Applied Mathematical Modelling 65 (2019) 415–427

Contents lists available at ScienceDirect

Applied Mathematical Modelling

journal homepage: www.elsevier.com/locate/apm

Sampled-data-based dissipative control of T-S fuzzy systemsR

Hong-Bing Zeng a, Kok Lay Teo b, Yong He c, Wei Wang a,b,∗
a
School of Electrical and Information Engineering, Hunan University of Technology, Zhuzhou 412007, China
b
Department of Mathematics and Statistics, Curtin University, Perth, WA 6102, Australia
c
School of Automation, China University of Geosciences, Wuhan 430074, China

a r t i c l e i n f o a b s t r a c t

Article history: This paper investigates the problem of sampled-data-based dissipative control of Takagi–
Received 10 February 2018 Sugeno (T-S) fuzzy systems. The sampling period is assumed to be varying within an inter-
Revised 18 July 2018
val. By making full use of realistic information about the whole sampling interval, we pro-
Accepted 14 August 2018
pose a novel complete sampling-interval-dependent looped function, which depends not
Available online 24 August 2018
only on the interval from x(t) to x(tk ) but also on the interval from x(t) to x(tk+1 ). Based on
Keywords: this function, a sufficient condition is established, ensuring the considered system strictly
Fuzzy systems (Q, S, R ) − γ -dissipative. Furthermore, a approach to design desired sampled-data con-
Dissipative control trollers is proposed. It is applied to solve the problems of a truck-trailer and an inverted
Sampled-data control pendulum. From the results obtained, we see that this approach is highly effective.
Looped function
© 2018 Elsevier Inc. All rights reserved.

1. Introduction

The signal produced by each sensor is indispensable to be sampled before it can be processed by a digital device.
This difficult task is now achievable due to the advancement of high-speed computers. As a consequence, sampled-data
control has attracted considerable interest among control community since the last decades [1–3]. Many interesting results
have been reported in the literature. Some examples are given below. In [4], an approach is proposed for the design of
a robust H∞ filter for uncertain systems based on sampled measurements such that a given H∞ performance index is
satisfied. In [5], an input delay approach is put forward to investigate the robust stabilization of sampled-data systems.
Based on the framework of the input delay approach, a time-dependent Lyapunov functional approach is introduced to
investigate the stability of sampled-data systems in [6]. In [7], improved conditions are obtained for linear sampled-data
systems through the use of Wirtinger’s inequality and a discontinues Lyapunov function. In [8], a procedure for the design
of a robust H∞ sampled-data controller is proposed for sampled-data systems based on probabilistic sampling. In [9], a
stochastic sampled-data approach is developed for the design of a distributed H∞ filter in sensor networks. Recently, a free
matrix-based inequality approach proposed in [10] and [11] was applied in [12] to obtain a sampled-data synchronization
control for a class of chaotic systems with much improved efficiency. In [13], the stability of uncertain sampled-data
systems with incremental delay is investigated by using looped-functions. In [14], a free-matrix-based time-dependent

R
This work was supported by the National Natural Science Foundation of China (61741308, 61672225, 61703153), the National Science Fund of Hunan
Province (2015JJ5021), the Scientific Research Fund of Hunan Provincial Education Department (15B067,16C0475), and the Aid Program for Science and
Technology Innovative Research Team in Higher Educational Institutions of Hunan Province.
∗
Corresponding author.
E-mail addresses: [email protected] (H.-B. Zeng), [email protected] (K.L. Teo), [email protected] (Y. He), [email protected] (W. Wang).

https://doi.org/10.1016/j.apm.2018.08.012
0307-904X/© 2018 Elsevier Inc. All rights reserved.
416 H.-B. Zeng et al. / Applied Mathematical Modelling 65 (2019) 415–427

discontinuous Lyapunov approach is proposed to investigate the stability of sampled-data system. In [15], an improved
stability condition is established by means of a new two-sided looped function.
On the other hand, the use of Takagi–Sugeno (T-S) fuzzy models is popular amongst researchers. A main reason is that a
nonlinear system can be described rather accurately as a set of local linear models mingled together via fuzzy membership
functions. As a consequence, it can be analyzed by methods similar to those available from linear system theory (see
[16–19] and the references therein). In previous literature, it is observed that a lot of effort has been made to theoretical re-
search and application techniques of T-S fuzzy systems. For example, the questions on the stabilization of T-S fuzzy systems
is investigated in [20–23], and the design of H∞ controller for T-S fuzzy systems is studied in [24]. For the problems of
passivity and passification of stochastic T-S fuzzy systems, they are being considered in [25]. In [26], the design of H∞ filter
for T-S fuzzy systems is investigated. The fault detection of T-S fuzzy systems is under consideration in [27–29]. By using
the input delay approach, the stabilization of T-S fuzzy systems with nonuniform uncertain sampling is studied in [30]. In
[31], a method for the design of robust H∞ sampled-data controller is devised for uncertain T-S fuzzy systems. In [32], the
stabilization of output-feedback polynomial fuzzy-model-based control systems is investigated. In [33], the problem of H∞
tracking control is investigated for sampled-data networked control system based on T-S fuzzy model. In [34], the problem
of output-feedback control is addressed for T-S fuzzy delta operator systems with time-varying delay via an input-output
approach. In [35], the focus is on the mixed H∞ and passivity-based control for a class of stochastic T-S fuzzy systems
with Markovian switching and probabilistic time-varying delays. In [36], an approach to design sampled-data-based state
feedback controller is presented for a class of Takagi–Sugeno fuzzy systems. In [37], the problem of finite-time resilient
reliable sampled-data control is investigated for fuzzy systems with randomly occurring uncertainties.
As is well known, dissipative theory plays an important role in the analysis and design of control systems adopting
input-output descriptions through energy-related consideration [38]. It serves as a powerful tool in characterizing crucial
system behaviours and therefore attracts an increasing interest amongst researchers in the control community. For instance,
sufficient conditions are derived in [39] to guarantee the dissipativity of singular delay systems. In [40] and [41], the ques-
tions on dissipativity for local field neural networks and static neural networks with time-delay are analyzed respectively.
An augmented system approach is presented for the design of static output-feedback dissipative control in [42]. In [43], the
problem finite-time dissipative control is investigated for a class of Markovian jump systems with actuator saturation and
partly unknown transition rates. In [44], a decentralized event-triggered dissipative control is developed for systems with
the entries of the system outputs having several physical properties. In [45], the problem of reliable dissipative control is
studied for T-S fuzzy systems with Markov jumping parameters. In [46], a similar problem is investigated for T-S fuzzy
systems with variable sampling, where sufficient conditions are derived and a design method for sampled-data controller
is provided. However, in the construction of the Lyaponov functionals, the focus is only on the interval from x(t) to x(tk ),
while the information on the interval from x(t) to x(tk+1 ) is completely neglected. As a result, the outcomes are rather
conservative, which motivates current research.
In this paper, we investigate the question on dissipative control of T-S fuzzy systems through the use of sampled-data
control. The contributions can be summarized as follows: (1) a novel complete sampling-interval-dependent looped function
is proposed, which depends not only on the interval from x(t) to x(tk ) but also on the interval from x(t) to x(tk+1 ); (2) an
improved free-matrix-based integral inequality is employed to deal with the derivative of the looped function; (3) a general
dissipativity condition is presented to design desired sampled-data controllers. This approach is applied to the studies of a
truck-trailer system and an inverted pendulum systems. Its outcomes are highly satisfactory.
Notation: Notations to be introduced below will be used throughout the paper. The superscripts ‘−1’ and ‘T’ denote,
respectively, the inverse and transpose of a matrix; Rn and Rm×n denote, respectively, the n-dimensional Euclidean space
and the set of all m × n real matrices. N denotes the set of natural numbers. A symmetric matrix Q > 0 (Q ≥ 0) means that
Q is positive definite (positive semidefinite). Sym{X } = X + X T . λmax (R) (λmin (R)) denotes the maximum (minimum) of the
eigenvalue of a real symmetric matrix R. I and 0 represent the identity matrix and a zero matrix, respectively. L2 [0, +∞ )
represents the space of square-integrable vector functions over [0, +∞ ). The symmetric terms in a symmetric matrix are
denoted by ‘∗ ’, and diag{} denotes a block-diagonal matrix. || · || stands for the Euclidean norm of a vector and its induced
norm of a matrix.

