Piloted Simulation of Fault Detection, Isolation and Reconfiguration Algorithms for a Civil
Transport Aircraft
S. Ganguli, G. Papageorgiou, S. Glavaški, M. ElgersmaHoneywell Advanced Technology GNC
Presented by: G. Papageorgiou
SAE ConferenceOctober 2005
#NCC-1-334 with NASA Langley Research Center#NAS1-00107 with NASA Langley Research Center
2 AIAA GNC 2005
Aircraft Control Surfaces
• Commanded Control Surfaces (via autopilot):- Aileron Difference- Average Elevator- Rudder
3 AIAA GNC 2005
Piloted Simulation Setup
4 AIAA GNC 2005
IFD vs Matlab: Comparative plots
5 AIAA GNC 2005
CUPRSys Overview
6 AIAA GNC 2005
CUPRSys Algorithms – Aircraft Model• Express aircraft dynamics as sum of nominal nonlinear function and a linear
combination of (nonlinear) basis functions.
),( uxfx ),(),(0 uxHbuxfx
• Aircraft equations:
)()(
)(11
13
MMJJJ
FFgevv
T
Tm
Tvx ][ TTu ][
),,(~00
)()()(
1
1
0,11
0,1
3 vHbBJ
IFFJJJ
FFgevv m
T
Tm
),,(~ bcbdiagB
||||
||||
||||2
21 ~),,(
v
v
vv
SBSS
vvb
106H
IM
7 AIAA GNC 2005
Flight Conditions
Reduction of effectiveness faults and various maneuvers
8 AIAA GNC 2005
Matlab Simulation (Low Cruise Pitch Down)
- No Fault-. 75% Fault.. Reconfigured- Command
Pilot modeledby Prop. Gain
9 AIAA GNC 2005
Flight Card
10 AIAA GNC 2005
Low Cruise – 10 deg Pitch Down (No fault)
11 AIAA GNC 2005
Low Cruise – 10 deg Pitch Down (75% e fault)
12 AIAA GNC 2005
Low Cruise – 10 deg Pitch Down (Reconfiguration)
Larger command
13 AIAA GNC 2005
Low Cruise – 10 deg Pitch Down (Reconfiguration)
Smaller command
14 AIAA GNC 2005
Cooper Harper Ratings & Pilot Workload - LC
Low Cruise
|| ||2pedalforce
ecolumnforcwheelforce
LP HPs1
workloadpilot
(5 rad/s) (2 rad/s)*
* R. Mercadante, “Piloted Simulation Verification of a Control Reconfiguration for a Fighter Aircraft under Impairment”, AGARD No. 456, Toulouse, France, 1989
15 AIAA GNC 2005
FDI Performance
• Performance measured by:- False-alarms
1 LC pitch-up maneuver, and during flare tasks (ground effects not modeled?)
- Missed detection (none, but sensitivity to small faults not tested)- Accuracy of identified fault
16 AIAA GNC 2005
Lessons Learnt and Recommendations
• Limitations- CUPRSys uses and sensors – typically not available- Feel system model not available for design
• Current deficiencies of CUPRSys- On-board aircraft model uses exact replica of Engine Model- H-matrix and Threshold Functions vary with flight condition
• Gain reconfiguration vs control re-allocation- CUPRSys restricted to gain reconfiguration (commanding
through autopilot)- Control authority of additional surfaces can restore flying qualities
17 AIAA GNC 2005
Conclusions & Future Work
• Piloted simulations conducted at LaRC suggest- Robust control law- Promising FDIR capabilities (need more validation sims with
control re-allocation)
• Future Work- Utilize control allocation- Accommodate sensor dynamics and noise- Accommodate turbulence- Expanded set of failures
18 AIAA GNC 2005
Pilot Cueing
19 AIAA GNC 2005
Integration Flight Deck (IFD)
• Piloted Simulations were conducted at the LaRC IFD facility.
