2005 American Control Conference
5
June 8-10, 2005. Portland, OR, USA
Reliable H∞ Aircraft Flight Controller Design Against Faults
With State/Output Feedback
Le Feng, Jianliang Wang Engkee Poh
School of Electrical & DSO National Laboratories
Electronic Engineering Singapore, 118230
Nanyang Technological University pengkee@dso.org.sg
Singapore, 639798
{pg04094824, ejlwang}@ntu.edu.sg
Abstract— This paper presents a novel H∞ formulation to
design a reliable tracking controller for a discrete LTI aircraft
system against actuator outage faults and control surface
impairment modelled as polytopic uncertainties. Both the
state-feedback and output-feedback cases are considered. The
approach is based on multiobjective optimization using several
parameter dependent Lyapunov functions, each corresponding
to a vertex of the uncertainty polytope. The main advantage of
this approach with respect to the other well-known techniques
is the reduced conservativeness. In the output-feedback case,
a new Linear Matrix Inequality (LMI) formulation is derived
and solved by an iterative LMI (ILMI) algorithm. Finally, the
application to a nonlinear F-16 aircraft model with actuator
outage faults and control surface impairment illustrates the
effectiveness of the proposed approach.
Keywords: H∞ Control; Flight Control; Linear Matrix In-
equalities; Lyapunov Function; Outage Fault; Control Surface
impairment; State/Output Feedback
I. I NTRODUCTION
Reliability, maintainability and survivability are three
main issues in the aircraft flight controller design area.
Reliability means that the controller should optimize the
performance in normal condition and meet certain closed-
loop stability and performance demands in the presence
of a set of given faults simultaneously. Such system is
called a reliable control system which is particularly suitable
in safety-critical systems, such as aircraft, nuclear power
plants where safety and reliability are sometimes of more
importance than good performance solely. In the literature
to date, reliable control has attracted considerable amount
of attention ([1]-[2]), and several applications have been
developed ([3]-[4]).
A various reliable flight control system design conception
named self-repairing flight control system (SRFCS) was
proposed in ([5], [6]). SRFCS is regarded as an active ap-
proach which includes real-time fault detection and isolation
(FDI) procedure, and control system reconfiguration proce-
dure in the presence of faults. Such active approach provides
satisfying robustness and reliability for aircraft. However,
identifying the fault and reconfiguring the controller may
require time delay, which may not be available in many
safety-critical applications.
Another reliable flight control system design conception
0-7803-9098-9/05/$25.00 ©2005 AACC 2664
ThB11.
Fang Liao
Tamasek Laboratories
National University of Singapore
Singapore, 117508
Email: tsllf@nus.edu.sg
is to design a fixed controller which can tolerate a set
of faults by exploiting the inherent redundancy of aircraft
control surface actuators. As no FDI and/or controller
reconfiguration is needed, this passive approach provides
promising fault handling ability in practical applications
([7], [4], [8]) and has received considerable attention.
In this paper, the reliable H∞ flight tracking controller
design problem is studied, in the presence of actuator outage
faults and/or control surface impairment. For control surface
impairment fault, each fault is defined as a vertex to a
polytopic uncertain system, and the design goal is to find
a controller to stabilize this uncertain system. Comparing
with approaches that use only one Lyapunov function on all
vertices in controller design, parameter dependent Lyapunov
functions, one for each and every vertex of the uncertainty
polytope, are employed in this paper in order to reduce the
controller’s conservativeness. Multiobjective optimization
methodology is applied to optimize the tracking perfor-
mance during normal condition when there is no fault and
to maintain acceptable degradations when faults occur.
The paper is organized as follows. Section II presents mod-
els of outage faults and control surface impairment, together
with the formulation of reliable tracking problem. In Section
III, both state feedback and output feedback designs are
discussed and a new LMI formulation is derived. In Section
IV, a nonlinear F-16 aircraft model with mentioned faults is
used to illustrate the effectiveness of the proposed approach.
Finally, conclusion is given in Section V.
II. FAULT MODEL AND RELIABLE TRACKING PROBLEM
FORMULATION
Following [4], we present the discrete-time aircraft model
as follows:


