This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIE.2015.2404784, IEEE Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 1

Fault Reconstruction and Fault-tolerant Control via

Learning Observers in Takagi-Sugeno Fuzzy
Descriptor Systems with Time Delays
Qingxian Jia, Wen Chen, Member, IEEE, Yingchun Zhang, and Huayi Li

Abstract—This paper addresses the problems of observer- process in characterizing a wide class of practical systems
based fault reconstruction and fault-tolerant control for Takagi- including electric and electronic systems, aircraft systems,
Sugeno fuzzy descriptor systems subject to time delays and biological systems, etc. Due to the universal existence of
external disturbances. A novel fuzzy descriptor learning observer
is constructed to achieve simultaneous reconstruction of system inherent nonlinearities in industrial and commercial processes,
states and actuator faults. Sufficient conditions for existence nonlinear descriptor systems and their applications should
of the proposed observer are explicitly provided. Utilizing the be specifically considered. On the other hand, it is well
reconstructed fault information, a reconfigurable fuzzy fault- recognized that the Takagi-Sugeno (T-S) fuzzy model can
tolerant controller based on the separation property is designed approximate a highly complicated nonlinear system using
to compensate for the impact of actuator faults on system per-
formance by stabilizing the closed-loop system. In addition, the a family of fuzzy IF-THEN rules that represent the local
design of the fault reconstruction observer and the fault-tolerant linear state-space models [10], [11], [12]. Therefore, taking
controller is formulated in terms of linear matrix inequalities that advantages of both descriptor and T-S formalisms in the field
can be conveniently solved using convex optimization techniques. of modelling, it is significantly efficient for T-S Descriptor
At last, simulation results on a truck trailer system are presented Systems (TSDS’s) to describe many practical processes.
to verify the effectiveness of the proposed approaches.
During the past decade, TSDS’s have attracted consider-
Index Terms—Fault reconstruction, Fault-tolerant control, T- able attention in both research and practical applications. A
S fuzzy descriptor systems, Learning observers, Linear matrix few efforts have been reported in stabilization [13], output-
inequalities.
feedback control [14], H∞ control [15], and sliding mode
control [16], just to name a few. Nevertheless, to the best
I. I NTRODUCTION of our knowledge, few research activities have been aimed
at fault diagnosis and FTC in TSDS’s. Reference [17] is an
T HE increasing complexity of modern engineering sys-
tems will correspondingly increase the possibility of
system faults and/or failures. The occurrence of sensor, actu-
existing work for designing an Unknown Input Observer (UIO)
with disturbance-decoupling ability to perform actuator fault
detection in discrete-time TSDS’s. An observer-based reliable
ator or component failures may dramatically degrade system
control method is developed in [18] for continuous-time
performance and even result in catastrophic system collapse.
TSDS’s with time delays where a fuzzy Luenberger observer
As a response to high requirement for system safety, reliability,
is utilized to estimate system states. Literature [19] addresses
and survivability, fault diagnosis and Fault-Tolerant Control
the issue of robust dissipative FTC against actuator faults and
(FTC) for dynamic systems have been attractive subjects of
parametric uncertainties in the LMI optimization framework.
many investigations in control community and have received
An observer-based active FTC strategy is investigated in
considerable attention during the past few decades, and a great
[20] where a proportional multi-integral observer is proposed
deal of research progress has been made [1], [2], [3], [4], [5],
for actuator fault reconstruction with an assumption that the
[6], [7], [8] and the references therein.
faults and their derivatives are norm-bounded. Additionally,
Descriptor systems are also referred to as singular systems,
observer-based fault diagnosis and FTC in T-S fuzzy systems
differential-algebraic systems, generalized state-space systems,
using descriptor techniques have also been studied in [21],
and implicit systems [9]. They arise from a natural modeling
[22], [23]. Although several fault diagnosis and FTC schemes
Manuscript received March 30, 2014; revised April 7, 2014, August 29, for TSDS’s have been proposed, they are still challenging
2014, October 15, 2014 and December 24, 2014; accepted January 5, 2015. issues, and are under investigation.
Copyright (c) 2015 IEEE. Personal use of this material is permitted. Recently, researchers have been paying much attention to
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending a request to [email protected] Learning Observer (LO)-based fault reconstruction [24], [25],
Qingxian Jia and Huayi Li are with Research Center of Satellite Technol- [26], [27], [28], [29]. Its main advantage is that it could accu-
ogy, Harbin Institute of Technology, Harbin, P. R. China, 150001. (e-mail: rately reconstruct time-varying faults, especially fast-varying
[email protected], [email protected])
Wen Chen is with Division of Engineering Technology, Wayne State faults. At present, few studies have been focused on LO design
University, Detroit, MI, USA, 48202. (e-mail: [email protected]) in T-S fuzzy systems. Therefore, there is a strong incentive
Yingchun Zhang is with Research Center of Satellite Technology, Harbin for us to investigate fault reconstruction and FTC for TSDS’s
Institute of Technology, Harbin, P. R. China, 150001. He is also with
Aerospace Dongfanghong HIT Satellite Company Ltd, Shenzhen, P. R. China, via LOs. In view of these, we will address the problems
518057 (e-mail: [email protected]) of fuzzy LO-based fault reconstruction and reconfigurable

0278-0046 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIE.2015.2404784, IEEE Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 2

FTC for continuous-time TSDS’s subject to time delays and Hence, the overall fuzzy model achieved by fuzzy blending
external disturbances. In this paper, the main contributions that of each individual plant rule is described by
are worth emphasizing are summarized in the following three
aspects:  r
X

1) A novel Fuzzy Descriptor Learning Observer (FDLO)



 E ẋ(t) = µi (z(t)){Ai x(t) + Ahi x(t − h)
is constructed to achieve simultaneous reconstruction of i=1
(2)
system states and actuator faults, and sufficient condi- 

 + Bi u(t) + Dd(t) + Fi fa (t)}

tions for existence of the FDLO are provided explicitly. 
y(t) =Cx(t)
2) The FDLO is designed to completely decouple ex-
ternal disturbances from fault reconstruction based on
disturbance-decoupling techniques. where µi (z(t)) is defined as
3) The FDLO and the Fuzzy Fault-Tolerant Controller q
(F2 TC) are independently designed based on the sep- ωi (z(t)) Y
µi (z(t)) = Pr , ωi (z(t)) = µij (zj (t)) (3)
aration property [30], [31]. It is worth noting that the i=1 ωi (z(t)) j=1
separation property has not been developed in any LOs
and LO-based FTC.
where µij (·) is the grade of the membership function of µij .
In addition, a systematic design method for the FDLO and the Pr
It is assumed in this paper that P i=1 ωi (z(t)) > 0 with
F2 TC is suggested using LMI optimization techniques; thus, r
ωi (z(t)) ≥ 0, ∀t. Therefore, we have i=1 µi (z(t)) = 1 with
it is convenient to calculate their gain matrices. 0 ≤ µi (z(t)) ≤ 1, ∀t.
Throughout the article, A > 0 (A < 0) denotes that
In order to address the main results, the following assump-
A is positive (negative) definite, λmin (A) and λmax (A) are
tions are made.
the minimum and maximum eigenvalues of A, respectively,
A† denotes a generalized inverse of A, k · k and k · k∞ Assumption 1:
 
