You are on page 1of 5

Adaptive Control Allocation for Flight Systems with Stuck Actuator

Failures
Man Wang1 , Jianying Yang2
1. China Center for Information Industry Development (CCID) , Beijing, 100036, China
E-mail: dream26@pku.edu.cn
2. Science and Technology on Aerospace Intelligent Control Laboratory, The State Key Laboratory for Turbulence and Complex
Systems, Department of Mechanics and Engineering Science, College of Engineering, Peking University, Beijing 100871, China
E-mail: yjy@pku.edu.cn

Abstract: This paper proposes an adaptive control allocation method for flight systems in the presence of stuck actuator
failures and model mismatch. During the flight, some unknown actuators are stuck at unknown values and at unknown
time instants, which badly influence the system stability. The method can adaptively allocate the control power to the
operating actuators based on the failure pattern. More realistic situation that incorporates model mismatch caused by
icing, modeling errors and damage is also be considered in this paper. The traditional reference model is not appropriate,
and an adaptive reference model is designed to solve this problem. The effectiveness of the proposed method is simulated
by a longitudinal linear model of the NASA Generic Transport Model. The simulation results show that this method gives
better tracking error performance.

Key Words: Fault tolerant control, control allocation, adaptive control, flight control

1 INTRODUCTION that can manage actuator redundancy for different control


strategies handling actuator failures [13]. Many kinds of
Actuator failures may adversely influence the stability and approaches on control allocation have been developed. The
performance of flight systems. Fault tolerant control (FTC) general approaches incorporate weighted pseudo-inverse
has been studied as a method for accommodating failures, [13], dynamic control allocation [14], interior-point algo-
thus guaranteeing the reliability and safety of flight sys- rithms [15] and daisy chain control allocation [16], etc. CA
tems. Many approaches have been studied to achieve FTC is widely used in the aerospace system. Zhou introduced
[1]-[6], such as linear parameter-varying designs [1]-[3], two reconfigurable control allocation schemes and illus-
feedback linearization methods [5], sliding mode control trated them with an unmanned aerial vehicle under stuck
methods [6]-[8]. When the occurrence, pattern and extent actuator failures in [17]. Liu demonstrated an adaptive con-
of failures are all uncertain, it is difficult for flight systems trol allocation method which cooperated the actuators as a
to achieve FTC. Generally, model reference adaptive con- group and adaptively allocated the equivalent control signal
trol (MRAC) can adjust the controller along with the vary- to each actuator in [18]. Casavola used an ‘ad hoc’ online
ing system to achieve the desired stability and performance. parameter estimator and a control allocation algorithm to
Tao and Joshi proposed an adaptive state feedback control realize control reconfiguration adaptively [19].
scheme to compensate for actuator failures characterized
In this paper we utilize an adaptive control allocation
by fixed or varying values in [9]. In addition to actuator
method to deal with actuator redundancy while simultane-
failures, there may be parameter deviations or model mis-
ously compensating for the model mismatch by using an
match in the actual system due to icing, modeling errors
adaptive reference model. A combination of control alloca-
and damage. The model mismatch may alter the structure
tion and adaptive control is explored to guarantee the stabil-
of the system so that the signal boundedness and asymp-
ity and good tracking performance of the system with stuck
totic tracking performance are no longer guaranteed. Nu-
actuators and model mismatch. This method can adaptively
merous literatures have focused on this problem [10]-[12].
allocate the control power to the remaining actuators ac-
Tao and Joshi solved this problem by an adaptive reference
cording to the effectiveness level of each actuator, which
model in [10].
can make the tracking performance better. This is the nov-
In the aircraft design process, designers often add redun-
elty of this paper compared to the work in [20]. This paper
dant actuators to the aircraft. Although the adaptive con-
mainly focuses on adaptive control allocation for the sys-
trol method proposed by [9] is able to compensate for actu-
tem with model mismatch in system matrix.
ator failures, there is no actuator redundancy management
in their research. Control allocation (CA) is one approach The rest of this paper is organized as follows. Section II
formulates the system in the presence of actuator failures
This work is supported by Research project fund 17-163-11-ZT-003- and model mismatch and introduces the adaptive reference
018-01, Joint fund of the Ministry of education of China(6141A020223). model scheme briefly. Section III gives a detailed descrip-

