ratings. The LP allocator on explicit system identification utilized, the faded average was rated with level I handling qualities while elevators are commanded to their maximum extent the other two schemes were on avexage given level I I and any excess control is passed to symmehic handling qualities.
ailerons and spoilers. The LP allocator also utilizes the optimal surfaces first before transitioning to secondary surfaces, thus it has similar dead-bands to American Institute of Aeronautics and Astronautics , GEN 3 EXPERIMENT
COOPER - HARPER RATINGS
L A T E R A
L
1 2 3 4 Tail Control Wing Control Two Engine Out Tactical Descent Failure Failure Takeoff Cascading Failure SCENARIO L N G
I
T
U D I
N
A L 1 2 3 4 Tail Control Wing Control Out Tactical Descent Two Engine Failure Failure Take-off Cascading Failure SCENARIO Figure 3. GEN 3 C-17 COOPER HARPER Ratings American Institute of Aeronautics and Astronautics ' .
[ 5 ] Kaneshige, J. and Gundy-Burlet, K. Integrated Conclusions Neural Flight and Propulsion Control System, The results presented in the previous sections AIAA 2001-4386, August 2001.
demonstrate the effectiveness of the neural flight control system controlling a transport-class vehicle [ q Kaneshige, J., Bull, J., Kudzia, E., and Burcham, under a wide range of failure conditions. The generic F., Propulsion Control with Flight Director system can also help to reduce the high cost associated Guidance as an Emergency Flight Control System, with avionics development since it does not require AIAA 99-3962, August 1999 gain-scheduling or explicit system identification.
In general, the results demonstrate that under [7] Burcham, Frank W., Jr., John J. Burken, Trindel normal fight conditions, the neural system can achieve A. Maine, and C. Gordon Fullerton, Development performance, which is comparable to the a i r c r a f t ' s and Flight Test o f an Emergency Flight Control conventional system. The neural flight control system System Using Only Engine Thrust on an MD-11 can also provide additional control authority under Transport Airplane, NASA TP-97-206217, Oct.
damaged or failure conditions. For the cable-hydraulic 1997.
aircraft, significant improvements in handling qualities were provided by the NFCS system, with the piIots [8] Blake, Matthew W., The NASA Advanced advising that the pitch system augmentation was the Concepts Flight Simulator: A Unique Transport most critical. Retrofits of the aircraft, however, would Aircrafi Research Environment, ATAA-96-35 18- probably be cost prohibitive.
CP.
Results demonstrate that choice of control reallocation technique can significantlyimprove damage adaptation under various failure conditions.
Minimization of dead-bands is key to producing good handling qualities, and should be carefully considered in future research. The pilots generally preferred the table-driven scheme for the simple axis by axis failure scenarios with control dead-bands. In the final, highly coupled failure, the pilots generally preferred the LP allocation scheme. This suggests that integration of parameter identification techniques, vehicle health monitoring information or incorporation of control surface blending into the cost function will distinctly improve the handling qualities of the LP allocation scheme, and should be pursued in future research. The results also imply that aircraft structural design must be carefully evaluated when utilizing control surfaces in non-traditional manners.
References [l] Rysdyk, Rolf T., and Anthony J. Calise, Fault Tolerant Flight Control via Adaptive Neural Network Augmentation, AIAA 98-4483, August 1998.
[2] Totah, Joseph J., David J. Kinney, John T.
Kaneshige, and Shane Agabon, An Integrated Vehicle Modeling Environment, AIAA 99-4106, August 1999.
[3] Norgaard, M., Jorgensen, C., and Ross, J., Neural Network Prediction o f New Aircraft Design Coefficients, NASA TM-112197, May 1997.
[4] Kim, B., and Calise, A., Nonlinear Flight Control Using Neural Networks, AIAA Journal of Guidance, Navigation, and Control, Vol. 20, No.
1. 1997.
American Institute of Aeronautics and Astronautics