represent Euclidean and infinity norms of a vector or matrix, E
rank = n. (4)
respectively, ∗ means symmetric term, In is an identity matrix C
of size n, C+ denotes the set of all complex numbers with non
negative real part, r = {1, · · · , r}, and ∀ means “for all”.
Assumption 2: The triple matrix (E, Ai , C) is R-detectable
[32], ∀i ∈ r; that is,
II. S YSTEM F ORMULATION
 
sE − Ai
The considered continuous-time TSDS is composed of a rank = n, s ∈ C+ , ∀i ∈ r. (5)
C
set of fuzzy IF-THEN rules, each of which represents a local
linear input-output relationship of the nonlinear system. The
ith rule of this T-S fuzzy descriptor model is in the following Assumption 3:
form:
Plant Rule i: IF z1 (t) is Mi1 and · · · zq (t) is Miq , THEN  
E D
rank = n + rank(D). (6)

E ẋ(t) =Ai x(t) + Ahi x(t − h) + Bi u(t) C 0



 + Dd(t) + Fi fa (t)
(1)
 y(t) =Cx(t) Remark 1: Assumptions 1 implies that there exists a full-
†

 

x(t) =φ(t), t ∈ [−h, 0]   E
row rank matrix T1 T2 = such that
C
where x(t) ∈ Rn , u(t) ∈ Rm and y(t) ∈ Rp ; they repre-
sent system states, control inputs, and measurement outputs,  
E
respectively. External disturbances d(t) ∈ Rν , and actuator
 
T1 T2 = In . (7)
C
faults fa (t) ∈ Rm and could be constant or time-varying.
Herein, z(t) = [z1 (t), · · · , zq (t)]T , representing the premise
variable vector that is supposed to be measurable. Fuzzy It also guarantees the impulse observability of the triple
sets Mij (i = 1, · · · , r; j = 1, · · · , q) are characterized by matrix (E, Ai , C), ∀i ∈ r [32]. Both Assumptions 1 and
the membership function, and r and q are the number of 2 are necessary conditions for existence of the designed
IF-THEN rules and of the premise variables, respectively. observer in the latter section. Assumption 3 is presented
Constant matrix E is singular, i.e. rank(E) = s < n. Initial for disturbance-decoupling design of the proposed observer.
vector φ(t) is a continuous function when t ∈ [−h, 0] where It is worth noting that such an assumption is common in
time delay h is a positive constant scalar, and φ(t) ∈ Rn . the design of disturbance-decoupling observer for descriptor
Without loss of generality, it is supposed that matrices Fi and systems. Similar assumptions can also be found in [17], [32]
C are of full column rank and of full row rank, respectively. and references therein.

0278-0046 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIE.2015.2404784, IEEE Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 3

III. FAULT RECONSTRUCTION BASED ON A FUZZY According to [34], a solution of (13) exists if and only if
DESCRIPTOR LEARNING OBSERVER  
E D  
A. Design of a fuzzy descriptor learning observer E D
rank  C 0  = rank . (14)
C 0
In order to reconstruct actuator faults in the TSDS (2), a In 0
novel FDLO is constructed as follows Under Assumption 3, it is easily verified that (14) holds.
r
X Therefore, a general solution for [T H] [34] can be obtained
ω̇(t) = µi (z(t)){T Ai x̂(t) + T Ahi x̂(t − h) as
i=1
 E D †
 
+ T Bi u(t) + L1i (y(t) − ŷ(t))
  
T H = In 0
C 0
+ L2i (y(t − h) − ŷ(t − h)) + T Fi fˆa (t)}, (8a)   † ! (15)
x̂(t) =ω(t) + Hy(t), (8b) E D E D
−N In+p −
C 0 C 0
ŷ(t) =C x̂(t), (8c)
r
X where N ∈ Rn×(n+p) , representing an arbitrary matrix. In ad-
fˆa (t) =K1 fˆa (t − τ ) + µi (z(t))K2i (y(t) − ŷ(t)) (8d) dition, to guarantee fault detectability by the proposed FDLO,
i=1 matrix T should be selected such that rank(T Fi ) = rank(Fi ),
where ω(t) ∈ R and x̂(t) ∈ Rn ; they are observer states
n ∀i ∈ r.
and the estimated states, respectively. Signals ŷ(t) and ŷ(t − To establish stability and convergence of the presented
h) are estimated system outputs at the sampling time t and FDLO such that accuracy of fault reconstruction can be
t − h, respectively; ŷ(t) ∈ Rp , and ŷ(t − h) ∈ Rp . Fault guaranteed, the following assumption is needed.
reconstruction signal fˆa (t) ∈ Rm , which is updated by both Assumption 4: Assume that kf˜a (t)k∞ ≤ kf , where the
its previous information at t − τ and current output estimation vector f˜a (t) := fa (t) − K1 fa (t − τ ) and kf is a small positive
error. Parameter τ is the learning interval [29]. Diagonal matrix constant.
K1 = diag{σ1 , · · · , σm } with σi ∈ (0, 1], and T , H, K1 , L1i , Using Assumption 4, one obtains
L2i and K2i are gain matrices with appropriate dimensions to r
X
be determined later. ef (t) = K1 ef (t − τ )− µi (z(t))K2i Cex (t)+ f˜a (t). (16)
Assumption 1 implies that there exist gain matrices T and i=1
H such that Remark 3: In the FDLO (8), learning interval τ can be
T E + HC = In . (9) adjusted to guarantee fault-reconstructing accuracy. It could
State estimation errors, output estimation errors, and fault be selected large if a fault is constant or slowly varying;
reconstruction errors can be represented by ex (t) = x(t) − otherwise, it should be chosen small. Specially, in sampled-
x̂(t), ey (t) = y(t) − ŷ(t), and ef (t) = fa (t) − fˆa (t), data control systems, it can be taken as the sampling interval,
respectively. By using (9), the following estimation error or as an integer multiple of the sampling interval. Assumption
dynamics can be obtained as 4 and equation (16) imply that, for each i ∈ m, σi is selected
close to 1; in other words, gain matrix K1 is designed close to
r
X Im , and learning interval τ is selected to be sufficiently small.
ėx (t) = µi (z(t)){(T Ai − L1i C)ex (t) + (T Ahi
Then, kf˜a (t)k∞ can be guaranteed to be sufficiently small
i=1
(10) for both constant and time-varying faults. Therefore, under
− L2i C)ex (t − h) + T Dd(t) + T Fi ef (t)}
Assumption 4, the proposed FDLO-based fault-reconstructing
ey (t) =Cex (t). scheme can be applied to reconstruct time-varying faults,
On the basis of the disturbance-decoupling techniques pro- especially fast-varying faults.
posed in [17] and [33], matrix T is selected such that Remark 4: The fault-constraining condition described in
Assumption 4 in this paper is less restrictive compared with
T D = 0; (11) those in [20] and [35] because the norm-bounded conditions
hence, estimation error dynamics (10) can be simplified as on the derivatives of faults are not required in this paper.
r
Remark 5: Compared with the fuzzy unknown input ob-
X servers proposed in [17] and [33], just two gain matrices T and
ėx (t) = µi (z(t)){(T Ai − L1i C)ex (t)
H in (8a) and (8b) and two equality constraints (9) and (11)
i=1
(12) are involved for disturbance-decoupling purpose. It is therefore
+ (T Ahi − L2i C)ex (t − h) + T Fi ef (t)}
easier to design the FDLO.
ey (t) =Cex (t).
Remark 2: In order to design gain matrices T and H B. Stability Analysis of the FDLO
such that constraints (9) and (11) are simultaneously satisfied, In this subsection, the stability and convergence of the
an augmented matrix equation composing of (9) and (11) is FDLO (8) are proved, and detailed design of observer gain
organized as matrices is discussed.
Theorem 1: Considering the TSDS shown in (2). If there
 