2. Preliminaries

Consider a T-S fuzzy model described by the following IF-THEN rules:

✦ Plant Rule i: IF δ 1 (t) is Fi1 , δ 2 (t) is Fi2 and  and δ p (t) is Fip , THEN

x˙ (t ) = Ai x(t ) + B1i u(t ) + D1i ω (t )
(1)
z(t ) = Ci x(t ) + B2i u(t ) + D2i ω (t )

where i ∈ R = {1, 2, . . . , r}, r is the number of IF-THEN rules, Fi1 , Fi2 , . . . , Fip are the fuzzy sets, and δ 1 (t), δ 2 (t), . . . , δ p (t)
are the premise variables. x(t ) ∈ Rn and u(t ) ∈ Rm are the state and control input, respectively. ω (t ) ∈ Rq is the disturbance
input belonging to L2 [0, +∞ ), z(t ) ∈ Rs is the control output. Ai , B1i , D1i , Ci , B2i , and D2i are constant matrices with
appropriate dimensions.
 H.-B. Zeng et al. / Applied Mathematical Modelling 65 (2019) 415–427 417

Denote δ (t ) = [δ1 (t ) δ2 (t ) ··· δ p (t )], and

p
μi (δ (t )) = Fi j (δ j (t )),
j=1

μi (δ (t ))
hi (δ (t )) = r ,
j=1 μ j (δ (t ))

where Fij (δ j (t)) are the grade of memberships of δ j (t) in Fij . It is obvious that μi (δ (t)) ≥ 0 for all t. Thus, it can be obtained

that hi (δ (t)) ≥ 0 and ri=1 hi (δ (t )) = 1 for all t. Consequently, the resultant fuzzy system is inferred as a weighted average
of the local models:

x˙ (t ) = A(t )x(t ) + B1 (t )u(t ) + D1 (t )ω (t )
(2)
z(t ) = C (t )x(t ) + B2 (t )u(t ) + D2 (t )ω (t )

where

r 
r
A(t ) = hi (δ (t ))Ai , C (t ) = hi (δ (t ))Ci ,
i=1 i=1

r 
r
B1 (t ) = hi (δ (t ))B1i , B2 (t ) = hi (δ (t ))B2i ,
i=1 i=1

r 
r
D1 (t ) = hi (δ (t ))D1i , D2 (t ) = hi (δ (t ))D2i .
i=1 i=1

The control signal is generated by a zero-order holder. Choose the following state feedback controller for the
system (1):
u(t ) = Kx(tk ), tk  t < tk+1 , k ∈ N, (3)
where K ∈ Rm×n is the gain matrix of state feedback controller, x(tk ) is the measurement of x(t) at the sampling instant tk ,
limk→+∞ tk = +∞. Without loss of generality, the sampling period is assumed to be variable and the distance between any
two consecutive sampling instants is assumed to be within an interval, i.e.
tk+1 − tk = hk ∈ {hm , hM }, (4)
and hm and hM with hM ≥ hm > 0 represent, respectively, the upper and the lower bounds of sampling periods.
Substituting (3) into (2), the corresponding closed-loop system is obtained as follow:
✦ Closed-Loop System:

x˙ (t ) = A(t )x(t ) + B1 (t )Kx(tk ) + D1 (t )ω (t )
(5)
z(t ) = C (t )x(t ) + B2 (t )Kx(tk ) + D2 (t )ω (t ) tk  t < tk+1 .

Remark 1. Sampled-data-based control has widely applied in practical systems since a continuous-time plant is usually con-
trolled with a digital device. In contract with conventional continuous-time systems, the analysis of sampled-data systems
incorporating full-time information leads to challenging control problems with a rich mathematical structure. A sampled-
data system can be modeled by a simplified discrete-time system obtained by discretizing the plant. However, this discrete
model does not capture the inter-sample behavior of the real system. By contrast, the system model given in (5) fully cap-
ture the sample behavior thus has wide applications in the analysis of sampled-data systems.

In the following, we introduce some dissipative properties. Define the energy supply rate function for system (5) as
follows:
r ( ω , z ) = z T Qz + 2z T S ω + ω T Rω , (6)
where Q, S and R are real matrices with Q = R= QT , RT .
Without loss of generality, it is assumed that Q  0, and denote
1
Q˜ − = (−Q ) 2 . Then, we introduce definition of dissipativity as give below.