x(k + 1)= Ax(k) + B u u(k) + B w w(k)
(1)
y(k)= Cx(k), x(0)= x0
where x(k) = [u, w, q, v, p, r]T ∈ n is the state, u(k) ∈
m is the control input, w(k) ∈ h is the bounded input
disturbance and y(k) ∈ p is the output.
To study the reliable flight control and tracking problem in
case of faults, the fault models must be established first.
There are two types of faults to be considered in this paper.
Let uF (k) represent the control surface input vector after 3) In the case of actuator outage faults and/or control
actuator outage faults have occurred. Then, the following surface impairment, the output signal Sy(k) tracks
actuator fault model is adopted in this paper: the reference signal r(k) without steady-state error
and with an acceptable degradation in tracking per-
uF (k) = ωL u(k), L = 0, 1, . . . , lp , lp
2m − 1 (2)
formance.
where ωL ∈ m×m is the scaling factor satisfying In order to obtain a reliable flight controller with zero
ωL ∈ Ω  {diag[ωL1 , ωL2 , . . . , ωLm ] steady state tracking error for step input, we introduce
(3) integral action on the continuous-time tracking error ė(t) =
ωLj = 0 or 1, j = 1, 2, . . . , m}
−SCx(t) + r(t). Using a sampling time of 0.01s, the fol-
From equations (2) and (3), it is clear that ωLj = 0 (1
lowing discrete-time vertex system model can be obtained.
j
m) represents the outage fault case of the j th actuator ⎧ u w
and ωLj = 1 corresponds to normal condition of the j th ⎨xa(k +1) = Aai xa (k) + Bai ωL u(k) + Bai wa (k)
control channel. We also define L = 0 to correspond to the ya (k) = Cai xa (k) (8)
⎩
normal condition, i.e., ω0 = Im×m . z(k) = C1 xa (k) + D1 u(k)
Another fault we are dealing with here is the control surface where the augmented state is xa (k) = [eT (k) xT (k)]T and
impairment, which is characterized by the percentage loss the disturbance vector is wa (t) = [rT (k) wT (k)]T , the ma-
of the total control surface area. Unlike actuator outage trix C1 , D1 are constant matrices of appropriate dimension,
faults of the form (2), control surface impairment will u
while z(k) is the regulated output. The matrices Aai , Bai ,
change the aerodynamic characteristics of the aircraft. For w
Bai and Cai are the discretization of the corresponding
different percentage control surface losses, corresponding continuous-time vertex system models.
aircraft models can be obtained and defined as vertices to a Then the following discrete-time model with polytopic
polytopic uncertain system. Then, for arbitrary percentage uncertainty will be used to model with actuator fault (2)-(3)
loss on such control surface, aircraft model can be obtained and control surface impairment fault (4)-(5).
from these models by linear interpolation. To describe this, ⎧
the following polytopic uncertainties are adopted: ⎨xa(k +1) = Aa(θ)xa(k) + Bau(θ)ωL u(k) + Baw(θ)wa (k)
l l ya (k) = Ca(θ)xa (k) (9)
A(θ) = i=0 Ai θi , B u (θ) = i=0 Biu θi , ⎩
l  (4) z(k) = C1 xa (k) + D1 u(k)
l
C(θ) = i=0 Ci θi , B w (θ) = i=0 Biw θi where
where the parameter θ = [θ0 , θ1 , . . . , θl ]T satisfying l l
  Aa (θ) = i=0 Aai θi , Bau (θ) = i=0 Bai
u
θi ,
l l  (10)
 Ca (θ) = i=0 Cai θi , w l w
Ba (θ) = i=0 Bai θi
l+1
θ∈Θ θ∈ : θi  0, θi = 1 (5)
i=0 If we obtain a controller to stabilize the augmented system
Vertex matrices Ai , Biu , Biw ,
Ci (i = 0, . . . , l) are known (9), then it also stabilizes the original system (6) and
matrices with i = 0 corresponding to the normal condition guarantees the steady state error to be zero.
and i = 1, . . . , l, the various vertex fault conditions. III. H∞ F LIGHT CONTROLLER DESIGN USING
Hence, integrating the actuator outage faults (2) and con- PARAMETER - DEPENDENT LYAPUNOV FUNCTIONS
trol surface impairment (4), the aircraft dynamics can be
For convenience, we first give the customary H∞ perfor-
characterized as mance indices over an infinite time horizon. Consider the


x(k + 1) = A(θ)x(k) + B u (θ)uF (k) + B w (θ)w(k) system (9) and a real number γ00 > 0, the exogenous signal
(6) wa is attenuated by γ00 if, assuming x(0) = 0,
y(k) = C(θ)x(k), x(0)= x0
Consider discrete-time aircraft model (6) with both actuator