  E D  
T H = In 0 . (13) exist positive definite symmetric matrices P ∈ Rn×n , Q ∈
C 0

0278-0046 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIE.2015.2404784, IEEE Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 4

Rn×n , Q1 ∈ Rm×m and K1 ∈ Rm×m , as well as matrices Further, with the help of (18) and (23), (22) can be simpli-
K2i ∈ Rm×p , Y1i ∈ Rn×p and Y2i ∈ Rn×p such that, ∀i ∈ r, fied as
  r
X
Ξ1 P T Ahi − Y2i C
< 0, (17) V̇ (t) ≤ µi (z(t)){eTx (t)[P (T Ai−L1i C) + (T Ai−L1i C)T
∗ −Q i=1
× P + Q]ex (t)+2eTx (t)P (T Ahi −L2i C)ex (t − h)
(T Fi )T P = α0 K2i C, (18) − eT (t − h)Qex (t − h) + 2α0 eT (t − τ )K T f˜a (t)
x f 1
+ α0 ef (t − τ )K1 K1 ef (t − τ ) + α0 fa (t)f˜a (t)
T T ˜T
and
− 0 eTf (t)Q1 ef (t) − eTf (t − τ )Q1 ef (t − τ )}.
α0 α1 K1T K1 − Q1 ≤ 0 (19) (24)

where Ξ1 = P T Ai + (T Ai )T P − Y1i C − C T Y1iT + Q, For any scalar α > 0, 2X T Y ≤ (1/α)X T X +αY T Y ; thus,
α0 = (1 + 0 )λmax (Q1 ), and α1 = 1 + 1 with 0 ≥ 0 and (24) can be further transformed into
1 > 0, then the presented FDLO (8) can realize that state r
X
eTx (t) eTx (t − h) eTf (t)
 
estimation error and fault reconstruction error are uniformly V̇ (t) ≤ µi (z(t))
ultimately bounded. Observer gain matrices can be accordingly i=1
 
determined by L1i = P −1 Y1i and L2i = P −1 Y2i , ∀i ∈ r. Ξ1 P T Ahi − Y2i C 0
Proof: Consider the following Lyapunov-Krasovskii func- × ∗ −Q 0 
tional candidate: ∗ ∗ −0 Q1 (25)
 
ex (t)
Z t α0 α1 ˜T
T
V (t) =ex (t)P ex (t) + eTx (s)Qex (s)ds ×  ex (t − h)  + f (t)f˜a (t)
α1 − 1 a
t−h ef (t)
Z t (20)
+ T
ef (s)Q1 ef (s)ds +eTf (t − τ )[α0 α1 K1T K1 − Q1 ]ef (t − τ )
t−τ
where Y1i = P L1i , Y2i = P L2i and α1 = 1 + 1 with 1 > 0.
According to (17), we have
The derivative of the Laypunov candidate with respect to  
time can be derived as Ξ1 P T Ahi − Y2i C 0
r
Ξ= ∗ −Q 0  < 0. (26)
∗ ∗ −0 Q1
X
V̇ (t) ≤ µi (z(t)){eTx (t)[P (T Ai − L1i C) + (T Ai
i=1 If conditions (19) and (26) hold, it follows from (25) that
− L1i C)T P + Q]ex (t) + 2eTx (t)P T Fi ef (t)
(21) V̇ (t) ≤ −ξkζ(t)k2 + δ, (27)
+2eTx (t)P (T Ahi −L2i C)ex (t − h)−eTx (t − h)
× Qex (t − h)−0 eTf (t)Q1 ef (t)+α0 eTf (t)ef (t) where ξ = λmin (−Ξ), ζ(t) = [eTx (t) eTx (t − h) eTf (t)]T and
δ = αα10−1
α1 2
kf .
− eTf (t − τ )Q1 ef (t − τ )}
It follows that V̇ (t) < 0 for ξkζ(t)k2 > δ , which means
that the trajectory of ζ(t) that is outside of a small set Ψ =
where α0 = (1 + 0 )λmax (Q1 ) with 0 ≥ 0.
{ζ(t)|kζ(t)k2 ≤ δ/ξ} will converge to this set Ψ according
to the Lyapunov stability theory; thus, the ζ(t) is uniformly
Substituting (16) into (21) leads to
ultimately bounded.
Xr It can be noted that the condition (17) is an LMI that
V̇ (t) ≤ µi (z(t)){eTx (t)[P (T Ai −L1i C) + (T Ai −L1i C)T P can be conveniently solved using standard LMI tools while
i=1 the condition (18) is a matrix equality that is difficult to be
+ Q]ex (t)+2eTx (t)P (T Ahi − L2i C)ex (t − h)+2eTx (t) directly solved using the MATLAB LMI toolbox [23], [28],
[29]. To tackle this difficult problem, we rewrite linear equality
× [P T Fi − α0 (K2i C)T ]K1 ef (t − τ ) − eTx (t)[2P T Fi
condition (18) as
−α0 (K2i C)T ]K2i Cex (t)+2eTx (t)[P T Fi−α0 (K2i C)T ]
((T Fi )T P − α0 K2i C)T ((T Fi )T P − α0 K2i C) < γIn (28)
× f˜a (t)+α0 eTf (t − τ )K1T K1 ef (t − τ )+α0 f˜aT (t)f˜a (t)
− eTx (t − h)Qex (t − h) + 2α0 eTf (t − τ )K1T f˜a (t) where γ is a positive scalar. By using Schur complement
T T lemma [36], (28) is equivalent to
− 0 ef (t)Q1 ef (t) − ef (t − τ )Q1 ef (t − τ )}.
−γIm (T Fi )T P − α0 K2i C
 