978-1-7281-0106-4/19/$31.00 2019
c IEEE 3933
tion of the adaptive control allocation scheme for the sys- where Am ∈ Rn×n , Bm ∈ Rn ; xm ∈ Rn is the reference
tem with actuator failures and model mismatch. The effec- model state; r ∈ R is a bounded reference input. The ref-
tiveness of this method is demonstrated by a longitudinal erence model is assumed to be able to capture the desired
linear model of the NASA Generic Transport Model in Sec- performance of the plant.
tion IV, and some concluding remarks are given in Section However, if the aircraft suffers a failure, the matching con-
V. ditions (5) have to be modified as
2 PROBLEM FORMULATION A0 + b 0 K 1 T = A m ; b0 K 2 = B m
(7)
b0 K3 = −B0 σ ū
Consider a linear time-invariant system described by
where K3 is a scalar utilized to compensate for the constant
ẋ(t) = (A0 + δa )x(t) + B0 u(t) (1) actuator value and the ideal control allocation gain α∗ has
to be changed to another vector α satisfying B0 (I − σ)α =
where A0 , δa ∈ Rn×n , B0 ∈ Rn×m , x(t) ∈ Rn and
b0 .
u(t) ∈ Rm . A0 , B0 represent the nominal values of the
In addition to actuator failures, the aircraft may suffer ic-
system parameters, and δa represents unknown parameter
ing or other damage which can result in model mismatch in
deviation or model mismatch. During the operation, the
system matrix for the actual situation. Under this circum-
actuators u(t) may suffer failures, which can be modeled
stance, the matching conditions (7) cannot be satisfied for
as [9]
any K1 , K2 , K3 which indicates that the desired closed-
uj (t) = ūj , t ≥ tj loop performance of the plant is no longer guaranteed. Ac-
(2)
j ∈ J = {j1 , j2 , . . . , jp } ⊆ {1, 2, . . . , m} cording to [10], we utilize the following adaptive reference
model to solve this problem.
where the failure pattern J, the failure value ūj , and the
moment that the failure occurs tj are all unknown. The ẋm = AM xm + Bm r (8)
failure value ūj is assumed to be a constant which indicates where AM is a time-varying matrix incorporating the es-
that the aircraft control surface is locked at some unknown timate of model mismatch. The system matrix AM can
fixed value. The actual control input to the system can be be represented as an affine function of a parameter vec-
described as [9] nρ

