International Journal of Systems Science

ISSN: 0020-7721 (Print) 1464-5319 (Online) Journal homepage: http://www.tandfonline.com/loi/tsys20

Robust fast adaptive fault estimation and tolerant

control for T-S fuzzy systems with interval time-
varying delay

Hui Li, Fuqiang You, Fuli Wang & Shouping Guan

To cite this article: Hui Li, Fuqiang You, Fuli Wang & Shouping Guan (2017): Robust fast
adaptive fault estimation and tolerant control for T-S fuzzy systems with interval time-varying delay,
International Journal of Systems Science, DOI: 10.1080/00207721.2017.1282057

To link to this article: http://dx.doi.org/10.1080/00207721.2017.1282057

Published online: 10 Feb 2017.

Submit your article to this journal

Article views: 3

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=tsys20

Download by: [The UC San Diego Library] Date: 14 February 2017, At: 18:57
INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE, 
http://dx.doi.org/./..

Robust fast adaptive fault estimation and tolerant control for T-S fuzzy systems with
interval time-varying delay
Hui Li , Fuqiang You, Fuli Wang and Shouping Guan
Control Theory and Control Engineering, School of Information Science and Engineering, Northeastern University, Shenyang, China

ABSTRACT ARTICLE HISTORY

This paper studies the problem of fault estimation (FE) and the active fault tolerant control (FTC) for Takagi– Received  August 
Sugeno (T-S) fuzzy systems with interval time-varying delay and norm-bounded external disturbance. Based Accepted  January 
on the fast adaptive fault estimation (FAFE) algorithm, our attention focuses on designing an adaptive
KEYWORDS
observer-based controller to guarantee the filtering error system to be asymptotically stable and satisfy Fast adaptive fault
theH∞ performance index. By constructing a new Lyapunov–Krasovskii functional including the information estimation; linear matrix
of the lower and upper delay bounds, the sufficient delay-dependent conditions have been established to inequality; active fault
guarantee the existence of adaptive observer-based controller in terms of linear matrix inequalities (LMIs). tolerant control; interval
Compared with the constant delay and time-varying delay, the interval time-varying delay is the less con- time-varying delay; delay
servative form. Furthermore, we make full use of the information of the delay and no terms are ignored dependence
when the stability of the system is analysed. In addition, the results for the systems with time-varying struc-
tured uncertainties are established. The results of the active FTC are showed in terms of LMIs. Finally, two
examples are given to verify the effectiveness of the proposed method.

1. Introduction length of delays, the delay-dependent methods are generally less

Over the past 10 years, there has been rapidly growing interest conservative than the delay-independent ones. Extensive results
in fuzzy logic control of nonlinear systems, and there have been have been proposed, for example, Gassara, Hajjaji, and Chaa-
a lot of successful applications on the practical systems, such bane (2010) proposes a single-step condition which depends on
as balancing and swing-up of an inverted pendulum on a cart the size of the delay to overcome the drawback of two-step lin-
and the computer-simulated truck–trailer system. The main fea- ear matrix inequality (LMI) approach. A method of robustH∞
ture of Takagi–Sugeno (T-S) fuzzy model is to express the local reliable control for uncertain T-S fuzzy systems with unavail-
dynamics of each fuzzy rule by the linear system model, and in able states and time-varying delay is developed. For a class of
terms of IF–THEN rules, such that these models can approxi- discrete-time T-S fuzzy Itô stochastic systems with time-varying
mate a wide class of nonlinear systems. Therefore, the conven- delays, Su, Shi, Wu, and Nguang (2013) considers the problem
tional linear system theory can be applied to analyse and design of inducedl2 filter design. For a class of discrete-time T-S fuzzy
the nonlinear systems. time-delay systems, Su, Shi, Wu, and Basin (2014) investigates
It is well known that time delay frequently occurs in many the problem of reliable filter design with strict dissipativity. Gas-
engineering systems, such as long transmission lines in pneu- sara et al. (2010), Su et al. (2013) and Su et al. (2014) do not
matic systems, rolling mills, nuclear reactors, hydraulic systems consider the fault signal.
and manufacturing processes (Richard, 2003). Among the time- Due to the increasing demands for performance, reliability
delay types, the interval time-varying delay is often encountered, and safety in industrial processes, fault detection and identi-
i.e. the delay varies in an interval in which the lower bound fication (FDI), fault estimation (FE) and fault tolerant control
is not restricted to be zero (Jiang & Han, 2005). The interval (FTC) techniques have received considerable attentions. Among
time-varying delay has been identified from many practical sys- the techniques to estimate fault, the fast adaptive fault estima-
tems. A typical example of dynamical system with interval time- tion (FAFE) algorithm is proposed to improve the rapidity and
varying delay is networked control systems (Balasubramaniam, the accuracy of the FE (Zhang, Jiang and Shi, 2009b), which has
Krishnasamy, & Rakkiyappan, 2011; Park, Ko, & Jeong, 2011). been applied to all kinds of systems, such as polytopic linear
Therefore, it is of great significance to study systems with inter- parameter-varying system in Zhang, Jiang, and Chen (2009a),
val time-varying delay. The existence of time delay may induce linear time-varying delay system in Zhang and Jiang (2008) and
instability, oscillation and poor performance of systems. The T- neutral system in Zhang, Jiang, and Shumsky (2009d). The FE
S fuzzy model is a valid way to deal with the nonlinear system. can provide the information for FTC to achieve the reliability
Therefore, for the nonlinear systems with time delay, we can of the system. FDI, FE and FTC have become popular topics
use the T-S fuzzy model to analyse. The results about the time and receive considerable attentions. Karimi, Chadli, Zhang, and
delay can be divided into two parts: delay independent and delay Shi (2014) focuson theoretical and practical aspects on new and
dependent. Because of making full use of the information on the emerging trends in FDI/FTC. In Wang, Jing, Karimi, and Chen

CONTACT Hui Li lihui_neu@.com; [email protected]

©  Informa UK Limited, trading as Taylor & Francis Group
2 H. LI ET AL.

(2015), the robust fault tolerantH∞ control problem of active (2014) and Sun et al. (2014) do not consider the interval of time-
suspension systems with finite-frequency constraint is investi- varying delay.
gated. Both the actuator faults and external disturbances are Moreover, to the best of our knowledge, only few results
considered in the controller synthesis. A robustH∞ dynamic out- have appeared to analyse robust FE and FTC for the T-S fuzzy
put feedback controller with fault tolerant ability is proposed. In system with interval time-varying delay, actuator fault and
Du, Jiang, Shi, and Karimi (2014), the problem of fault detec- norm-bounded external disturbance. The similar problem is
tion for continuous-time switched systems under asynchronous researched in Gassara, El Hajjaji, and Chaabane (2014). Gas-
switching is investigated. In Yin, Wang, and Karimi (2014), a sara et al. (2014) deal with the problem of FTC for continuous-
robust data-driven fault detection approach is proposed with time T-S fuzzy systems with interval time-varying delay by using
application to a wind turbine benchmark. The nonlinearity, adaptive observer. To reduce the conservation of the result,
unknown disturbances and significant measurement noise are Jessen’s integral inequality is applied to deal with the cross-
considered. Accordingly, a variety of important results about T- product items. The proposed approach simplifies the design of
S fuzzy systems with fault have been reported in the literature. an active FTC and gives only one step to estimate the state vec-
Aouaouda, Chadli, Shi, and Karimi (2015) consider robust sen- tor, the actuator fault and the controller gains. Compared with
sor fault detection observer design for uncertain and disturbed Gassara et al. (2014), there are many differences in this paper.
discrete-time T-S systems using H− /H∞ criterion. In Zhang, At first, we consider the system with external disturbances and
Jiang, and Staroswiecki (2010), the problem of robust FE and uncertainties which are not considered in Gassara et al. (2014).
FTC for T-S fuzzy systems with actuator faults is addressed. Suf- What’s more, the Lyapunov–Krasovskii functional in this paper
ficient conditions for the existence of both the fuzzy-augmented is different from Gassara et al. (2014), which includes both lower
FE observer and the observer-based dynamic output feedback and upper bounds of delay. Last but not the least, when deal-
FTC are given in terms of LMIs. For discrete-time T-S fuzzy ing with the cross-product items in this paper, free-weighting
systems, Jiang, Zhang, and Shi (2011) address the same prob- matrices (He, Wu, She, & Liu, 2004) are provided to express
lem. For discrete-time T-S fuzzy systems, Zhang, Jiang, and Shi the relationships between the terms in the Leibinz–Newton for-
(2012) studies the problem of robust FE observer design via mula. We make full use of the information of the delay and no
piecewise Lyapunov functions, and gives the forms of full-order terms are ignored when the stability of the system is analysed.
FE observer and reduced-order FE observer. Sami and Patton Therefore, although the problem of FTC for continuous-time
(2013) describe a novel fault tolerant tracking control strategy T-S fuzzy systems with interval time-varying delay is similar,
based on robust FE for T-S fuzzy systems with simultaneous the form of system and the methods to use are different. The
actuator and sensor faults. However, Jiang et al. (2011), Sami and research works on the above systems still have both practical
Patton (2013), Zhang et al. (2010) and Zhang et al. (2012) do not utility and theoretical significance.
consider the effect of time delay. In this study, inspired by the FE method in Zhang et al.
Therefore, the problems of robust FE and FTC for T-S fuzzy (2009b), the FAFE algorithm is applied for the above system.
systems with simultaneous fault and time delay have attracted First of all, the problem of FE is addressed, where the new
extensive attentions in recent years. Ahmadizadeh, Zarei, and Lyapunov–Krasovskii functional including the information of
Karimi (2014) only investigate the problem of robust fault detec- the upper and the lower bounds of the time delay is constructed.
tion design for a class of uncertain T-S models with unknown When analysing the stability of the system, we make full use
input and time-varying delay. The main objective of the paper of the information of the delay and no terms are ignored. The
(Sakthivel, Vadivel, Mathiyalagan, & Arunkumar, 2014) is to improved sufficient delay-dependent conditions are established
design a robust state feedback reliableH∞ control for a class of in terms of the LMIs. What’s more, we extend the results to the
uncertain T-S fuzzy systems with actuator failures and time- system with uncertainties. In addition, the problem of active
varying delay. For T-S fuzzy descriptor systems subject to time FTC is addressed. Finally, two examples are given to show the
delays and external disturbances, Jia, Chen, Zhang, and Li (2015) effectiveness of the proposed method.
address the problems of observer-based fault reconstruction and This paper is organised as follows. The FE is presented in
FTC. In Zhang, Jiang, and Shi (2009c), a novel fuzzy FAFE algo- Section 2. Section 3 presents the main results for the FE filters
rithm is proposed to enhance the performance of FE, and active which guarantee the filtering error system to be asymptotically
FTC is researched for time-delay T-S fuzzy systems. But the stable with a prescribed H∞ performance. A delay-dependent
external disturbances or uncertainties are not considered. Tong, criterion is established to reduce the conservatism of design-
Yang, and Zhang (2011) considers the observer-based dynamic ing procedure. Section 4 presents the main results for the active
output feedback FTC to compensate for the effect of unknown FTC. In Section 5, two examples are given to illustrate that the
sensor faults for T-S fuzzy systems with time delay. But the time proposed methods are valid. Finally, we conclude the paper in
delay is constant. For a class of T-S fuzzy systems with time- Section 6.
varying state delay and actuator faults, Huang and Yang (2014)
are concerned with the problem of robust FE and FTC. To make
the sufficient conditions less conservative, based on the (k − 2. Problem formulation
1)-step FE information, a novel k-step FE observer is proposed Consider a T-S fuzzy system with interval time-varying delay.
to construct the k-step fault error dynamics. UnderH∞ perfor- The ith rule of the system is expressed in the following IF–THEN
mance constraint, Sun, Wang, and He (2014) address the prob- rules:
lem of robust FE for T-S nonlinear systems with actuator, sensor
faults and time-varying state delay. However, Huang and Yang Plant Rulei(i = 1, 2, . . . , r):
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 3

IF z 1 (t ) is μi1 and, . . ., and z p (t ) is μip , THEN The derivative of f (t ) with respect to time is norm bounded,
i.e. f˙ (t ) ∈ L2 [0, ∞):

