See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/272375909

Anti-Windup Based Dynamic Output Feedback Controller Design with

Performance Consideration for Constrained Takagi–Sugeno Systems

Article  in  Engineering Applications of Artificial Intelligence · April 2015

DOI: 10.1016/j.engappai.2015.01.005

CITATIONS READS

27 234

3 authors:

Anh-Tu Nguyen Antoine Dequidt

Université Polytechnique Hauts-de-France Université Polytechnique Hauts-de-France
56 PUBLICATIONS   349 CITATIONS    54 PUBLICATIONS   229 CITATIONS   

SEE PROFILE SEE PROFILE

Michel Dambrine
Université Polytechnique Hauts-de-France
138 PUBLICATIONS   2,806 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Fuzzy Model-Based Approaches for Nonlinear Systems View project

Robust Control for Intelligent Vehicles View project

All content following this page was uploaded by Anh-Tu Nguyen on 12 October 2017.

The user has requested enhancement of the downloaded file.

 Engineering Applications of Artificial Intelligence 40 (2015) 76–83

Contents lists available at ScienceDirect

Engineering Applications of Artificial Intelligence

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

Anti-windup based dynamic output feedback controller design

with performance consideration for constrained Takagi–
Sugeno systems
AnhTu Nguyen, Antoine Dequidt, Michel Dambrine n
LAMIH Laboratory UMR 8201, CNRS-University of Valenciennes and Hainaut-Cambrésis, UVHC, Le Mont Houy, 59313 Valenciennes Cedex, France

art ic l e i nf o a b s t r a c t

Article history: This paper proposes an LMI-based method to guarantee the closed-loop regional stability and
Received 9 July 2014 performance for Takagi–Sugeno systems that are subject to input saturation, state constraints, and also
Received in revised form amplitude-bounded disturbance. Based on the Lyapunov stability tool, the proposed method provides
28 November 2014
conditions to simultaneously design the dynamic output feedback controller and its anti-windup
Accepted 13 January 2015
compensator. By solving a convex optimization problem, these conditions are derived such that a
tradeoff between the upper bound on the nominal ℒ2 gain for exogenous disturbance and the minimal
Keywords: size of the domain of attraction can be found. This method is simple and systematic, allowing dealing
Nonlinear systems with a very large class of constrained nonlinear systems. The effectiveness of the proposed method is
Takagi–Sugeno fuzzy model
illustrated with numerical examples.
Dynamic output feedback controller
& 2015 Elsevier Ltd. All rights reserved.
Anti-windup compensator
ℒ2 performance
Domain of attraction

1. Introduction For technological or economic reasons, the state variables are

Over the past two decades, Takagi–Sugeno (T–S) fuzzy models
(Takagi and Sugeno, 1985) have been intensively studied in the
control community (Tanaka and Wang, 2001). It is motivated by
the fact that these models have general approximation capability
for complex dynamical system (Cao and Rees, 1997). Moreover,
under weak conditions, a nonlinear model can be, globally or more
often regionally, rewritten on a fuzzy T–S form. Stability analysis
or controller synthesis is then facilitated due to their polytopic
structure (Tanaka and Wang, 2001). As a consequence, this
approach has now become a very attractive research topic in
control theory (Feng, 2006). Stability analysis of a given T–S
system is investigated in most cases via the direct Lyapunov
method through the use of a quadratic Lyapunov function; the
derived stability conditions are expressed as linear matrix inequal-
ities (LMIs) (Boyd et al., 1994) for which efficient solvers are
available. For the controller design, the choice of a parallel
distributed compensation (PDC) control law is usually done for
T–S models (Tanaka and Wang, 2001). An abundant literature is
available on this nonlinear state feedback control law; see e.g. Feng
(2006) for a quick overview.
n
Corresponding author. Tel.: þ 33 3 27 51 14 71; fax: þ 33 3 27 51 13 16.
E-mail addresses: nguyen.trananhtu@gmail.com (A. Nguyen),
antoine.dequidt@univ-valenciennes.fr (A. Dequidt),
michel.dambrine@univ-valenciennes.fr (M. Dambrine).
not all measured in most real-world applications. In order to deal
with this practical problem, output feedback control must be used.
In general, observer-based control scheme is proposed for uncon-
strained T–S systems (Liu and Zhang, 2003; Guerra et al., 2006).
However, the design problem becomes much more complicated
when state and/or input constraints have to be explicitly consid-
ered. As highlighted in Ding (2009), this control issue is not well
addressed in the literature.
Due to physical/technical limitations and/or safety constraints,
actuator saturation is unavoidable in almost all real applications.
This phenomenon can severely degrade the closed-loop system
performance and, in some cases, may lead to system instability.
Motivated by this practical control aspect, a great deal of effort has
been focused on saturated systems (Tarbouriech et al., 2011). In
the literature, several methods now exist to handle saturation
effects, but the most popular and effective one remains the anti-
windup (AW) approach (see for instance Kothare et al. (1994)).
LMI-based synthesis of anti-windup compensators has been pro-
posed recently to synthesize either static (Gomes da Silva and
Tarbouriech, 2005; Mulder et al., 2009) or dynamic anti-windup
compensators (Cao et al., 2002; Grimm et al., 2003; Hu et al.,
2008); an overview of these results can be found in the survey
(Tarbouriech and Turner, 2009). Most of these works deal with a
two-step method in which the controller and the AW strategy are
designed separately. This method often proves to be satisfactory
but it has however some drawbacks. First, only sub-optimal

http://dx.doi.org/10.1016/j.engappai.2015.01.005
0952-1976/& 2015 Elsevier Ltd. All rights reserved.
 A. Nguyen et al. / Engineering Applications of Artificial Intelligence 40 (2015) 76–83 77