tor ρ ∈ Rnρ , that is, AM [ρ(t)] = Am + ρi (t)AMi ,
u(t) = v(t) + σ[ū − v(t)] = (I − σ)v(t) + σ ū (3) 1
where ρi (t) ∈ [ρimin , ρimax ]. The quadratic stability can
where v(t) ∈ Rm is the commanded control input, σ be achieved based on the following Lemma.
represents the failure pattern and is denoted as σ = Lemma 1: [10] The uncertain linear system (8) is robust
diag{σ1 , σ2 , . . . , σm }, where σi = 1 if the ith actuator quadratic stable if there exist positive definite matrices P =
has failed and σi = 0 otherwise. It is assumed that there is P T and Q = QT such that
at least one operable actuator in the system, which guaran- ATM i P +P AM i < − Q (9)
tees that the system remains controllable after the failure.
In this paper, we mainly focus on the case that all actuators Since the system matrix AM is time-varying, the stabil-
have similar physical characteristics. This condition sug- ity of the reference model must be guaranteed. A param-
gests that there is sufficient actuator redundancy for failure eter projection proposed in [20] can be utilized to ensure
compensation. Following [18], the redundant actuators can that the parameter estimates remain within the stable re-
be combined as a group and the commanded input can be gion during the update procedure.
defined as As mentioned previously, the failure pattern, value and
v(t) = α̂(t)v̂0 (t) (4) occurrence of time are all unknown. For accommoda-
tion of the uncertain failure, the adaptive control signal
where v̂0 (t) and α̂(t) are the group commanded input and should be utilized. Denoting θ̂ = [K̂1T K̂2 K̂3 ]T and
the adaptive control allocation gain respectively. The com- ω = [xT r 1]T , the adaptive control law can be de-
ponents of the adaptive control allocation gain α̂(t) = scribed as
[α̂1 (t), α̂2 (t), . . . , α̂m (t)] are related to the effectiveness of
the actuators and they can be updated online. v̂0 = K̂1T x + K̂2 r + K̂3 = θ̂T ω (10)
For the nominal plant with no failure, the ideal control al- where K̂1 ∈ Rn , K̂2 ∈ Rm and K̂3 ∈ R are the estimates
location gain is denoted as α∗ ∈ Rm which can be de- of K1 , K2 and K3 .
cided based on the prior knowledge of the actuators, and Remarks:
b0 = B0 α∗ is defined as the equivalent input matrix for the
(1) The existence of the new desired allocation gain α is
group actuators. In this paper, the pair (A0 , b0 ) is assumed
valid which has been proved in [18]. The value of
to be stabilizable which indicates that there exist a matrix
the new allocation gain α may be unknown due to the
K1 ∈ Rn and a scalar K2 ∈ R such that
uncertainty of the failure. For the actual plant, it can
A0 + b 0 K 1 T = Am ; b0 K 2 = B m (5) be updated adaptively to accommodate the failure.

and the reference model can be described as (2) Since the length of the paper is restricted, the details
of the parameter projection are not presented in our
ẋm = Am xm + Bm r (6) paper. People can found the details in [20].

3934 The 31th Chinese Control and Decision Conference (2019 CCDC)
3 ADAPTIVE CONTROL ALLOCATION FOR proof: Define
THE SYSTEM WITH MODEL MISMATCH n
 
This section mainly introduces the adaptive control alloca- V = eT P e+ γi−1 α̃i2 + θ̃T Γ−1
θ θ̃ +
T −1
δ̃ai Γa δ̃ai (19)
tion method for the system with model mismatch in system i∈J
/ i=1

matrix. In addition to utilizing model reference adaptive Differentiating (19) with respect to time, we get
control to compensate for the failure, we update the control  
allocation gain online to allocate the group control signal V̇ = 2eT P AM e − δ̃a x + B0 (I − σ)α̃v̂0 + b0 θ̃T ω
to each actuator based on the weight coefficients.  
 −1 ˙ T −1 ˙
n
T −1 ˙
Combing (1), (3), (4) and (10), the closed-loop system can +2 γ α̃i α̂i + θ̃ Γ θ̂ +
i θ δ̃ Γ δ̂ ai ai a
be described as i∈J
/ i=1
  
˙
= −eT Qe − 2trace δ̃aT P exT − Γ−1 a δ̂ a
ẋ = (A0 + δa )x + B0 (I − σ)v + B0 σ ū 