∞

∞

z(k)
2 < γ00
wa (k)
2 (11)
outage fault (2) and control surface impairment (4), the k=0 k=0
reliable controller design problem in this paper is to find
Hereafter, we will design state/output feedback reliable
a controller such that:
flight controllers in an H∞ framework by using parameter-
1) The closed-loop system is robustly stable for all ωL ∈ dependent Lyapunov functions.
Ω and θ ∈ Θ.
2) In the event of no control surface actuator outage A. State Feedback Controller Design
faults, the output signal Sy(k) tracks the reference For the augmented polytopic uncertain system (9), fault
signal r(k) without steady-state error, that is: model (2)-(4), consider the static state-feedback controller
lim e(k) = 0, e(k) = r(k) − Sy(k) (7) u(k) = Ks xa (k) (12)
k→∞
and with optimized closed-loop performance. Here The closed-loop system of system (9) is given by
S is a known matrix of appropriate dimension to

xa (k +1)=[Aa(θ)+Bau (θ)ωL Ks ]xa(k)+Baw (θ)wa(k)
determine which outputs are required to track the (13)
z(k)=[C1 + D1 Ks ]xa (k)
reference signal.
2665
 Theorem 3.1: Consider the discrete linear system (9). where
There exists a static state-feedback control law of type (12),
such that the closed-loop system (13) is stable and the H∞ Z11 =QLi − G − GT − 2GT G0 − 2GT0 G + 2GT0 G0
u u
constraint (11) is fulfilled for all faults in (2)-(4), if there Z22 =−QLi −Bai ωL KCai (Bai ωL K0 Cai )T − Bai
u
ωL K0
exist matrices QLi = QTLi > 0, non-singular matrix G and
u T u u
Cai (Bai ωL KCai ) +Bai ωL K0 Cai (Bai ωL K0 Cai )T
V for i = 0, . . . , l, and L = 0, 1, . . . , lp , lp
2m − 1, Z33 =−I −D1 KCai (D1 K0 Cai )T −D1 K0 Cai (D1 KCai )T
⎡ ⎤ +D1 K0 Cai (D1 K0 Cai )T
QLi − G − GT ∗ ∗ ∗
u
⎢ Aai G + Bai ωL V −QLi ∗ ∗ ⎥ Proof: By using the Schur lemma, (21) can be
⎣ C1 G + D1 V 0 −I ∗ ⎦<0 (14) rewritten as follows:
0 w T
(Bai ) 0 −γLi I ⎡ ⎤
Z11 ∗ ∗ ∗
The state feedback gain Ks is given by A
⎢ ai G Z22 ∗ ∗ ⎥
⎣ C1 G 0 Z33 ∗ ⎦+
Ks = V G−1 (15) 0 (Bai w T
) 0 −γLi I
⎡ T ⎤
G G ∗ ∗ ∗
Proof: Given QLi = QTLi > 0 and the addi- ⎢ u u
B ω KCai × ⎥
tional non-singular matrix G of suitable dimension, (GT − ⎢Bai ωL KCai G aiu L ∗ ∗⎥ +
⎢ (Bai ωL KCai )T ⎥ (22)
QLi )Q−1 ⎣ 0 ∗⎦
Li (G − QLi ) is nonnegative definite, and conse- 0 0
quently, 0 0 0 0
⎡ ⎤
0 < G + GT − QLi
GT Q−1 Li G (16) T
G G ∗ ∗ ∗
⎢ 0 0 ∗ ∗⎥
holds true. Hence, the following LMI holds, ⎣ ⎦<0
D1 KCai G 0 D1 KCai (D1 KCai )T ∗
⎡ ⎤
−GT Q−1Li G ∗ ∗ ∗ 0 0 0 0
u
⎢Aai G + Bai ωL V −QLi ∗ ∗ ⎥
⎣ C G +D V < 0 (17)
1 i 1 0 −I ∗ ⎦ which is, in turn, equivalent to,
0 w T
(Bai ) 0 −γLi I ⎡ ⎤
QLi − G − GT ∗ ∗ ∗
u
where V = Ks G. Then, as G is nonsingular and ⎢(Aai + Bai ωL KCai )G −QLi ∗ ∗ ⎥
⎣ (C + D KC )G −I ∗ ⎦
+
G−1 always exists, we can multiply (17) from the 1 1 ai 0
w T
left by diag{QLiG−T , I, I, I} and from the right by 0 (Bai ) 0 −γLi I
⎡ ⎤
diag{G−1 QLi , I, I, I}, the following LMI is obtained (G − G0 ) T
∗ ∗ ∗
⎡ ⎤ ⎢×(G − G0 ) ⎥
−QLi ∗ ∗ ∗ ⎢ u u ⎥
u ⎢ (Bai ωL KCai − Bai ωL K0 Cai )× ⎥
⎢Aai QLi + Bai ωL WLi −QLi ∗ ∗ ⎥ ⎢ 0 u u T ∗ ∗ ⎥+
⎣ C1 QLi + D1 WLi −I ∗ ⎦<0 (18) ⎢ (Bai ωL KCai − Bai ωL K0 Cai ) ⎥
0 ⎣ 0 0 0 ∗⎦
w T
0 (Bai ) 0 −γLi I
0 0 0 0
where WLi =Ks QLi . This is nothing but the H∞ formu- ⎡ T ⎤
(G − G0 )
∗ ∗ ∗
lation based on one Lyapunov function in ([9]). ⎢×(G − G0 ) ⎥
⎢ ⎥
⎢ 0 0 ∗ ∗⎥
B. Output Feedback Controller Design ⎢ KC − D K C ⎥< 0 (23)
⎢ (D1 ai 1 0 ai )× ⎥
⎣ 0 0 T ∗⎦
Consider the static output-feedback controller (D KC − D K C )
1 ai 1 0 ai
0 0 0 0
u(k) = Kya (k) = KCa (θ)xa (k) (19)
Because the second and third terms in (23) is nonnegative
Assume that Ca (θ) is of full row rank. The closed-loop definite, then (23) (and hence, (21)) is a sufficient condition
system of system (9) becomes to the first term of (23). By Theorem 3.1, this implies