Definition 1. [47] System (5) is said to be strictly (Q, S, R ) − γ -dissipative if, for any γ > 0, such that
 tf  tf
r (ω (t ), z(t ))dt  γ ω (t )T ω (t )dt, ∀t f > t0 (7)
t0 t0

under zero initial condition.

The objective of this paper is to design a controller of form (3) for system (5) such that (7) is satisfied.
Before presenting the main results of the paper, let us introduce the following inequality, which is helpful to reduce the
conservativeness of the derived results.
418 H.-B. Zeng et al. / Applied Mathematical Modelling 65 (2019) 415–427

Lemma 1. [2] Let x : [α , β ] → Rn be a differentiable signal. For a vector ξ ∈ Rm , symmetric matrices R = RT > 0, R ∈ Rn×n , and
N1 , N2 ∈ Rn×m , the following inequality holds:
 β  
( β − α )2
− x˙ (s )T Rx˙ (s )ds  (β − α )ξ T N1T R−1 N1 + N2T R−1 N2 ξ
α 3
  β 
+2ξ T N1T (x(β ) − x(α ) ) − 2N2T x(s )ds
α

+2(β − α )ξ T
N2T [x (β ) + x(α )]. (8)
Remark 2. The inequality given in Lemma 1 is an improved version of the free-matrix-based inequality in [10]. Since
β β
α x (s )ds, instead of β −α α x (s )ds, is included in inequality (8), which provides much advantages for deriving the main
1

result of this paper.

3. Main results

In this section, we shall establish solvability of the sampled-data dissipative control problem formulated in the previous
section. For brevity, the following nomenclature are used for vectors and matrices.
 t
v1 (t ) = x(s )ds,
tk

v3 (t ) = x(t ) − x(tk ), v4 (t ) = x(t ) − x(tk+1 ),

T
η1 (t ) = v3 (t )T v1 (t )T ,
T
η2 (t ) = v4 (t )T v2 (t )T ,
T
η3 = x(tk )T x(tk+1 )T ,
T
η (t ) = x(t )T η3T v1 (t )T v2 (t )T x˙ (t )T ω (t )T ,
e j = 0n×( j−1)n In 0n×(6− j )n 0n×q , j = 1, . . . , 6
e7 = 0q×6n Iq .
For the case that the controller is given as a constant, a dissipativity condition is presented as follows.

Theorem 1. Supposed that there exist matrices P > 0, R1 > 0, R2 > 0, Q1 = Q1T , Q2 , Q3 = Q3T , Q4 , U1 , U2 , U3 , X, Z = Z T , M1i , M2i ,
N1i , N2i , i ∈ R, and a scalar γ > 0, such that (9) and (10) are satisfied for hk ∈ {hm , hM },
⎡   ⎤
1i hk N1Ti hk hM N2Ti Q
T ˜T
2i −
⎢ ∗ −R2 0 0 ⎥
⎣ ∗ ∗ −3R2 0
⎦ < 0, (9)
∗ ∗ ∗ −I

⎡   ⎤
2i hk M1Ti hk hM M2Ti Q
T ˜T
2i −
⎢ ∗ −R1 0 0 ⎥
⎣ ∗ ∗ −3R1 0
⎦ < 0, (10)
∗ ∗ ∗ −I
where

1i = 1 + 2i + hk 3i ,
= + + hk 4i ,
2i 1
 2i

1 = Sym eT1 Pe6 − T
1 Q2 2 + T
4 Q4 2 + T
6 Xe6 + eT6 X 7 + M1Ti 6 − 2M2Ti e4 − N1Ti 7 − 2N2Ti e5
T T
− 1 Q1 1 + 4 Q3 4 ,
 
2i = Sym T
10 1i − T
2i Se7 + eT7 (γ I − R )e7 ,
 T T
 T
3i = Sym 1 Q1 3 + 3 Q2 2 + N2i 9 + 2Z 2 + eT6 R1 e6 ,
 T T
 T
4i = Sym 4 Q3 5 + 5 Q4 2 + M2i 8 − 2Z 2 + eT6 R2 e6 ,
T
1 = eT1 − eT2 eT4 ,
T
2 = eT2 eT3 ,
 H.-B. Zeng et al. / Applied Mathematical Modelling 65 (2019) 415–427 419

T
3 = eT6 eT1 ,
T
4 = eT1 − eT3 eT5 ,
T
5 = eT6 −eT1 ,
T
6 = eT1 − eT2 ,
T
7 = eT1 − eT3 ,
8 = e1 + e2 ,
9 = e1 + e3 ,
T
10 = eT1U1 + eT2U2 + eT6U3 ,
1i = Ai e1 + B1i Ke2 + D1i e7 − e6 ,
2i = Ci e1 + B2i Ke2 + D2i e7 .
Then, system (5) is strictly (Q, S, R ) − γ -dissipative.

Proof. Construct the following complete interval-dependent looped function for system (5):

7
V (t ) = V j (t ), t ∈ [tk , tk+1 ), k ∈ N, (11)
j=1

where
V1 (t ) = xT (t )P x(t ),
V2 (t ) = (tk+1 − t )η1T (t )[Q1 η1 (t ) + 2Q2 η3 ],
V3 (t ) = (t − tk )η2T (t )[Q3 η2 (t ) + 2Q4 η3 ],
V4 (t ) = 2vT3 (t )X v4 (t ),
V5 (t ) = (tk+1 − t )(t − tk )η3T Z η3 ,
 t
V6 (t ) = (tk+1 − t ) x˙ (s )T R1 x˙ (s )ds,
tk
 tk+1
V7 (t ) = −(t − tk ) x˙ (s )T R2 x˙ (s )ds.
t
Taking the derivative of (11) along the solution of system (5) yields
V˙ 1 (t ) = 2η (t )T eT1 Pe6 η (t ),

V˙ 2 (t ) = −η (t )T T
1 ( Q1 1 + 2Q2 2 ) + 2(tk+1 − t )

× T
1 Q1 3 + T
3 Q2 2 η (t ),

V˙ 3 (t ) = η (t ) T T
4 ( Q3 4 + 2Q4 2 ) + 2(t − tk )