(22)
< 0. (29)
∗ −In
If the condition (18) holds, one obtains The design problem of (17) and (18) is now converted into
the following minimization problem:
− eTx (t)[2P T Fi − α0 (K2i C)T ]K2i Cex (t)
(23)
= − α0 eTx (t)(K2i C)T K2i Cex (t) ≤ 0. Minimize γ, subject to (17) and (29). (30)

0278-0046 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIE.2015.2404784, IEEE Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 5

The above minimization problem can be solved by using the IV. D ESIGN OF FAULT- TOLERANT CONTROLLERS BASED
Solvers minx in the LMI toolbox of MATLAB. Alternatively, ON FUZZY DESCRIPTOR LEARNING OBSERVERS
a sufficient small positive scalar γ can be selected such that In this section, using reconstructed fault information pro-
matrices P and K2i can be computed to make (T Fi )T P vided by the FDLO, we will construct a reconfigurable F2 TC
approximate to α0 K2i C with satisfactory precision by solving scheme such that closed-loop TSDS (2) is stable. Before
(17) and (29). presenting the main results, the following assumptions are
The results proposed in Theorem 1 can be readily extended made.
to the case of constant actuator faults, as claimed by the Assumption 5: B1 = · · · = Br = B.
following corollary. Assumption 6: rank(B Fi ) = rank(B), ∀i ∈ r.
Corollary 1: Assume that K1 = Im . If there exist positive Remark 8: Assumption 5 is a common assumption for
definite symmetric matrices P ∈ Rn×n and Q ∈ Rn×n , and system (2) and similar assumptions can be found in [16], [20]
matrices K2i ∈ Rm×p , Y1i ∈ Rn×p and Y2i ∈ Rn×p to and the references therein. Assumption 6 implies there exists
satisfy conditions (17) and (18) with α0 = 1, ∀i ∈ r, then a nonzero matrix F̄i ∈ Rm×m such that Fi = B F̄i , ∀i ∈ r
the presented FDLO (8) guarantees asymptotic reconstruction [35], [39]. Further, it is obtained that
of constant actuator faults. Moreover, observer gain matrices
can be computed as L1i = P −1 Y1i and L2i = P −1 Y2i , ∀i ∈ r. (In − BB † )Fi = (In − BB † )B F̄i = 0, ∀i ∈ r. (33)
Proof: The proof of Corollary 1 is similar to that of In order to compensate for the fault effect on the TSDS
Theorem 1. The detailed proof is thus omitted. such that the faulty system is as close as possible to the pre-
Remark 6: If K1 = Im , then f˜a (t) ≡ 0, kf = 0 and δ = 0 fault system, a state-feedback-based reconfigurable F2 TC is
when constant actuator faults occur in the considered system constructed as follows
(2). Thus, via Corollary 1, the FDLO can achieve asymptotic
reconstruction of constant actuator faults without requirement u(t) = un (t) + uf (t) (34)
for Assumption 4. In other words, the FDLO can guarantee where un is the normal state-feedback controller in fault-free
unbiased reconstruction of constant actuator faults. condition, and uf is additional controller which is used to
Remark 7: It should be noted that necessary conditions for accommodate actuator faults. Specifically, controllers un and
the solvability of the proposed FDLO design using Theorem uf are expressed as follow
r
1 and Corollary 1 must be satisfied. In reference to [37], [38], X
the following necessary conditions for (17) and (18) to be un (t) = − µi (z(t))K3i x̂(t), (35)
i=1
solvable ∀i ∈ r can be provided in terms of invariant zeros as r
X
follows and uf (t) = −B †
µi (z(t))Fi fˆa (t) (36)
1) rank(CT Fi ) = rank(T Fi ) = m, ∀i ∈ r. i=1

2) No invariant zeros of (T Ai , T Fi , C) lie in C+ , ∀i ∈ r. where K3i ∈ Rm×n , ∀i ∈ r, representing the controller gain
The above two conditions mean that the pair (T Ai , C) is matrices to be determined later.
detectable, ∀i ∈ r; that is, Therefore, the closed-loop TSDS without external distur-
bances becomes
r
 
sIn − T Ai
= n, s ∈ C+ , ∀i ∈ r.
X
rank (31) E ẋ(t) = µi (z(t)){Ai x(t)−BK3i x̂(t)+Ahi x(t − h)
C
i=1
It can also be understood that (31) is a necessary condition + (In − BB † )Fi fˆa (t) − Fi fˆa (t) + Fi fa (t)}
for (17) and (18) to hold ∀i ∈ r. In what follows, we will r
(37)
X
further discuss necessary conditions for (31) to hold such that = µi (z(t)){(Ai − BK3i )x(t)
the applicable scope of the proposed FDLO can be clarified. i=1
Based on (31), the following derivation can be obtained: + Ahi x(t − h) + ∆i (t)}
      T
sIn − T Ai s(T E + HC) − T Ai where ∆i (t) = BK3i Fi [ex (t) ef (t)] .
rank = rank
C C To design controller gain matrices K3i , ∀i ∈ r, such that

sT E − T Ai
 
T 0

sE − Ai
 the closed-loop TSDS with faults is still stable, the following
=rank = rank (32) theorem is provided.
C 0 Ip C
   Theorem 2: Suppose that Theorem 1 holds. If there exist
sE − Ai
≤min rank(T ) + p, rank positive definite symmetric matrices X ∈ Rn×n and Q2 ∈
C
Rn×n , a matrix Mi ∈ Rm×n , and a positive scalar ηc such
Therefore, if the condition (31) holds, the triple matrix that, ∀i ∈ r, the following relations hold,
(E, Ai , C) should be R-detectable, ∀i ∈ r. Meanwhile, it is X T E T = EX ≥ 0, (38)
necessary that rank(T ) ≥ n − p. In addition, according to
(11) and above condition 1), the following conditions hold: and  
Θ11 Ahi X
rank(T ) ≤ n − ν, rank(T ) ≥ m, and p ≥ m. Therefore,  ∗ ηc In − Q2 0 <0 (39)
rank[D Fi ] ≤ n, ∀i ∈ r. ∗ ∗ −(ηc In + Q2 ) −1