solution can be achieved since the controller and its AW compen- which actually provides the less-conservative results among all non-
sator are designed separately (Sawada et al., 2009). Second, the quadratic approaches found in the open literature.
effect of the nominal controller on the closed-loop performance In this paper, we address a novel one-step method to design
under saturation is completely ignored (Mulder et al., 2009). To simultaneously a DOFC and an AW compensator for a disturbed T–S
overcome these drawbacks, an alternative solution called one-step system subject to control input and system state constraints. The
method, which designs simultaneously the controller and the AW proposed approach has some special features deserving particular
compensator, has been proposed. Among the few works existing in attention: quadratic boundedness of the trajectories (see Ding (2009)
the literature, we can notably cite (Mulder et al., 2009) in which an and references therein) is ensured for any admissible initial condition
LMI solution ensuring global stability and performance is pro- and disturbance signal, as well as a maximal gain for the unsaturated
posed for systems that are stable in open loop; Sawada et al. system. Note that this performance may be ensured regionally for the
(2009) propose a regional approach based on a change of variable; saturated case as in Dai et al. (2009); however, this decoupling allows
whereas Dai et al. (2009) rely on the parameter elimination reducing the conservatism of the results. Moreover, it will be shown
approach of Skelton et al. (1998). Both latter approaches are based that the control design can be formulated as a multi-objective LMI
on some preliminary results proposed in Hu et al. (2006). optimization problem. In such a way, the obtained controller can solve
To date, a large number of works on AW-based design are available the tradeoff between some predefined closed-loop requirements.
for linear systems (Tarbouriech et al., 2011), but very few works deal The paper is organized as follows. Section 2 describes the design
with nonlinear cases, especially when exogenous disturbance signals problem and recalls some preliminaries results. The main result is
are actively present (Gomes da Silva Jr. et al., 2013). In the T–S control stated in Section 3. In Section 4, a constructive control design is
framework, there are a couple of works devoted to the analysis or presented as a multi-objective LMI optimization problem. The results
control design of saturated systems (for instance Cao and Lin (2003), of the paper are effectively illustrated through an example in Section
Tseng and Chen (2006), Du and Zhang (2009), Bezzaoucha et al. 5. Finally, Section 6 gives some concluding remarks.
(2013), Ariño et al. (2010)). However, very few papers are dedicated to The notations and terminology used in this paper are standard. For
AW synthesis for T–S systems: in Ting and Chang (2011), the authors an integer number r, Ωr denotes the set f1; 2; …; r g. ℝ þ ¼ ½0; 1Þ is the
addressed a two-step approach to deal with a continuous time-delay set of non-negative real numbers. ℒ2e ðℝn Þ denotes the extended
T-S systems; in Zhang et al. (2009), an interesting one-step approach ℒ2 -space composed of measurable functions f : ½t 0 ; 1Þ↦ℝn such that
based on piecewise fuzzy AW dynamic output feedback controller RT RT 1=2
t 0 ‖f ðt Þ‖ dt o1; 8 T Zt 0 ; and ‖f ‖2;T ¼ ð t 0 ‖f ðt Þ‖ dtÞ : xðiÞ is the
2 2
(DOFC) for discrete-time T–S systems has been proposed; note that
ith element of a vector x. x g y, with x; y A ℝn means that
these results seem to be valid only for systems that are stable in open-
xðiÞ  yðiÞ 4 0 for all iA Ωn . X ðiÞ denotes the ith row of a matrix X,
loop since no admissible sets of initial conditions are defined; and, at
last, Song et al. (2011) which extends the approach proposed in Gomes and symðXÞ ¼ X þ X T (for square matrices). X 4 0 means that X is a
da Silva and Tarbouriech (2005) to the case of T–S systems. It is symmetric, positive-definite matrix. I denotes the identity matrix of
noteworthy that an important point is neglected in all these results: appropriate dimension, and ðnÞ stands for matrix blocks that can be
besides control input saturation, the T–S model is only valid on a given deduced by symmetry in a partitioned matrix. For P A ℝnn such that
 
subset of the state space. This is of course true for any model of real- P 4 0; ℰðP Þ denotes the ellipsoid x A ℝn : xT Px r 1 . For any value of
world systems, but is fundamental in the writing of a T–S model using their arguments, the nonlinear functions η1 ; …; ηr are said to verify
Pr
the nonlinear sector decomposition approach (Tanaka and Wang, the convex sum property if ηi Z 0; 8 i A Ωr and i ¼ 1 ηi ¼ 1. The
2001). This validity domain can be represented by some constraints following notations are occasionally used:
on the state variables (see Example 2 for illustration of this fact). It is !1
Xr   Xr   X r X r    
particularly important to consider explicitly these state constraints in Yθ ¼ ηi θ Y i ; Y θ 1 ¼ ηi θ Y i ; Z θθ ¼ ηi θ ηj θ Z ij
the control design to ensure a good behavior of the closed-loop system i¼1 i¼1 i¼1j¼1
in response to disturbances. This fact has been very recently empha- ð1Þ
sized in Nguyen (2013), Nguyen et al. (2014), Klug et al. (2015). The  
two first cited references concern stabilization of nonlinear switching where Y i , Z ij are matrices of appropriate dimensions and ηi θ , i A Ωr
systems under saturation. The latter one deals with the design of a are functions sharing the convex sum property.
dynamic output feedback control for nonlinear systems represented in
Takagi–Sugeno form. Apart from the fact that we consider the effect of
2. Problem definition and preliminaries results
bounded disturbances on the design, a fundamental difference with
the work in Klug et al. (2015) is that they considered discrete-time
2.1. Control problem definition
systems, whereas this paper concerns continuous-time systems. This
allows them to use elegantly non-quadratic Lyapunov functions for the
2.1.1. Closed-loop system description
control law design. Multiple or non-quadratic Lyapunov functions are
Consider the following fuzzy T–S model described by (Tanaka
also developed in the continuous-time case, allowing one to obtain
and Wang, 2001) valid on a polyhedral domain:
less conservative conditions for stability analysis (Tanaka et al., 2003;
8
Mozelli et al., 2009). Concerning synthesis of control laws, the > Xr   
>
> x_ ¼ ηi θ Ai x þ Bui u þ Bwi w
situation becomes much more complex and remains open due to >
>
>
> i¼1
the presence of the time-derivative of the membership functions in >
>
>
< Xr   
the stability conditions. Results thus far developed lead to very z¼ ηi θ C zi x þ Dzu zw
i u þ Di w ð2Þ
complex design conditions or to controllers requiring the inversion >
> i¼1
>
>
>
> Xr   
in real time of a parameter-dependent matrix (see for instances >
>
Guelton et al. (2009), Guerra et al. (2012), Bouarar et al. (2013)). The
>y¼
>
: ηi θ C yi x þ Dyw
i w
i¼1
goal of this paper is to obtain a controller that may be easily
implemented in practice. For this reason, the obtained results rely where x A P x  ℝnx , u A ℝnu , w A ℝnw , y A ℝny and θ A ℝk are respec-
on the quadratic stabilization approach. However, as will be shown in tively, the state, the control input, the disturbance, the measured
Section 5 through numerical example, in the context of input- output and the scheduling variable vectors of the system. The
saturated T–S systems, the proposed results may be competitive in regulated output vector z A ℝnz is used for performance purposes.
some sense with respect to those presented in Guerra et al. (2012) For iA Ωr , the real constant matrices of appropriate dimensions Ai ,
78 A. Nguyen et al. / Engineering Applications of Artificial Intelligence 40 (2015) 76–83

z y yw

Bui , Bw zu zw
i , C i , Di , Di , C i , Di represent the set of r local linear Define xTcl 9 xT xc T , the constrained closed-loop system is obtained
subsystems and the nonlinear scalar functions ηi θ satisfy the from (2) and (10):
convex sum property. For system (2), we consider the following 8
assumptions. > X r X r       
>
>
>
>
x_ cl ¼ ηi θ ηj θ Aij xcl þ Bwij w  Bψi þ Rψ Ecij ψ ðuc Þ
<
Assumption 1. The scheduling variable vector θ A ℝk is assumed i ¼ 1j ¼ 1
ð12Þ
to be a function of all measured signals of interest (states, external >
> Xr X r     
>
>
> z¼ ηi θ ηj θ ℂij xcl þ Dwij w  Dψi ψ ðuc Þ
disturbances, etc.) with the exception of the control input value u: : i ¼ 1j ¼ 1