20 AIAA GNC 2005
Acknowledgement
• Thanks to the NASA Team for their support, encouragement and various helpful discussions:- Pat Murphy- Steve Derry- Gus Taylor- Rob Rivers- Tom Bundick- Christine Belcastro
21 AIAA GNC 2005
www.honeywell.com
22 AIAA GNC 2005
NASA Aviation Safety & Security Program• NASA AvSSP
- $500 million*- Reduce commercial aviation accident rate by 80% by 2007*
(* http://www.nasa.gov/centers/langley/news/factsheets/AvSP-factsheet.html)
http://avsp.larc.nasa.gov/program.html
23 AIAA GNC 2005
AMASF Program Overview
• Phase I (mid-sized commuter aircraft)- FDI technologies for selected failures (stuck/floating actuators,
reduction of control surface effectiveness) + icing- Pilot Cueing strategies- Control Reconfiguration
• Phase II (mid-sized civil transport aircraft)- Transition of algorithms to new aircraft- Algorithms + display integrated to CUPRSys- Failure type: reduction of control surface effectiveness
• Phase III- Piloted simulation at LaRC
24 AIAA GNC 2005
CUPRSys Algorithms – Reconfigurable CLAW
• Based on Dynamic Inversion
Cxyuxgxfx
)()(
• Desired Dynamics
• Control Law (under certain assumptions)
),( yyTy cdes
))(),(())(( 1 xCfyyTxCgu c
DI P + I
r
e
a
K
*c
c
Cp
desVg
dtd
des
des
rqp
)sin(
)(sin)1(
yVg
zVg
nrnq
p
co
c
+ feedforward
25 AIAA GNC 2005
CUPRSys Algorithms – Reconfigurable CLAW• Controller bandwidths
- High dynamic pressure (High/Low Cruise): [p C* ] = [2.0 1.25 1.0] rad/s
- Low dynamic pressure (near Approach): [p C* ] = [2.5 0.75 0.75] rad/s
• Inceptor Scalings- Wheel (85 deg): 0.25 (deg/s)/deg- Column (-9.2 to 13.3 deg): 2.00 (deg/s)/deg- Pedal (4 inch): 0.02 rad/inch
• Anti-windup (software limiting)
fiKb 1/s Kb
desyInversion
u
fc
cyy
Kaw
sat lim
y
26 AIAA GNC 2005
CUPRSys Algorithms – Fault Detection
• Residual Generator
Angular Acceleration
EstimatorNoise
Rejection
dtdu
tmeasuremen residualerr
LPc
c
c
Cp
* abs( )
• Threshold Function
threshold
Bias
Turbulence Rejection
z
y
x
nnn
Scaled, Added
Scaled, Added
27 AIAA GNC 2005
CUPRSys Algorithms – Fault Isolation
• RLS Estimator:
After acquiring k samples:
Over-determined linear algebra problem:
Weighted Least Squares problem:
Solved using standard RLS Estimation algorithm with forgetting factor.
),(),(0 uxHbMuxfx I
)],(,),,([)],(,),,([
111
0111011
kkkk
kkIk
uxbuxbGuxfxuxfxMF
)( 0 HHHGF kk
),(),0( 0 kkk GWHGHF
28 AIAA GNC 2005
CUPRSys Algorithms – Fault Isolation
• H-Matrix Convergence Criterion
• H-Matrix Update- H-matrix for FD (Residual Generator)- H-matrix for CLAW
• Signal Injection- Trade-off between sufficient excitation time and quick FDIR- Simultaneous doublet commands (4 sec) in all three axes
0.5 deg/s p 0.5 deg/s C*
1 deg
ji
jiHdtd
,
),( ji
jiH,
),(and
29 AIAA GNC 2005
High Cruise – 10 deg Pitch Down (No fault)
+ lightturbulence
30 AIAA GNC 2005
High Cruise – 10 deg Pitch Down (75% e fault)
31 AIAA GNC 2005
High Cruise – 10 deg Pitch Down (Reconfiguration)
32 AIAA GNC 2005
Cooper Harper Ratings & Pilot Workload
High Cruise
|| ||2pedalforce
ecolumnforcwheelforce
LP HPs1
workloadpilot