0278-0046 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIE.2015.2404784, IEEE Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 6

where Θ11 = Ai X + XATi − BMi − MiT B T , then the F2 TC where Ω11 = Ai X + XATi − BK3i X − (K3i X)T B T +
designed in (34) can ensure that the state vector of the overall X(ηc In + Q2 )X. By using Schur complement lemma, (47)
system (37) is uniformly ultimately bounded. Controller gain is equivalent to (39).
matrices can therefore be determined by K3i = Mi X −1 , ∀i ∈ Based on Corollary 1 and Theorem 2, the following corol-
r. lary can be readily obtained for the design of F2 TC against
Proof: Consider the following Lyapunov-Krasovskii func- constant actuator faults.
tional candidate: Corollary 2: Suppose that Corollary 1 holds. If there exist
Z t positive definite symmetric matrices X ∈ Rn×n and Q2 ∈
W (t) = xT (t)E T P1 x(t)+ xT (s)Q2 x(s)ds+η0 V (t) (40) Rn×n , a matrix Mi ∈ Rm×n , and a positive scalar ηc such
t−h
that, ∀i ∈ r, conditions (38) and (39) hold; then the F2 TC
where η0 is a positive scalar. (34) can ensure that the overall system (37) is asymptotically
Define stable. Moreover, controller gain matrices can be computed as
E T P1 = P1 E ≥ 0. (41) K3i = Mi X −1 , ∀i ∈ r.
Proof: The proof of Corollary 2 is similar to that of
Substituting (27) and (41) into the derivative of the Lya- Theorem 2; it is therefore omitted.
punov candidate leads to
Remark 9: Based on the separation property [30], [31],
Xr the FDLO and the F2 TC are independently designed, and the
Ẇ (t) ≤ µi (z(t)){xT (t)[P1 (Ai −BK3i )+(Ai − BK3i )T P1 overall system (37) is stable. Via such a design methodology,
i=1 the design problems of the FDLO and of the F2 TC are
+ Q2 ]x(t) + 2xT (t)P1 Ahi x(t − h) + 2xT (t)P1 ∆i (t) respectively formulated into a series of LMIs that can be
− xT (t − h)Q2 x(t − h) − η0 ξkζ(t)k2 + η0 δ}. conveniently solved using LMI optimization techniques. In
(42) addition, it can be noted that external disturbance d(t) is not
included in the proof of Theorem 2 and Corollary 2. They are
It is assumed that the TSDS under the nominal state- omitted because they have little influence on our main results;
feedback controller (35) is asymptotically stable in the absence if included, they will increase the complexity of proof.
of actuator faults and external disturbances. It implies that Remark 10: This work is motivated by our previous results
there exists a positive scalar ηc such that, ∀i ∈ r, on [28] where an LO-based fault-reconstruction approach is
xT (t)[P1 (Ai −BK3i )+ (Ai − BK3i )T P1 + Q2 ]x(t) proposed in Linear Parameter-Varying (LPV) systems. The
major differences between the two studies are summarized in
+ 2xT (t)P1 Ahi x(t − h) − xT (t − h)Q2 x(t − h) (43)
2
the following three aspects. First, this paper is focused on
< − ηc kx̄(t)k TSDS’s with additive actuator faults and time delays while
where x̄(t) = [xT (t) xT (t − h)]T . [28] is targeted at polytopic LPV systems subject to loss
Substituting (43) into (42) gives of actuator effectiveness. Second, this paper presents a new
FDLO that is designed to completely decouple external distur-
Ẇ (t) < −ηc kx(t)k2 +ηf kx(t)kkē(t)k−η0 ξkē(t)k2 +η0 δ (44) bances from fault reconstruction using disturbance-decoupling
techniques. In [28], a robust LO is designed to attenuate the
where ηf = k[2P1 BK3i 2P1 Fi ]k and ē(t) = [eTx (t) eTf (t)].
effect of external disturbances using H∞ technique. Third, this
paper addresses the problem of FDLO-based reconfigurable
If η0 ≥ ηf2 /ηc ξ, then it follows from (44) that FTC while FTC is not considered in [28].
ηc η0 ξ
Ẇ (t) < − kx(t)k2 − kē(t)k2 + η0 δ, (45)
2 2 V. S IMULATION STUDIES
2
therefore, the proposed F TC can ensure that the overall In this section, a truck trailer system [40] is simulated to
system (37) is stable no matter if actuator faults occur or not. demonstrate the effectiveness of the proposed FDLO-based
approaches. The dynamic model of the considered truck trailer
To guarantee stability of the closed-loop TSDS (37), the system is described by
above analysis indicates that gain matrices K3i , ∀i ∈ r,
υ t̄ υ t̄ υ t̄
should be designed to make conditions (41) and (43) hold ẋ1 (t) =− a x1 (t)−(1−a) x1 (t − h)+ u(t)
simultaneously. If X = P1−1 , then (41) is equivalent to (38). Ll t0 Ll t0 lt0
In addition, the inequality (43) is also equivalent to υ t̄ υ t̄
ẋ2 (t) =a x1 (t) + (1 − a) x1 (t − h) (48)
  Ll t0 Ll t0
Π11 P1 Ahi υ t̄ υ t̄ υ t̄
<0 (46) ẋ3 (t) = sin[x2 (t)+a x1 (t)+(1−a) x1 (t − h)]
∗ ηc In − Q2 t0 2Ll 2Ll
where Π11 = P1 (Ai −BK3i )+(Ai −BK3i )T P1 +Q2 +ηc In . where x1 (t) is the angle difference between the truck and
If post- and pre-multiplying (46) by diag{X, In }, then it is trailer, x2 (t) is the angle of the trailer, x3 (t) is the vertical
obtained that position of rear end of the trailer, u(t) is the steering angle, a
is the retarded coefficient, υ is the constant speed of backing
 
Ω11 Ahi
<0 (47) up, l is the length of the truck, and Ll is the length of the
∗ ηc In − Q2

0278-0046 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIE.2015.2404784, IEEE Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 7

trailer. The truck-trailer parameters are taken as: a = 0.7, obtained by solving (17) and (29) using the MATLAB LMI
l = 2.8, Ll = 5.5, υ = −1.0, t̄ = 2.0, t0 = 0.5, and h = 0.5. toolbox
To obtain a T-S descriptor representation, we introduce the 
1.8483 −0.2903 0 0


following variable: 
 −0.2618 1.0426 0 0


L11 = ,
 

x4 (t) = x2 (t) − a Lυl tt̄0 x1 (t) − (1 − a) Lυl tt̄0 x1 (t − h).