Assumption 2. The input vector u is subject to symmetric magni-

tude limitations: where
umaxðiÞ r uðiÞ r umaxðiÞ ; umaxðiÞ 4 0; 8 i A Ωnu : ð3Þ 2 3 2 3 " #  
Ai þ Bui Dc C yj Bui C cj Bw u c yw
i þ Bi D Dj Bui 0
Aij ¼ 4 5; Bijw ¼4 5; Biψ ¼ ; Rψ ¼
Bci C yj Acij Bci Dyw
j 0 I
Assumption 3. The disturbance signal w is assumed to belong to
the set W δ defined by    h i
c y ψ
    ℂij ¼ C zi þDzu
i D Cj Dzu c
i Cj ; Dijw ¼ Dzw zu c yw
i þDi D Dj ; Di ¼ Dzu ð13Þ
W δ ¼ w A ℒ2e ℝnw : wT ðt ÞRwðt Þ r δ; 8 t Z t 0 ;
i
ð4Þ

where the matrix R 4 0 and the bound δ 4 0 are known. and the controller output is given as
X
r   
Assumption 4. The validity domain P x of (2) is a polyhedral set uc ¼ ηi θ K i xcl þ K wi w ; ð14Þ
defined by i¼1
n o h i
c y c
P x ¼ x A ℝnx : xT h1 r 1; 8 k A Ωq ;
k c yw
ð5Þ where K i ¼ D C i C i and K w i ¼ D Di .
Using notations (1), the closed-loop system (12) is rewritten as
k
where the q vectors h1 A ℝnx are given.  
ψ
θθ w  Bθ þ Rψ Eθθ ψ ðuc Þ
x_ cl ¼ Aθθ xcl þ Bw c
Consider now the unconstrained dynamic output feedback
controller (DOFC) in the form of (Tanaka and Wang, 2001): ψ
8 θθ w  Dθ ψ ðuc Þ
z ¼ ℂθθ xcl þ Dw ð15Þ
> X r X r     ij Xr  
>
> x_ c ¼ η θ η θ A x þ ηi θ Bic y þ v; xc ð0Þ ¼ 0
>
> i j c c and the DOFC output (14) is rewritten as
< i ¼ 1j ¼ 1 i¼1
ð6Þ uc ¼ K θ xcl þ K w
θ w: ð16Þ
>
> X r   i
>
> uc ¼ ηi θ C c xc þDc y
>
:
i¼1

2.1.2. Control problem definition

where xc A ℝnx , uc A ℝnu are, respectively, the state and the output
Local stability and performance specifications of the closed-
vectors of the controller. The additional term v is introduced to
loop system (12) will be presented in terms of Lyapunov analysis
minimize the undesirable degradation of closed-loop performance
tools. Our goal is to propose a systematic method to design a DOFC
caused by input saturation. The controller (6) has to be designed to
together with its AW compensator of the form (10) such that the
guarantee the local stability and some performance requirements
closed-loop system satisfies the following properties.
for the closed-loop system. Because of the input limitation, the
actual control signal injected into the system is subject to the
Property 1. [Regional Quadratic Stability]. For w ¼ 0; there exist a
saturation effect:
quadratic function V ðxcl Þ ¼ xTcl Pxcl with P 4 0 and a real number
u ¼ sat ðuc Þ; ð7Þ τ1 4 0 such that V_ ðxcl Þ o  τ1 V ðxcl Þ along the trajectory of the closed-
where each component of the saturation function sat ð U Þ is given loop system for any initial augmented state in the ellipsoid ℰðP Þ. This
by fact means that, for any xcl ðt 0 Þ A ℰðP Þ, the corresponding trajectory of
     the closed-loop system will converge exponentially to the origin with
sat ðuc ÞðiÞ 9 sign ucðiÞ min ucðiÞ ; umaxðiÞ ; iA Ωnu : ð8Þ a rate less than τ1 =2.
Then, the interactions between the system (2) and the constrained
Property 2. [State Constraints]. For any initial augmented state in
controller are given as
ℰðP Þ and any disturbance signal in W δ , the trajectory of the closed-
r X
X r   loop system (12) remains in the polyhedral region described by linear
u ¼ sat ðuc Þ; v¼ ηi θ Ecij ðsat ðuc Þ  uc Þ: ð9Þ
inequalities:
i¼1j¼1
n o
P xcl ¼ xcl A ℝ2nx : hk xcl r 1; 8 k A Ωq ;
T
From (6) and (9), the DOFC combined with the anti-windup strategy ð17Þ
can be expressed as h iT
8 kT
where hk ¼ h1
>
>
X r X r      Xr   0 (implying that the system state x remains in
> x_ c ¼
>
> ηi θ ηj θ Acij xc  Ecij ψ ðuc Þ þ ηi θ Bci y the validity domain P x of the model (2)).
< i ¼ 1j ¼ 1 i¼1
ð10Þ
>
> X r   Property 3. [Finite ℒ2e -Gain Performance]. For xcl ðt 0 Þ ¼ 0; and for
>
> uc ¼ ηi θ C ci xc þDc y
>
: any w A W δ such that no saturation occurs, there exists a positive real
i¼1
number γ such that ‖z‖2;T f r γ ‖w‖2;T f for all T f Z t 0 :
where Ecij are the anti-windup gains to be designed and ψ ðuc Þ 9uc 
sat ðuc Þ. The ith component of the decentralized dead-zone nonlinear-
ity ψ ðuc Þ is defined as 2.2. Preliminaries
(  
0 if uðiÞ  rumaxðiÞ
ψ ðiÞ ðuÞ ¼     ð11Þ In this section, some important preliminaries results needed for
uðiÞ  sign uðiÞ umaxðiÞ if uðiÞ  4 umaxðiÞ
design problem in Section 3 will be presented.
 A. Nguyen et al. / Engineering Applications of Artificial Intelligence 40 (2015) 76–83 79