× T
4 Q3 5 + T
5 Q4 2 η (t ),
V˙ 4 (t ) = 2η (t )T T
6 Xe6 − eT6 X 7 η (t ),
V˙ 5 (t ) = η (t ) [(tk+1 − t ) − (t − tk )]
T T
2Z 2 η (t ),
V˙ 6 (t ) = (tk+1 − t )x˙ (t )T R1 x˙ (t ) + J1 ,
V˙ 7 (t ) = (t − tk )x˙ (t )T R2 x˙ (t ) + J2 ,
t tk+1
with J1 = − tk x˙ (s )T R1 x˙ (s )ds, J2 = − t x˙ (s )T R2 x˙ (s )ds.
Applying Lemma 1, we get
   
hM
2  
J1 ≤ η (t )T (t − tk ) M1T R−1
1 M1 + M2T R−1
1 M2 + Sym M1T 6 − 2M2 e4 + (t − tk )M2
T
8 η (t ), (12)
3
   
hM T −1
2  
J2 ≤ η (t ) T
(tk+1 − t ) N1T R−1
2 N1 + N R N2 − Sym N1T 7 + 2N2T e5 − (tk+1 − t ) N2T 9 η (t ), (13)
3 2 2
where

r 
r
Ml = hi (δ (t ))Mli , Nl = hi (δ (t ))Nli , l = 1, 2; i ∈ R ,
i=1 i=1
420 H.-B. Zeng et al. / Applied Mathematical Modelling 65 (2019) 415–427

For any matrices U1 , U2 , and U3 with appropriate dimension, it follows from system (5) that
0 = 2[x(t )TU1 + x(tk )TU2 + x˙ (t )TU3 ]
×[−x˙ (t ) + A(t )x(t ) + B1 (t )Kx(tk ) + D1 (t )ω (t )]
= 2η (t )T T
10 1 η (t ). (14)
r
i=1 hi (δ (t )) 1i . Then, adding the right-hand side of (14) to V (t ) and using (12) and (13), we obtain, for
where 1 = ˙
t ∈ [tk , tk+1 ),

V˙ (t ) − r (ω, z ) + γ ωT (t )ω (t )
t − t t − tk

 ηT (t ) k+1 Ξ̄1 + Ξ̄2 η (t ), (15)
hk hk
where
 2

hM T −1
Ξ̄1 = ¯ 1 + hk N1T R−1 N1 + N R N2 − 2T Q 2 ,
2
3 2 2
 2

hM
Ξ̄2 = 2 + hk M1 R1 M1 +
¯ T −1 T −1
M2 R1 M2 − 2T Q 2
3
with

r
¯1 = 1 + hi (δ (t ))( 2i + hk 3i ),
i=1

r
¯2 = 1 + hi (δ (t ))( 2i + hk 4i ),
i=1

r

2 = hi (δ (t )) 2i .
i=1

Thus, it follows, respectively, from (9) and (10) that

⎡   ⎤
¯1
2 Q−
T ˜T
hk N1T hk hM N2T
⎢ ∗ −R2 0 0 ⎥
⎣ ∗ ∗ −3R2 0
⎦ < 0, (16)
∗ ∗ ∗ −I
⎡   ⎤
¯2
2 Q−
T ˜T
hk M1T hk hM M2T
⎢ ∗ −R1 0 0 ⎥
⎣ ∗ ∗ −3R1 0
⎦ < 0, (17)
∗ ∗ ∗ −I

meaning that Ξ̄1 < 0 and Ξ̄2 < 0 in the sense of Schur complement. Thus, we have
V˙ (t ) − r (ω, z ) + γ ω (t )T ω (t )  0. (18)
Integrating (18) from t0 to tf yields
 tf  tf
γ ω (t )T ω (t )dt − r (ω (t ), z(t ))dt  −V (t f ) + V (t0 ). (19)
t0 t0

Consider the zero initial condition, it is clear that (7) holds. Thus, system (5) is strictly (Q, S, R ) − γ -dissipative. This
completes the proof. 

Remark 3. Notice that V j (tk ) = V j (tk+1 ) = 0, j = {2, 3, . . . , 7}, which satisfies the looped function condition given in [48]. In
[46], only the information from x(tk ) and x(t) was taken into consideration. By contrast, function (11) additionally considers
the information from x(t) to x(tk+1 ), through the introduction of two new terms, V3 (t) and V7 (t). Therefore, the conditions
derived in the paper are less conservative than [46].

Remark 4. In [46], the sufficient conditions being derived only depend on the upper bound of the sampling period. In
contrast with the sufficient conditions obtained in [46], Theorem 1 depends on both the upper bound and the lower bound.
In fact, by setting hm = 0 in Theorem 1, the corresponding conditions become upper-bound-dependent and lower-bound-
independent. Theorem 1 can also be applied to deal with constant sampling by simply setting hm = hM .

Remark 5. Theorem 1 presents sufficient conditions for ensuing the (Q, S, R )-dissipativity for the closed-loop system (5).
If choosing Q = −γ −1 I, S = I, R = 2γ I, Theorem 1 reduces to a bounded real lemma; if choosing Q = 0, S = I, R = 2γ I,
 H.-B. Zeng et al. / Applied Mathematical Modelling 65 (2019) 415–427 421

Theorem 1 reduces to a passive condition; if choosing Q = −γ −1 α I, S = (1 − α )I, R = 2γ I, where α ∈ [0, 1], Theorem 1 be-
comes sufficient conditions for mixed H∞ and passivity.

With given controller gain, the conditions given Theorem 1 are a set of linear matrix inequalities (LMIs), which is
easily solved by some existing LMI tools. However, when the controller gain is unknown, these conditions become a set of
nonlinear matrix inequalities, which cannot be checked directly by using existing numerical softwares. For the problem of
dissipative control, we shall present the following sufficient conditions for the existence of such a state feedback controller.