 0 0 0.9945 0 

0 0 0 0.9945
In reference to [35] and [40], the following fuzzy rules can be 
1.8483 −0.2903 0 0


established: 
 −0.2619 1.0426 0 0


υ t̄ υ t̄ L12 = ,
 
Rule 1: IF z(t) = x2 (t) + a 2L l
x1 (t) + (1 − a) 2L l
x1 (t − h) 
 0 0 0.9945 0 

is about 0, THEN 0 0 0 0.9945
(
E ẋ(t) =A1 x(t) + Ah1 x(t − h) + B1 u(t) + Dd(t)
   
0.1089 0 0 0 0.1090 0 0 0
(49)
−0.1089  −0.1089 0 0 0
   
y(t) =Cx(t)  0 0 0  
L21=  , L22= ,
   

 0 0 0 0  

 0 0 0 0 

υ t̄ υ t̄
Rule 2: IF z(t) = x2 (t) + a 2L l
x1 (t) + (1 − a) 2L l
x1 (t − h) 0 0 0 0 0 0 0 0
is about π or −π, THEN  
( K21 = 103  −1.1263 −0.2067 0 0 ,
E ẋ(t) =A2 x(t) + Ah2 x(t − h) + B2 u(t) + Dd(t) K22 = 103 −1.1263 −0.2067 0 0 .
(50)
y(t) =Cx(t) In the simulation, the learning interval and the sampling
where time are set as 0.005s and 0.001s, respectively. The initial
condition φ(t) and the initial state x(0) are set as sinπt, t ∈
−a Lυl tt̄0
 

1 0 0 0


0 0 0  [−0.5, 0], and [0.5π 0.75π − 5.0 1.77]. A constant fault and
υ t̄
a Ll t0 0 0 0 
   
 0 1 0  0  a time-varying fault are respectively created as
E= , A1 = ,
   
0 0 1 0 υ 2 t̄2 υ t̄



 −a 
2Ll t0
 t0 0 0 


2 t ≥ 8s

2cosπt + 1 t ≥ 8s
0 0 0 0

υ t̄
−a Ll t0 1 0 −1

fa (t) = , and fa (t) =
0 other 0 other.
−(1 − a) Lυl tt̄0 0 0 0 
 
 υ t̄ 

 lt0  Simulation results on the FDLO-based fault reconstruction
 (1 − a) υ t̄ 0 0 0 
 
 0  are illustrated in Figs. 1-6. Figs. 1-4 show original system
 
Ll t0
Ah1 =

 (1 − a) υ t̄
2 2

 , B 1 = ,
0 0 0   0  states and their estimates under a fault-free condition while
  
 2L l t 0
−(1 − a) Lυl tt̄0 0 0 0 0
 
Figs. 5 and 6 illustrate the reconstructed signals of a constant

υ t̄
−a Ll t0 0 0 0 

 υ t̄  fault and a time-varying fault, respectively. Figs. 1-4 reveal

υ t̄  lt0  that the designed FDLO can accurately track original system
a Ll t0 0 0 0 
 
 
0  states. As comparison, fault-reconstructing signals using the
A2 =

ϕυ 2 t̄2 ϕυ t̄

 , 2 
B = 
 0 
,


 −a 2L t
l 0 t0
0 0 
   conventional adaptive fault estimation (CAFE) algorithm and
0
 
−a Lυl tt̄0 1 0 −1 the fast adaptive fault estimation (FAFE) algorithm proposed
−(1 − a) Lυl tt̄0 0 0 0 
 


0

in [35] are depicted in Figs. 5 and 6. It is observed that,
 (1 − a) υ t̄ 0 0 0   0  compared with the CAFE, the FDLO designed using Theorem
   
Ll t0
Ah2 = , D=  υ t̄  ,
   
ϕυ 2 t̄2 1 not only provides better rapidity of fault reconstruction,
 (1 − a) 2L t 0 0 0
   
  t0 
 l 0
υ t̄

0 but also achieves more accurate reconstruction of the time-
−(1 − a) Ll t0 0 0 0
varying actuator fault. It can also be noted that both the FAFE
C = I4 , ϕ = 10t π , and d(t) = sin(z(t)) − z(t).
0
and the FDLO have the same rapidity and accuracy for fault
Here, actuator faults are considered. It is assumed that reconstruction. However, integrators are employed in FAFE
F1 = B1 and F2 = B2 . Membership functions µ1 (z(t)) and for updating mechanisms. The problems are that derivatives
µ2 (z(t)) are defined as follows of measurement outputs are usually not available and the
derivatives themselves sometimes may not exist at some time
 µ1 (z(t))
 =  instants [24]. This will limit the applicability of the FAFE.
1 1
1− 1+exp(−3(z(t)−0.5π)) · 1+exp(−3(z(t)+0.5π)) , The proposed FDLO-based approach circumvents the above
µ2 (z(t)) = 1 − µ1 (z(t)). constraints because the algebraic learning algorithm is used
for updating learning laws. Another benefit is the calculation
FDLO design: First, gain matrices T and H can be obtained
is cheap because of the algebraic learning algorithm in the
by solving (15)
   
FDLO. All of these make it possible for the FDLO to be
0.5 0 0 0 0.5 0 0 0 easily implemented in practice.
FDLO-based F2 TC design: Choosing Q2 = 20I4 , ηc = 1,
   
 0 0.5 0 0   0 0.5 0 0 
T =  , H= .
   

 0 0 0 0 


 0 0 1.0 0 
 and solving (38) and (39) in Theorem 2 using MATLAB LMI
0 0 0 0 0 0 0 1.0 toolbox, the following gain matrices are obtained
Further, choosing 0 = 0, α1 = 1.02, Q1 = 5, and K1
 
K31 = 103  −0.3138 2.5470 −0.6193 −0.1231 ,
is selected as 0.99 to satisfy (19) in Theorem 1. Choosing K32 = 103 −0.3142 2.7518 −0.7336 −0.1230 .
γ = 10−6 and Q = 5I4 , the following gain matrices can be

0278-0046 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIE.2015.2404784, IEEE Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 8

4
10
System state x1

Actuator fault fa(t) and its reconstruction

Estimated state x1
System state x1(t) and its estimate

2
5

0
0

−2 Actual fa(t)
−5
FDLO
CAFE
FAFE
−4
−10 0 5 10 15 20
0 2 4 6 8 10
t(seconds)
t(seconds)
Fig. 5. Reconstructed signals of a constant fault using the CAFE algorithm,
Fig. 1. System state x1 (t) and its estimate under a fault-free condition. the FAFE algorithm, and the FDLO via Theorem 1, respectively.

4
3

Actuator fault fa(t) and its reconstruction

System state x2
Estimated state x2
System state x2(t) and its estimate

2
1.5

0
0

−2 Actual fa(t)
−1.5
FDLO
CAFE
FAFE
−4
0 5 10 15 20
−3
0 2 4 6 8 10 t(seconds)
t(seconds) Fig. 6. Reconstructed signals of a time-varying fault using the CAFE
Fig. 2. System state x2 (t) and its estimate under a fault-free condition. algorithm, the FAFE algorithm, and the FDLO via Theorem 1, respectively.

1
in fault−free condition
4
with time−varying fault
System state x
3 with FDLO−based FTC
Estimated state x3
System output response y1(t)
System state x3(t) and its estimate

0.5

−4 −0.5

−1
5 10 15 20
−8 t(seconds)
0 2 4 6 8 10
t(seconds)
Fig. 7. System output response y1 (t) with the F2 TC using fault information
Fig. 3. System state x3 (t) and its estimate under a fault-free condition. shown in Fig. 6.

0.2
10 in fault−free condition
System state x4 with time−varying fault
Estimated state x with FDLO−based FTC
System state x4(t) and its estimate

System output response y2(t)

0.1
5

0
0

−0.1
5 10 15 20
−5
0 2 4 6 8 10 t(seconds)
t(seconds)
Fig. 8. System output response y2 (t) with the F2 TC using fault information
Fig. 4. System state x4 (t) and its estimate under a fault-free condition. shown in Fig. 6.

