Adaptive Controller on the NASA AirSTAR Fli ht T t V hi l Flight

Fli ht T t f L Ad ti C t llFlight Test of L1 Adaptive Controller on the NASA AirSTAR

Fli ht T t V hi lFlight Test Vehicle

I M G E i XIrene M. GregoryDynamic Systems and Control Branch

NASA Langley Research Center

Enric XargayDept. Aerospace Engineering

University of Illinois at Urbana-Champaign

Naira HovakimyanDept. Mechanical Engineering

University of Illinois at Urbana-Champaign

Chengyu CaoDept. Mechanical Science & Engineering

University of Connecticut y p gy

Aerospace Control and Guidance Systems Committee Meeting 106L J ll CALa Jolla, CA

October 13-15, 2010

Subscale Flight Testing

IRAC h i f d l f t l f il d d iIRAC research is focused on loss-of-control, failure and damage scenarios, and their mitigation though the application of adaptive control.

Actual Transport Aircraft Loss of Control

Control law objectives:

Under adverse conditions: Actual Transport Aircraft Loss-of-Control

Accident Data

• Keep aircraft in the Extended flight envelope

• Return to Normal Flight Envelope

Control actions within 2-4 seconds of failure onset are critical

Normal

⇒ Transient performance

⇒ Fast adaptation

ExtendedFlt envelope

o aFlt envelope

IRAC = Integrated Resilient Aircraft Control project, part of Aviation Safety Program1

AirSTAR GTM aircraftHigh risk flight conditionsHigh-risk flight conditions,

some unable to be tested in target application environment.

5.5 % geometrically and dynamically scaled model82 in wingspan, 96 in length, 49.6 lbs (54 lbs full), 53 mph stall speedModel angular response is 4.26 faster than full scaleModel velocity is 4.26 times slower than regular scale

Dr. Irene M. Gregory 2

Why an all-adaptive design?Typically direct adaptive controllers have been used as an augmentation toTypically direct adaptive controllers have been used as an augmentation to a robust baseline controller.

Great deal of interest in seeing what adaptation can do as a standalone control strategycontrol strategy

Stress the L1 methodology and architecture to better understand the practical performance

the controller behavior at the limits

With a well designed robust baseline controller it is more difficult to assess how much adaptation is helping in providing stability and performance p p g p g y probustness and how much is the baseline controller contributing.

Baseline controller may hide potential deficiencies of the adaptive controller.

Robust Baseline

L1

AFCS

L1 AdaptiveControl Law


Consider nonlinear system dynamics

MIMO Nonlinear System with Cross-Coupling

Consider nonlinear system dynamics

Can be expressed as the following system:

( ) ( , ) ( , ) ( )x t f x t g x t u t= +

Can be expressed as the following system:

( )( ) ( )( ) ( ) ( ) ( ) ( ) ( ) ( ) (0)x t A x t B t f x t z t t B f x t z t t x xμ= + + + =

Matcheduncertainty

Unmatcheduncertainty

( )( ) ( )

( )( ) ( ) ( ) ( )( )1 2 0

0 0

( ) ( ) ( ) ( ), ( ), ( ), ( ), , (0)

( ) , , , , , (0)

m m um

z z z z z

x t A x t B t f x t z t t B f x t z t t x x

z t g x t t x t g x t x t t x x

μ= + + + =

= = =

Unmodeled( ) ( ) ( ),

( ) ( )

as G s u s

y t Cx t

μ =

=

UnmodeleddynamicsActuator

dynamics

Control objective:Design an adaptive state feedback control law u(t) to ensure that the system output response y(t) tracks the output response ym(t) of the desired systemdesired system


( ) ( ) ( )

( ) ( )m m m m g

m m m

x t A x t B K r t

y t C x t

= +

=

L1 All-adaptive Controller Architecture

Control augmentation system α command p β commandControl augmentation system – α-command, p-β commandDesigned at 1 flight condition – 80-KEAS (~4.5 deg AoA)