Theorem 2. Suppose that for given scalars ε 1 , ε 2 , there exist matrices P̄ > 0, R̄1 > 0, R̄2 > 0, Q̄1 = Q̄1T , Q̄2 , Q̄3 = Q̄3T , Q̄4 , Ū , X̄ ,
Z̄ = Z̄ T , L, M̄1i , M̄2i , N̄1i , N̄2i , i ∈ R, and a scalar γ > 0, such that (20) and (21) given below are satisfied for hk ∈ {hm , hM },
⎡   ⎤
¯ 1i hk N̄1Ti hk hM N̄2Ti ¯ T Q˜ T−
2i
⎢ ∗ −R̄2 0 0 ⎥
⎣ ⎦ < 0, (20)
∗ ∗ −3R̄2 0
∗ ∗ ∗ −I
⎡   ⎤
¯ 2i hk M̄2Ti hk hM M̄2Ti ¯ T Q˜ T−
2i
⎢ ∗ −R̄1 0 0 ⎥
⎣ ⎦ < 0, (21)
∗ ∗ −3R̄1 0
∗ ∗ ∗ −I
where
¯ 1i = ¯ 1 + ¯ 2i + hk ¯ 3i ,
¯ 2i = ¯ 1 + ¯ 2i + hk ¯ 4i ,
 
¯ 1 = Sym eT P̄ e6 − T Q̄2 + T
+ T
+ eT6 X̄ + M̄1Ti − 2M̄2Ti e4 − N̄1Ti − 2N̄2Ti e5
1 1 2 4 Q̄4 2 6 X̄ e6 7 6 7
T T
− 1 Q̄1 1 + 4 Q̄3 4 ,
 
2i = Sym ¯ T
10
¯ 1i − ¯ T Se7
2i + eT7 (γ I − R )e7 ,
 T T
 T
3i = Sym 1 Q̄1 3 + 3 Q̄2 2 + N̄2i 9 + 2 Z̄ 2 + eT6 R̄1 e6 ,
 T T
 T
4i = Sym 4 Q̄3 5 + 5 Q̄4 2 + M̄2i 8 − 2 Z̄ 2 + eT6 R̄2 e6 ,
T
¯ 10 = eT1 + ε1 eT2 + ε2 eT6 ,
¯ 1i = AiŪ T e1 + B1i Le2 + D1i e7 − Ū T e6 ,
¯ 2i = CiŪ T e1 + B2i Le2 + D2i e7 .
and j , j = {1, 2, . . . , 9}, are defined as that in Theorem 1. Then system (5) is strictly (Q, S, R ) − γ -dissipative. Moreover, the
state feedback controller gain is given by
K = LŪ −T . (22)
Proof. Define U2 = ε1U1 , U3 = ε2U1 , Ū = L = KŪ T , P̄ = Ū PŪ T , R̄1 = Ū R1Ū T , R̄2 = Ū R2Ū T , Q̄l = H1 Ql H1T , l = {1, 2, 3, 4},
U1−1 ,
X̄ = Ū X Ū T , H1 = diag{Ū , Ū }, H2 = diag{Ū , Ū , Ū , Ū , Ū , Ū , I}, Z̄ = H1 ZH1T , M̄ pi = Ū M pi H2T , N̄ pi = Ū N pi H2T , p = 1, 2, and H3 =
diag{Ū , Ū , Ū , Ū , Ū , Ū , I, Ū , Ū , I}. Pre- and post-multiplying (9) and (10) with H3 and H3T , respectively, we obtain (20) and
(21). This completes the proof. 

Remark 6. Theorem 2 provides a sufficient condition for the existence of a sampled-data controller such that system (5) is
strictly (Q, S, R ) − γ -dissipative. This condition is formulated in terms of LMIs, which can be easily solved. The optimal
dissipativity performance γ ∗ can be obtained through setting κ = −γ and minimizing κ subject to (20) and (21).

4. Numerical examples

In this section, two examples are provided to show the effectiveness of the proposed techniques and its superiority over
others.

Example 1. Consider the following truck-trailer model given in [46]:

⎧
⎨x˙ 1 (t ) = − Lt̄0 x1 (t ) + lt̄0 u(t ) + 0.1ω1 (t ) + 0.1ω2 (t )
⎪ v̄t̄ v̄t̄

x˙ 2 (t ) = Lv̄t̄t̄ x1 (t ) + 0.1ω1 (t ) + 0.1ω2 (t ) (23)

⎩x˙ (t ) = v̄t̄ sin x (t ) + v̄t̄ x (t )
⎪ 0

3 t̄0 2 2L 1

where x1 (t) is the difference of angle between the truck and trailer, x2 (t) and x3 (t) represent the angle of the trailer and its
vertical position of rear end, respectively. The model parameters are given as v̄ = −1, t̄0 = 0.5, t̄ = 2, L = 5.5, and l = 2.8.
422 H.-B. Zeng et al. / Applied Mathematical Modelling 65 (2019) 415–427

Table 1
γ max for hm = 0 and different hM .
hM 0.05 0.15 0.25 0.35 NoV∗

Wu et al. [46] 0.9971 0.9311 0.8193 – 78

Theorem 2 1.0533 1.0126 0.9380 0.7436 276

Table 2
γ max for hM = 3.5 and different hm .
hm 0.1 0.20 0.30 0.35

Theorem 2 0.8031 0.8448 0.8715 0.8798

Choose δ (t ) = x2 (t ) + (v̄t̄ /2L )x1 (t ) and define the membership functions

 sin(δ (t ))−ḡδ (t )
if δ (t ) = 0
h1 (δ (t )) = δ (t )(1−ḡ)
1 if δ (t ) = 0
and h2 (δ (t )) = 1 − h1 (δ (t )), where ḡ = 10−2 /π . Then, this system can be represented by the following T-S fuzzy model:
Plant Rule 1: IF δ (t) is about 0, THEN
x˙ (t ) = A1 x(t ) + B11 u(t ) + D11 ω (t )
Plant Rule 2: IF δ (t) is about ± π , THEN
x˙ (t ) = A2 x(t ) + B12 u(t ) + D12 ω (t )
T T
where x(t ) = x1 (t ) x2 (t ) x3 (t ) , ω (t ) = ω1 (t ) ω2 (t ) , and
⎡ ⎤ ⎡ ⎤
− Lv̄t̄t̄ 0 0 − Lv̄t̄t̄ 0 0
⎢ ⎥ ⎢ ⎥
0 0

A1 = ⎣ v̄t̄
Lt̄0
0 0⎦, A2 = ⎣ Lv̄t̄t̄ 0 0⎦
0
2 2
v̄2 t̄ 2 v̄t̄ 0 ḡv̄ t̄ ḡv̄t̄
2Lt̄0 t̄0 2Lt̄0 t̄0
0
⎡ v̄t̄ ⎤ ⎡ ⎤
0.1 0.1
l t̄0
B11 = B12 = ⎣ 0 ⎦, D11 = D12 = ⎣0.1 0.1⎦
0 0 0
The parameters of the output z(t) are chosen as follows:
 
0.1 0.2 0.2
C1 = C2 =
0.3 0.2 0.1
   
1.2 0.15 0
B21 = B22 = , D21 = D22 =
1.1 0 0.2
For making comparison, we choose the same parameters as those given in [46], i.e.,
     