the closed-loop stability and H∞ performance (11) of the
xa(k+1)=[Aa(θ)+Bau(θ)ωLKCa(θ)]xa(k)+Baw(θ)wa (k)
(20) output feedback system (20).
z(k) =[C1 + D1 KCa(θ)]xa (k)
Remark 3.3: Theorem 3.2 gives a sufficient condition to
We have the following approach for output feedback case. achieve closed-loop stability and H∞ performance via a
Theorem 3.2: Consider the discrete linear system (9). static output feedback control law. In fact, (21) is not an
There exists a static output-feedback control law of type LMI. However, if G0 and K0 are given, then (21) is an LMI
(19) such that the closed-loop system (20) is stable and
the H∞ constraint (11) is fulfilled for all faults in (2)- in G, K and QLi , i = 0, . . . , l, and L = 0, 1, . . . , lp , lp

(4), if there exist matrices QLi = QTLi > 0, non-singular 2m −1. The conservativeness of this sufficient condition lies
matrix G, G0 and K, K0 such that for i = 0, . . . , l, and in the differences between G − G0 and K − K0 . By using
L = 0, 1, . . . , lp , lp
2m − 1, the following iterative algorithm, this conservativeness can
⎡ ⎤ be minimized.
Z11 ∗ ∗ ∗ ∗ ∗
⎢Aai G Z22 ∗ ∗ ∗ ∗⎥ Algorithm:
⎢C G Z33 ∗ ∗ ∗⎥
⎢ 1 0 ⎥ 1) Calculate a stabilizing static output feedback controller K00 .
⎢ 0 w T
−γLi I ∗ ∗⎥ <0 (21)
⎢ (Bai ) 0 ⎥ If there is no such controller can be found, then stop and
⎣ G (B u ω KC )T 0 0 −I ∗⎦ the algorithm fails to get a solution. The detail can be found
ai L ai
G 0 (D1 KCai )T 0 0 −I from Algorithm 5.2 of [10] and is omitted here for brevity.
2666
 2) Choose performance upper bounds γLi for i = 1, . . . , l
and L = 1, . . . , lp , lp
2m − 1, with K00 from Step
1, minimize γ00 (the performance upper bound in normal
condition) subject to the inequality (14). This gives G00 .
3) At the j th (j = 1, 2, . . .) iteration, set G0 = Gj−1
0 , K0 =
K0j−1 and minimize γ00 j
subject to linear matrix inequality
(21). This gives Gj , K j and γ00 j
. 4. ω3= diag{1, 1, 0, 1, 1} → Left aileron outage fault
j j−1
4) If |γ00 − γ00 | <  where  > 0 is a given error tolerance,
the obtained K j is the optimal output feedback controller
Kopt . Otherwise, set j = j + 1 and return to Step 3.
Remark 3.4: As the LMI (21) is solvable, the given
algorithm is convergent. In the case of state feedback
control, the method proposed here is based on LMIs instead
of ILMIs. Compared with [11], where only one common
Lyapunov function is adopted, the approach developed in
this paper employs several parameter dependent Lyapunov
functions, each one corresponding to a different vertex of
the uncertainty polytope. By this method, the conservative-
ness can be reduced theoretically. The design and simulation
in the Section IV also illuminates the advantage.
IV. D ESIGN AND SIMULATION
In modern aircraft flight controller design, more and more