L1controller

1

m m m m cmd

q

x A x BZ ZZ

VV Vδα

δ

α αδ

= +

⎛ ⎞ ⎛ ⎞+⎛ ⎞ ⎛ ⎞⎜ ⎟ ⎜ ⎟= +⎜ ⎟ ⎜ ⎟⎜ ⎟ ⎜ ⎟

e

e

q

VV Vq q MM M δα

δ= +⎜ ⎟ ⎜ ⎟⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠ ⎜ ⎟⎜ ⎟ ⎝ ⎠⎝ ⎠

L1controller

0 0

0 1

pa r

aa r

Lp p L LY Y YV

δ δβ

δ δ

δβ β

δ

⎛ ⎞⎛ ⎞⎛ ⎞ ⎛ ⎞⎜ ⎟ ⎛ ⎞⎜ ⎟⎜ ⎟ ⎜ ⎟⎜ ⎟= − + ⎜ ⎟⎜ ⎟⎜ ⎟ ⎜ ⎟⎜ ⎟ ⎝ ⎠⎜ ⎟ ⎜ ⎟ ⎜ ⎟

0a r

ra rr

Vr r N NN N

δ δ

δ δβ

δ⎜ ⎟⎜ ⎟⎜ ⎟ ⎜ ⎟⎜ ⎟ ⎝ ⎠⎜ ⎟ ⎜ ⎟ ⎜ ⎟⎜ ⎟⎝ ⎠ ⎝ ⎠ ⎝ ⎠⎝ ⎠

Flight Control Law Evaluation Matrix I

Task 1st straight leg 2nd straight leg Turns

1Latency Injection (5ms / 5 sec)

Fault EngagedRoll Doublet

Fault EngagedPitch Doublet

Fault Engaged

2Δ(Cmα & Clp ) ≈

00%Fault EngagedRoll Doublet


Disengage Fault


- 50%Fault EngagedRoll Doublet


Disengage Fault50% Roll Doublet Pitch Doublet


- 75%Fault EngagedRoll Doublet


Disengage Fault

Δ(Cmα & Clp ) ≈ Fault Engaged Fault Engaged

5 -100% (neutrally stable)

Fault EngagedRoll Doublet


Disengage Fault


-125% (unstable)Fault EngagedRoll Doublet


Disengage Fault125% (unstable) Roll Doublet Pitch Doublet

7High AoACapture

No FaultNo Doublet

No FaultNo Doublet

N/A


Cmα – degraded by 2 inboard elevator segments 50% reduction in pitch control effectiveness Clp – degraded by spoilers

Flight Control Law Evaluation Matrix II

Task Downwindstraight leg Upwind straight leg Turns

1 Offset LandingAchieve

good trim

No fault 1st – Practice landing

2nd - Evaluation landing N/A

2Offset Landing

Neutrally stable: Δ(Cmα & Clp ) ≈ -100%

Achieve good trim

Fault EngagedEvaluation landing

Disengage Fault

3Offset Landing

Unstable:Δ(Cmα & Clp ) ≈ -125%

Achieve good trim

Fault EngagedEvaluation landing

Disengage Fault


Cmα – degraded by 2 inboard elevator segments 50% reduction in pitch control effectiveness Clp – degraded by spoilers

L1 Support Tasks on Modeling – Sept. 2100

2ndTask 1st straight leg 2nd

straight leg

Flat turn: 2 deg/s ramp up to desired β value hold for the remainder of straight leg

1 β - Vane Calibrationhold for the remainder of straight leg

1 deg/s ramp up to desired β value;Repeat[ 2, 4, 6, 8]β = ± ± ± ±

8β = ±

2 V C lib ti

Variable α strategy: set engine RPM, let αincrease for remainder of straight leg

C t t t t t i RPM il t R t

8β = ±

2 α - Vane Calibration Constant α strategy: set engine RPM, pilot acquires and holds target α for remainder of straight leg

Repeat

3Unsteady

Aerodynamics

Post-Stall High α Tracking: Attain stable flight at α=18, inject wave train:(1) Step [18 15 15 20 20 15]3 Aerodynamics

Modeling(1) Step [18 15, 15 20, 20 15](2) Schroeder sweep(3) Variable frequency Sinusoid


GTM T2 – Flight Test Evaluation (June 2010)L1 all-adaptive flight control law provides performance/stability for nominal andL1 all-adaptive flight control law provides performance/stability for nominal and

impaired aircraft• Not an augmentation to a baseline controller, that provides nominal aircraft

performance

Flight Control Law related tasks during June 2010 deployment :• Injected longitudinal and lateral stick doublets for each fault, continuous stick doublets

on straight legs during latency fault • Latency fault: starting at 20 msec, continuous increase in latency (5 msec every 5 sec)

carried through the turns until aircraft is neutrally stable or unstable – want graceful performance degradation

Robust to 0 125 sec additional time delay [0 147 total time delay]Robust to 0.125 sec additional time delay [0.147 total time delay]• Simultaneous longitudinal and lateral stability degradation (Cma/Clp):

50%: nominal performance75%: small degradation of performance in roll75%: small degradation of performance in roll100%: small degradation of performance in pitch, larger degradation in roll125%: large amplitude roll with pitch doublet

Modeling Tasks:gL1 used for β-sweep in flat turn maneuver


GTM T2 - Flight Test Evaluation (June 2010)