2.2.1. Generalized sector bound condition for input h saturation

i 3. Main results
i i
Given the matrices K i A ℝnu 2nx and Gi ¼ G1 G2 A ℝnu 2nx ,
with Gi1 A ℝnu nx , Gi2 A ℝnu nx , for i A Ωr , we define the polyhedral The theorem below provides LMI conditions to design the
set P u as follows: DOFC together with its anti-windup strategy of the form (10)
which solves the control problem defined in Section 2.1.2.
P u ¼ \ ri ¼ 1 Ξ ðK i  Gi Þ; ð18Þ
where Theorem 1. Given positive real numbers δ and τ1 , assume there
     exists positive definite matrices P 11 A ℝnx nx , X 11 A ℝnx nx , positive
Ξ ðK i  Gi Þ 9 xcl A ℝ2nx :  K iðlÞ  GiðlÞ xcl  rumaxðlÞ ; 8 l A Ωnu : diagonal matrices Si A ℝnu nu , matrices W ij A ℝnx nu , U i A ℝnu nx ,
ð19Þ V i A ℝnu nx , A^ ij A ℝnx nx , ℬ
^ i A ℝnx ny , C^ i A ℝnu nx and D
^ A ℝnu ny for
ði; jÞ A Ωr  Ωr , and positive real numbers τ2 , γ such that
The following lemma is an extended version of Lemma 1 of Gomes " #
da Silva and Tarbouriech (2005). X 11 I
4 0; ð24Þ
I P 11
Lemma 1. Consider the function ψ ðuc Þ defined in (11) with uc
defined in (16). If xcl A P u \ P xcl , then the following condition is
2 3
verified: X I X 11 h1
k

!  1" " r #  # 6 11 7
X X X r 6 n k 7 Z 0; 8 k A Ωq ð25Þ
r
xcl 4 P 11 h1 5
ψ T ðuc Þ ηi Si ψ ðuc Þ  ηi Gi ηi K wi r 0; ð20Þ
i ¼ 1 i ¼ 1 w n n 1
i¼1
2 3
where Si A ℝnu nu are any positive diagonal matrix and η1 ; ⋯; ηr are X 11 n n
scalar functions satisfying the convex sum propriety. 6 I P n 7
6  11 7 Z 0; 8 ði; lÞ A Ωr
4 5
b
C iðlÞ  U iðlÞ b i
DC y V iðlÞ u2maxðlÞ
Proof. The proof of this result can be performed in the same ðlÞ
fashion as the one given in Nguyen (2013).
 Ωnu ð26Þ
2.2.2. Other preliminary results
Some other results that will be useful in the proof of our main τ1  τ2 δ 4 0 ð27Þ
result are recalled below. n
Ψ ii o 0; 8 i A Ωr
Lemma 2. (Tuan et al., 2001) Let ϒ ij , ði; jÞ A Ωr  Ωr be symmetric
matrices of appropriate dimensions and η1 ; …; ηr be a family of 2
Ψ þ Ψ ij þ Ψ ij o0; 8 ði; jÞ A Ωr  Ωr and iaj ð28Þ
P
r Pr r 1 ii
numbers satisfying the convex sum property. Condition n
ηi ηj ϒ ij o 0 is satisfied if i¼1j¼1
Λii o 0; 8 i A Ωr
(
ϒ ii o 0; 8 i A Ωr 2
ð21Þ Λ þ Λij þ Λij o 0; 8 ði; jÞ A Ωr  Ωr and ia j ð29Þ
r 1 ii
r  1ϒ ii þ ϒ ij þ ϒ ij o0; 8 ði; jÞ A Ωr  Ωr
2
and ia j
where Ψ ij in (28) and Λij in (29) are given as follows:
2 3
Remark 1. Note that more performing relaxation result exists in ℍij Iij U Ti  Bui Sj ^ yw þ Bw
Bui DD j i
Liu and Zhang (2003), Ariño and Sala (2007). However, Lemma 2 6 7
6 n J V Ti  W ij P 11 Bw ^ yw 7
i þ ℬi Dj 7
constitutes a good tradeoff between complexity and conservatism Ψ ij 9 6
6
ij
7 ð30Þ
6 n n  2Sj ^ yw
DD 7
since it does not require the introduction of auxiliary variables. 4 i 5
n n n  τ2 R
Lemma 3. (Boyd et al., 1994) The ellipsoid ℰðP Þ is included in the
polyhedral set P u defined in (18) if and only if 2 3
Lij Mij ^ yw þ Bw
Bui DD C zi X 11 þ Dzu ^
j i i Cj
6 7
6 n ℕij P 11 Bw ^ yw C zi þ Dzu ^ y 7
i þ ℬi Dj i DC j
   T Λij 9 6
6
7
7 ð31Þ
K iðlÞ  GiðlÞ P  1 K iðlÞ  GiðlÞ r u2maxðlÞ ; 8 ði; lÞ A Ωr  Ωnu ; ð22Þ 6 n n  γI Dzw þ D zu ^ yw 7
DD
4 i i j 5
n n n  γI

with

   
ℍij 9 sym Ai X 11 þBui C^ j þ τ1 X 11 ; ^ T þ Ai þ Bu DC
Iij 9 A ^ y þ τ1 I; ^ i C j þ τ1 P 11 ;
Jij 9 sym P 11 Ai þ ℬ
ij i j y
  T
  ð32Þ
Lij 9 sym Ai X 11 þ Bui C^ j ; ^ þ Ai þBu DC
Mij 9 A ^ y; ℕij 9sym P 11 Ai þ ℬ^ iCj :
ij i j y

Let P 12 and X 12 be two nonsingular matrices satisfying the

condition:
where K iðlÞ , GiðlÞ are, respectively, the lth rows of the matrices K i
and Gi . P 11 X 11 þ P 12 X T12 ¼ I: ð33Þ
Lemma 4. (Boyd et al., 1994) The ellipsoid ℰðP Þ is included in the
polyhedral set P xcl defined in (17) if and only if Then, the DOFC together with its anti-windup compensation (10)
given by, 8 ði; jÞ A Ωr  Ωr :
 
" # 1
Ecij ¼ P 12 W ij Sj 1 P 11 Bui
P hk
T Z 0; 8 k A Ωp : ð23Þ
hk 1 ^
Dc ¼ D
80 A. Nguyen et al. / Engineering Applications of Artificial Intelligence 40 (2015) 76–83