ẋ(t ) = Ai x(t ) + Adi x(t − τ (t )) + Bi u(t ) + E i f(t ) + F i w(t )
y(t ) = Cx(t ) f˙ (t ) ≤ fmax ; 0 ≤ fmax < ∞
(1)
where x(t ) is the state vector, u(t ) is the control input vector,
Assumption 2.2 (Gassara et al., 2014):
y(t ) is the output vector, f (t ) represents the additive actuator
fault and w(t ) is the disturbance, which is assumed to belong
to L2 [0, ∞). Ai , Adi , Bi , Ei , Fi and C are known constant real rank(CBi ) = nu , i = 1, . . . , r
matrices of appropriates dimensions. It is supposed that matri-
Lemma 2.1 (Wang, Xie, & de Souza, 1992):
ces Ei are of full-column rank, the pairs (Ai , Bi ) are control-
lable, and the pairs (Ai , Ci ) are observable. z j (t )( j = 1, . . . , p) For a symmetric positive definite matrix P, the following
and μi j (i = 1, 2, . . . , r; j = 1, 2, . . . , p) are, respectively, the inequality holds:
premise variable and the fuzzy sets that are characterised by
membership function. r is the number of IF–THEN rules, and p 2xT y ≤ xT Px + yT P−1 y; x, y ∈ Rn
is the number of the premise variables. τ (t ) is an interval time-
varying delay satisfying Lemma 2.2 (Xie & Souza, 1992):

0 ≤ τm ≤ τ (t ) ≤ τM , 0 ≤ τ̇ (t ) ≤ τd (2) Consider a negative definite matrix  < 0. Given a symmet-

ric matrix X of appropriate dimension such that XT X < 0,
where τm and τM are the lower and upper delay bounds. then ∃λ ∈ R+ such that
The overall fuzzy model achieved by fuzzy blending of each
individual plant rule (local model) is given by XT X ≤ −2λX − λ2 −1

⎧ Lemma 2.3 (Yu & Lien, 2008):

⎪
⎪ 
r
⎪
⎨ẋ(t ) = hi (z(t ))[Ai x(t ) + Adi x(t − τ (t )) LetU,W and Zbe real matrices of appropriate dimensions
i=1 (3) with Z satisfying Z =ZT , then
⎪
⎪ + Bi u(t ) + E i f(t ) + F i w(t )]
⎪
⎩
y(t ) = Cx(t )
Z + UF(t )W + WT FT (t )UT < 0, FT (t )F(t ) ≤ I
where z(t ) = [z 1 (t ), . . . , z p (t )], hi (z(t )) = σi (z(t ))/
r p if and only if there exists a scalar λ > 0 such that
i=1 σi (z(t )), σ i (z(t )) = j=1 μ i j (z j (t )) is the membership
function. It is obvious that the fuzzy weighting functionshi (z(t ))
satisfy Z + λ−1 UT U + λWT W < 0

⎧ r A block diagram of this paper is shown in Figure 1.

⎨
⎪
hi (z(t )) = 1 In order to estimate the state and the fault of system (4), the
⎪
⎩ i=1 state and FE observer will be constructed in Section 3. We can
0 ≤ hi (z(t )) ≤ 1 obtain the error dynamics and the estimation of the fault by the
deduction. On the basis of the obtained online fault estimation
For simplicity, we introduce the following notations: information, we will design an active fault tolerant controller to
guarantee stability when the system with fault in Section 4.

r 
r
hi = hi (z(t )), A(h) = hi Ai , Ad (h) = hi Adi , B(h) 3. Fault estimation
i=1 i=1 In order to estimate the state and the fault of system (4), the fol-

r r 
r
lowing FE observer is constructed:
= hi Bi , E(h) = hi Ei , F(h) = hi Fi
i=1 i=1 i=1 ⎧
⎪ ˙ ) = A(h)x̂(t ) + Ad (h)x̂(t − τ (t )) + B(h)u(t )
x̂(t
⎪
⎪
⎨
Then the T-S fuzzy model (3) can be rewritten as + E(h)f̂(t ) − L(h)(ŷ(t ) − y(t ))
(5)
⎪
⎪ ŷ(t ) = Cx̂(t )
⎧ ⎪
⎩˙
⎨ẋ(t ) = A(h)x(t ) + Ad (h)x(t − τ (t )) f̂(t ) = −R̄(h)(ėy (t ) + ey (t ))
+ B(h)u(t ) + E(h)f(t ) + F (h)w(t ) (4)
⎩
y(t ) = Cx(t ) where x̂(t ) is the observer state, ŷ(t ) is the observer output vec-
tor, fˆ (t ) is an estimation of actuator fault f (t). L(h) and R̄(h)
In order to obtain our main result in this paper, two assump- are the observer gains to be designed L(h) = ri=1 hi Li , R̄(h) =
tions and three lemmas are given: r
i=1 hi R̄i . Symmetric positive definite matrix  is the learning
Assumption 2.1 (Gassara et al., 2014): rate.
4 H. LI ET AL.

u (t ) x(t ) xˆ (t )
xr (t ) Controller State
System
Gain Estimation
-
fˆ (t )

Fault Residual
Compensation Production

Fault
Diagnosis

Figure . A block diagram of this paper.

Denoting ex (t ) = x̂(t ) − x(t ), ey (t ) = ŷ(t ) − y(t ), ef (t ) = Theorem 3.1:

f̂(t ) − f(t ), the error dynamics are given by
Consider system (6), under Assumptions 2.1 and 2.2, for the
given positive scalars τm , τM , τd , ε, α and γ , the error dynamic
⎧ system (6) is asymptotically stable with (w(t ) = 0), while
⎪
⎪ ėx (t ) = (A(h) − L(h)C)ex (t ) + Ad (h)ex (t − τ (t ))
⎨ satisfying a prescribed H∞ performance (7), if there exist matri-
+ E(h)ef (t ) − F (h)w(t ) (6) ces P > 0, Q1 > 0, Q2 > 0, G > 0, R > 0, Z > 0,L(h)
⎪
⎪
⎩ and free-weighting matrices M = [ MT1 MT2 MT3 MT4 MT5 ]T ,
ey (t ) = Cex (t )
N = [ NT1 NT2 NT3 NT4 NT5 ]T such that the following conditions
hold:

The proposed robust FE observer design is converted to the

problem of seeking the gain matrix L(h) such that:
ETi P = R̄i C (8)
(i) the error dynamic system (6) with interval time-varying 
 3ii < 0 i = 1, 2, . . . , r (9)
delay is asymptotically stable (when w(t ) = 0);
(ii) the following H∞ performance is satisfied:  3ij +  3ji < 0, i, j = 1, 2, . . . , r, i < j (10)
3ii < 0 i = 1, 2, . . . , r (11)
3ij + 3ji < 0, i, j = 1, 2, . . . , r, i < j (12)
L L
2
e f (t ) dt ≤ γ 2 w(t )2 dt (7)
0 0 where

for all L > 0,γ > 0 and w(t ) ∈ L2 [0, ∞) under zero initial con-
ditions.
⎡ ⎤
 311 312 −MT3 + N1  314 −PFi − MT5 τM N1  317  318
⎢ ∗ 322 MT − NT + N2 324 MT − NT √ √
⎢ 3 3 5 5 τM N2 τM ATdi P τmM ATdi P ⎥
⎥
⎢ ∗ ∗ −Q2 + N3 + N3 N4
T T
N5 T
τM N3 0 0 ⎥
⎢ √ √ ⎥
⎢ ∗ ∗ ∗ 344 T
Ei PFi τM N4 τM Ei P
T
τmM Ei P ⎥
T
 ⎢
 3ij = ⎢ √ √ ⎥<0 (13)
⎢ ∗ ∗ ∗ ∗ −γ 2 I τM N5 − τM FTi P − τmM FTi P ⎥ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ −τM (R + Z) 0 0 ⎥
⎢ ⎥
⎣ ∗ ∗ ∗ ∗ ∗ ∗ −2αP + α R 2
0 ⎦
∗ ∗ ∗ ∗ ∗ ∗ ∗ −2αP + α 2 Z
⎡  ⎤
 311 312 −MT3 + N 1  314 −PF i − MT5 τM M1  317  318
⎢ ∗ 322 MT − N T + N 2 324 MT − N T τM M2 √ √
⎢ 3 3 5 5 τM ATdi P τmM ATdi P ⎥ ⎥
⎢ ∗ ∗ −Q2 + N 3 + N 3 N 4
T T
N5 T
τM M3 0 0 ⎥
⎢ √ √ ⎥
⎢ ∗ ∗ ∗ 344 T
E i PF i τM M4 τM E i P
T
τmM E i P ⎥
T
⎢
3i j = ⎢ √ √ ⎥<0 (14)
⎢ ∗ ∗ ∗ ∗ −γ 2 I τM M5 − τM F Ti P − τmM F Ti P ⎥ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ −τM R 0 0 ⎥
⎢ ⎥
⎣ ∗ ∗ ∗ ∗ ∗ ∗ −2αP + α R 2
0 ⎦
∗ ∗ ∗ ∗ ∗ ∗ ∗ −2αP + α Z 2
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 5

By considering the performance (7), we obtain


 311 = PAi − Yi C + ATi P − CT YTi + εQ1 + Q2 − M1 − MT1

 314 = −ATi PEi + CT YTi Ei − MT4 V̇ + eTf (t )ef (t ) − γ 2 wT (t )w(t )
√
 317 = τM (ATi P − CT YTi ) ≤ eTx (t )Pėx (t ) + ėTx (t )Pex (t ) + εeTx (t )Q1 ex (t )
√
 318 = τmM (ATi P − CT YTi ) − ε(1 − τd )eTx (t − τ (t ))Q1 ex (t − τ (t ))
312 = PAdi + M1 − N1 − MT2 + eTx (t )Q2 ex (t ) − eTx (t − τM )Q2 ex (t − τM )
322 = −ε(1 − τd )Q1 + M2 + MT2 − N2 − NT2 t
+ τM ėTx (t )Rėx (t ) − ėTx (t )Rėx (t )ds
324 = −ATdi PEi + MT4 − NT4 t−τM
344 = I + G − 2ETi PEi t−τ (t )
+ τmM ėTx (t )Zėx (t ) − ėTx (t )Zėx (t )ds
t−τM

+ ėTf (t )−1 ef (t ) + eTf (t )−1 ėf (t )

Then the observer gain matrices can be obtained as follows: + eTf (t )ef (t ) − γ 2 wT (t )w(t ) (19)

in which
Li = P−1 Yi (15)
ėTf (t )−1 ef (t ) + eTf (t )−1 ėf (t )
˙ ˙
= (f̂(t ) − ḟ(t ))T −1 ef (t ) + ef (t )T −1 (f̂(t ) − ḟ(t ))
and the FE can be obtained that = [−R1C(ėx (t ) + ex (t )) − ḟ(t )]T −1 ef (t ) (20)
T −1
+ef (t )  [−R1C(ėx (t ) + ex (t )) − ḟ(t )]
= −2ef (t )T R1 (h)C(ėx (t ) + ex (t )) − 2ef (t )T −1 ḟ(t )

r t
f̂(t ) = − hi (z(t ))R̄i ( ey (τ )dτ + ey (t )) (16)
i=1 tf From Lemma 2.1 and Assumption 2.1, we can obtain that

− 2ef (t )T −1 ḟ(t ) ≤ eTf (t )Gef (t ) + ḟ T (t )−1 G−1 −1 ḟ(t )

where t f denotes the instant when fault occurs.
≤ eTf (t )Gef (t ) + δ (21)
Proof: Consider the following Lyapunov function:

where δ = fmax
2
λmax (−1 G−1 −1 ) > 0
t So we can get that
V = eTx (t )Pex (t ) + ε eTx (t )Q1 ex (t )ds
t−τ (t )
t 0 t V̇ + eTf (t )ef (t ) − γ 2 wT (t )w(t )
+ eTx (t )Q2 ex (t )ds + ėTx (t )Rėx (t )dsdθ t
t−τM −τM t+θ ≤ ξ T (t )1 ξ (t ) + τM ėTx (t )Rėx (t ) − ėTx (t )Rėx (t )ds
−τm t t−τM
+ ėTx (t )Zėx (t )dsdθ + eTf (t )−1 ef (t ) (17) t−τ (t )
−τM t+θ + τmM ėTx (t )Zėx (t ) − ėTx (t )Zėx (t )ds + δ (22)
t−τM

By taking the derivation of V , we have where

 T
ξ (t ) = eTx (t ) eTx (t − τ (t )) eTx (t − τM ) eTf (t ) wT (t )
V̇ ≤ eTx (t )Pėx (t )
+ ėTx (t )Pex (t )
+ εeTx (t )Q1 ex (t )
− ε(1 − τd )ex (t − τ (t ))Q1 ex (t − τ (t ))
T
1 =
⎡ ⎤
+ eTx (t )Q2 ex (t ) − eTx (t − τM )Q2 ex (t − τM ) 111 PAd (h) 0 −(A(h) − L(h)C)T PE(h) −PF(h)
t ⎢ ∗ −ε(1 − τ )Q 0 −ATd (h)PE(h) ⎥
⎢ d 1 0 ⎥
+ τM ėTx (t )Rėx (t ) − ėTx (t )Rėx (t )ds ⎢ ⎥
⎢ ∗ ∗ −Q2 0 0 ⎥
t−τM ⎢ ⎥
⎣ ∗ ∗ ∗ I + G − 2ET (h)PE(h) ET (h)PF(h)⎦
t−τ (t )
+ τmM ėTx (t )Zėx (t ) − ėTx (t )Zėx (t )ds ∗ ∗ ∗ ∗ −γ 2 I
t−τM
111 = P(A(h) − L(h)C) + (A(h) − L(h)C)T P + εQ1 + Q2
+ ėTf (t )−1 ef (t ) + eTf (t )−1 ėf (t ) (18)
6 H. LI ET AL.