−0.04 0 1.1 0.5 3 0
Q= ,S= ,R=
0 −1 3 2 0 1

Firstly, we discuss the case of variable sampling. By using Theorem 2 with ε1 = ε2 = 0.5 and hm = 0, the maximum
of dissipativity performance for different upper bounds of sampling periods, hM , is obtained and summarized in Table 1
along with the results provided in [46]. It is clearly observed from Table 1 that Theorem 2 provides a larger dissipativity
performance than that obtained in [46]. It is also summarized from the table that less conservative results are obtained
at the cost of larger computational consumption. To show the relationship between the dissipativity performance and the
lower bound of sampling periods, given hM = 0.35, the maximum of dissipativity performance for different lower bounds of
sampling periods is calculated and listed in Table 2. It is concluded from Table 2 that the index of dissipativity performance
increase with the increase of hm . On the other hand, given the dissipativity performance index γ = 0.8193 and hm = 0, the
maximum sampling period obtained by Theorem 2 is hM = 0.32, while maximum sampling period given in [46] is 0.25. The
result is increased by 28%. It is obtained by (22) that the corresponding state feedback controller is
K = [9.8894 −7.6927 1.3140]
T
Choosing the disturbance input ω (t ) = [sin(0.1t ) exp(−0.1t ) sin(0.1t ) exp(−0.1t )] and the initial condition
T
x(0 ) = [−0.5π −0.75π −10] , the corresponding state response and control input of the system under the designed
controller are depicted in Figs. 1 and 2. Clearly, this system is stabilized by the given controller.
 H.-B. Zeng et al. / Applied Mathematical Modelling 65 (2019) 415–427 423

Fig. 1. State response of system (23) in the case of variable sampling.

Fig. 2. Control input u(t) in the case of variable sampling.

Table 3
γ max for different hm = hM .
hM 0.05 0.15 0.25 0.35

Theorem 2 1.0570 1.0303 0.9892 0.8798

Next, we discuss the case of constant sampling, i.e. hm = hM = h̄. Choose the upper bound of sampling periods as the
same as those given in Table 1. The maximum of dissipativity performance for different sampling periods is computed
and listed in Table 3. We can conclude from Tables 1 and 3 that, for the same upper bounds of the sampling period,
constant sampling gives rise to better dissipativity performance than that of variable sampling. Furthermore, it is observed
from these two tables that a larger optimal dissipativity performance requires a smaller sampling period. In other word,
the dissipativity performance can be improved by the increase of sampling frequency. Furthermore, given the dissipativity
performance index γ = 0.8193, the maximum sampling period obtained by Theorem 2 is h̄ = 0.37 and the corresponding
424 H.-B. Zeng et al. / Applied Mathematical Modelling 65 (2019) 415–427

Fig. 3. State response of system (23) in the case of constant sampling.

Fig. 4. Control input u(t) in the case of constant sampling.

state feedback controller is

K = 6.8471 −4.2335 0.7185
Under the designed controller, the corresponding state response and control input of the system are depicted in Figs. 3 and
4. It is shown in Fig. 3 that this system is stabilized by the given controller. These demonstrate the advantages and the
superiority of the proposed method.

Example 2. Consider the problem of balancing and swing-up of an inverted pendulum on a cart. The dynamics of the
pendulum motion is given by
⎧
⎨x˙ 1 (t ) = x2 (t ) + 0.1ω1 (t ) 2
⎪
x˙ 2 (t ) = (g sin(x1 (t )) − amlx2 (t ) sin(2x1 (t ))/2
(24)
⎩−a cos(x1 (t ))u(t ))(4l/3 − aml cos (x1 (t )))
−1
⎪ 2

+0.1ω2 (t )
where x1 (t ) ∈ (−π /2, π /2 ) denotes the angle (in radians) of the pendulum from the vertical, and x2 (t) is the angular veloc-
ity. g = 9.8 m/s2 is the gravity constant, m is the mass of the pendulum, M is the mass of the cart, 2l is the length of the
 H.-B. Zeng et al. / Applied Mathematical Modelling 65 (2019) 415–427 425

Fig. 5. State response of system (24).

pendulum, u(t) is the force applied to the cart (in Newtons), and a = 1/(m + M ). In this example, we choose m = 2.0 {kg},
M = 8.0 kg, and 2l = 1.0 m [49].
Choose δ (t ) = x1 (t ) and define the membership functions

1 − π2 δ (t ) if 0  δ (t ) < π2
h1 (δ (t )) =
1 + π2 δ (t ) if − π2 < δ (t ) < 0
and h2 (δ (t )) = 1 − h1 (δ (t )). Then, this system can be represented by the following T-S fuzzy model:
Plant Rule 1: IF x1 (t) is about 0, THEN
x˙ (t ) = A1 x(t ) + B11 u(t ) + D11 ω (t )
 
Plant Rule 2: IF x1 (t) is about ± π2 |x1 (t )| < π , THEN
2
x˙ (t ) = A2 x(t ) + B12 u(t ) + D12 ω (t )
T T
where x(t ) = x1 (t ) x2 (t ) , ω (t ) = ω1 (t ) ω2 (t ) , and
   
0 1 0 1
A1 = g , A2 = 2g
4l/3−aml
0 π (4l /3−aml β 2 ) 0
   
0 0
B11 = , B12 = aβ
− 4l/3a−aml − 4l /3−aml β2
D11 = D12 = 0.1I.
and β = cos(88◦ ). Membership functions for Rule 1 and Rule 2 are respectively

1 − π2 θ (t ) if 0  θ (t ) < π2
h1 (θ (t )) =
1 + π2 θ (t ) if − π2 < θ (t ) < 0
and h2 (θ (t )) = 1 − h1 (θ (t )).
The parameters of the output z(t) are chosen as follows:
 
0.1 0.2
C1 = C2 =
0.1 0.2
 
0.1
B21 = B22 = , D21 = D22 = 0.1I
0
In this example, we choose
     
−0.04 0 2 0.5 3 0
Q= ,S= ,R=
0 −1 1 1 0 2
426 H.-B. Zeng et al. / Applied Mathematical Modelling 65 (2019) 415–427

Fig. 6. Control input u(t) of system (24).