 
C ci ¼ C^ i  Dc C yi X 11 X 12
T ℒ2e -gain from the admissible disturbance w to the regulated
  output z will be less than γ if
1 ^
Bci ¼ P 12 ℬi  P 11 Bui Dc 2 T 3
    Aθθ P þ PAθθ PBw θθ ℂTθθ
1 ^ T 6   7
Acij ¼ P 12 Aij  P 11 Ai þ Bui Dc C yj X 11 P 12 Bci C yj X 11  P 11 Bui C cj X T12 X 12 6 n  γI Dw
T 7
o 0: ð42Þ
4 θθ 5
ð34Þ n n  γI
solves the control problem stated in Section 2.1.2. Applying again Lemma  2 and then using a congruence transforma-
tion with diag Π 1 ; I; I , it can be found that (29) implies inequality
T
Proof sketch. Conditions (24) and (33) imply the existence of two
matrices P 22 and X 22 such that the block matrices P and X given by (42). This fact shows that Property 3 holds which completes
Scherer et al. (1997) the proof.
" # " #
P 11 P 12 X 11 X 12
P¼ ; X¼ ð35Þ
P T12 P 22 X T12 X 22 4. Anti-windup based dynamic output feedback controller
synthesis
1
are definite positive and X ¼ P . Note also that the same proper-
ties imply that matrices P 12 and X 12 are regular. Let us introduce The feasibility problem stated in Theorem 1 provides the
the matrices conditions to design the controller satisfying some predefined
" # " #
X 11 I I P 11 requirements in Section 2.1.2. In this section, the design method
Π1 9 XT 0 ; Π 2 9 PΠ 1 ¼ 0 PT : ð36Þ will be formulated as a multi-objective optimization problem. Two
12 12
objectives are considered: maximization of the disturbance rejec-
tion and of the estimated domain of attraction.
  In order to obtain a sufficiently large estimated domain of
By congruence transformation with diag Π 1 ; I ; the inequality
T
attraction, we can consider, for example, the reference ellipsoid set
(23) is shown to be equivalent to (25). This implies that the X R  ℝ2nx containing the origin:
ellipsoid ℰðP Þ is included in the polyhedral set P xcl defined in (17).  
Similarly, by Schurcomplement
 lemma and congruence transfor- X R ¼ xcl A ℝ2nx : xTcl P 0 xcl r 1 ; ð43Þ
mation with diag Π 1 ; I ; the inequality (22) is shown to be
T
where P 0 4 0: Then, finding the largest ellipsoid ℰðP Þ can be
equivalent to (26), so that the ellipsoid ℰðP Þ is included in the formulated as the following optimization problem:
T
polyhedral set P u defined in (18) with Gi1 9 V i and Gi2 9 U i X 12 
T
Gi1 X 11 X 12 . max τ subject to : τX R D ℰðP Þ: ð44Þ
τ;P 4 0
By Lemma 2, the inequalities (28) imply that Ψ θθ 9
P
r Pr     It is worth noting that X R defines the directions in which we want
ηi θ ηj θ Ψ ij o 0. Using a congruence transformation with to maximize ℰðP Þ. Condition τX R  ℰðP Þ is equivalent to P 0 Z τ2 P.
i¼1j¼1
  By Schur complement lemma, this condition is equivalent to, with
diag Π 1 ; Sθ 1 ; I , this inequality is proved to be equivalent to
T
X ¼ P  1 and μ 9 1=τ2 :
2   3  
ψ
ATθθ P þ PAθθ þ τ1 P GTθ Sθ 1  P Bθ þ Rψ Ecθθ PBw μP 0 I
6 θθ 7 Z 0: ð45Þ
6 7 I X
6 n  2Sθ 1 Sθ 1 K w 7 o0: ð37Þ
4 θ 5
n n  τ2 R Note that maximization τ amounts to minimize
 μ. Pre- and post-
multiplying condition (45) by diag I; Π 2 and its transpose, the
T
h i
Pre- and post-multiplying (37) by xTcl ψ T ðuc Þ wT and its trans- condition (45) is equivalent to
2 3
pose, the following condition can be obtained after some algebraic I P 11
6 μP 0 0 P T 7
manipulations: 6 12 7
 6 7 Z 0: ð46Þ
h i x  6 X I 7
_ 4 11 5
V ðxcl Þ þ τ1 xcl Pxcl  τ2 w Rw  2ψ ðuc ÞSθ ψ ðuc Þ  Gθ K θ
1
T T T w cl
o 0: n
w I P 11
ð38Þ
Now, based on the result of Theorem 1, the following LMI
Since ℰðP Þ  P u \ P xcl , by Lemma 1, condition (38) implies that the
optimization problem solves the control design problem stated in
following condition is verified along the trajectories of the closed-
Section 2.1.2 while maximizing the disturbance rejection level and
loop system (12):
the estimated domain of attraction of the system (12):
V_ ðxcl Þ þ τ1 xTcl Pxcl  τ2 wT Rw o 0: ð39Þ  
min λγ þ 1  λ μ; ð47Þ
^ ij ;ℬ
P 11 ;X 11 ; S; U i ; V i ; W i ; A ^ i ;C^ i ;D;
^ τ2 ;γ ;μ
From (27) and (39), it follows that:
   
V_ ðxcl Þ þ τ1 xTcl Pxcl 1 þ τ2 δ  wT Rw o 0: ð40Þ such that the following LMI conditions hold: γ 4 0, μ 40; LMIs
(24)–(29); LMI (46).
Satisfaction of (40) ensures the ellipsoid ℰðP Þ is a robustly The positive weighting factor 0 r λ r 1 is chosen according to
positively invariant set (Blanchini, 1999) with respect to the desired trade-off between the closed-loop performance (charac-
closed-loop system (15). In addition, since ℰðP Þ  P xcl , these facts terized by γ ) and the size of the domain of attraction ℰðP Þ of the
prove Property 2. closed-loop system (characterized by μ).
Moreover, when w ¼ 0 (or w vanishes), it follows from (39) that
8 xcl A ℰðP Þ: Remark 2. It is important to stress that Eq. (33) has an infinite
number of solutions and the choice of matrices P 12 , X 12 is totally
V_ ðxcl Þ o  τ1 xTcl Pxcl ; ð41Þ
irrelevant for closed-loop performance (El Ghaoui and Scorletti,
which proves Property 1. 1996). As remarked in Dai et al. (2009), to date this degree of
Furthermore, in the absence of input saturation, i.e. ψ ðuc Þ ¼ 0, freedom is not yet exploited for the selection of the most desirable
by the well-known Bounded Real Lemma (Scherer et al., 1997), the controller. However, we have clearly pointed out in (46) that the
 A. Nguyen et al. / Engineering Applications of Artificial Intelligence 40 (2015) 76–83 81