From Newton–Leibniz formula, we can get the following t

equations: ≤ ξ T (t )2 ξ (t ) − ėTx (t )Rėx (t )ds
t−τ (t )

t−τ (t )
t − ėTx (t )(Z + R)ėx (t )ds
2ξ (t )M[ex (t ) − ex (t − τ (t )) −
T
ėTx (t )ds] =0 (23) t−τM
t−τ (t ) t−τ (t )
t−τ (t )
+ 2ξ T (t )N ėTx (t )ds
2ξ (t )N[ex (t − τ (t )) − ex (t − τM ) −
T
ėTx (t )ds] =0 t−τM
t−τM
(24) t
+ 2ξ T (t )M ėTx (t )ds + δ
t−τ (t )

Adding (23) and (24) to (22), we can obtain that  t

≤ (1 τM )( ξ T (t )2 ξ (t )ds+2τM ξ T (t )M
t−τ (t )
t t
V̇ + eTf (t )ef (t ) − γ 2 wT (t )w(t ) × ėTx (t )ds − τM ėTx (t )Rėx (t )ds)
t t−τ (t ) t−τ (t )
≤ ξ T (t )1 ξ (t ) + τM ėTx (t )Rėx (t ) − ėTx (t )Rėx (t )ds t−τ (t )
t−τM 
t−τ (t )
+ (1 τM )( ξ T (t )2 ξ (t )ds+2τM ξ T (t )N
t−τM
+ τmM ėTx (t )Zėx (t ) − ėTx (t )Zėx (t )ds
t−τM t−τ (t ) t−τ (t )
t−τ (t ) × ėTx (t )ds − τM ėTx (t )(Z + R)ėx (t )ds) + δ
− 2ξ (t )N[ex (t − τ (t )) − ex (t − τM ) −
T
ėTx (t )ds] t−τM t−τM
t−τM  T   
t  t ξ (t ) 2 τM M ξ (t )
− 2ξ T (t )M[ex (t ) − ex (t − τ (t )) − ėTx (t )ds] + δ ≤ (1 τM ) ds
t−τ (t ) t−τ (t ) ėx (t ) ∗ −τM R ėx (t )
≤ ξ T (t )1 ξ (t ) + τM ξ T (t )T1 R1 ξ (t )  T  
 t−τ (t ) ξ (t ) 2 τM N
+ τmM ξ T (t )T1 Z1 ξ (t ) − ξ T (t )( + T )ξ (t ) + (1 τM )
t t−τ (t ) t−τM ėx (t ) ∗ −τM (Z + R)
− ėTx (t )Rėx (t )ds − ėTx (t )Zėx (t )ds  
t−τM t−τM ξ (t )
t−τ (t ) × ds + δ (26)
ėx (t )
+ 2ξ T (t )N ėTx (t )ds
t−τM
t
(25) where 2 = 1 + τM1 R1 +  τmM 1Z1 −  − 
T T T
+ 2ξ T (t )M ėTx (t )ds + δ 
2 τ M M  2 τM N
t−τ (t ) Denoting 3 = ∗ −τM R , 3 = ∗ −τM (Z + R) , β(t ) =
 
ξ (t ) 
ėx (t ) , ε1 = λmin (−3 ) > 0, ε2 = λmin (− 3 ) > 0, we can
where obtain that

  V̇ + eTf (t )ef (t ) − γ 2 wT (t )w(t )

 = M −M + N −N 0 0 , 1  t
= [ A(h) − L(h)C Ad (h) 0 E(h) −F(h) ] ≤ −(ε1 τM ) β(t )T β(t )ds
t−τ (t )
 t−τ (t )
− (ε2 τM ) β(t )T β(t )ds + δ
We can further get that t−τM
 t
≤ [(−ε1 − ε2 ) τM ] β(t )T β(t )ds + δ
t−τM
V̇ + eTf (t )ef (t ) − γ 2 wT (t )w(t ) ≤ −(ε1 + ε2 )β(t )2 + δ (27)
≤ ξ T (t )2 ξ (t )
t t−τ (t )
If β(t )2 ≥ δ/(ε1 + ε2 ) holds, then it has V̇ + eTf (t )ef (t ) −
− ėTx (t )Rėx (t )ds − ėTx (t )Zėx (t )ds
t−τM t−τM γ wT (t )w(t ) < 0. We can conclude that (7) holds for all L > 0
2

t−τ (t ) and w(t ) ∈ L2 [0, ∞) under zero initial conditions. In addition,

+ 2ξ T (t )N ėTx (t )ds by choosing the same Lyapunov function as (17) and the similar
t−τM line in the earlier deduction, we can easily obtain that the time
t
+ 2ξ T (t )M ėTx (t )ds + δ derivative of V along the solution of error dynamics (6) with
t−τ (t ) (w(t ) = 0) satisfies V̇ < 0, which indicates the asymptotic sta-
bility of systems (6) with (w(t ) = 0).
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 7

From ε1 = λmin (−3 ) > 0 ε2 = λmin (− 3 ) > 0, we can conclude that
⎡ ⎤
311 312 −MT3 + N1 314 −PF(h) − MT5 τM N1 317 318
⎢ √ √ ⎥
⎢ ∗ 322 MT3 − NT3 + N2 324 MT5 − NT5 τM N2 τM ATd (h)P τmM ATd (h)P ⎥
⎢ ⎥
⎢ ∗ ∗ −Q2 + N3 + NT3 NT4 NT5 τM N3 0 0 ⎥
⎢ √ √ ⎥
⎢ ∗ ∗ ∗ 344 E (h)PF(h)
T
τM N4 τM E (h)P
T
τmM E (h)P ⎥
T
⎢ ⎥
 3 = ⎢ √ √ ⎥<0 (28)
⎢ ∗ ∗ ∗ ∗ −γ 2 I τM N5 − τM FT (h)P − τmM FT (h)P⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ −τM (R + Z) 0 0 ⎥
⎢ ⎥
⎢ −1 ⎥
⎣ ∗ ∗ ∗ ∗ ∗ ∗ −PR P 0 ⎦
−1
∗ ∗ ∗ ∗ ∗ ∗ ∗ −PZ P

⎡ ⎤
311 312 −MT3 + N1 314 −PF(h) − MT5 τM M1 317 318
⎢ ∗ 322 MT − NT + N2 √ √
⎢ 3 3 324 MT5 − NT5 τM M2 τM ATd (h)P τmM ATd (h)P ⎥
⎥
⎢ ∗ ∗ −Q2 + N3 + NT3 NT4 NT5 τM M3 0 0 ⎥
⎢ √ √ ⎥
⎢ ∗ ∗ ∗ 344 E (h)PF(h)
T
τM M4 τM ET (h)P τmM ET (h)P ⎥
3 = ⎢
⎢ ∗ √ √ ⎥<0 (29)
⎢ ∗ ∗ ∗ −γ 2 I τM M5 − τM FT (h)P − τmM FT (h)P⎥ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ −τM R 0 0 ⎥
⎢ ⎥
⎣ ∗ ∗ ∗ ∗ ∗ ∗ −PR−1 P 0 ⎦
∗ ∗ ∗ ∗ ∗ ∗ ∗ −PZ−1 P

where

311 = P(A(h) − L(h)C) + (A(h) − L(h)C)T P + εQ1 + Q2 − M1 − MT1

314 = −(A(h) − L(h)C)T PE(h) − MT4
√
317 = τM (A(h) − L(h)C)T P
√
318 = τmM (A(h) − L(h)C)T P
312 = PAd (h) + M1 − N1 − MT2
322 = −ε(1 − τd )Q1 + M2 + MT2 − N2 − NT2
324 = −ATd (h)PE(h) + MT4 − NT4
344 = I + G − 2ET (h)PE(h)

We can find that the forms of (28) and (29) are not LMIs. So we are now in a position to transform the conditions (28)
and (29) in terms of LMIs, which can be solved efficiently by using existing solvers such as LMI toolbox in the MATLAB
software.
Based on Lemma 2.2, we can obtain that ∃α ∈ R+ such that

− PR−1 P ≤ −2αP + α 2 R (30)

− PZ−1 P ≤ −2αP + α 2 Z (31)

Denoting Y(h) = PL(h), we can obtain the following LMIs forms:

⎡ ⎤
 311 312 −MT3 + N1  314 −PF(h) − MT5 τM N1  317  318
⎢ √ √ ⎥
⎢ ∗ 322 MT3 − NT3 + N2 324 MT5 − NT5 τM N2 τmM ATd (h)P ⎥
τM ATd (h)P
⎢ ⎥
⎢ ∗ ∗ −Q2 + N3 + NT3 NT4 NT5 τM N3 ⎥
⎢ 0 0 ⎥
⎢ √ √ ⎥
⎢ ∗ ∗ ∗ 344 ET (h)PF(h) τM N4 τM ET (h)P τmM ET (h)P ⎥
 ⎢ ⎥
3 =⎢ √ √ ⎥<0 (32)
⎢ ∗ ∗ ∗ ∗ −γ 2 I τM N5 − τM FT (h)P − τmM FT (h)P⎥
⎢ ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ −τM (R + Z) 0 0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ −2αP + α 2 R ⎥
⎣ 0 ⎦
∗ ∗ ∗ ∗ ∗ ∗ ∗ −2αP + α Z 2
8 H. LI ET AL.

⎡ ⎤
 311 312 −MT3 + N1  314 −PF(h) − MT5 τM M1  317  318
⎢ ∗ 322 MT − NT + N2 √ √
⎢ 3 3 324 MT5 − NT5 τM M2 τM ATd (h)P τmM ATd (h)P ⎥
⎥
⎢ ∗ ∗ −Q2 + N3 + NT3 NT4 NT5 τM M3 0 0 ⎥
⎢ √ √ ⎥
⎢ ∗ ∗ ∗ 344 E (h)PF(h)
T
τM M4 τM ET (h)P τmM ET (h)P ⎥
3 = ⎢
⎢ ∗ √ √ ⎥<0 (33)
⎢ ∗ ∗ ∗ −γ 2 I τM M5 − τM FT (h)P − τmM FT (h)P⎥ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ −τM R 0 0 ⎥
⎢ ⎥
⎣ ∗ ∗ ∗ ∗ ∗ ∗ −2αP + α 2 R 0 ⎦
∗ ∗ ∗ ∗ ∗ ∗ ∗ −2αP + α 2 Z

in An, Wen, Gan, and Li (2011) and Li and Yang (2012). When
where the delay is constant, the results will reduce to the forms in
Zhang et al. (2009c). So the interval time-varying delay is the
 311 = PA(h) − Y(h)C + AT (h)P − CT YT (h) + εQ1 less conservative form.
+ Q2 − M1 − MT1 Remark 3.3:

 314 = −AT (h)PE(h) + CT YT (h)E(h) − MT4 When the delay is zero, the system (4) will be the following
√ form:
 317 = τM (AT (h)P − CT YT (h))
√ 
 318 = τmM (AT (h)P − CT YT (h)) ẋ(t ) = A(h)x(t ) + B(h)u(t ) + E(h)f(t ) + F (h)w(t )
y(t ) = Cx(t )
If (32) and (33) hold, we can obtain that
Theorem 3.1 will change to the following form:
Under Assumptions 2.1 and 2.2, for the given positive scalar

r γ , the error dynamic system is asymptotically stable with

3 = h2i (z(t )) 3ii (w(t ) = 0), while satisfying a prescribed H∞ performance (7),
i=1 if there exist matrices P > 0, G > 0,L(h) such that the follow-

r 
r
ing conditions hold:
+ hi (z(t ))h j (z(t ))( 3ij +  3ji ) < 0 (34)
i=1 i< j ETi P = R̄i C

r
3 = h2i (z(t ))3ii  3ii < 0 i = 1, 2, . . . , r
i=1

r 
r
 3ij +  3ji < 0, i, j = 1, 2, . . . , r, i < j
+ hi (z(t ))h j (z(t ))(3ij + 3ji ) < 0 (35)
i=1 i< j
where
⎡ ⎤
which imply that (9)–(12) hold. The proof is completed. PAi − Yi C + ATi P − CT YTi −ATi PEi + CT YTi Ei −PFi
⎢ ⎥
 3ij =⎣ ∗ I + G − 2ETi PEi ETi PFi ⎦ < 0
Remark 3.1 (Gassara et al., 2014): ∗ ∗ −γ I
2