= (A0 + δa )x + B0 (I − σ)α̂v̂0 + B0 σ ū +2 α̃i eT P v̂0 bi + γi−1 α̂˙ i
= (A0 + δa )x + b0 v̂0 + B0 (I − σ)α̃v̂0 + B0 σ ū / 
i∈J 
˙
= (A0 + δa + b0 K1 T )x + b0 K2 r + B0 (I − σ)α̃v̂0 +2 θ̃T ωeT P b0 + Γ−1 θ θ̂
+b0 θ̃T ω (20)
(11) Substituting (16), (17) and (18) into (20), it is easy to obtain
where α̃ = α̂ − α and θ̃ = θ̂ − θ are the estimated errors of
the allocation gain and controller gain respectively. V̇ = −eT Qe ≤ 0 (21)
According to the matching condition (5), (11) can be writ- Since V (t) is a positive definite function and V̇ (t) ≤ 0, it is
ten as easy to obtain V (t) ∈ L∞ , which indicates that e(t), x(t),
ẋ = (Am + δa )x + Bm r + B0 (I − σ)α̃v̂0 + b0 θ̃T ω (12) θ̂, δ̂a , α̂i are all bounded, and e(t) ∈ L2 . Based on the error
dynamic equation (15) and the closed-loop signal bounded-
Following [20], the modified reference model is defined as ness, we can get ė(t) ∈ L∞ , and thus limt→∞ e(t) = 0.
˙ ˙
Due to e(t) ∈ L2 L∞ , θ̂, δ̂ a , α̂˙ i ∈ L2 L∞ and
ẋm = (Am + δ̂a )xm + Bm r (13) x, r ∈ L∞ , all signals and estimated parameters are
bounded, realizing limt→∞ [x(t) − xm (t)] = 0.
where δ̂a is the estimate of δa . The stablity of the modified
reference model can be guaranteed by Lemma 1 and the 4 DESIGN EXAMPLE
parameter projection [20]. A longitudinal linear model of the NASA Generic Trans-
The system matrix AM mentioned previously is defined as port Model is utilized to demonstrate the advantages of
AM = Am + δ̂a (14) the proposed method in this paper. The state variables are
the true airspeed V (ft/s), angle of attack αα (rad), pitch
Denoting the tracking error as e = x − xm , (12) and (13) rate q (rad/s), and pitch angle θθ (rad), which can be de-
T
yield scribed as x = [VT , αα , q, θθ ] . The control inputs in-
clude left outboard elevator δelob (deg), left inboard eleva-
ė = AM e − δ̃a x + B0 (I − σ)α̃v̂0 + b0 θ̃T ω (15) tor δelib (deg), right outboard elevator δerob (deg) and right
inboard elevator δerib (deg), which can be represented as
Compared to the adaptive laws in [20], the following theo- T
u = [δelob , δelib , δerob , δerib ] .
rem incorporates an adaptive control allocation law, which The system matrices are
can allocate the virtual control signal adaptively when the ⎡ ⎤
aircraft suffers a failure. −0.045 −8.9632 0.0349 −32.1740
⎢ −0.0035 −2.7429 0.9514 0 ⎥
Theorem 1: For the system in (12), the control law (10) A=⎢ ⎣ −0.0056 −42.6233 −3.5616


with the controller parameter adaptive law 0
0 0 1 0
˙
θ̂ = −Γθ ωeT P b0 (16) ⎡ ⎤
−0.0110 −0.0110 −0.0110 −0.0110
⎢ −0.0012 −0.0012 −0.0012 −0.0012 ⎥
the model mismatch estimation law B=⎢ ⎣ −0.1962 −0.1962 −0.1962 −0.1962 ⎦

˙
δ̂ a = Γa P exT (17) 0 0 0 0

and the control allocation adaptive law We set the initial value of α(t) as [1, 1, 1, 1]T . Once a fail-
ure occurs, α(t) will allocate the control signal adaptively.
α̂˙ i = −γi eT P v̂0 bi , i = 1, . . . m, i ∈
/J (18) We utilize LQR method to design the nominal controller
K1 and assume K2 = 1, satisfying
where α̂i denotes the ith element of α̂i ; Γa ∈ Rn×n and
Γθ ∈ R(n+2)×(n+2) are positive definite and symmetric; A0 + b 0 K 1 T = Am ; b0 K 2 = B m
γi is a positive scalar, guarantee that all the closed-loop The stable region is
signals are bounded and limt→∞ [x(t) − xm (t)] =0 in the
presence of actuator failures and model mismatch. ρ ∈ [−0.4938, 0.4938]

The 31th Chinese Control and Decision Conference (2019 CCDC) 3935
The reference command is described as

−5 deg, 1s ≤ t < 8s 5
r(t) =
0, else

δelob(deg)
0

−5
The aircraft suffers an unknown failure pattern such that
−10
the first two actuators get stuck at their current location at 0 2 4 6 8 10 12 14 16 18
Time(s)
1.01s.
5
The model mismatch matrix is
⎡ ⎤

δelib(deg)
0
0 0 0 0
⎢ 0 0 0 0 ⎥
−5

δa = ⎢⎣ 0 0 0

0 ⎦ −10
0 2 4 6 8 10 12 14 16 18

0.1 0 0 −0.1 Time(s)

20

Obviously, δa will not be compensated for any K1 and K2 .