challenging maneuvers such as high angle of attack flight
or high roll rate flight under a low air speed condition
should be considered. Under these conditions, significant
In this example, the following possible control surface
actuator faults are considered:
1. ω0= diag{1, 1, 1, 1, 1} → Normal condition;
2. ω1= diag{0, 1, 1, 1, 1} → Left elevator outage fault
3. ω2= diag{1, 0, 1, 1, 1} → Right elevator outage fault
5. ω4= diag{1, 1, 1, 0, 1} → Right aileron outage fault
6. ω5= diag{0, 1, 0, 1, 1} → Left elevator and left
aileron outage faults;
7. ω6= diag{0, 1, 1, 0, 1} → Left elevator and right
aileron outage faults;
8. ω7= diag{1, 0, 0, 1, 1} → Right elevator and left
aileron outage faults;
9. ω8= diag{1, 0, 1, 0, 1} → Right elevator and right
aileron outage faults;
By using the proposed approach in Section III-A and
Section III-B, a reliable state feedback tracking controller
and a reliable output feedback tracking controller can be
designed to tolerate the above outage faults.
Figure 1 and Figure 2 shows the linear simulation results
of the reliable static state/output feedback controller, for
all eight actuator outage faults, ω1 − ω8 . The tracking
commands are unit steps.

right elevator (degree)

left elevator (degree)

kinematic and inertial couplings are prevalent and must be 0.5 0.5

handled properly. Therefore, three variables: stability axis 0 0

roll rate µ̇rat , angle of attack α and sideslip angle β, must −0.5 −0.5

−1 −1
be stabilized and tracked closely. 0 2 4 6 8 10 0 2 4 6 8 10

right aleron (degree)

left aleron (degree)

2 t (s) 2 t (s)
In this section, two examples of reliable tracking control are
given to demonstrate the proposed method. The nonlinear 0 0

dynamic model for this F-16 aircraft can be found in [4]. −2 −2

The following parameters are used for the two examples, 4
0 2 4
t (s)
6 8 10
mu rat (degree/s) 2
0 2 4
t (s)
6 8 10
rudder (degree)

where S determines the outputs required to track. 2 1

⎡ ⎤
0 1 0 0 0   0 0

I 0 03×3
S=⎣0 0 0 1 0⎦, C1= 3×3 3×6 , D1=
−2 −1
0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)
03×3 03×6 0.1I3×3 1.5 1.5
alpha (degree)

beta (degree)

0 0 0 0 1 1 1
m
γLi=1, i = 1, . . . , l, L = 1, 2, . . . , lp , lp
2 − 1 0.5 0.5

0 0
0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)

A. Example 1 (Actuator Outage Faults)

Fig. 1. Response curves of the normal and actuator outage fault cases
First, only actuator outage faults are considered. The trim with reliable state feedback controller
condition is as follows,
right elevator (degree)
left elevator (degree)

0.5 0.5

T = 0.1464, Xcg = 0.35, H = 500m, 0 0

Vt = 152.4m/s, α = 2.3703◦ , β = 0.0◦ , −0.5 −0.5

(24)
δel = −0.5831◦ , δer = −0.5831◦ , δal = 0.0◦ , −1
0 2 4 6 8 10
−1
0 2 4 6 8 10
right aleron (degree)