In (27), we can see that if f˙ (t ) = 0, that is, f1max = 0, the

state x(t ) of the system, the state estimation error ex (t ) and the Then the observer gain matrices can be obtained as follows:
FE error e f (t )converge asymptotically to zero. If f (t ) is time Li = P−1 Yi
varying, these variables converge to a ball, centred at the ori-
gin, defined by terms δ,ε1 and ε2 , The radius of this set can be and the FE can be obtained that
minimised by a choice of parameter  that minimises δ without 
r  t 
changing ε1 and ε2 . It thus improves the accuracy of the estima- f̂(t ) = − hi (z(t ))R̄i ey (τ )dτ + ey (t )
tion. i=1 tf

Remark 3.2:
where t f denotes the instant when fault occurs.
The main advantage of Theorem 3.1 which we propose is The proof is similar to Theorem 3.1, which is omitted here.
to reduce the conservativeness by considering interval time-
Remark 3.4:
varying delay. Among the time-delay types, the interval time-
varying delay is often encountered. The delay varies in an We will discuss the computational complexity of Theorem
interval in which the lower bound is not restricted to be 3.1. There are 18 unknown matrices in Theorem 3.1.
zero (Jiang &Han, 2005). The interval time-varying delay They are P, Q1 , Q2 , G, R, Z, L, M1 , M2 , M3 , M4 , M5
τ (t ) is differentiable and bounded with a constant delay- , N1 , N2 , N3 , N4 , N5 and R̄. We apply LMI toolbox in the
derivative bound. When the lower bound of the time delay τm MATLAB software to solve the conditions in Theorem 3.1. The
is zero, the interval time-varying delay will change to the forms time is short and only some seconds.
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 9

where

⎡ ⎤
˜  311 
˜ 312 −MT + N1  314 −PFi − MT τM N1  317  318 PD
⎢ ∗ 
3
˜ 322 MT − NT + N2 324 MT − NT
5
√ √ ⎥
⎢ 3 3 5 5 τM N2 τM ATdi P τmM ATdi P 0 ⎥
⎢ ∗ ∗ −Q2 + N3 + N3 N4
T T T
τM N3 ⎥
⎢ N5
√
0
√
0 0 ⎥
⎢ ∗ ∗ ∗ 344 T
τM N4 τM Ei P
T
τmM Ei P −Ei PD ⎥
T T
⎢ Ei PFi
√ √ ⎥
˜ = ⎢ ∗ ∗ ∗ ∗ −γ 2 I τM N5 − τM FTi P − τmM FTi P ⎥<0 (43)
3ij ⎢ 0 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ −τM (R + Z) ⎥
⎢ 0 0
√
0 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ −2αP + α R 2
τM PD ⎥
⎢ 0
√ ⎥
⎣ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −2αP + α 2 Z τmM PD⎦
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −λI
⎡  ⎤
˜ ˜ 
 311 312 −M3 + N1  314 −PFi − M5 τM M1
T T 
 317 
 318 PD
⎢ ∗  ˜ 322 MT − NT + N2 324 MT − NT τM M2 √ √ ⎥
⎢ 3 3 5 5 τM ATdi P τmM ATdi P 0 ⎥
⎢ ∗ ∗ −Q2 + N3 + N3 N4
T T T
τM M3 ⎥
⎢ N5
√
0
√
0 0 ⎥
⎢ ∗ ∗ ∗ 344 T
τM M4 τM Ei P
T
τmM Ei P −Ei PD ⎥
T T
⎢ Ei PFi
√ √ ⎥
˜ 3ij = ⎢ ∗ ∗ ∗ ∗ −γ 2 I τM M5 − τM FTi P − τmM FTi P ⎥<0 (44)
⎢ 0 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ −τM R ⎥
⎢ 0 0
√
0 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ −2αP + α R 2
τM PD ⎥
⎢ 0
√ ⎥
⎣ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −2αP + α 2 Z τmM PD⎦
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −λI
˜ 
 311 = PAi − Yi C + Ai P − CT YTi + εQ1 + Q2
T
Extend the results to the system with norm-bounded uncer-
tainty. Consider a T-S fuzzy system with time-varying delay and −M1 − MT1 + λET1i E1i
parametric uncertainties:
⎧ ˜ 312 = PAdi + M1 − N1 − MT + λET E2i
⎪ r 
⎪
⎨ẋ(t ) = hi (z(t ))((Ai + Ai (t ))x(t ) + (Adi 2 1i
i=1
(36) ˜ 322 = −ε(1 − τd )Q1 + M2 + MT − N2 − NT + λET E2i

⎪
⎪ + Adi (t ))x(t−τ (t ))+Bi u(t )+E i f(t ) + F i w(t )) 2 2 2i
⎩
y(t ) = Cx(t )
where Ai (t ) and Adi (t ) are the norm-bounded time-varying Then the observer gain matrix can be obtained as follows:
uncertainties which represent parametric uncertainties in the
system and are assumed to belong to the following sets: Li = P−1 Yi (45)

[ Ai (t ), Adi (t )] = Di F1i (t ) [E1i , E2i ] (37) and the FE can be obtained that

where Di ,E1i and E2i are known constant matrices of appropri- 

r t
fˆ (t ) = − hi (z(t ))R̄i ( ey (τ )dτ + ey (t )) (46)
ate dimensions and F1i (t ) is an unknown matrix function with tf
i=1
Lebesgue measurable elements satisfying FT1i (t )F1i (t ) ≤ I∀t>0.
Based on Theorem 3.1, we can extend the results to the sys- where t f denotes the instant when fault occurs.
tem (36) with time-varying structured uncertainties.
Proof: Replace Ai , Adi in error dynamics (6) with Ai +
Theorem 3.2: Di F1i (t )E1i ,Adi + Di F2i (t )E2i , respectively. Then, error dynam-
Consider system (36). Under Assumptions 2.1 and ics (6) for the uncertain system (36) is equivalent to the following
2.2, for the given positive scalars τm , τM , τd , ε, α, λ and condition:
γ , the error dynamic system (6) is asymptotically stable 
with (w(t ) = 0), while satisfying a prescribed H∞ perfor- 3ij + UTi F1i (t )Vi + VTi FT1i (t )Ui < 0
(47)
mance (7), if there exist matrices P > 0, Q1 > 0, Q2 >  3ij + UTi F1i (t )Vi + VTi FT1i (t )Ui < 0
0, G > 0, R > 0, Z > 0,L(h) and free-weighting matrices √ √
M = [ MT1 MT2 MT3 MT4 MT5 ]T , N = [ NT1 NT2 NT3 NT4 NT5 ]T where Ui = [ DT P 0 0 −DT PEi 0 0 τM DT P τmM DT P ],
such that the following conditions hold: Vi = [ E1i E2i 0 0 0 0 0 0 ]. By Lemma 2.3, a necessary and
ETi P = R̄i C (38) sufficient condition to satisfy inequality (47) for the uncertain
system (36) is that there exists a scalar λ > 0, such that
˜  3ii < 0 i = 1, 2, . . . , r (39) 
3ij + λ−1 UTi Ui + λVTi Vi < 0
(48)
˜  3ij + ˜  3ji < 0, i, j = 1, 2, . . . , r, i < j (40)  3ij + λ−1 UTi Ui + λVTi Vi < 0

˜ 3ii < 0
 i = 1, 2, . . . , r (41) holds. Applying Schur complements, we have that (48) is equiv-
alent to (43) and (44). The proof is completed.
˜ 3ij + 
 ˜ 3ji < 0, i, j = 1, 2, . . . , r, i < j (42)
10 H. LI ET AL.

4. Fault accommodation
For simplicity, we make B = E in the T-S fuzzy system (3). On the basis of the obtained online fault estimation information, we
design a fault tolerant controller to guarantee stability in the presence of faults. The active FTC is


r
u(t ) = hi (z(t ))Ki x̂(t ) − f̂(t ) (49)
i=1


where Ki are the controller gains to be designed and K(h) = ri=1 hi (z(t ))Ki .
Substituting (49) into (3) and (6), we obtain the dynamics of the closed-loop system and the state estimation error:

ẋ(t ) = (A(h) + B(h)K(h))x(t ) + Ad (h)x(t − τ (t )) − B(h)K(h)ex (t ) + B(h)ef (t ) + F (h)w(t ) (50)

ėx (t ) = (A(h) − L(h)C)ex (t ) + Ad (h)ex (t − τ (t )) + B(h)ef (t ) − F (h)w(t ) (51)

The problem of active FTC for the closed-loop fuzzy system is converted to design the gain matrices L(h) and K(h) such that:

(i) the closed-loop fuzzy system (50) with interval time-varying delay is asymptotically stable (when w(t ) = 0);
(ii) The following H∞ performance is satisfied:

L L
2
y(t ) dt ≤ γ̄ 2 w(t )2 dt (52)
0 0

for all L > 0, γ̄ > 0 and w(t ) ∈ L2 [0, ∞) under zero initial conditions.
Theorem 4.3:

Consider systems (50) and (51), under Assumptions 2.1 and 2.2, for the given positive scalars τm , τM , τd , ε, α and γ̄ , if there
exist matrices X = P−1 
1 > 0, P2 > 0, Z1 > 0, Z2 > 0, Z3 > 0, Z4 > 0, Z5 > 0, Z6 > 0, G > 0, Y i , Wi and free-weighting matri-
                 
ces M = [ M 1 M 2 M 3 M 4 M 5 M 6 M 7 M 8 ] , N = [ N 1 N 2 N 3 N 4 N 5 N 6 N 7 N 8 ] such that the following condi-
T T T T T T T T T T T T T T T T T T

tions hold:

BTi P2 = R̄i C (53)

 3ii < 0 i = 1, 2, . . . , r (54)

 3ij +  3ji < 0, i, j = 1, 2, . . . , r, i < j (55)

3ii < 0 i = 1, 2, . . . , r (56)

3ij + 3ji < 0, i, j = 1, 2, . . . , r, i < j (57)

where
⎡ ⎤
 A X 0 314 M1 − N1 N1 Bi X Fi X τM N1 0 0 XCT
⎢ 311 di ⎥
⎢ ∗ 322 0 −M2 M2 − N2 N2 0 0 τM N2 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ ∗ ∗ −Z6 −M3 M3 − N3 N3 0 0 τM N3 0 0 0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ¯ ¯ 345 ¯ 346
 ¯  −P2 Fi − M T
 τ 
MN 4 ¯ ¯ 0 ⎥
⎢ 344 347 8 3410 3411 ⎥
⎢ ∗ ∗ ∗ ∗ ¯ 355 ¯ 356
 ¯ 357 M − N
  T  T 
τM N 5 ¯ 3510 ¯ 3511 0 ⎥
 ⎢ 8 8 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ¯ 366
 N T7 N T8 τM N 6 0 ⎥
 3ij =⎢ 0 0 ⎥<0 (58)
⎢ ¯ ¯ ¯ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ 377 T
Bi P2 Fi τM N 7 3710 3711 0 ⎥
⎢ ¯ 3810 ¯ 3811 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −γ̄ I 2
τM N 8  0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −τM (Z3 + Z4 ) 0 0 0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ¯ 31010 0
 0 ⎥
⎢ ⎥
⎣ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗  ¯ 31111 0 ⎦
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −I
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 11

⎡ ⎤
 A X 0 314 M1 − N1 N1 Bi X Fi X τM M1 0 0 XCT
⎢ 311 di ⎥
⎢ 322 0 −M2 M2 − N2 N2 0 0 τM M2 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ ∗ −Z6 −M3 M3 − N3 N3 0 0 τM M3 0 0 0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ¯ ¯ 345 ¯
346 ¯
347 −P2 Fi − M T8