δerob(deg)
10
The tracking errors of the system with and without adaptive
0
control allocation (ACA) are shown in Fig. 1. For space
−10
saving, this paper only presents the actual control inputs, 0 2 4 6 8 10 12 14 16 18
Time(s)
the model mismatch estimates and the control allocation
estimates of the method with ACA in Fig. 2-Fig. 3. 20

δerib(deg)
When the aircraft suffers a failure, the allocation estimates
begin to adjust adaptively. The estimation of δa always 0

keeps within the stable region during the adaption and ap- −20
0 2 4 6 8 10 12 14 16 18
proaches the true value. From Fig. 2, we can find that the Time(s)

first two actuators get stuck and get out of control. Ob-
viously, the method without ACA exhibits large tracking
errors. By contrast, the tracking errors of the method with Figure 2: The control input of the system with ACA.
ACA are significantly reduced by using the gain adaption
law (18). The method with ACA improves the efficiency of 8
α (t)
the adaption and reduces the complexity of the controller 6
1
α2(t)
design.
α(t)

4 α (t)
3
2 α (t)
4

0
0.1 0 2 4 6 8 10 12 14 16 18
no α(t) Time(s)
with α(t) Figure 3: The allocation gain of the system with ACA.
V (ft/s)

0
T

−0.1
0 2 4 6 8 10 12 14 16 18
Time(s) 5 CONCLUSION
1
no α(t) In this paper we addressed the problem in adaptive con-
αα(deg)

0.5 with α(t) trol allocation of systems with model mismatch. We real-
0 located the control power to each actuator according to the
−0.5 failure pattern in addition to adjusting the controller when
0 2 4 6 8 10 12 14 16 18
Time(s) the aircraft experienced a failure. The actual flight systems
10 with model mismatch was taken into consideration in this
no α(t)
paper. An adaptive reference model was utilized to com-
p(deg/s)

5 with α(t)
pensate for the model mismatch. Finally, simulation of the
0
NASA Generic Transport Model presented better tracking
−5
0 2 4 6 8 10 12 14 16 18 error performance. Future work will concentrate on exten-
Time(s)
sion to the case with actuator saturation errors.
0.5
no α(t)
with α(t)
θθ(deg)

0 0.2
δa44
0.1
−0.5 δa41
δ (t)

0 2 4 6 8 10 12 14 16 18 0
a

Time(s)
−0.1

−0.2
0 2 4 6 8 10 12 14 16 18
Figure 1: The tracking errors of the system with ACA and Time(s)
without ACA. Figure 4: The estimation of δa in the reference system with
ACA.