δar = 0.0◦ , δr = 0.0◦

left aleron (degree)

2 t (s) 2 t (s)

0 0
where T is the engine throttle, H is the altitude, Vt is the
total airspeed, Xcg is the center of gravity location, δel , δer , −2
0 2 4
t (s)
6 8 10
−2
0 2 4
t (s)
6 8 10

δal , δar and δr are inputs to the model (i.e., the left and 4 2
mu rat (degree/s)
rudder (degree)

2 1
right elevators, the left and right ailerons and the rudder). 0 0
Then, the F-16 aircraft with independent control surfaces −2 −1
can be approximately modelled as follows, 1.5
0 2 4
t (s)
6 8 10
1.5
0 2 4
t (s)
6 8 10

alpha (degree)

beta (degree)

x(k + 1) = Ax(k) + B u ωL u(k) + B w w(k) 1 1

(25)
y(k) = Cx(k) L = 0, . . . , 8
0.5 0.5

0 0
0 2 4 6 8 10 0 2 4 6 8 10

where x(k) = [u, w, q, v, p, r]T , y(k) = [q, µ̇rat , r, α, β]T ,

t (s) t (s)

w(k) represents the vertical gust disturbance and A, B u , Fig. 2. Response curves of the normal and actuator outage fault cases
B w , C are system matrices trimmed from nonlinear aircraft with reliable output feedback controller
model and discretized with a sampling period of 0.01s.
2667
For comparison, we designed two standard (without consid- disturbance and
 
ering the actuator outage faults, state feedback and output A(θ) =  3i=0 Ai θi , B u (θ) =  3i=0 Biu θi ,
(27)
feedback) controllers, by using the same LMI formulation. C(θ) = 3i=0 Ci θi , B w (θ) = 3i=0 Biw θi
The simulation results are shown by Figure 3 and 4.
By using the proposed approach in Section III-A and
III-B, a reliable state feedback tracking controller and a

right aleron (degree)right elevator (degree)

left elevator (degree)

0.5 1

0 0 reliable output feedback tracking controller can be designed

−0.5 −1
to tolerate the above control surface impairment. Figure 5
−1 −2
0 2 4 6 8 10 0 2 4 6 8 10
and 6 shows the linear simulation results of the reliable
left aleron (degree)

1 t (s) 4 t (s)

0 2 static state/output feedback controller, for normal case and

−1

−2
0

−2
3 fault cases. The tracking commands are unit steps.
0 2 4 6 8 10 0 2 4 6 8 10
4 t (s) 2 t (s)
mu rat (degree/s)
rudder (degree)

0 0.2

elevator (degree)
2

aleron (degree)
1 −1 0
0
−2 −0.2
−2 0
0 2 4 6 8 10 0 2 4 6 8 10 −3 −0.4
2 t (s) 2 t (s)
alpha (degree)

beta (degree)

−4 −0.6
1 1 0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)
0 0 3 1.5

mu rat (degree/s)
rudder (degree)
−1 −1
0 2 4 6 8 10 0 2 4 6 8 10 2 1
t (s) t (s)

1 0.5

Fig. 3. Response curves of the normal and actuator outage fault cases 0 0
0 2 4 6 8 10 0 2 4 6 8 10
with standard state feedback controller 1.5
t (s)
1.5
t (s)
right aleron (degree)right elevator (degree)
left elevator (degree)

0.5 1

alpha (degree)

beta (degree)
0 0 1 1

−0.5 −1
0.5 0.5
−1 −2
0 2 4 6 8 10 0 2 4 6 8 10
0 0
left aleron (degree)

1 t (s) 4 t (s) 0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)
0 2

−1 0

−2 −2
Fig. 5. Response curves of the normal condition and control surface
0 2 4 6 8 10 0 2 4 6 8 10
4 t (s) 2 t (s) impairment with reliable state feedback controller
mu rat (degree/s)
rudder (degree)

0 0.2
2
elevator (degree)

aleron (degree)
1 −1 0
0
−2 −0.2
−2 0
0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s) −3 −0.4
2 2
alpha (degree)

beta (degree)

−4 −0.6
1 1 0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)
0 0 3 mu rat (degree/s) 1.5
rudder (degree)