When hm = hM = 15ms, the optimal dissipativity performance γ ∗ obtained by Theorem 2 is 0.7339, and the cor-
responding controller gain matrix is given as K = [2012.5 649.8]. Supposing that the disturbance input ω (t ) =
T T
[sin(0.1t ) exp(−0.5t ) sin(0.1t ) exp(−0.5t )] and the initial condition x(0 ) = [π /4 0] , the corresponding state re-
sponse and control input of the system under the designed controller are depicted in Figs. 5 and 6. It is shown in the
figures that the balance of the inverted pendulum can be implemented by the given controller. This further verifies the
effectiveness of the proposed method.

5. Conclusion

In this paper, the problem of sampled-data-based dissipative control of T-S fuzzy systems has been investigated. A novel
complete sampling-interval-dependent looped function has been proposed, which takes advantage of the useful information
about the realistic sampling interval. Sufficient conditions have been established to ensure that the considered system
is strictly dissipative. Based on the derived conditions, a scheme to design desired sampled-data controller is proposed.
Finally, two illustrative examples are given to show the applicability of the approach proposed. The outcome is much less
conservative when compared with those obtained in the literature. The approach proposed in this paper can be extended
to analysis and synthesis of systems such as networked control systems [50], discrete-time systems [51], and stochastic
systems [52]. The proposed design scheme does not concern the premise variables. Future work will fucus on the cases of
unmeasurable premise variables, failures in design or signal conditions [53–55].

References

[1] Y. Wu, H. Su, P. Shi, Z. Shu, Z. Wu, Consensus of multiagent systems using aperiodic sampled-data control, IEEE Trans. Cybern. 46 (9) (2015) 2132–2143.
[2] H.B. Zeng, K.L. Teo, Y. He, H. Xu, W. Wang, Sampled-data synchronization control for chaotic neural networks subject to actuator saturation, Neuro-
computing 260 (2017) 25–31.
[3] N. Noroozi, A. Khayatian, H.R. Karimi, Semiglobal practical integral input-to-state stability for a family of parameterized discrete-time interconnected
systems with application to sampled-data control systems, Nonlinear Anal. Hybrid Syst. 17 (2015) 10–24.
[4] P. Shi, Filtering on sampled-data systems with parametric uncertainty, IEEE Trans. Autom. Control 43 (1998) 1022–1027.
[5] E. Fridman, A. Seuret, J.P. Richard, Robust sampled-data stabilization of linear systems: an input delay approach, Automatica 40 (2004) 1441–1446.
[6] E. Fridman, A refined input delay approach to sampled-data control, Automatica 46 (2010) 421–427.
[7] K. Liu, E. Fridman, Wirtinger’s inequality and Lyapunov-based sampled-data stabilization, Automatica 48 (2012) 102–108.
[8] H. Gao, J. Wu, P. Shi, Robust sampled-data H∞ control with stochastic sampling, Automatica 45 (2009) 1729–1736.
[9] B. Shen, Z. Wang, X. Liu, A stochastic sampled-data approach to distributed H∞ filtering in sensor networks, IEEE Trans. Circ. Syst. I Reg. Pap. 58 (2011)
2237–2246.
[10] H.B. Zeng, Y. He, M. Wu, J. She, Free-matrix-based integral inequality for stability analysis of systems with time-varying delay, IEEE Trans. Automat.
Control 60 (2015a) 2768–2772.
[11] H.B. Zeng, Y. He, M. Wu, J. She, New results on stability analysis for systems with discrete distributed delay, Automatica 60 (2015b) 189–192.
[12] H.B. Zeng, J.H. Park, S.P. Xiao, Y. Liu, Further results on sampled-data control for master-slave synchronization of chaotic Lur’e systems with time delay,
Nonlinear Dyn. 82 (2015c) 851–863.
[13] A. Seuret, C. Briat, Stability analysis of uncertain sampled-data systems with incremental delay using looped-functionals, Automatica 55 (2015)
274–278.
[14] T.H. Lee, J.H. Park, Stability analysis of sampled-data systems via free-matrix-based time-dependent discontinuous Lyapunov approach, IEEE Trans.
Autom. Control 62 (2017) 3653–3657.
[15] H.B. Zeng, K.L. Teo, Y. He, A new looped-functional for stability analysis of sampled-data systems, Automatica 82 (2017) 328–331.
 H.-B. Zeng et al. / Applied Mathematical Modelling 65 (2019) 415–427 427