0278-0046 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIE.2015.2404784, IEEE Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 9

0.5
R EFERENCES
in fault−free condition
with time−varying fault
with FDLO−based FTC
[1] M. Blanke, M. Kinnaert, J. Lunze, and M. Staroswiecki, Diagnosis and
System output response y3(t) fault-tolerant control. Springer-Verlag, 2006.
0
[2] S. X. Ding, Model-based fault diagnosis techniques-Design schemes,
algorithms, and tools. Springer-Verlag, 2008.
[3] S. Yin, S. X. Ding, X. C. Xie, and H. Luo, “A review on basic data-
driven approaches for industrial process monitoring,” IEEE Trans. Ind.
−0.5
Electron., vol. 61, no. 11, pp. 6414–6428, 2014.
[4] Z. W. Gao and S. X. Ding, “Fault estimation and fault-tolerant control
for descriptor systems via proportional, multiple-integral and derivative
observer design,” IET Control Theory & Applications, vol. 1, no. 5, pp.
−1
1208–1218, 2007.
5 10
t(seconds)
15 20 [5] ——, “Actuator fault robust estimation and fault-tolerant control for a
class of nonlinear descriptor systems,” Automatica, vol. 43, no. 5, pp.
Fig. 9. System output response y3 (t) with the F2 TC using fault information 912–920, 2007.
shown in Fig. 6. [6] D. Ichalal, B. Marx, J. Ragot, and D. Maquin, “New fault tolerant control
0.5
strategies for nonlinear Takagi-Sugeno systems,” International Journal
in fault−free condition of Applied Mathematics and Computer Science, vol. 22, no. 1, pp. 197–
with a time−varying fault 210, 2012.
with the FDLO−based FTC
[7] A. Haghani, T. Jeinsch, and S. X. Ding, “Quality-related fault detection
System output response y4(t)

0.25
in industrial multimode dynamic processes,” IEEE Trans. Ind. Electron.,
vol. 61, no. 11, pp. 6446–6453, 2014.
0
[8] S. Yin, H. Luo, and S. X. Ding, “Real-time implementation of fault-
tolerant control systems with performance optimization,” IEEE Trans.
Ind. Electron., vol. 61, no. 5, pp. 2402–2411, 2014.
[9] G. R. Duan, Analysis and Design of Descriptor Linear Systems.
−0.25
Springer, New York, 2010.
[10] H. Moddi and M. Farrokhi, “Robust observer design for sugeno systmes
with incremental quadratic nonlinearity in the consequent,” Journal of
−0.5
5 10 15 20 Applied Mathematics and Computer Science, vol. 23, no. 4, pp. 711–
t(seconds)
723, 2013.
Fig. 10. System output response y4 (t) with the F2 TC using fault information [11] Y. H. Chang, W. S. Chan, and C. W. Chang, “T-S fuzzy model-based
shown in Fig. 6. adative dynamic surface control for ball and beam system,” IEEE Trans.
Ind. Electron., vol. 60, no. 6, pp. 2251–2263, 2013.
Based on the real-time information on the reconstructed [12] X. H. Chang and G. H. Yang, “Nonfragile H∞ filter design for T-S
time-varying fault as shown in Fig. 6, system output re- fuzzy systems in standard form,” IEEE Trans. Ind. Electron., vol. 61,
no. 7, pp. 3448–3458, 2014.
sponses with the F2 TC are demonstrated in Figs. 7-10. It [13] C. Lin, Q. G. Wang, and T. H. Lee, “Stability and stabilization of a
can be observed that the FDLO-based FTC strategy can class of fuzzy time-delay descriptor systems,” IEEE Trans. Fuzzy Syst.,
effectively accommodate the effect of actuator faults on sys- vol. 14, no. 4, pp. 542–551, 2006.
[14] P. Liu, W. T. Yang, and C. E. Yang, “Robust observer-based output
tem performance. All of simulation results claim that the feedback control for fuzzy descriptor systems,” Expert Systems with
proposed FDLO-based fault reconstruction and FTC schemes Applications, vol. 40, no. 11, pp. 4503–4510, 2013.
are effective for TSDS’s subject to time delays and external [15] H. Gassara, A. EI Hajjaji, M. Kchaou, and M. Chaahane, “Observer
based (Q, V, R)-α-dissipative control for TS fuzzy descriptor systems
disturbances. with time delay,” Journal of the Franklin Institute, vol. 351, no. 1, pp.
187–206, 2014.
VI. C ONCLUSIONS [16] C. S. Han, G. J. Zhang, L. G. Wu, and Q. S. Zeng, “Sliding mode
control of T-S fuzzy descriptor systems with time-delay,” Journal of the
In this article, FDLO-based fault reconstruction and FTC Franklin Institute, vol. 349, no. 4, pp. 1430–1444, 2012.
for TSDS’s subjects to time delays and external disturbances [17] B. Marx, D. Koenig, and J. Ragot, “Design of observers for Takagi-
Sugeno descriptor systems with unknown inputs and application to fault
have been investigated. A new FDLO has been constructed to diagnosis,” IET Control Theory & Applications, vol. 1, no. 5, pp. 1487–
simultaneously reconstruct system states and actuator faults 1495, 2007.
based on disturbance-decoupling techniques. Utilizing real- [18] Y. H. Yuan and G. M. Zhang, “Observer-based reliable controller
designs for T-S fuzzy descriptor systems with time-delay,” in Seventh
time information on reconstructed faults, a state-feedback- International Conference on Fuzzy Systems and Knowledge Discovery,
based F2 TC is designed to accommodate the effect of actuator Yantai, China, Aug 2010, pp. 248–253.
faults on system performance. An advantage of the proposed [19] B. Y. Zhu, Q. D. Zhang, Q. L. Zhang, and S. Y. Xing, “Dissipative fault-
tolerant control for T-S fuzzy descriptor systems,” in Chinese Control
approach is that the design problem of the FDLO and the Conference, Xi’an, China, July 2013, pp. 3375–3380.
F2 TC can be reformulated into the problem of LMIs such [20] H. Hamdi, M. Rodrigues, C. Mechmeche, and N. Benhadjbraiek, “Ob-
that they can be conveniently solved using the LMI toolbox server based fault tolerant control for Takagi-Sugeno nonlinear descrip-
tor systems,” in International Conference on Control, Engineering and
in MATLAB. The practical example has clearly verified the Information Technology, Sousse,Tunisia, 2013, pp. 52–57.
effectiveness of the proposed FDLO-based method for fault [21] Z. W. Gao, X. Y. Shi, and S. X. Ding, “Fuzzy state/disturbance observer
reconstruction and FTC. It should be pointed out that the design for T-S fuzzy systems with application to sensor fault estimation,”
coupling problem resulting from the design of the F2 TC is IEEE Trans. Syst., Man, Cybern. B, Cybern., vol. 38, no. 3, pp. 875 –
880, 2008.
inevitable due to the independent design of the FDLO and the [22] M. Chadli, A. Abdo, and S. X. Ding, “H− /H∞ fault detection filter
F2 TC based on the separation property. It may have significant design for discrete-time Takagi-Sugeno fuzzy system,” Automatica,
impact on fault-tolerant performance, especially when fast- vol. 49, no. 7, pp. 1996–2005, 2013.
[23] M. Liu, X. Cao, and P. Shi, “Fuzzy-model-based fault-tolerant design for
varying faults occur. To tackle this difficult issue, an integrated nonlinear stochastic systems against simultaneous sensor and actuator
design will be our main focus in the near-future work. faults,” IEEE Trans. Fuzzy Syst., vol. 21, no. 5, pp. 789–799, 2013.