τM M 4 ¯ 
3410 ¯ 
3411 0 ⎥
⎢ 344 ⎥
⎢ ∗ ∗ ∗ ∗ ¯ 355 ¯ 356
 ¯ 357 M T − N T
 τM M 5 ¯ 3510
 ¯ 3511
 0 ⎥
⎢ 8 8 ⎥
⎢ ¯ 366 N T7 N T8 τM M 6 0 ⎥
3ij =⎢ ∗ ∗ ∗ ∗ ∗  0 0 ⎥<0 (59)
⎢ ¯ 377 ¯ 3710 ¯ 3711 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗  BTi P2 Fi τM M 7   0 ⎥
⎢ ¯ 3810 ¯ 3811 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −γ̄ 2 I τM M 8   0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −τM Z3 0 0 0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ¯ 31010
 0 0 ⎥
⎢ ⎥
⎣ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ¯ 31111
 0 ⎦
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −I
311 = Ai X + Bi Wi + XT ATi + WTi BTi + εZ5 + Z6 , 314 = −Bi Wi − M1 , 322 = −ε(1 − τd )Z5 ,
¯  = P2 Ai − Y i C + AT P2 − CT Y T + εZ1 + Z2 − M 4 − M T , 
 ¯ 345 = P2 Adi − M T + M 4 − N 4 ,  ¯ 346 = −M T + N 4 ,
¯
344
 
i
T  T
i
¯  √ 4
 T ¯ 
5 6
 347 = −Ai P2 B + C Y i Bi − M 7 , 3410 =
T T
τM (Ai P2 − C Y i ), 
T T
√ 3411
¯ 355 = −ε(1 − τd )Z1 +M T + M 5 − N T − N 5 ,  ¯ 356
= τmM (ATi P2 − CT Y Ti ),  5 5
T  ¯ T T T ¯
= M 6 − N 6 + N 5 , 357 = −Adi P2 Bi + M 7 − N 7 , 366 = −Z2 + N 6 + N 6
T T

¯ ¯ 3510 = √τM AT P2 , 
377 = G − 2BTi P2 Bi ,  ¯ 3511
√ √
¯ 3710 = τM B P2 , 
di
¯ 3810 = −√τM FT P, 
¯ 3711 = √τmM BT P2 ,  ¯ 3811 = −√τmM FT P2 ,
= τmM Adi P, 
T T
i i i i
¯ 31010 = −2αP2 + α 2 Z3 , 
 ¯ 31111 = −2αP2 + α 2 Z4 , M1
= XM 1 X,N1 = XN 1 X, M2 = XM 2 X, N2 = XN 2 X,M3 = XM 3 X, N3 = XN 3 X,
Z5 = XZ5 X, Z6 = XZ6 X

Then state x(t ) of the system, state estimation error ex (t ) and FE error e f (t ) are bounded. In this case, the gains of observer and
active FTC can be obtained as follows:
Li = P−1 
2 Y i, Ki = Wi X−1 (60)

Proof: Consider the following Lyapunov function:

t t 0 t
V  = xT (t )P1 x(t ) + eTx (t )P2 ex (t ) + ε eTx (t )Z1 ex (t )ds + eTx (t )Z2 ex (t )ds + ėTx (t )Z3 ėx (t )dsdθ
t−τ (t ) t−τM −τM t+θ
−τm t t t
+ ėTx (t )Z4 ėx (t )dsdθ + eTf (t )−1 ef (t ) + ε xT (t )Z5 x(t )ds + xT (t )Z6 x(t )ds (61)
−τM t+θ t−τ (t ) t−τM

Taking the derivation of V  , we consider the following inequality:

V̇  + yT (t )y(t ) − γ̄ 2 wT (t )w(t ) ≤ ẋT (t )P1 x(t ) + xT (t )P1 ẋ(t ) + eTx (t )P2 ėx (t ) + ėTx (t )P2 ex (t ) + εeTx (t )Z1 ex (t )
− ε(1 − τd )eTx (t − τ (t ))Z1 ex (t − τ (t )) + eTx (t )Z2 ex (t ) − eTx (t − τM )Z2 ex (t − τM )
t t−τ (t )
+ τM ėTx (t )Z3 ėx (t ) − ėTx (t )Z3 ėx (t )ds + τmM ėTx (t )Z4 ėx (t ) − ėTx (t )Z4 ėx (t )ds
t−τM t−τM

+ εxT (t )Z5 x(t ) − ε(1 − τd )xT (t − τ (t ))Z5 x(t − τ (t )) + xT (t )Z6 x(t ) − xT (t − τM )Z6 x(t − τM )
+ ėTf (t )−1 ef (t ) + eTf (t )−1 ėf (t ) + yT (t )y(t ) − γ̄ 2 wT (t )w(t ) (62)

in which
ėTf (t )−1 ef (t ) + eTf (t )−1 ėf (t )
˙ ˙
= (f̂(t ) − ḟ(t ))T −1 ef (t ) + ef (t )T −1 (f̂(t ) − ḟ(t ))
(63)
= [−R1C(ėx (t ) + ex (t )) − ḟ(t )]T −1 ef (t ) + ef (t )T −1 [−R1C(ėx (t ) + ex (t )) − ḟ(t )]
= −2ef (t )T R1 (h)C(ėx (t ) + ex (t )) − 2ef (t )T −1 ḟ(t )

From Lemma 2.1 and Assumption 2.1, we can obtain that

− 2ef (t )T −1 ḟ(t ) ≤ eTf (t )Gef (t ) + ḟ T (t )−1 G−1 −1 ḟ(t ) ≤ eTf (t )Gef (t ) + δ (64)
12 H. LI ET AL.

where δ = fmax
2
λmax (−1 G−1 −1 ) > 0
So we can get that
t
T ¯ 1 ξ  (t ) + τM ėT (t )Z3 ėx (t ) −
V̇  + yT (t )y(t ) − γ̄ 2 wT (t )w(t ) ≤ ξ  (t ) ėTx (t )Z3 ėx (t )ds
x
t−τM
t−τ (t )
+ τmM ėTx (t )Z4 ėx (t ) − ėTx (t )Z4 ėx (t )ds + δ (65)
t−τM

where
 T
ξ  (t ) = xT (t ) xT (t − τ (t )) xT (t − τM ) eTx (t ) eTx (t − τ (t )) eTx (t − τM ) eTf (t ) wT (t )
⎡ ⎤
¯ 111 P1 Ad (h) 0 −P1 B(h)K(h) 0 0 P1 B(h) P1 F(h)
⎢ ∗ ¯ 122 ⎥
⎢ 0 0 0 0 0 0 ⎥
⎢ ∗ ∗ −Z6 ⎥
⎢ 0 0 0 0 0 ⎥
⎢ ¯ 144 ¯ ⎥
¯1 =⎢ ⎢ ∗ ∗ ∗  P A (h) 0  −P F(h) ⎥
 2 d 147 2
⎥
⎢ ∗ ∗ ∗ ∗ ¯
155 0 −Ad (h)P2 B(h)
T
0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ −Z2 0 0 ⎥
⎢ ⎥
⎣ ∗ ∗ ∗ ∗ ∗ ∗ ¯
177 B (h)P2 F(h)⎦
T

∗ ∗ ∗ ∗ ∗ ∗ ∗ −γ̄ 2 I
¯ 111 = P1 (A(h) + B(h)K(h)) + (A(h) + B(h)K(h))T P1 + εZ5 + Z6 + CT C

¯ 122 = −ε(1 − τd )Z5



¯ 144 = P2 (A(h) − L(h)C) + (A(h) − L(h)C)T P2 + εZ1 + Z2



¯ 155 = −ε(1 − τd )Z1



¯ 147 = −(A(h) − L(h)C)T P2 B(h)



¯ 177 = G − 2BT (h)P2 B(h)



From Newton–Leibniz formula, we can get the following equations:

t
2ξ  (t )M [ex (t ) − ex (t − τ (t )) −
T
ėTx (t )ds] = 0 (66)
t−τ (t )
t−τ (t )
2ξ  (t )N  [ex (t − τ (t )) − ex (t − τM )
T
− ėTx (t )ds] = 0 (67)
t−τM

Adding (66) and (67) to (65), we can obtain that

V̇  + yT (t )y(t ) − γ̄ 2 wT (t )w(t )
t t−τ (t )
T ¯ 1 ξ  (t ) + τM ėT (t )Z3 ėx (t ) −
≤ ξ  (t ) ėTx (t )Z3 ėx (t )ds + τmM ėTx (t )Z4 ėx (t ) − ėTx (t )Z4 ėx (t )ds
x
t−τM t−τM
t−τ (t ) t
−2ξ  (t )N  [ex (t − τ (t )) − ex (t − τM ) − ėTx (t )ds] − 2ξ  (t )M [ex (t ) − ex (t − τ (t )) −
T T
ėTx (t )ds] + δ
t−τM t−τ (t )
T ¯ 1 ξ  (t ) + τM ξ  T (t ) T Z3  1 ξ  (t ) + τmM ξ  T (t ) T Z4  1 ξ  (t ) − ξ  T (t )( +  T )ξ  (t )
= ξ (t ) 1 1
t t−τ (t ) t−τ (t )
ėTx (t )Z4 ėx (t )ds + 2ξ  (t )N 
T
− ėTx (t )Z3 ėx (t )ds − ėTx (t )ds
t−τM t−τM t−τM
t
+2ξ  (t )M
T
ėTx (t )ds + δ (68)
t−τ (t )

where
   
 = 0 0 0 M −M + N −N 0 0 , 1 = 0 0 0 A(h) − L(h)C Ad (h) 0 B(h) −F(h)
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 13

We can further get that

V̇  + yT (t )y(t ) − γ̄ 2 wT (t )w(t )
t t−τ (t ) t−τ (t ) t
T ¯ 2 ξ  (t ) −
≤ ξ  (t ) ėTx (t )Z3 ėx (t )ds − ėTx (t )Z4 ėx (t )ds + 2ξ  (t )N 
T
ėTx (t )ds + 2ξ  (t )M
T
ėTx (t )ds + δ
t−τM t−τM t−τM t−τ (t )
t t−τ (t ) t−τ (t ) t
T ¯ 2 ξ  (t ) −
= ξ  (t ) ėTx (t )Z3 ėx (t )ds − ėTx (t )(Z3 + Z4 )ėx (t )ds + 2ξ  (t )N 
T
ėTx (t )ds + 2ξ  (t )M
T
ėTx (t )ds + δ
t−τ (t ) t−τM t−τM t−τ (t )
 t t t
≤ (1 τM )( ¯ 2 ξ  (t )ds+2τM ξ  (t )M
ξ  (t )
T T
ėTx (t )ds − τM
ėTx (t )Z3 ėx (t )ds)
t−τ (t ) t−τ (t ) t−τ (t )
 t−τ (t ) t−τ (t ) t−τ (t )
+(1 τM )( ¯ 2 ξ  (t )ds+2τM ξ  T (t )N 
ξ  (t )
T
ėTx (t )ds − τM ėTx (t )(Z3 + Z4 )ėx (t )ds) + δ
t−τM
  T     t−τM
t−τ (t )  
t−τM
T    
 t
ξ (t ) ¯ 2 τM M ξ (t )  ξ (t ) ¯2 τM N  ξ (t )
≤ (1 τM ) ds + (1 τM ) ds +δ
t−τ (t ) ėx (t ) ∗ −τM Z3 ėx (t ) t−τM ėx (t ) ∗ −τM (Z3 + Z4 ) ėx (t )
(69)
where ¯2 =¯ 1 + τM  T Z3  1 + τmM  T Z4  1 −  −  T
1 1
¯ 2 τM M
  ¯ τM N ξ  (t )
¯
Denoting 3 = [ ¯ 
],  3 = [ 2 ], β̄(t ) = [ ¯ 3 ) > 0, ε4 = λmin (−¯  3 ) > 0, we can
], ε = λmin (−
∗ −τM Z3 ∗ −τM (Z3 + Z4 ) ėx (t ) 3
obtain that

 t  t−τ (t )
V̇  + yT (t )y(t ) − γ̄ 2 wT (t )w(t ) ≤ −(ε3 τM ) β̄(t )T β̄(t )ds − (ε4 τM ) β̄(t )T β̄(t )ds + δ
t−τ (t ) t−τM
 t
≤ [(−ε3 − ε4 ) τM ] β̄(t )T β̄(t )ds + δ
t−τM
2
≤ −(ε3 + ε4 ) β̄(t ) +δ (70)

If β̄(t )2 ≥ δ/(ε3 + ε4 ) holds, then it has V̇  + yT (t )y(t ) − γ̄ 2 wT (t )w(t ) ≤ 0. We can conclude that (52) holds for all L > 0
and w(t ) ∈ L2 [0, ∞) under zero initial conditions. In addition, by choosing the same Lyapunov function as (61) and the similar
line in the earlier deduction, we can easily obtain that the time derivative of V  along the closed-loop fuzzy system (50) (when
w(t ) = 0) satisfies V̇ < 0, which indicates the asymptotic stability of systems (50) (when w(t ) = 0).
From ε3 = λmin (− ¯ 3 ) > 0 and ε4 = λmin (−¯  3 ) > 0, we can conclude that

⎡ ⎤
¯ 311 P1 Ad (h) 0
 ¯ 314 M 1 − N 1 N 1 P1 B(h)
 P1 F(h) τM M 1 0 0
⎢ ⎥
⎢ ∗ ¯ 322
 0 −M 2 M 2 − N 2 N 2 0 0 τM M 2 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ ∗ ∗ −Z6 −M 3 M 3 − N 3 N 3 0 0 τM M 3 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ¯ 344
 ¯ 345
 ¯ 346 
 ¯ 347 −P2 F(h) − M T τM M 4 ¯ 3410
 ¯ 3411
 ⎥
⎢ 8 ⎥
⎢ ¯ 355 ¯ 356 
¯ 357 ¯ 3510 ¯ 3511 ⎥
⎢ ∗ ∗ ∗ ∗   M T8 − N T8 τM M 5   ⎥
⎢ ⎥
¯3 =⎢
 ⎢ ∗ ∗ ∗ ∗ ∗ ¯ 366
 N T7 N T8 τM M 6 0 0
⎥
⎥<0 (71)
⎢ ⎥
⎢ ¯ 377 ¯ 3710 ¯ 3711 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗  BT (h)P2 F(h) τM M 7   ⎥
⎢ ⎥
⎢ ¯ 3810 ¯ 3811 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −γ̄ 2 I τM M 8   ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −τM Z3 ⎥
⎢ 0 0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −P2 Z−1 ⎥
⎣ 3 P2 0 ⎦
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −P2 Z−1
4 P2
14 H. LI ET AL.