matrix P 12 can be considered as a slack variable contributing to fact can be explained by the over-conservativeness of the unsatu-
enlarge the domain of attraction. For matrix inversion property, rated control approach in Guerra et al. (2012) which is not the case
the condition P 12 þ P T12 4 0 should be added to guarantee that P 12 when applying the parameter-dependent sector condition pro-
is nonsingular. Then X 12 is deduced from condition (33). Now, the posed in Lemma 1. Moreover, the degree of relaxation of the NQA
DOFC with its AW strategy (10) can be computed from (34). depends strongly on the control bound of the system. These
serious drawbacks of the NQA are inherently due to the handling
Remark 3. Note that the condition X 4 0 is equivalent to of the time-derivative of membership functions involved in the
1
P 11 4X 11 and if xð0Þ A ℰðP 11 Þ and xc ð0Þ ¼ 0, then we have derivation of the design conditions as stated in Section 1. Note that
xcl ð0Þ A ℰðP Þ. Since the set ℰðP Þ is robustly positively invariant with if closed-loop performance is considered for NQA, the obtained
respect to the closed-loop system, it follows that xcl ðtÞ A ℰðP Þ for all results are expected to be much more conservative (Guerra et al.,
w A W δ and t 4 0: Therefore, the trajectories x will never leave out 2012).
the projection of ℰðP Þ onto the planedefined  by xc ¼ 0. This
1
projection is nothing but the ellipsoid ℰ X 11 . Example 2. Consider the following nonlinear open-loop unstable
system:
8    
5. Illustrative examples >
> x_ ¼  x1 þ 0:1 þ 0:12x22 x2 þ 1:48 þ 0:16x32 u þ0:1w
> 1
>
< x_ ¼ x þ 0:1w
2 1
In this section, numerical examples are given for illustrating the ð49Þ
>
> z ¼ x2 þ u þ 0:1w
effectiveness of the proposed method. The implementations are >
>
: y ¼ x þ 0:2w
2
done using the SeDuMi solver and the YALMIP toolbox (Lofberg,
2004) with option sdpsettings(‘solver’,‘sedumi’,‘shift’,1e  4) for the Since the two nonlinearities f ðx2 Þ ¼ 0:1 þ0:12x22 and
resolution of all LMI problems. g ðx2 Þ ¼ 1:48 þ0:16x32 are unbounded functions of x2 , it is impos-
sible to obtain a T-S model that is globally equivalent to (49). One
Example 1. Consider the T–S system (2) adapted from Example 2
solution consists in restricting the range of evolution of variable x2 ,
in Guerra et al. (2012) with, 8 iA Ω2 :
for instance jx2 j r 1:5: We also consider an additional constraint on
       
2  10 1 a 5 b the first state variable, leading to the validity domain P x ¼
A1 ¼ ; Bu1 ¼ ; A2 ¼ ; Bu2 ¼ ;  
2 0 1 1 2 2 x A ℝ2 : jx1 jr 1:5; jx2 j r 1:5 . The control saturation limit will be
  taken as: umax ¼ 1.
0:1

Bw w
1 ¼ B2 ¼ ; C yi ¼ C zi ¼ 1 0 ; Dyw zw
i ¼ Di ¼ 0; Di
zu
On P x , functions f and g are such that
0:1 ( (
¼ 0; η1 ðx1 Þ ¼ ð1  sin ðx1 ÞÞ=2; η2 ðx1 Þ ¼ 1  η1 ðx1 Þ: ð48Þ f min r f ð U Þ r f max f ð U Þ ¼ ω1 f min þ ω2 f max
; ð50Þ
g min rg ð U Þ r g max g ð U Þ ¼ ω3 g min þ ω4 g max
The feasibility of (48) is checked using the proposed quadratic
anti-windup-based approach (QAWA) and the non-quadratic with f min ¼ 0:1; f max ¼ 0:37; g min ¼ 0:94; g max ¼ 2:02: The nor-
approach (NQA) (Theorem 3 in Guerra et al. (2012)) for all the malized nonlinear functions of the T–S model are given by
806 points of a 31  26 rectangle grid on the parameter space η1 ¼ ω1 ω3 ; η2 ¼ ω1 ω4 ; η3 ¼ ω2 ω3 ; η4 ¼ ω2 ω4 ; ð51Þ
ða; bÞ, a A ½  20; 25, b A ½0; 25. It is noteworthy that in Guerra et al.
(2012) static state-feedback controller is considered for undis- where
turbed T–S systems and closed-loop performance was not taken f max  f ð U Þ g max  gð U Þ
into account for the control design. Thus, it is not objective to ω1 ¼ ; ω2 ¼ 1  ω1 ; ω3 ¼ ; ω4 ¼ 1  ω3 :
f max  f min g max  g min
compare these two approaches since their control contexts are ð52Þ
different. However, checking the feasibility of their conditions
provides us an idea about the conservatism of each approach. For Then, with expressions (51) and (52), the nonlinear system (49)
this study, the following data are considered: δ ¼ 0:01, τ1 ¼ 0:5, can be exactly represented on the polyhedral set P x by the T–S
  model (2) with subsystem matrices given by, 8 i A Ω4
umax ¼ 10 and P x ¼ x A ℝ2 : jx1 j r π =2; jx2 j r π =2 . Note also that
     
all considered T–S systems are open-loop unstable. As can be  1  f min g min  1 f min
observed in Fig. 1, the feasibility space obtained with QAWA is A1 ¼ ; Bu1 ¼ ; A2 ¼ ;
1 0 0 1 0
larger than that of NQA for the considered grid. This interesting  
g max
T y

Bu2 ¼ ; Bw
i ¼ 0:1 0:1 ; C i ¼ C zi ¼ 0 1 ;
0
25    
 1  f max g min
A3 ¼ ; Bu3 ¼ ;
1 0 0
   
20  1 f max g max
A4 ¼ ; Bu4 ¼ ; Dyw zw zu
i ¼ 0:2; Di ¼ 0:1; Di ¼ 1: ð53Þ
1 0 0

15
In what follows, it will be shown that the proposed DOFC
b

satisfies all predefined properties stated in Section 2.1.2. To that

10 end, let us take τ1 ¼ 0:1, the reference shape X R ¼ ℰðP 0 Þ with
P 0 ¼ 100I and the amplitude-bounded
 disturbance is defined as
wðt Þ ¼ 0:5 sin 3π t þ π =4 , with δ ¼ 0:25. First, solving the convex
5 optimization problem (47) with the weighting factor λ ¼ 0, which
aims at maximizing the invariant set ℰðP Þ, yields
2 3
0:6459 0:0155 0:0406  0:0401
0
-20 -15 -10 -5 0 5 10 6 0:0155 0:6459  0:0401 0:0406 7
6 7
P¼6 7;
a 4 0:0406  0:0401 0:0111  0:0110 5
Fig. 1. Feasibility spaces obtained with QAWA ( þ ) and NQA (o). 0:0401 0:0406  0:0110 0:0111
82 A. Nguyen et al. / Engineering Applications of Artificial Intelligence 40 (2015) 76–83