0278-0046 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TIE.2015.2404784, IEEE Transactions on Industrial Electronics
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 10

[24] W. Chen and F. Chowdhury, “Simultaneous identification of tim-varying time-delay systems via linear Takagi-Sugeno fuzzy models,” Fuzzy Sets
parameters and estimation of system states using iterative learning and Systems, vol. 124, no. 2, pp. 213–229, 2001.
observers,” International Journal of Systems Science, vol. 38, no. 1,
pp. 39–45, 2007.
[25] Z. F. Gao, B. Jiang, P. Shi, and F. Y. Xu, “Fault-tolerant control for a
near space vehicle with a stuck actuator fault based on a Takagi-Sugeno
fuzzy model,” Proceedings of the Institution of Mechanical Engineers, Qingxian Jia received the B.E. degree in Au-
Part I: Journal of Systems and Control Engineering, vol. 224, no. 5, pp. tomation from Nanjing University of Technology,
587–598, 2010. Nanjing, China, and the M.S. degree in Control
[26] L. N. Yao, J. Qin, H. Wang, and B. Jiang, “Design of new fault diagnosis Science and Engineering from Harbin Institute of
and fault tolerant control scheme for non-Gaussian singular stochastic Technology, China, in 2008, 2010, respectively. Now
distruibution systems,” Automatica, vol. 48, no. 9, pp. 2305–2313, 2012. he is pursuing a Ph.D degree in Control Science
[27] W. Chen, W. T. Chen, M. Saif, H. Wu, and M. Li, “Simultaneous and Engineering, Harbin Institute of Technology.
fault isolation and estimation of Lithium-Ion batteries via synthesized His research interests include fault diagnosis, fault-
design of Luenberger and learning observers,” IEEE Trans. Control Syst. tolerant control and their applications.
Technol., vol. 22, no. 1, pp. 290–298, 2014.
[28] Q. X. Jia, W. Chen, Y. C. Zhang, and X. Q. Chen, “Robust fault
Wen Chen (M’09) received the Ph.D. degree from
reconstruction via learning observers in linear parameter-varying sys-
Simon Fraser University, Burnaby, BC, Canada,
tems subject to loss of actuator effectiveness,” IET Control Theory &
in 2004. From 2005 to 2007, he was a Postdoc-
Applications, vol. 8, no. 1, pp. 42–50, 2014.
toral Researcher with the University of Louisiana,
[29] Q. X. Jia, W. Chen, Y. C. Zhang, and H. Y. Li, “Fault reconstruction
Lafayette. After that period, he worked in industrial
for continuous-time systems via learning observers,” Asian Journal of
companies as a Control Systems Engineer. He joined
Control(to be published), 2014.
the Division of Engineering Technology,Wayne State
[30] H. H. Choi, “LMI-based nonlinear fuzzy observer-controller design for University, Detroit, MI, in 2009. His teaching and
uncertain MIMO nonlinear systems,” IEEE Trans. Fuzzy Syst., vol. 15, research interests are control systems, alternative
no. 5, pp. 956–971, 2007. energy storage, and fault diagnosis of industrial
[31] A. A. Chunodhar and M. R. Akella, “Switching angular velocity ob- systems.
server for rigid-body attitude stabilization and tracking control,” Journal
of Guidance, Control, and Dynamics, vol. 37, no. 3, pp. 869–878, 2014.
[32] D. Koenig and S. Mammar, “Design of proportional-integral observer
for unknown input descriptor systems,” IEEE Trans. Autom. Control,
vol. 47, no. 12, pp. 2057–2062, 2002. Yingchun Zhang received the B.E. degree in Au-
[33] M. Chadli and H. R. Karimi, “Robust observer design for unknown tomation from Tsinghua University, Beijing, China,
inputs Takagi-Sugeno models,” IEEE Trans. Fuzzy Syst., vol. 21, no. 1, and the M.S. degree in Control Science and Engi-
pp. 158–164, 2013. neering from Harbin Institute of Technology, China,
[34] D. Koenig, “Observer design for unknown input nonlinear descriptor in 1984, 1987, respectively. He is currently a
systems via convex optimization,” IEEE Trans. Autom. Control, vol. 51, full professor in Harbin Institute of Technology
no. 6, pp. 1047–1052, 2006. and a chief engineering of Shenzhen Aerospace
[35] K. Zhang, B. Jiang, and P. Shi, “A new approach to observer-based Dongfanghong HIT Satellite Company Ltd. His
fault-tolerant controller design for Takagi-Sugeno fuzzy systems with research interests include satellite fault diagnosis,
state delay,” Circuits System and Signal Processing, vol. 28, no. 5, pp. fault-tolerant control and micro-satellite design.
679–697, 2009.
[36] S. Boyd, L. E. Ghaoui, E. Feron, and V. Balakrishnan, Linear Matrix
Inequalities in Systems and Control Theory. Philadelphia, USA: SIAM,
1994.
[37] C. Edwards, “A comparison of sliding mode and unknown input ob-
servers for fault reconstruction,” in Proceedings of 43rd IEEE Confer- Huayi Li received the Ph. D degree in Control
ence on Decision and Control, Atlantis, Bahamas, Dec 2004, pp. 5279– Science and Engineering from Harbin Institute of
5284. Technology, China, in 2008. Currently, he is an
[38] C. Edwards, X. G. Yan, and S. K. Spurgeon, “On the solvability of the associate professor in Spacecraft Design in Harbin
constrained Lypunov problem,” IEEE Trans. Autom. Control, vol. 52, Institute of Technology, Harbin, China. His research
no. 10, pp. 1982–1987, 2007. interests include fault monitoring, satellite formation
[39] S. M. Tabatabaeipour and T. Bak, “Robust observer-based fault esti- and satellite electronic testing.
mation and accommodation of discrete-time piecewise linear systems,”
Journal of the Franklin Institute, vol. 351, no. 1, pp. 277–295, 2014.
[40] Y. Y. Cao and P. M. Frank, “Stability analysis and synthesis of nonlinear

0278-0046 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.