⎡ ⎤
¯ 311 P1 Ad (h)
 0 ¯ 314 M 1 − N 1 N 1 P1 B(h) P1 F(h) τM N 1 0 0
⎢ ∗ ¯ 322 0 −M 2 M 2 − N 2 N 2 0 0 τM N 2 0 0 ⎥
⎢ ⎥
⎢ ∗ ∗ −Z6 −M 3 M 3 − N 3 N 3 0 0 
τM N 3 0 0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ¯ 344 ¯ 345  ¯ 346 ¯ 347 −P2 F(h) − M T
 τM N 4 ¯ 3410 ¯ 3411 ⎥
⎢ 8 ⎥
⎢ ¯ ¯ 356 ¯ 357 ¯ 3510 ¯ 3511 ⎥
⎢ ∗ ∗ ∗ ∗ 355   M T − N T τM N 5   ⎥
¯ =⎢ ⎥
8 8
 ⎢ ∗ ∗ ∗ ∗ ∗ ¯
366 N T7 N T8 τM N 6 0 0 ⎥ < 0 (72)
3
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ¯ 377
 BT (h)P2 F(h) τM N 7 ¯ 3710 ¯ 3711 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −γ̄ 2 I τM N 8 ¯ 3810 ¯ 3811 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −τM (Z3 + Z4 ) 0 0 ⎥
⎢ ⎥
⎣ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −1
−P2 Z3 P2 0 ⎦
−1
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −P2 Z4 P2

where

¯ 311 = P1 (A(h) + B(h)K(h)) + (A(h) + B(h)K(h))T P1 + εZ5 + Z6 + CT C, 

 ¯ 314
 ¯ ¯
= −P1 B(h)K(h) − M 1 , 322 = −ε(1 − τd )Z5 , 344 = P2 (A(h) − L(h)C) + (A(h) − L(h)C)T P2 + εZ1
+Z2 − M 4 − M T4 ,  ¯ 345 = P2 Ad (h) − M T + M 4 − N 4 ,
¯ 3410 = √τM (A(h) − L(h)C)T P2 , 
5
¯ 346 = −M + N 4 , 
  T  ¯ 347 = −(A(h) − L(h)C)T P2 B(h) − M T ,  ¯ 3411
√ 6
¯
7
T  T  ¯
= τmM (A(h) − L(h)C) P2 , 355 = −ε(1 − τd )Z1 +M 5 + M 5 − N 5 − N 5 , 356
T

¯ 357 = −AT (h)PB(h) + M T − N T , 

= M T6 − N T6 + N 5 ,  ¯ 366 = −Z2 + N T + N 6 ,
¯ 377 = G − 2B (h)PB(h),  √d
¯ 3510 = τM A (h)P2 , 
7
√7
¯ 3511 = τmM AT (h)P,  ¯ 3710
6
 T T
√ ¯ 3711 = √τmM BT (h)P2 , 
d
¯ 3810 = −√τM FT (h)P, 
d
¯ 3811 = −√τmM FT (h)P2
= τM BT (h)P2 , 

We can find that the forms of (71) and (72) are not LMIs. So we are now in a position to transform the conditions (71) and (72)
in terms of LMIs, which can be solved efficiently by using existing solvers such as LMI toolbox in the MATLAB software.
Based on Lemma 2.2, we can obtain that ∃α ∈ R+ such that

− P2 Z−1
3 P2 ≤ −2αP2 + α Z3
2
(73)

− P2 Z−1
4 P2 ≤ −2αP2 + α Z4
2
(74)

Denoting Y (h) = P2 L(h), we can obtain the following LMI forms:

⎡¯ ¯ 314 M 1 − N 1 N 1 ⎤
311 P1 Ad (h) 0  P1 B(h) P1 F(h) τM N 1 0 0
⎢ ∗ ¯ 322 0 −M 2 M 2 − N 2 N 2 0 0 τM N 2 0 0 ⎥
⎢ ⎥
⎢ ∗ ∗ −Z −M 3 M 3 − N 3 N 3 0 0 τM N 3 0 0 ⎥
⎢ 6 ⎥
⎢ ∗ ∗ ∗ ¯ ¯ 345  ¯ 346 ¯
 −P − M T8 τM N 4 ¯ ¯  ⎥
⎢ 344 347 2 F(h) 3410 3411 ⎥
⎢ ∗ ∗ ∗ ∗ ¯
355  ¯ 356 ¯ 357
 M − N T
T
τM N 5 ¯ 3510 ¯ 3511 ⎥
⎢ ⎥
¯ =⎢ ⎥
8 8
 ⎢ ∗ ∗ ∗ ∗ ∗ ¯
366 N T7 N8T
τM N 6 0 0 ⎥<0 (75)
3
⎢ ¯ 377 ¯ 3710 ¯ 3711 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗  B (h)P2 F(h)
T
τM N 7   ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −γ̄ I
2
τM N 8 ¯ 3810 ¯ 3811 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −τM (Z3 + Z4 ) 0 0 ⎥
⎢ ⎥
⎣ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ¯
31010 0 ⎦
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ¯ 31111

 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 15

⎡¯ ¯ 314 M 1 − N 1 N 1 ⎤
311 P1 Ad (h) 0  P1 B(h) P1 F(h) τM M 1 0 0
⎢ ∗ ¯ 322 0 −M 2 M 2 − N 2 N 2 0 0 τM M 2 0 0 ⎥
⎢ ⎥
⎢ ∗ ∗ −Z6 −M 3 M 3 − N 3 N 3 0 0 τM M 3 0 0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ¯ ¯ 345  ¯ 346 ¯
 −P − M T8 τM M 4 ¯ ¯ ⎥
⎢ 344 347 2 F(h) 3410 3411 ⎥
⎢ ∗ ∗ ∗ ∗ ¯
355  ¯ 356 ¯ 357
 M T − N T τM M 5 ¯ 3510 ¯ 3511 ⎥
⎢ ⎥
¯3 =⎢ ⎥
8 8
 ⎢ ∗ ∗ ∗ ∗ ∗ ¯
366 N T7 N8T
τM M 6 0 0 ⎥<0 (76)
⎢ ¯ 377 ¯ 3710 ¯ 3711 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗  B (h)P2 F(h) τM M 7
T
  ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −γ̄ 2 I τM M 8 ¯ 3810 ¯
3811 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −τM Z3 0 0 ⎥
⎢ ⎥
⎣ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ¯ 31010
 0 ⎦
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ¯ 31111


where

¯  = P2 A(h) − Y (h)C + AT (h)P2 − CT Y T (h) + εZ1 + Z2 − M 4 − M T , 

 ¯
344 4 347
T T T ¯ 
= −A (h)P2 B(h) + C Y (h)B(h) − M 7 , 3410
T
√ ¯ √ ¯ 31010 = −2αP2 + α 2 Z3 , 
¯ 31111 = −2αP2 + α 2 Z4 .
= τM (AT (h)P2 − CT Y (h)),  τmM (AT (h)P2 − CT Y (h)), 
T T
3411 =

Pre- and post-multiplying inequalities (75) and (76) byP = diag([ P−1 −1 −1
1 P1 P1 I I I I I I I I ]), it follows that (75) and (76) are
equivalent to the following inequalities:
⎡ ⎤
311 Ad (h)X 0 314 M1 − N1 N1 B(h)X F(h)X τM M1 0 0 XCT
⎢ ⎥
⎢ ∗ 322 0 −M2 M2 − N2 N2 0 0 τM M2 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ ∗ ∗ −Z6 −M3 M3 − N3 N3 0 0 τM M3 0 0 0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ¯ ¯ 345 ¯
346 ¯
347 −P2 F(h) − M T8

τM M 4 ¯ 
3410 ¯ 
3411 0 ⎥
⎢ 344 ⎥
⎢ ∗ ∗ ∗ ∗ ¯ 355 ¯ 356
 ¯ 357
 M T8 − N T8 τM M 5 ¯ 3510
 ¯ 3511
 0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ¯ 366
 T
N T8 τM M 6 0 ⎥
3 = ⎢ N7 0 0 ⎥<0 (77)
⎢ ¯ 377 BT (h)P2 F(h) ¯ 3710 ¯ 3711 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗  τM M 7   0 ⎥
⎢ ¯ 3810 ¯ 3811 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −γ̄ 2 I τM M 8   0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −τM Z3 0 0 0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ¯ 31010
 0 0 ⎥
⎢ ⎥
⎣ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ¯ 31111
 0 ⎦
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −I
⎡ ⎤
 A (h)X 0 314 M1 − N1 N1 B(h)X F(h)X τM N1 0 0 XCT
⎢ 311 d ⎥
⎢ ∗ 322 0 −M2 M2 − N2 N2 0 0 τM N2 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ ∗ ∗ −Z6 −M3 M3 − N3 N3 0 0 τM N3 0 0 0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ¯ ¯ 345 ¯
346 ¯ 
347 −P2 F(h) − M 8 T 
τM N 4 ¯ 
3410 ¯ 
3411 0 ⎥
⎢ 344 ⎥
⎢ ∗ ∗ ∗ ∗ ¯ 355 ¯ 356
 ¯ 357
 M T8 − N T8 τM N 5 ¯ 3510 ¯ 3511
 0 ⎥
 ⎢ ⎥
⎢ ¯ 366 T T 
0 ⎥
3 =⎢ ∗ ∗ ∗ ∗ ∗  N7 N8 τM N 6 0 0 ⎥<0 (78)
⎢ ¯ 377 BT (h)P2 F(h) ¯ 3710 ¯ 3711 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗  τM N 7   0 ⎥
⎢ ¯ 3810 ¯ 3811 ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −γ̄ 2 I τM N 8   0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −τM (Z3 + Z4 ) 0 0 0 ⎥
⎢ ⎥
⎢ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ¯ 31010
 0 0 ⎥
⎢ ⎥
⎣ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ¯ 31111
 0 ⎦
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ −I

where

¯ 3 P ,  3 = P 
3 = P  ¯  P , 311 = A(h)X + B(h)W(h) + XT AT (h) + WT (h)BT (h) + εZ5 + Z6 , 314
3

= −B(h)W(h) − M1 , 322 = −ε(1 − τd )Z5 , M1 = XM 1 X,N1 = XN 1 X,M2 = XM 2 X, N2

= XN 2 X, M3 = XM 3 X,N3 = XN 3 X, Z5 = XZ5 X, Z6 = XZ6 X
16 H. LI ET AL.