3936 The 31th Chinese Control and Decision Conference (2019 CCDC)
REFERENCES [11] T. E. Gibson, L. G. Crespo and A. M. Annaswamy, “Adap-
tive Control of Hypersonic Vehicles in the Presence of
[1] R. R. Wang, H. Zhang and J. M. Wang, “Linear Parameter-
Modeling Uncertainties,” American Control Conference, St.
Varying based Fault-Tolerant Controller Design for a Class of
Louis, MO, pp. 3178-3183 (2009).
Over-Actuated Nonlinear Systems with Applications to Elec-
[12] S. M. Joshi, G. Tao and T. Khong, “A Direct Adaptive Con-
tric Vehicles,’’ IET Control Theory and Applications, Vol. 8,
trol Approach in the Presence of Model Mismatch,” AIAA
No. 9, pp. 705-717 (2014).
Guidance, Navigation, and Control Conference, Chicago,
[2] H. Zhang, X. J. Zhang and J. M. Wang, “Robust gain-
Illinois (2009).
scheduling energy-to-peak control of vehicle lateral dynam-
[13] H. Alwi and C. Edwards, “Fault tolerant control using slid-
ics stabilisation,’’ Vehicle System Dynamics, Vol. 52, No. 3,
ing modes with on-line control allocation,” Automatica, Vol.
pp. 309-340 (2014).
44, No. 7, pp. 1859-1866 (2008).
[3] R. R. Wang and H. Zhang, “Linear Parameter-Varying Con-
[14] O. Härkegård, “Dynamic Control Allocation Using Con-
troller Design for Four-Wheel Independently Actuated Elec-
strained Quadratic Programming,” Journal of Guidance,
tric Ground Vehicles With Active Steering Systems,’’ IEEE
Control, and Dynamics, Vol. 27, No. 6, pp. 1028-1034
Transactions on Control Systems Technology, Vol. 23, No. 4,
(2004).
pp. 1281-1296 (2014).
[15] J. Peterson and M. Bodson, “Interior-point algorithms for
[4] S. Y. Cao, L. Guo and X. Y. Wen, “Fault tolerant control
control allocation,” Journal of Guidance, Control, and Dy-
with disturbance rejection and attenuation performance for
namics, Vol. 28, No. 3, pp. 471-480 (2005).
systems with multiple disturbances,’’ Asian Journal of Con-
[16] J. M. Buffington and D. F. Enns, “Lyapunov Stability Anal-
trol, Vol. 13, No. 6, pp. 1256-1064 (2010).
ysis of Daisy Chain Control Allocation,” Journal of Guid-
[5] M. Wang and D. H. Zhou, “Fault tolerant control of feedback
ance, Control, and Dynamics, Vol. 19, No. 6, pp. 1226-1230
linearizable systems with stuck actuators,’’ Asian Journal of
(1996).
Control, Vol. 10, No. 1, pp. 74-87 (2008).
[17] Q. L. Zhou, Y. M. Zhang, C. A. Rabbath and J. Apkar-
[6] Y. H. Liu, Y. G. Niu, J. Lam and B. Y. Zhang, “Sliding Mode
ian, “Two reconfigurable control allocation schemes for un-
Control for Uncertain Switched Systems with Partial Actua-
manned aerial vehicle under stuck actuator failures,” AIAA
tor Faults,’’ Asian Journal of Control, (2015). Guidance, Navigation, and Control Conference, Toronto,
[7] X. Wang, “Incremental Sliding-Mode Fault-Tolerant Flight Ontario Canada (2010).
Control,” Journal of guidance, control, and dynamics, In [18] Y. Liu and L. G. Crespo, “Adaptive control allocation in the
press (2018). presence of actuator failures,” AIAA Guidance, Navigation,
[8] Y. Gao, “Sliding mode fault-tolerant control of uncertain sys- and Control Conference, Toronto, Ontario Canada (2010).
tem: A delta operator approach,” International journal of ro- [19] A. Casavola and E. Garone, “Fault-tolerant adaptive control
bust and nonlinear control, Vol. 27, No. 18, pp. 4173-4187 allocation schemes for overactuated systems,” International
(2017). Jornal Of Robust And Nonlinear Control, Vol. 20, No. 17,
[9] G. Tao, S. M. Joshi and X. L. Ma, “Adaptive State Feed- 1958-1980 (2010).
back and Tracking Control of Systems with Actuator Fail- [20] S. M. Joshi, P. Patre and G. Tao, “Adaptive Control of Sys-
ures,” IEEE Transactions on Automatic Control, Vol. 46, No. tems with Actuator Failures Using an Adaptive Reference
1, pp. 78-95 (2001). Model,” Journal of Guidance, Control, and Dynamics, Vol.
[10] S. M. Joshi, G. Tao and P. Patre, “Direct Adaptive Control 35, No. 3, pp. 938-949 (2012).
Using an Adaptive Reference Model,” International Journal
of Control, Vol. 84, No. 1, pp. 180-196 (2011).

The 31th Chinese Control and Decision Conference (2019 CCDC) 3937

You might also like