μ ¼ 0:0066; γ ¼ 4:7607: ð54Þ behaviors for all closed-loop trajectories initialized in P x and these
trajectories remain inside this validity domain.
The estimated domain of attraction of the closed-loop system (in Now, solving the convex optimization problem (47) with the
the absence of disturbance) S 1 9 ℰðP 11 Þ is depicted in Fig. 2. It is aim of minimizing the disturbance effects (λ ¼ 1) yields
noteworthy that this set is not invariant. However, since the 2 3
35:0764 23:4791 12:2088 0:2352
ellipsoid ℰðP Þ is robustly invariant for all w A W δ , if xð0Þ A S 1 and
6 23:4791 61:2175 0:3027 8:3083 7
xc ð0Þ ¼ 0 then the corresponding closed-loop trajectory will stay in 6 7
  P¼6 7;
1 4 12:2088 0:3027 6:1661  3:0119 5
the set S 2 9 ℰ X 11 , see Remark 3. It can be also observed in
0:2352 8:3083  3:0119 14:1328
Fig. 2 that the ellipsoid S 2 is maximized along the direction of the
polyhedral set S 3 9 P u \ P xcl which is, in turn, included in the μ ¼ 0:8938; γ ¼ 0:2432: ð55Þ
!
P4
S4 9 Ξ ηi ðKi Gi Þ . Fig. 4 shows the evolution of the performance variable for two
i¼1 designs of the controller. The first design (Case 1) corresponds to
Fig. 3 shows that the closed-loop system is stabilized and the the minimization of ℒ2e -gain performance whose solution is given
disturbance is well attenuated. By means of simulation it can be in (55), while in Case 2 this performance is not taken into account
also checked that the obtained controller provides stable
1.2
2
1
1.5
0.8 With L2 performance

Performance variable
1
Without L 2 performance
0.6
0.5

0 0.4
x2

-0.5 0.2

-1 0

-1.5 -0.2

-2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -0.4
0 5 10 15 20
x Time [s]
1

Fig. 2. Projection of ℰðP Þ onto the plane defined by xc ¼ 0 and system trajectories. Fig. 4. Performance variable of Case 1 (solid line) and Case 2 (dashed line).

1 3
u
2 uc
Controller output
System states

0
umax
1
x1
-1
x2 0

-2 -1
0 10 20 30 0 10 20 30
Time [s] Time [s]