If (77) and (78) hold, we can obtain that 5. Simulation results

In this section, two examples are given to demonstrate the effec-

r tiveness of the proposed methods. The T-S fuzzy systems have
 3 = h2i (z(t )) 3ii wide implication on the practical systems. For example, the
i=1 problem of balancing and swing-up of an inverted pendulum

r 
r on a cart is considered in Zhang et al. (2010). The equations of
+ hi (z(t ))h j (z(t ))( 3ij +  3ji ) < 0 (79) the pendulum motion are given by
i=1 i< j
ẋ1 (t ) = x2 (t )

r
(g sin(x1 (t ))−amlx22 (t ) sin(2x1 (t ))/2−a cos(x1 (t ))u(t ))
3 = h2i (z(t ))3ii ẋ2 (t ) = (4l/3−amlcos2 (x1 (t )))
i=1 y(t ) = x1 (t )

r 
r
+ hi (z(t ))h j (z(t ))(3ij + 3ji ) < 0 (80) where x1 (t ) is the angle (in radians) of the pendulum from verti-
i=1 i< j cal, x2 (t ) is the angular velocity (rad/s), u(t ) is the force applied
to the cart (in Newton), g = 9.8 m/s2 is the gravity constant, m
which imply that (54)–(57) hold. The proof is completed. is the pendulum mass, M is the cart mass, 2l is the pendulum
Remark 4.5: length, and a = 1/(m + M).
It can be used in the computer-simulated truck–trailer sys-
In order to further reduce the conservativeness of tem borrowed from Gassara et al. (2010):
stability criteria for the systems with interval time-
v t¯ v t¯ v t¯
varying delay, the derivative of Lyapunov functional ẋ1 (t ) = −a x1 (t ) − (1 − a) x1 (t − τ (t )) + u(t )
involves
t
τM ėTx (t )Z3 ėx (t ) − t−τM ėTx (t )Z3 ėx (t )ds and Lt0 Lt0 lt0
 v t¯ v t¯
t−τ (t ) T
τmM ėx (t )Z4 ėx (t ) − t−τM ėx (t )Z4 ėx (t )ds. However, when
T ẋ2 (t ) = a x1 (t ) + (1 − a) x1 (t − τ (t ))
Lt0 Lt0
analysing the given stability results, some terms are ignored v t¯ v t¯ v t¯
in (Huang, He, and Zhang (2011) and Lin, Wang, Lee, and ẋ3 (t ) = sin[x2 (t ) + a x1 (t ) + (1 − a) x1 (t − τ (t ))]
t0 2L 2L
Chen (2008), or some useful negative integral terms are lost
(e.g. Su, Chen, Lin, & Zhang, 2009). In this paper, we make full The constanta is the retarded coefficient, which satisfies the
use of the information aboutτ (t ) and use the method in Sun conditions a ∈ [0, 1]. The limits 1 and 0 correspond to no delay
et al. (2014). No terms are ignored. Therefore, the results in term and to a completed delay term, respectively.
this paper are obvious to be less conservative than the existing
results in Huang et al. (2011), Lin et al. (2008) and Su et al. Example 5.1
(2009). Based on the practical background, we consider the following
Remark 4.6: T-S fuzzy model with the parameters which are the same with
Gassara et al. (2014) and Sun et al. (2014):
It is hard to solve (8)–(12) (or (38)–(42), (53)–(57)) simulta- ⎧
⎪
⎪ r
neously. We can transform (8), (38) and (53) in Theorems 3.1, ⎪ ) = hi (z(t ))[Ai x(t ) + Adi x(t − τ (t ))
⎨ ẋ(t
3.2 and 4.3 into an optimisation problem (Zhang et al., 2009d): i=1
⎪
⎪ + Bi u(t ) + E i f(t ) + F i w(t )]
⎪
⎩
min η y(t ) = Cx(t )
 
ηI ETi P − R̄i C (81)
s.t. T >0 where
(ETi P − R̄i C) ηI
   
min η −0.1125 −0.0200 −0.1125 −1.5270
  A1 = , A2 =
ηI BTi P2 − R̄i C (82) 1 0 1 0
s.t. >0    
(BTi P2 − R̄i C)
T
ηI −0.0125 −0.0050 −0.0125 −0.2300
Aτ 1 = , Aτ 2 = ,
0 0 0 0
       
In order to make ETi P approximate to R̄i C and BTi P2 approx- 10 10 10 10
B1 = , B2 = , E1 = , E2 = ,
imate to R̄i C with a satisfactory precision, a sufficiently small 10 10 10 10
   
positive scalar ηshould be selected in advance to meet (81) and 0.01 0 0.01 0  
F1 = , F2 = ,C = 0 1
(82). 0.01 0 0.01 0
Remark 4.7:
The pair(Ai , C) is observable. The control input u(t ) is the
For simplification, in Section 4, we make B = E in the T-S unit step function.w(t ) is represented by band-limited white
fuzzy system (3) when we design a fault tolerant controller. So noise and the noise power is 0.001, whose dimension is 2 ×
the conditions of Theorem 4.3 are simplified, which results in the 1. The interval time-varying delay acting on the system state
conservativeness of the developed method. If B = E, the deduc- is given byτ (t ) = 0.2 sin(t ) + 0.3(τm = 0.1, τM = 0.5, τmM =
tion will become more complex. We will consider this condition 0.4, τd = 0.2). The parameter values α = 10, ε = 1, γ = γ̄ =
in the future work. 0.0715 are chosen.
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 17

(a)
2

f
1.5 the estimation of f

the fault and its estimation 0.5

−0.5

−1

−1.5

−2
0 2 4 6 8 10 12 14 16 18 20
time,s
(b) 2

f
1.5 the estimation of f

1
the fault and its estimation

0.5

−0.5

−1

−1.5

−2
3 3.2 3.4 3.6 3.8 4 4.2 4.4 4.6 4.8 5
time,s

Figure . (a) The fault f (t ) and its estimation; (b) partial enlarged figure.

When i = 1, by taking η = 1 × 10eps(−5) and solving the solutions:

conditions in Theorem 3.1, one can obtain the following solu-
tions:  
3.6195 −3.6195
P = 10eps(5) ,
−3.6195 3.7067
   
1.9550 −1.9550 1.1742 −1.2934
P= 10eps(5) , Q1 = 10eps(5) ,
−1.9550 2.0377 −1.2934 1.5238
   
9.1601 −9.0877 1.2865 −1.2944
Q1 = 10eps(4) , Q2 = 10eps(5) ,
−9.0877 9.5572 −1.2944 1.3502
   
0.9799 −0.9661 5.9329 −5.9654
Q2 = 10eps(5) , Z = 10eps(4) ,
−0.9661 1.0193 −5.9654 6.0750
   
3.6387 −3.6620 6.0424 −6.0315
Z= 10eps(4) , R = 10eps(4) ,
−3.6620 3.7637 −6.0315 6.1375
     
3.6338 −3.6429 8.7204 0.1631
R= 10eps(4) , R̄2 = 10eps(3) , L2 =
−3.6429 3.7556 0 2.8769
   
8.2712 1.7831
R̄1 = 10eps(3) , L1 =
0 2.8884 The membership functions for rules 1 and 2 are the follow-
ing:

When i = 2, by taking η = 1 × 10eps(−5) and solving the  

conditions in Theorem 3.1, one can obtain the following h1 (x2 (t )) = 1 − x2 (t )2 2.25, h2 (x2 (t )) = x2 (t )2 2.25
18 H. LI ET AL.
1

x1
0.8
x2

0.6

0.4

0.2

the state x
0

−0.2

−0.4

−0.6

−0.8

−1
0 2 4 6 8 10 12 14 16 18 20
time,s

Figure . The state x(t ) under the nominal control law when the system is fault free.

Assume that time-varying fault f (t ) is created as When i = 2, it can obtain the following solutions:

    
0 0 ≤t ≤3 238.1349 123.9767 6.6387 −6.6363
f (t ) = P1 = , P2 = ,
sin(3t − 9) + sin(4t − 12) 3 < t ≤ 20 123.9767 372.5113 −6.6363 6.8770
   
−0.0024 −0.0000 0.1318 −0.2859
Z1 = , Z2 = ,
0.01 0 −0.0000 −0.0024 −0.2859 0.6066
By taking the learning rate  = [ ] and the sampling    
0.01 0 0.7614 −0.7479 −0.0024 −0.0000
Z3 = , Z4 = ,
time 0.1 s, using the FAFE algorithm, the simulation result is −0.7479 0.7820 −0.0000 −0.0024
shown in Figure 2.  
0.4685 −0.0786
Figure 2 shows the system fault f (t ) and its estimation fˆ (t ). Z5 = 10eps(−3)
−0.0786 −0.3643
,
From Figure 2, we can see that the fault appears at 3 s. The FE  
0.8896 0.0422
fˆ (t ) can track the fault f (t )well and the estimation error is very Z6 = 10eps(−3) ,
0.0422 −0.1203
small. We can conclude that the FE is realised from the simula-    
tion result when the delay is interval time-varying. Therefore, 0.2407 −0.9708
R̄2 = , L2 = ,
Theorem 3.1 is valid. 0 1.7229
 
To proof the validity of fault accommodation controller, the −0.3344 0.5628
process of simulation is classified into three parts. Consider sys- K2 =
0.5698 0.2967
tem with the parameters which are the same with Example 5.1.
Part 1: we consider a nominal control law
The state of the system under the nominal control law is
shown in Figure 3.

r
Figure 3 shows the state x(t ) under the nominal control law
u(t ) = hi (z(t ))Ki x̂(t )
when the system is fault free. From Figure 3, we can see that the
i=1
nominal control law can make the state stable when the system
is fault free.
When the system is fault-free, we use the nominal control law Part 2: When the system includes the fault, we still use the
to control the system. When i = 1, it can obtain the following nominal control law to control the system. Assume that time-
solutions: varying fault f (t ) is created as
   
0.4608 0.0089 122.7035 −122.7011 
P1 = , P2 = , 0 0 ≤t ≤3
0.0089 0.0054 −122.7011 122.9403 f (t ) =
    200 sin(3t − 9) + 200 sin(4t − 12) 3 < t ≤ 20
2.1393 −2.1469 4.1810 −4.3444
Z1 = , Z2 = ,
−2.1469 2.1500 −4.3444 4.5157
    By taking the learning rate  = [
10 0
] and the sampling time
14.5077 −14.4942 4.3500 −4.3551 10 0
Z3 = , Z4 = ,
−14.4942 14.5289 −4.3551 4.3562 0.1 s, using the FAFE algorithm, the fault and its estimation are
   
0.4878 −0.1026 0.4881 −0.1119 shown in Figure 4. The estimation error of the fault is shown in
Z5 = , Z6 = ,
−0.1026 1.6978 −0.1119 0.7441 Figure 5. The state of the system under the nominal control law
      is shown in Figure 6. The estimation error of the state is shown
0.2392 0.0232 −0.6815 −0.0054
R̄1 = , L1 = , K1 = in Figure 7.
0 1.1486 0.6915 0.0134
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 19

(a)
400

f
the estimation of f
300

200

the falut and its estimation 100

−100

−200

−300

−400
0 2 4 6 8 10 12 14 16 18 20
time,s
(b) 400

f
the estimation of f

390

380
the falut and its estimation

370

360

350

340
3.3 3.32 3.34 3.36 3.38 3.4 3.42 3.44 3.46 3.48 3.5
time,s

Figure . (a) The fault f (t ) and its estimation; (b) partial enlarged figure.

From Figures 4–7, we can see that the scope of the FE error is Part 3: When the system includes the fault, we use the active
−1.8 to 1.8, the scope of the state x(t ) is −10 to 10 and the scope FTC
of the state estimation error is −0.5 to 0.5. We can conclude that
the FE error has a bad influence on the state estimation. If the 
r

nominal control law still be used to control the system including u(t ) = hi (z(t ))Ki x̂(t ) − fˆ (t )
fault, we cannot make the state be in good performance. i=1

1.5

1
the estimation error of fault

0.5

−0.5

−1

−1.5

−2
0 2 4 6 8 10 12 14 16 18 20
time,s

Figure . The estimation error of the fault f (t ).

20 H. LI ET AL.

(a)
20

x1
15 x2
the estimation of x1
the estimation of x2
10
the state and its estimation

−5

−10

−15

−20
0 2 4 6 8 10 12 14 16 18 20
time,s
(b) 4

x1
3
x2
the estimation of x1
2 the estimation of x2

1
the state and its estimation

−1

−2

−3

−4

−5

−6
3 3.1 3.2 3.3 3.4 3.5 3.6 3.7
time,s

Figure . (a) The state x(t ) under the nominal control law when the system includes fault; (b) partial enlarged figure.

0.5
the estimation error of x1
0.4 the estimation error of x2

0.3

0.2
the estimation error of state

0.1

−0.1

−0.2

−0.3

−0.4

−0.5
0 2 4 6 8 10 12 14 16 18 20
time,s

Figure . The estimation error of the state x(t ) under the nominal control law when the system includes fault.
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 21

(a)
20

15 x1
x2
the estimation of x1
the estimation of x2
10

the state and its estimation

5

−5

−10

−15

−20
0 2 4 6 8 10 12 14 16 18 20
time,s
(b)
0.5

0.4
x1
x2
0.3 the estimation of x1
the estimation of x2

0.2
the state and its estimation

0.1

−0.1

−0.2

−0.3

−0.4

−0.5
3 3.5 4 4.5 5 5.5 6 6.5 7
time,s

Figure . (a) The state x(t ) under the active fault tolerant control when the system includes fault; (b) partial enlarged figure.