FLT23: L1 all-adaptive FCL under light turbulence

100

stk-to-surfL1 all-adaptive1200

1400

stk-to-surfL1 all-adaptive

Research Pilot SPSP Research Pilot SPSP

60

80

eed,

kts

600

800

1000

ude,

ft

1

20

40Airs

pe0

200

400Alti

tu

0 200 400 600 800 1000 12000

Time, sec

0 200 400 600 800 1000 1200

-200

0

Time, sec

~12.5 minsHigh AOA flightof flightwith L1

High AOA flight

Post‐stall regimes

Nominal Aircraft Wave Train Responsecmd and p cmd wave trains (WT) enter as pilot stick commands

Pitch axis – α cmd doublet Roll axis p cmd doublet

α-cmd and p_cmd wave trains (WT) enter as pilot stick commandsPilot asked for hands off during WT – WT characterized by straight lines

4

6

8

sec

20

30

ec

ppcmd

φ

Pitch axis – α_cmd doublet Roll axis – p_cmd doublet

-2

0

2

axis

, deg

, deg

/s

-10

0

10

xis,

deg

, deg

/se φ

-10

-8

-6

-4

Pitc

h a

αcmd

αq

-40

-30

-20

Rol

l a

α-cmd response designed for pilot, not to the maximum potential of the control law [tracking doublet faster – too sensitive for the pilot ]

425 430 435 44010

Time, sec465 470 475 480

40

Time, sec

law [tracking doublet faster too sensitive for the pilot ]Roll rate is a very fast and challenging response [with no turbulence – smooth, fast response tracking the p_cmd doublet]


Latency ResponseLatency faultLatency fault

Carried through the turnsEngaged around 286 secondsThe maneuver was abandoned at 394 seconds due to persistent rollThe maneuver was abandoned at 394 seconds due to persistent roll rate oscillations of ± 20 deg/sec

Latency active Latency active

15

20

c

αcmd

αq

Latency active20 12510055 85

40

60

c

ppcmd

Latency active20 12510055 85

5

10

s, d

eg, d

eg/s

ec q

0

20

40

, deg

, deg

/sec φ

-5

0

Pitc

h ax

is

-40

-20

Rol

l axi

s


280 300 320 340 360 380 400-10

Time, sec

280 300 320 340 360 380 400

-60

Time, sec

Latency Fault Doublet Response10 6

α

6

8

k, d

eg

of a

ttack

4

5

k, d

eg

αcmd

α

2

4

Ang

le o

f atta

ck

αcmd

α

Angl

e o

2

3

Ang

le o

f atta

ck

40

310 315 320 325 330-2

0

Time, sec

20

345 350 355 360 3650

1

Time, sec

20

30

40

ec

ppcmd

0

10

20

/secrate

-10

0

10R

oll r

ate,

deg

/se

-30

-20

-10

Rol

l axi

s, d

eg, d

eg/

ppcmd

φ

Rol

l

13345 350 355 360 365-40

-30

-20

310 315 320 325 330

-50

-40

R

Cmα/Clp Degradation WT Response

Pitch axis α cmd doublet Roll axis p cmd doubletPitch axis – α_cmd doublet Roll axis – p_cmd doublet

5

10

g/se

c

20

30

40

/sec

ppcmd

φCmα/Clp

0

ch a

xis,

deg

, deg

-10

0

10

ll axi

s, d

eg, d

eg/Cmα/Clp

75%50% long cntrl

580 585 590 595 600-10

-5

Time sec

Pitc

αcmd

αq

625 630 635 640-40

-30

-20

Time sec

Ro

Time, sec Time, sec

Neutrally stable5

10

deg/

sec

αcmd

αq

10

20

30

40

eg/s

ec

ppcmd

φ

Neutrally stableCmα/Clp

100%50% long cntrl

-5

0

Pitc

h ax

is, d

eg, d

-20

-10

0

10

Rol

l axi

s, d

eg, d

e


675 680 685 690-10

-5

Time, sec

P

710 715 720 725

-40

-30

Time, sec

R

125%Cmα/Clp Degradation WT Response

Pitch axis α cmd doublet Roll axis responsePitch axis – α_cmd doublet Roll axis response

15

20

ec

60

80

100

c

ppcmd

φ

125%

0

5

10

xis,

deg

, deg

/se

20

0

20

40

xis,

deg

, deg

/se φ

15

-10

-5

Pitc

h ax

αcmd

αq -80

-60

-40

-20

Rol

l ax

740 750 760 770-15

Time, sec

740 750 760 770

Time, sec

Pilot called “knock it off” but did not abandon the control law

Test engineer simply flipped the switch to turn off the stability degradation fault and the controller recovered its nominal performance immediately.