0.5 20

x c1
Performance variable

Controller states

0 0 x c2

-0.5 -20
z
w
-1 -40
0 10 20 30 0 10 20 30
Time [s] Time [s]
Fig. 3. Closed-loop system trajectories and disturbance attenuation (left) and corresponding control signal response and evolution of the controller state of the DOFC (right).
 A. Nguyen et al. / Engineering Applications of Artificial Intelligence 40 (2015) 76–83
in the control design: controller parameters are obtained solving
the feasibility LMIs of Theorem 1 without condition (29).
From this figure, the interest of the consideration of ℒ2e -gain
performance in the control design is clearly confirmed.
6. Concluding remarks
In this paper, a novel LMI-based approach has been proposed to
design simultaneously a dynamic output feedback controller
together with an anti-windup compensator for nonlinear systems
under input and state constraints and subject to bounded persis-
tent disturbances. Based on the Lyapunov stability theory, condi-
tions for the stability in closed-loop are given, which allows
formulating the design of the control as a multi-objective convex
optimization problem. In this way, the controller and anti-windup
compensator gains can be efficiently computed with semidefinite
programming solvers. By means of examples, the paper has
pointed out a very interesting fact, namely, the combination of a
generalized sector condition with quadratic Lyapunov approach
for input-saturated T–S systems leads to simple and efficient
conditions which may be competitive with respect to recent
non-quadratic approaches. This is particularly true when the
control bounds are not excessively large. It has been also shown
that the resulting controller satisfies several predefined closed-
loop properties. The proposed approach can be applied to a large
class of constrained nonlinear systems and the obtained control-
lers are without real difficulties in terms of implementation for
real-time applications.
Acknowledgments
This work was developed in the framework of the CERVIFER
project and of the International Campus on Safety and Intermod-
ality in Transportation. This project is supported in part by the
European Community (through the FEDER European Funds for
Regional Development) and the Nord Pas-de-Calais Region. The
authors gratefully acknowledge the support of these institutions.
Appendix A. Supplementary information
Supplementary data associated with this article can be found in
the online version at http://dx.doi.org/10.1016/j.engappai.2015.01.
005.
References
Ariño, C., Pérez, E., Sala, A., 2010. Guaranteed cost control analysis and iterative
design for constrained Takagi–Sugeno systems. Eng. Appl. Artif. Intell. 23,
1420–1427.
Ariño, C., Sala, A., 2007. Relaxed LMI conditions for closed-loop fuzzy systems with
tensor-product structure. Eng. Appl. Artif. Intell. 20 (8), 1036–1046.
Bezzaoucha, S., Marx, B., Maquin, D., Ragot, J., 2013. Stabilization of nonlinear
systems subject to actuator saturation. In: Proceedings of the IEEE International
Conference on Fuzzy Systems.
Blanchini, F., 1999. Set invariance in control. Automatica 35 (11), 1747–1767.
Bouarar, T., Guelton, K., Manamanni, N., 2013. Robust non-quadratic static output
feedback controller design for Takagi–Sugeno systems using descriptor redun-
dancy. Eng. Appl. Artif. Intell. 26 (2), 739–756.
Boyd, S., Ghaoui, L.E., Feron, E., Balakrishnan, V., 1994. Linear Matrix Inequalities in
System and Control Theory. Society for Industrial and Applied Mathematics
(SIAM), Philadelphia.
Cao, Y.Y., Lin, Z., 2003. Robust stability analysis and fuzzy-scheduling control for
nonlinear systems subject to actuator saturation. Trans. Fuzzy Syst. 11 (1),
57–67.
Cao, Y., Lin, Z., Ward, D., 2002. An anti-windup approach to enlarging domain of
attraction for linear systems subject to actuator saturation. IEEE Trans. Autom.
Control 47 (1), 140–145.
Cao, S.G., Rees, N.W.F.G., 1997. Analysis and design for a class of complex control
systems, Parts I and II. Automatica 33 (6), 1017–1039.
View publication stats
83
Dai, D., Hu, T., Teel, A.R., Zaccarian, L., 2009. Output feedback design for saturated
linear plants using deadzone loops. Automatica 45 (12), 2917–2924.
Ding, B., 2009. Quadratic boundedness via dynamic output feedback for con-
strained nonlinear systems in Takagi–Sugeno's form. Automatica 45,
2093–2098.
Du, H., Zhang, N., 2009. Fuzzy control for nonlinear uncertain electrohydraulic active
suspensions with input constraint. IEEE Trans. Fuzzy Syst. 17 (2), 343–356.
El Ghaoui, L., Scorletti, G., 1996. Control of rational systems using linear-fractional
representations and linear matrix inequalities. Automatica 32 (9), 1273–1284.
Feng, G., 2006. A survey on analysis and design of model-based fuzzy control
systems. IEEE Trans. Fuzzy Syst. 14 (5), 676–697.
Gomes da Silva Jr., J.M., Castelan, E.B., Corso, J., Eckhard, D., 2013. Dynamic output
feedback stabilization for systems with sector-bounded nonlinearities and
saturating actuators. J. Frankl. Inst. 350, 464–484.
Gomes da Silva, J.M., Tarbouriech, S., 2005. Anti-windup design with guaranteed
regions of stability: an LMI-based approach. IEEE Trans. Autom. Control 50 (1),
106–111.
Grimm, G., et al., 2003. Antiwindup for stable linear systems with input saturation:
an LMI-based synthesis. IEEE Trans. Autom. Control 48 (9), 1509–1525.
Guelton, K., Bouarar, T., Manamanni, N., 2009. Robust dynamic output feedback
fuzzy Lyapunov stabilization of Takagi–Sugeno systems – a descriptor redun-
dancy approach. Fuzzy Sets Syst. 160 (19), 2796–2811.
Guerra, T.-M., Bernal, M., Guelton, K., Labiod, S., 2012. Non-quadratic local
stabilization for continuous-time Takagi–Sugeno models. Fuzzy Sets Syst. 201,
40–54.
Guerra, T.M., Kruszewski, A., Vermeiren, L., Tirmant, H., 2006. Conditions of Output
stabilization for nonlinear models in the Takagi–Sugeno's form. Fuzzy Sets Syst.
157 (9), 1248–1259.
Hu, T., Teel, A.R., Zaccarian, L., 2006. Stability and performance for saturated
systems via quadratic and non-quadratic Lyapunov functions. IEEE Trans.
Autom. Control 51 (11), 1770–1786.
Hu, T., Teel, A.R., Zaccarian, L., 2008. Anti-windup synthesis for linear control
systems with input saturation: achieving regional, nonlinear performance.
Automatica 44, 512–519.
Klug, M., Castelan, E., Leite, V., Silva, L., 2015. Fuzzy dynamic output feedback
control through nonlinear Takagi–Sugeno models. Fuzzy Sets Syst 263, 92–111.
Kothare, M.V., Campo, P.J., Morari, M., Nett, C.N., 1994. A unified framework for the
study of anti-windup designs. Automatica 30 (12), 1869–1883.
Liu, X., Zhang, Q., 2003. New approaches to controller designs based on fuzzy
observers for T–S fuzzy systems via LMI. Automatica 39 (9), 1571–1582.
Lofberg, J., 2004. YALMIP: a toolbox for modeling and optimization in MATLAB. In:
Proceedings of the IEEE International Symposium on Computer Aided Control
Systems Design, Taipei.
Mozelli, L.A., Palhares, R.M., Souza, F.O., Mendes, E.M.A.M., 2009. Reducing
conservativeness in recent stability conditions of T–S fuzzy systems. Auto-
matica 45 (6), 1580–1583.
Mulder, E.F., Tiwari, P.Y., Kothare, M.V., 2009. Simultaneous linear and anti-windup
controller synthesis using multiobjective convex optimization. Automatica 45,
805–811.
Nguyen, A.-T., 2013. Advanced Control Design Tools for Automotive Applications
(Ph.D. thesis). University of Valenciennes.
Nguyen, A.-T., Dambrine, M., Lauber, J., 2014. Lyapunov-based robust control design
for a class of switching non-linear systems subject to input saturation:
application to engine control. IET Control Theory Appl. 8 (17), 1789–1802.
Sawada, K., Kiyamaa, T., Iwasaki, T., 2009. Generalized sector synthesis of output
feedback control with anti-windup structure. Syst. Control Lett. 58 (6),
421–428.
Scherer, C., Gahinet, P., Chilali, M., 1997. Multiobjective output feedback control via
LMI optimization. IEEE Trans. Autom. Control 42 (7), 896–911.
Skelton, R.E., Iwasaki, T., Grigoriadis, K., 1998. A Unified Algebraic Approach to
Linear Control Design. Taylor & Francis, New York.
Song, X., et al., 2011. Robust stabilization of state delayed TS fuzzy systems with
input saturation via dynamic anti-windup fuzzy design. Int. J. Innov. Comput.
Inf. Control 7 (12), 6665–6676.
Takagi, T., Sugeno, M., 1985. Fuzzy identification of systems and its applications to
modeling and control. Trans. Syst. Man Cybern. 15 (1), 116–132.
Tanaka, K., Hori, T., Wang, H.O., 2003. A multiple Lyapunov function approach to
stabilization of fuzzy control systems. IEEE Trans. Fuzzy Syst. 11 (4), 582–589.
Tanaka, K., Wang, H.O., 2001. Fuzzy Control Systems Design and Analysis: a Linear
Matrix Inequality Approach. Wiley-Interscience, New York: Wiley.
Tarbouriech, S., Garcia, G., Gomes da Silva Jr., J.M., Queinnec, I., 2011. Stability and
Stabilization of Linear Systems with Saturating Actuators. Springer-Verlag,
London.
Tarbouriech, S., Turner, M.C., 2009. Anti-windup design: an overview of some
recent advances and open problems. IET Control Theory Appl. 3 (1), 1–19.
Ting, C.S., Chang, Y.N., 2011. Robust anti-windup controller design of time-delay
fuzzy systems with actuator saturations. Inf. Sci. 181, 3225–3245.
Tseng, C.-S., Chen, B.-S., 2006. H1 fuzzy control design for nonlinear systems
subject to actuator saturation. In: Proceedings of the IEEE International
Conference on Fuzzy Systems.
Tuan, H., Apkarian, P., Narikiyo, T., Yamamoto, Y., 2001. Parameterized linear matrix
inequality techniques in fuzzy control system design. Trans. Fuzzy Syst. 9 (2),
324–332.
Zhang, T., Feng, G., Liu, H., Lu, J., 2009. Piecewise fuzzy anti-windup dynamic output
feedback control of nonlinear processes with amplitude and rate actuator
saturations. IEEE Trans. Fuzzy Syst. 17 (2), 253–264.