When i = 1, solving the conditions in Theorem 4.3, one can When i = 2, solving the conditions in Theorem 4.3, one can
obtain the following solutions: obtain the following solutions:
   
    0.1591 −0.0000 2.5227 −2.5227
P1 = , P2 = 10eps(3) ,
0.1380 0.0000 2.9871 −2.9871 −0.0000 2.2477 −2.5227 2.6083
P1 = , P2 = 10eps(5) ,    
0.0000 1.1297 −2.9871 2.9880 578.1658 −652.0627 615.0615 −631.8623
  Z1 = , Z2 =
−652.0627 780.6130 −631.8623 676.1559
1.2883 −1.2932    
Z1 = 10eps(5) , 377.0706 −375.2059 337.5367 −341.6409
−1.2932 1.2984 Z3 = .Z4 = ,
  −375.2059 385.0473 −341.6409 352.6667
1.4113 −1.4125    
Z2 = 10eps(5) , =
6.6164 4.2516
, Z6 =
6.6143 4.2489
,
−1.4125 1.4140 Z5
4.2516 6.5468 4.2489 6.5434
       
5.2189 −5.2175 85.5833 −0.1387 −3.8483 −55.9635
Z3 = 10eps(4) , R̄2 = , L2 = , K2 =
−5.2175 5.2173 0 2.5892 −3.9614 0
   
5.2045 −5.2046 8.3779 5.5050 10 0
Z4 = 10eps(4) , Z5 = , By taking the learning rate  = [ ] and the sampling time
−5.2046 5.2055 5.5050 8.3736 10 0
    0.1 s, using the FAFE algorithm, the state of the system under the
8.3758 5.5049 87.6173
Z6 = , R̄1 = , active FTC is shown in Figure 8.
5.5049 8.3736 0 From Figure 8, we can see that the scope of the state x(t ) is
    −0.5 to 0.5. Comparing Figure 6 with Figure 8, we can conclude
1.4137 −4.2192 −35.9061
L1 = , K1 = that the active FTC can make the state be better performance
2.5542 −4.3851 0
than the nominal control law when the system includes fault.
22 H. LI ET AL.

The state in Figure 8 is similar to Figure 3 and the error is from Shouping Guan received his PhD degree in automatic control from North-
the FE error. If the FE error is reduced, the state in Figure 8 is eastern University in 1995. He is currently a professor in Northeastern Uni-
closer to the state in Figure 3. In conclusion, when actuator fail- versity, and his research interests are in process optimal control, intelligent
control, monitoring and diagnosis in biochemical process, and other batch-
ures occur, the stability of the closed-loop system with the nom- like industrial control applications.
inal fuzzy controller is not even guaranteed, whereas the closed-
loop system using the active fault tolerant fuzzy controller still
operates correctly and maintains. ORCID
Hui Li http://orcid.org/0000-0001-8378-1698

6. Conclusion
This paper studies the problem of FE and the active FTC via References
adaptive observer-based controller for T-S fuzzy systems with Ahmadizadeh, S., Zarei, J., & Karimi, H.R. (2014). A robust fault detection
interval time-varying delay. By constructing a new Lyapunov– design for uncertain Takagi-Sugeno models with unknown inputs and
Krasovskii functional, the sufficient delay-dependent LMIs time-varying delays, Nonlinear Analysis: Hybrid Systems, 11, 98–117.
An, J., Wen, G., Gan, N., & Li, R. (2011). A delay-derivative-dependent
including the information of the lower and upper delay bounds
approach to robust H filter design for uncertain systems with time-
have been established to guarantee the existence of adaptive varying distributed delays, Journal of the Franklin Institute, 348(2), 179–
observer-based controller. Compared with the constant delay 200.
and time-varying delay, the interval time-varying delay is the Aouaouda, S., Chadli, M., Shi, P., & Karimi, H.R. (2015). Discrete-time
less conservative form, i.e. time delay that varies in an interval H − /H  sensor fault detection observer design for nonlinear systems
with parameter uncertainty. International Journal of Robust and Non-
in which the lower bound is not restricted to be zero. Further-
linear Control, 25(3), 339–361.
more, we make full use of the information of the delay and no Balasubramaniam, P., Krishnasamy, R., & Rakkiyappan, R. (2011). Delay-
terms are ignored when the stability of the system is analysed. interval-dependent robust stability results for uncertain stochastic sys-
Two examples are given to show that our method is valid to esti- tems with Markovian jumping parameters. Nonlinear Analysis: Hybrid
mate the fault using the FAFE algorithm, and the closed-loop Systems, 5(4), 681–691.
Du, D., Jiang, B., Shi, P., & Karimi, H.R. (2014). Fault detection for
system using the active fault tolerant fuzzy controller still oper-
continuous-time switched systems under asynchronous switching.
ates correctly and maintains. International Journal of Robust and Nonlinear Control, 24(11), 1694–
1706.
Gassara, H., El Hajjaji, A., & Chaabane, M. (2014). Adaptive fault toler-
Acknowledgments ant control design for Takagi-Sugeno fuzzy systems with interval time-
The authors would like to thank the AE and reviewers for their care- varying delay. Optimal Control Applications and Methods, 35(5), 609–
ful reading, constructive comments and fruitful suggestions that certainly 625.
improved the quality of the paper. Gassara, H., Hajjaji, A.E., & Chaabane, M. (2010). Observer-based robust
H reliable control for uncertain T-S fuzzy systems with state time
delay. IEEE Transactions on Fuzzy Systems, 18(6), 1027–1040.
Disclosure statement He, Y., Wu, M., She, J.-H., & Liu, G.-P. (2004). Delay-dependent robust sta-
No potential conflict of interest was reported by the authors. bility criteria for uncertain neutral systems with mixed delays. Systems
& Control Letters, 51(1), 57–65.
Huang, S.-J., He, X.-Q., & Zhang, N.-N. (2011). New results on H filter
Funding
design for nonlinear systems with time delay via T-S fuzzy models.
This work was supported by the National Natural Science Foundation of IEEE Transactions on Fuzzy Systems, 19(1), 193–199.
China [grant number 61304219]; National Natural Science Foundation of Huang, S.-J., & Yang, G.-H. (2014). Fault tolerant controller design for T-
China [grant number 61533007; the State Key Laboratory of Synthetical S fuzzy systems with time-varying delay and actuator faults: A k-step
Automation for Process Industries Fundamental Research Funds [grant fault-estimation approach. IEEE Transactions on Fuzzy Systems, 22(6),
number 2013ZCX02-04]. 1526–1540.
Jia, Q., Chen, W., Zhang, Y., & Li, H. (2015). Fault reconstruction and fault-
tolerant control via learning observers in Takagi-Sugeno fuzzy descrip-
Notes on contributors tor systems with time delays. IEEE Transactions on Industrial Electron-
Hui Li was born in 1990. She received her BS degree from Shenyang Uni- ics, 1–1.
versity of Technology in 2013. She received her MS degree from Northeast- Jiang, B., Zhang, K., & Shi, P. (2011). Integrated fault estimation and accom-
ern University in 2015. She is currently a PhD candidate with the Depart- modation design for discrete-time Takagi-Sugeno fuzzy systems with
ment of Automation in Northeastern University. Her main research inter- actuator faults. IEEE Transactions on Fuzzy Systems, 19(2), 291–304.
ests include fault diagnosis and FTC for nonlinear systems. Jiang, X., & Han, Q.-L. (2005). On H control for linear systems with inter-
val time-varying delay. Automatica, 41(12), 2099–2106.
Fuqiang You was born in 1978. He received the BS and MS degrees from Karimi, H.R., Chadli, M., Zhang, L., & Shi, P. (2014). Fault detection, iso-
Harbin Institute of Technology, Harbin, China, in 2000 and 2002, respec- lation, and tolerant control of vehicles using soft computing methods.
tively. He received the PhD degree from Shanghai Jiaotong University, IET Control Theory & Applications, 8(9), 655–657.
Shanghai, China, in 2005. He is an associate professor of School of Informa- Li, X.-J., & Yang, G.-H. (2012). Fault detection filter design for stochastic
tion Science and Engineering in Northeastern University, Shenyang, China. time-delay systems with sensor faults. International Journal of Systems
His primary research interests include fault diagnosis of time-delay systems Science, 43(8), 1504–1518.
and H∞ filtering. Lin, C., Wang, Q.-G., Lee, T.H., & Chen, B. (2008). H filter design for
nonlinear systems with time-delay through T-S fuzzy model approach.
Fuli Wang was born in 1957. He received his PhD degree from Northeastern IEEE Transactions on Fuzzy Systems, 16(3), 739–746.
University in 1987. He is a professor and adviser of School of Information Park, P., Ko, J.W., & Jeong, C. (2011). Reciprocally convex approach to sta-
Science and Engineering in Northeastern University, Shenyang, China. His bility of systems with time-varying delays. Automatica, 47(1), 235–238.
research interests include control application and new remote monitoring Richard, J.-P. (2003). Time-delay systems: An overview of some recent
and control. advances and open problems. Automatica, 39(10), 1667–1694.
 INTERNATIONAL JOURNAL OF SYSTEMS SCIENCE 23

Sakthivel, R., Vadivel, P., Mathiyalagan, K., & Arunkumar, A. (2014). Xie, L., & Souza, C.E.D. (1992). Robust H control for linear systems with
Fault-distribution dependent reliable H control for Takagi-Sugeno norm-bounded time-varying uncertainty. IEEE Transactions on Auto-
fuzzy systems. Journal of Dynamic Systems, Measurement, and Control, matic Control, 37(8), 1188–1191.
136(2), 021021. Yin, S., Wang, G., & Karimi, H.R. (2014). Data-driven design of robust fault
Sami, M., & Patton, R.J. (2013). Active fault tolerant control for nonlin- detection system for wind turbines. Mechatronics, 24(4), 298–306.
ear systems with simultaneous actuator and sensor faults. International Yu, K.-W., & Lien, C.-H. (2008). Stability criteria for uncertain neutral
Journal of Control, Automation and Systems, 11(6), 1149–1161. systems with interval time-varying delays. Chaos, Solitons & Fractals,
Su, X., Shi, P., Wu, L., & Basin, M.V. (2014). Reliable filtering with strict dis- 38(3), 650–657.
sipativity for T-S fuzzy time-delay systems. IEEE Transactions on Cyber- Zhang, K., & Jiang, B. (2008). Analysis and design of adaptive fault estima-
netics, 44(12), 2470–2483. tion for time-varying delay systems. In The 27th Chinese Control Con-
Su, X., Shi, P., Wu, L., & Nguang, S.K. (2013). Induced L2 filtering of ference (pp. 38–42). Kunming: IEEE.
fuzzy stochastic systems with time-varying delays. IEEE Transactions Zhang, K., Jiang, B., & Chen, W. (2009a). An improved adaptive fault esti-
on Cybernetics, 43(4), 1251–1264. mation design for polytopic LPV systems with application to helicopter
Su, Y., Chen, B., Lin, C., & Zhang, H. (2009). A new fuzzy H filter design models. In The 7th Asian Control Conference (pp. 1108–1113). Hong
for nonlinear continuous-time dynamic systems with time-varying Kong: IEEE.
delays. Fuzzy Sets and Systems, 160(24), 3539–3549. Zhang, K., Jiang, B., & Shi, P. (2009b). Fast fault estimation and accommo-
Sun, C., Wang, F., & He, X. (2014). Robust fault estimation for Takagi- dation for dynamical systems. IET Control Theory & Applications, 3(2),
Sugeno nonlinear systems with time-varying state delay. Circuits, Sys- 189–199.
tems, and Signal Processing, 34(2), 641–661. Zhang, K., Jiang, B., & Shi, P. (2009c). A new approach to observer-based
Tong, S., Yang, G., & Zhang, W. (2011). Observer-based fault-tolerant con- fault-tolerant controller design for Takagi-Sugeno fuzzy systems with
trol against sensor failures for fuzzy systems with time delays. Inter- state delay. Circuits, Systems & Signal Processing, 28(5), 679–697.
national Journal of Applied Mathematics and Computer Science, 21(4), Zhang, K., Jiang, B., & Shi, P. (2012). Fault estimation observer design for
617–627. discrete-time Takagi-Sugeno fuzzy systems based on piecewise Lya-
Wang, R., Jing, H., Karimi, H.R., & Chen, N. (2015). Robust fault- punov functions. IEEE Transactions on Fuzzy Systems, 20(1), 192–200.
tolerant H control of active suspension systems with finite- Zhang, K., Jiang, B., & Shumsky, A. (2009d). A new criterion of fault estima-
frequency constraint. Mechanical Systems and Signal Processing, 62–63, tion for neutral delay systems using adaptive observer. Acta Automatica
341–355. Sinica, 35(1), 85–91.
Wang, Y., Xie, L., & de Souza, C.E. (1992). Robust control of a class Zhang, K., Jiang, B., & Staroswiecki, M. (2010). Dynamic output feedback
of uncertain nonlinear systems. Systems & Control Letters, 19(2), fault tolerant controller design for Takagi-Sugeno fuzzy systems with
139–149. actuator faults. IEEE Transactions on Fuzzy Systems, 18(1), 194–201.