−1 −1
0 2 4 6 8 10 0 2 4 6 8 10 2 1
t (s) t (s)

1 0.5

Fig. 4. Response curves of the normal and actuator outage fault cases 0 0
0 2 4 6 8 10 0 2 4 6 8 10
with standard output feedback controller t (s) t (s)
1.5 1.5
alpha (degree)

beta (degree)

From these figures above, it can be seen that, although the 1 1

standard controllers perform almost the same as the reliable 0.5 0.5

controllers in the normal condition, the reliable controllers 0

0 2 4 6 8 10
0
0 2 4 6 8 10

give much better tracking performance in the presence of t (s) t (s)

any of the eight actuator outage faults. Meanwhile, while Fig. 6. Response curves of the normal condition and control surface
using the ILMI method, the output feedback controller can impairment with reliable output feedback controller
perform as good as the state feedback controller.
For comparison, we designed two standard (without consid-
B. Example 2 (Control Surface Impairment) ering the control surface impairment, state/output feedback)
In this example, control surface impairment (up to 75% controllers, by using the same LMI formulation. The sim-
efficiency loss to the elevator, symmetrically) is considered. ulation results are shown by Figure 7 and 8.
Since the aerodynamic characteristics of the aircraft will
change nonlinearly from normal condition to 75% loss of From Figure 7 and 8, the same conclusion as that of section
elevator, in order to achieve better modelling accuracy in IV-A can be drawn. Furthermore, to verify the effectiveness,
interpolating polytope vertices, we consider two additional the reliable output feedback controller is applied to a nonlin-
vertices (25% and 50% efficiency loss to the elevator, ear F-16 aircraft model to perform a Herbst-like maneuver,
symmetrically) besides the existing two. Then, the aircraft which consists of a large angle-of-attack pull-up, followed
model with control surface impairment can be written as

by a stability axis roll. For comparison, the standard output
x(k + 1) = A(θ)x(k) + B u (θ)u(k) + B w (θ)w(k) feedback controller is also tested. The simulation result is
(26)
y(k) = C(θ)x(k)
given by Figure 9 and 10. In these simulations, a 30% loss
where x(k) = [u, w, q, v, p, r]T , y(k) = [q, µ̇rat , r, α, β]T , of the elevator is introduced at 12s, followed by a 60% loss
u(k) = [δe , δa , δr ]T , w(k) represents the vertical gust at 15s and a 70% loss at 18s.
2668
 4 0.2

elevator (degree)
30 0.5

aleron (degree)
2 0

elevator (degree)

aleron (degree)
20
0 −0.2
10 0
−2 −0.4
0
−4 −0.6
0 2 4 6 8 10 0 2 4 6 8 10 −10 −0.5
t (s) t (s) 0 10 20 30 40 50 0 10 20 30 40 50
3 1.5 t (s) t (s)
0.5 6

mu rat (degree/s)
rudder (degree)

mu rat (degree/s)
2

rudder (degree)
1 0 4
1 −0.5
2
0.5 −1
0
−1.5 0
−1 0
0 2 4 6 8 10 0 2 4 6 8 10 −2 −2
0 10 20 30 40 50 0 10 20 30 40 50
t (s) t (s) t (s) t (s)
2 1.5 15 0.1
alpha (degree)

beta (degree)

alpha (degree)
1.5

beta (degree)
1 10 0.05

1 5 0
0.5
0.5 0 −0.05

0 0 −5 −0.1
0 2 4 6 8 10 0 2 4 6 8 10 0 10 20 30 40 50 0 10 20 30 40 50
t (s) t (s) t (s) t (s)
’−−’ : reference command ’−’: tracking signal

Fig. 7. Response curves of the normal condition and control surface

impairment with standard state feedback controller Fig. 9. Nonlinear simulation of Herbst-like maneuver by reliable output
4 0.2 feedback controller
1
elevator (degree)

aleron (degree)

elevator (degree)
2 0 20

aleron (degree)
0.5
0 −0.2
0 0
−2 −0.4
−0.5
−20
−4 −0.6
0 2 4 6 8 10 0 2 4 6 8 10 −1
0 10 20 30 40 50 0 10 20 30 40 50
t (s) t (s)
t (s) t (s)
3 1.5
0.5 6
mu rat (degree/s)
rudder (degree)

mu rat (degree/s)
rudder (degree)
2 0 4
1
−0.5
1 2
0.5 −1
0 0
−1.5
−1 0 −2 −2
0 2 4 6 8 10 0 2 4 6 8 10 0 10 20 30 40 50 0 10 20 30 40 50
t (s) t (s) t (s) t (s)
2 1.5 15 0.5

alpha (degree)
alpha (degree)

beta (degree)
beta (degree)