The pilot proceeded to fly into a typical aggressive turn less than 10


p p y yp ggseconds after the fault was terminated (~ 770 seconds)

L1 Adaptive FCL – High AOA Flight

Open-loop aircraft tends to aggressively roll off between 13deg and 15deg AOA and exhibits significant degradation in pitch stabilityL1 all-adaptive controller significantly improved pilot’s ability to fly the aircraft

t hi h l f tt k d d d hi kl d

Flt 23: L1 all adapt

at high angles of attack and decreased his workload. Robust to control deflection saturation

Flt 25: stick - surfaceFlt 23: L1 all-adapt

α, p-β cmd60

pφ

60

pφ

20

40

xis,

AoA

φα

20

40

xis,

AoA

φα

-20

0R

oll a

x

-20

0

Rol

l ax


830 835 840 845 850 855 860

-40

Time, sec

590 595 600 605

-40

Time, sec

High AoA Tracking – Unsteady Aero

Modeling unsteady aerodynamics by emulating the dynamic motion in the wind tunnel – determining efficacy of GTM to be a “flying wind tunnel”Target AoA = 18 deg – post-stallTarget AoA 18 deg post stallInjected inputs for L1 FCL to track – Step, Schroeder, Sinusoids

S h d I t Si id I t

18

20

22

αcmd

α18

20

22

αcmd

α

Schroeder Input Sinusoids Input

12

14

16

18

of a

ttack

, deg

α

12

14

16

18

of a

ttack

, deg

α

6

8

10A

ngle

o

6

8

10

Ang

le o


800 810 820 830 8404

Time, sec

740 750 760 770 7804

Time, sec

Angle of Attack Vane CalibrationStall occurs between 12 and 13 deg AoAStall occurs between 12 and 13 deg AoA

14 14

Variable AoA Strategy Constant AoA Strategy

10

12

ack,

deg

αcmd

α10

12

ack,

deg

4

6

8

Ang

le o

f atta

4

6

8

Ang

le o

f atta

αcmd

α

860 880 900 920 9402

4

Time, sec

1

750 800 8502

4

Time, sec

α

1

long stk

0

0.5

lot I

nput

long stklat stick

0

0.5ot

Inpu

t

long stklat stick

Dr. Irene M. Gregory 18860 880 900 920 940-1

-0.5

Pi

750 800 850-1

-0.5

Pilo

Sideslip Angle Vane Calibration0

2

Flat turn – hold target sideslipMinimize lateral axis excursions

-6

-4

-2

0

r def

lect

ions

, deg

2

720 740 760 780 800 820

-12

-10

-8

Time, sec

Rud

de

Rudcmd

Rud

25

-4

-2

0

angl

e, d

eg

Time, sec

5

10

15

20

n de

flect

ions

, deg

-8

-6

4

Side

slip

a

βcmd 720 740 760 780 800 820-10

-5

0

Time sec

Aile

ron

Ailcmd

AilR

8

10

φ

720 740 760 780 800 820-10

Time, sec

βTime, sec

2

0

2

4

6

8

nk a

ngle

, deg

Adequate

Dr. Irene M. Gregory 19720 740 760 780 800 820-10

-8

-6

-4

-2

Time sec

Ban

Offset Landing

I iti l ltit d i 100 ftInitial altitude is 100 ftInitial offset is 75 ft from centerlinePerformance boundaries:

Desired – |bank| =+-10; |gamma|=+-1; landing box = 3’x10’Adequate - |bank| =+-20; |gamma|=+-3; landing box = 9’x30’

Flying qualities ratings taken for nominal, neutrally stable, unstable y g q g yairplane

Insert HUD video herenominal

Insert HUD video hereNeutrally stable

Insert HUD video here

CHR 4 CHR 4


Insert HUD video hereUnstable

CHR 7

L1 Adaptive Controller Flight Test Summary

An all adaptive controller that provides both nominal aircraft performance andAn all adaptive controller that provides both nominal aircraft performance and takes care of large changes in aircraft dynamics

• No gain scheduling, no baseline to assist

One controller designed at nominal flight condition (80KEAS, 4 deg AOA) to provide satisfactory FQ and robustness

Controller able to handle large additional latency in the system (robustness measure) and provide nominal to slightly degraded performance for stability degraded cases (doublet tracking with neutrally stable aircraft).

Controller provided predictable response to the pilot under stability degradation and graceful performance degradation once nominal response was unachievable

Improved response in post-stall flight providing controllable aircraft to the pilot and facilitating safe return to normal flight

The classical tradeoff between robustness to system latency vs. performance was found to be consistent with the theory.


QUESTION?


Documents

Adaptive Controller on the NASA AirSTAR Fli ht T t V hi l Flight