[16] H. Hua, Y. Liu, J. Lu, J. Zhu, A new impulsive synchronization criterion for T-S fuzzy model and its applications, Appl. Math. Modell. 37 (2013)
8826–8835.
[17] D. Gong, H. Zhang, Z. Wang, J. Liu, Synchronization analysis for complex networks with coupling delay based on T-S fuzzy theory, Appl. Math. Modell.
36 (2012) 6215–6224.
[18] G. Feng, A survey on analysis and design of model-based fuzzy control systems, IEEE Trans. Fuzzy Syst. 14 (2006) 676–697.
[19] J. Zhang, P. Shi, Y. Xia, Robust adaptive sliding-mode control for fuzzy systems with mismatched uncertainties, IEEE Trans. Fuzzy Syst. 18 (2010)
700–711.
[20] C. Lin, Q. Wang, T. Lee, Stability and stabilization of a class of fuzzy time-delay descriptor systems, IEEE Trans. Fuzzy Syst. 14 (2006) 542–551.
[21] B. Chen, X. Liu, S. Tong, C. Lin, Observer-based stabilization of T-S fuzzy systems with input delay, IEEE Trans. Fuzzy Syst. 16 (2008) 652–663.
[22] L. Wu, X. Su, P. Shi, J. Qiu, Stability analysis and stabilization of discrete-time T-S fuzzy time-varying delay systems, IEEE Trans. Syst. Man Cybern. Part
B Cybern. 41 (2011) 273–286.
[23] H. Wu, H. Li, New approach to delay-dependent stability analysis and stabilization for continuous-time fuzzy systems with time-varying delay, IEEE
Trans. Fuzzy Syst. 15 (2007) 482–493.
[24] Z. Wu, P. Shi, H. Su, J. Chu, Reliable H∞ control for discrete-time fuzzy systems with infinite-distributed delay, IEEE Trans. Fuzzy Syst. 20 (2012) 22–31.
[25] J. Liang, Z. Wang, X. Liu, On passivity and passification of stochastic fuzzy systems with delays: the discrete-time case, IEEE Trans. Syst. Man Cybern.
Part B 40 (2010) 964–969.
[26] C. Lin, Q. Wang, T. Lee, B. Chen, H∞ filter design for nonlinear systems with time-delay through T-S fuzzy model approach, IEEE Trans. Fuzzy Syst. 16
(2008) 739–746.
[27] S.K. Nguang, P. Shi, S. Ding, Fault detection for uncertain fuzzy systems: an LMI approach, IEEE Trans. Fuzzy Syst. 15 (2007) 1251–1262.
[28] L. Wu, D.W.C. Ho, Fuzzy filter design for It oˆ stochastic systems with application to sensor fault detection, IEEE Trans. Fuzzy Syst. 17 (2009) 233–242.
[29] H. Dong, Z. Wang, J. Lam, H. Gao, Fuzzy-model-based robust fault detection with stochastic mixed time delays and successive packet dropouts, IEEE
Trans. Syst. Man Cybern. Part B Cybern. 42 (2012) 365–376.
[30] H. Gao, T. Chen, Stabilization of nonlinear systems under variable sampling: a fuzzy control approach, IEEE Trans. Fuzzy Syst. 15 (2007) 972–983.
[31] J. Yoneyama, Robust H∞ control of uncertain fuzzy systems under time-varying sampling, Fuzzy Sets Syst. 161 (2010) 859–871.
[32] H.K. Lam, Stabilization of nonlinear systems using sampled-data output-feedback fuzzy controller based on polynomial-fuzzy-model-based control
approach, IEEE Trans. Syst. Man Cybern. Part B Cybern. 42 (2012) 258–267.
[33] H.Q. Xiao, Y. He, M. Wu, S.P. Xiao, J.H. She, New results on H∞ tracking control based on the T-S fuzzy model for sampled-data networked control
system, IEEE Trans. Fuzzy Syst. 23 (2015) 2439–2448.
[34] H. Li, Y. Gao, P. Shi, X. Zhao, Output-feedback control for T-S fuzzy delta operator systems with time-varying delays via an input-output approach,
IEEE Trans. Fuzzy Syst. 23 (2015) 1100–1112.
[35] R. Sakthivel, S. Selvi, K. Mathiyalagan, P. Shi, Reliable mixed H∞ and passivity-based control for fuzzy Markovian switching systems with probabilistic
time delays and actuator failures, IEEE Trans. Cybern. 45 (2015) 2720–2731.
[36] R. Sakthivel, K. Boomipalagan, Y.K. Ma, M. Muslim, Sampled-data reliable stabilization of T-S fuzzy systems and its application, Complexity 21 (2016)
518–529.
[37] R. Sakthivel, T. Saravanakumar, Y.K. Ma, S.M. Anthoni, Finite-time resilient reliable sampled-data control for fuzzy systems with randomly occurring
uncertainties, Fuzzy Sets Syst. 329 (2017) 1–18.
[38] B. Brogliato, R. Lozano, B. Maschke, O. Egeland, Dissipative Systems Analysis and Control: Theory and Applications, Springer-Verlag, London, 2007.
[39] Z. Feng, J. Lam, H. Gao, α -dissipativity analysis of singular time-delay systems, Automatica 47 (2011) 2548–2552.
[40] H.B. Zeng, Y. He, P. Shi, M. Wu, S.P. Xiao, Dissipativity analysis of neural networks with time-varying delays, Neurocomputing 168 (2015a) 741–746.
[41] H.B. Zeng, J.H. Park, C.F. Zhang, W. Wang, Stability and dissipativity analysis of static neural networks with interval time-varying delay, J. Frankl. Inst.
352 (2015b) 1284–1295.
[42] Z. Feng, J. Lam, Z. Shu, Dissipative control for linear systems by static output feedback, Int. J. Syst. Sci. 44 (2013) 1566–1576.
[43] Y. Ma, X. Jia, D. Liu, Finite-time dissipative control for singular discrete-time Markovian jump systems with actuator saturation and partly unknown
transition rates, Appl. Math. Modell. 53 (2018) 49–70.
[44] X.M. Zhang, Q.L. Han, A decentralized event-triggered dissipative control scheme for systems with multiple sensors to sample the system outputs,
IEEE Trans. Cybern. 46 (2016) 2745–2757.
[45] J. Tao, R. Lu, P. Shi, H. Su, Z.G. Wu, Dissipativity-based reliable control for fuzzy Markov jump systems with actuator faults, IEEE Trans. Cybern. 47
(2017) 2377–2388.
[46] Z. Wu, P. Shi, H. Su, R. Lu, Dissipativity-based sampled-data fuzzy control design and its application to truck-trailer system, IEEE Trans. Cybern. 23
(2015) 1669–1679.
[47] D. Hill, P. Moylan, The stability of nonlinear dissipative systems, IEEE Trans. Autom. Control 21 (1976) 708–711.
[48] A. Seuret, A novel stability analysis of linear systems under asynchronous samplings, Automatica 48 (2012) 177–182.
[49] H.O. Wang, K. Tanaka, M.F. Griffin, An approach to fuzzy control of nonlinear systems: stability and design issues, IEEE Trans. Fuzzy Syst. 4 (1996)
14–23.
[50] R. Lu, H. Cheng, J. Bai, Fuzzy-model-based quantized guaranteed cost control of nonlinear networked systems, IEEE Trans. Fuzzy Syst. 23 (2015)
567–575.
[51] Y. Wei, J. Qiu, H.R. Karimi, Reliable output feedback control of discrete-time fuzzy affine systems with actuator faults, IEEE Trans. Circ. Syst. I Reg. Pap.
64(1) 170–181.
[52] H. Wang, X. Liu, K. Liu, H.R. Karimi, Approximation-based adaptive fuzzy tracking control for a class of nonstrict-feedback stochastic nonlinear time-de-
lay systems, IEEE Trans. Fuzzy Syst. 23 (5) (2015) 1746–1760.
[53] L. Zhang, E.K. Boukas, L. Baron, Fault detection for discrete-time Markov jump linear systems with partially known transition probabilities, Int. J.
Control 83 (8) (2010) 1564–1572.
[54] S. Rathinasamy, H.R. Karimi, M. Joby, S. Santra, Resilient sampled-data control for Markovian jump systems with adaptive fault-tolerant mechanism,
IEEE Trans. Circ. Syst. II Express Briefs 64 (11) (2017) 1312–1316.
[55] M. Chadli, H.R. Karimi, Robust observer design for unknown inputs Takagi–Sugeno models, IEEE Trans. Fuzzy Syst. 21 (1) (2013) 158–164.