1.5 10
1
1 5 0
0.5
0.5 0

0 0 −5 −0.5
0 2 4 6 8 10 0 2 4 6 8 10 0 10 20 30 40 50 0 10 20 30 40 50
t (s) t (s)
t (s) t (s)
’−−’ : reference command ’−’: tracking signal

Fig. 8. Response curves of the normal condition and control surface Fig. 10. Nonlinear simulation of Herbst-like maneuver by standard output
impairment with standard output feedback controller feedback controller

Clearly, from Figure 9 and 10, it can be seen that the

[2] C. S. Hsieh, “Reliable control design using a two stage linear
standard controller can not stabilize the aircraft in the quadratic reliable control,” Control Theory and Applications, IEE
presence of more than 60% efficiency loss of elevator, Proceedings, vol. 150, no. 1, pp. 77–82, 2003.
while the reliable controller can still work properly with [3] Q. Zhao and J. Jiang, “Reliable state feedback control system design
against actuator failures,” Automatica, vol. 34, pp. 1267–1272, 1998.
acceptable degradation in performance. [4] F. Liao, J. L. Wang, and G. H. Yang, “Reliable robust flight tracking
Summarizing these two examples, it is noted that the control: an LMI approach,” IEEE Transactions on Control Systems
proposed reliable controller design method can significantly Technology, vol. 10, pp. 76–89, 2002.
[5] N. E. Wu and T. Chen, “Reliability prediction for self-repairing flight
improve the system performance in the presence of faults, control systems,” Decision and Control, Proceedings of the 35th
without much sacrifice of performance in normal condition. IEEE, vol. 1, pp. 126–131, 1996.
[6] Y. Wang, S. S. Hu, and Y. Xiong, “Some basic problems about
V. C ONCLUSION self-repairing flight control,” Intelligent Control and Automation,
Proceedings of the 3rd World Congress, vol. 5, pp. 3161 – 3164,
This paper discusses the reliable flight controller design 2000.
problems subject to H∞ constraint. The actuator outage [7] G. H. Yang, J. L. Wang, and Y. C. Soh, “Reliable H∞ controller
faults and control surface impairment have been considered design for linear systems,” Automatica, vol. 37, pp. 717–725, 2003.
[8] G. H. Yang, J. L. Wang, and K. Y. Lou, “Reliable state feedback
by using polytopic fault models. To reduce the controller’s control synthesis for uncertain linear systems,” 6th International
conservativeness, the proposed approach is based on several Conference on Control, Automation, Robotics and Vision (ICARCV
parameter dependent Lyapunov functions, each one corre- 2000), 2000.
[9] G. F. Giancarlo, F. A. Cuzzola, D. MIgnone, and M. Morari,
sponding to a different vertex of the uncertainty polytope. “Analysis and control with performace of piecewise affine and hybrid
Furthermore, both state feedback and output feedback cases systems,” Swiss Federal Institute of Technology, Tech. Rep., 2000.
are discussed in this paper. By using an ILMI based [10] F. Liao, J. L. Wang, and G. H. Yang, “Reliable H2 static output-
feedback tracking control against aircraft wing/control surface im-
method, the output feedback case can be solved. Finally, pairment,” Proceedings of the international conference on Physics
the proposed method is applied to a nonlinear F-16 aircraft and Control, pp. 112–117, Aug 2003, St. Petersburg, Russia.
model and the effectiveness is illuminated. [11] ——, “Reliable aircraft flight control using static output feedback,”
Proceedings of AIAA Guidance, Navigation, and Control Conference,
R EFERENCES August 2002, Monterey, California, USA, AIAA-2002-4449.

[1] G. H. Yang, J. Lam, and J. L. Wang, “Reliable H∞ control for

affine nonlinear systems,” IEEE TRANSACTIONS ON AUTOMATIC
CONTROL, vol. 43, no. 8, pp. 1112–1117, 1998.

2669