©istockphoto.

com/metamorworks

Fuzzy Control Systems:

Past, Present and Future
Anh-Tu Nguyen
Laboratory LAMIH UMR CNRS 8201, Université Polytechnique
Hauts-de-France, Valenciennes, France

Tadanari Taniguchi
IT Education Center, Tokai University, Hiratsuka, Japan

Luka Eciolaza
Department of Robotics and Automation, Mondragon University, Arrasate, Spain

Víctor Campos and Reinaldo Palhares

Department of Electronics Engineering, Federal University of Minas Gerais, Belo Horizonte, Brazil

Michio Sugeno
Tokyo Institute of Technology, Yokohama, Japan

Abstract—More than 40 years after fuzzy logic control edge. Due to the lack of a systematic framework to study
appeared as an effective tool to deal with complex processes, Mamdani fuzzy systems, we have witnessed growing interest in
the research on fuzzy control systems has constantly evolved. fuzzy model-based approaches with Takagi-Sugeno fuzzy sys-
Mamdani fuzzy control was originally introduced as a model- tems and singleton-type fuzzy systems (also called piecewise
free control approach based on expert’s experience and knowl- multiaffine systems) over the past decades. This paper reviews

Digital Object Identifier 10.1109/MCI.2018.2881644

Date of publication: 10 January 2019 Corresponding Author: Anh-Tu Nguyen (Email: [email protected])

56 IEEE Computational intelligence magazine | FEbruary 2019 1556-603x/19©2019ieee

the key features of the three above types of fuzzy systems. consequents. However, T-S fuzzy systems do not have linguistic
Through these features, we point out the historical rationale for variables since only functional membership functions (MFs) are
each type of fuzzy systems and its current research mainstreams. used without labels. PMA systems are a simplified special case
However, the focus is put on fuzzy model-based approaches of both previous types of fuzzy systems. Indeed, PMA systems
developed via Lyapunov stability theorem and linear matrix are Mamdani fuzzy systems with singleton consequents and
inequality (LMI) formulations. Finally, our personal viewpoint they can be obtained from T-S fuzzy systems when only the
on the perspectives and challenges of the future fuzzy control constant terms in the consequents are present. While Mamdani
research is discussed. fuzzy systems are deeply concerned with fuzzy set and logic,
T-S fuzzy systems are only concerned with fuzzy set in their

F
I. Introduction premises. PMA systems stand independently of fuzzy set and
uzzy control was initiated by Mamdani [1] in 1974 stim- logic since there is no need to use neither linguistic variables
ulated by the Zadeh’s two seminal papers on fuzzy algo- nor MFs. In the case of PMA systems, these functions in the
rithms [2] in 1968 and linguistic analysis [3] in 1973. In premises are of a triangular shape and only play roles as param-
these papers, Zadeh presented a method of system mod- eters for interpolation which are not necessarily interpreted as
eling based on fuzzy IF-THEN rules with linguistic variables. membership functions in the conventional “fuzzy” sense. How-
The first application of fuzzy logic control was performed by ever, PMA systems derived from Mamdani may keep their sta-
Mamdani and Assilian on a laboratory steam engine [4] which tus in fuzzy control and linguistic labels could be assigned to
led to a great impact on the fuzzy control research. Indeed, singletons if necessary. Note that all three types of fuzzy systems
many fuzzy control systems have been proposed since the pub- are known to have general approximation capability for any
lication of the original paper in 1975. Generally speaking, there nonlinear functions [9]. However, compared to two other types,
are three types of fuzzy systems as classified by Sugeno in [5] T-S fuzzy modeling can drastically reduce the number of fuzzy
according to the consequent parts of IF-THEN rules. First, rules, especially for high dimensional complex systems [5].
Mamdani-type fuzzy systems [4] are defined by IF-THEN During the first ten years since Mamdani’s successful applica-
rules associated with linguistic variables as tion of fuzzy logic control, researchers were faced with a lot of
criticisms from the conventional control theorists as reported in
Rule R i: IF x 1 is M i1, x 2 is M i2, f, x m is M im [5], [10], [11].The main reason is that there was no stability analy-
THEN y is H i, i = 1, 2, f, n sis available for fuzzy control at that time [12]. To answer these
criticisms, T-S fuzzy systems were newly introduced in 1985.
where H i and M ij, j = 1, 2, f, m, are fuzzy sets, n is the Since T-S fuzzy systems are associated with linear models in their
number of fuzzy rules. Second, Takagi-Sugeno (T-S) fuzzy sys- consequents, model-based stability analysis and control design can
tems [6] are with functional consequents be performed for fuzzy systems using conventional Lyapunov-
based approaches as firstly shown in [13]. T-S fuzzy modeling can
Rule R i: IF x 1 is M i1, x 2 is M i2, f, x m is M im be used to represent exactly a nonlinear system in a compact set
THEN y is f i (x 1, x 2, f, x m), i = 1, 2, f, n. of the state space, where the nonlinearities are embedded in
membership functions [14]. Therefore, it is generally impossible to
The function f i ($) is usually linear as follows: make a non-conservative stability analysis of a given nonlinear
system with T-S fuzzy model-based approaches. More clearly, T-S
f i (x 1, x 2, f, x m) = b i + a i1 x 1 + a 2i x 2 + f + a im x m, fuzzy systems belong to a class of polytopic uncertain systems
whose uncertainties are caused by the nonlinearity of the MFs. As
where the coefficients b i, a ij, with j = 1, f, m, are constants. such, T-S fuzzy model-based stability analysis always remains con-
Third, singleton-type fuzzy systems [5] are with fuzzy rules of servative [15]. PMA models were then presented to overcome this
the following form: drawback of T-S fuzzy models. As shown in [5], PMA models are
Rule R i: IF x 1 is M i1, x 2 is M i2, f, x m is M im fully parametric like linear systems, though they are approximate
models of nonlinear systems in their nature. Because of this
THEN y is b i, i = 1, 2, f, n
advantage, it is theoretically possible to derive necessary and sufficient
where b i is a singleton, i.e., a real number. A singleton-type fuzzy stability conditions for PMA systems just as in the case of linear
system is recently called piecewise multiaffine (PMA) system in systems [16]. This cannot be the case for other types of fuzzy sys-
[7] since its input-output relation with respect to an affine-in- tems. Since T-S fuzzy systems and PMA systems include linear
control system is found to be a multiaffine function. This type of systems as a special case, it is expected that any theoretical frame-
fuzzy systems is used to be called PB (piecewise bilinear, more work developed for these both types of fuzzy systems would be
precisely biaffine) systems because for a two-dimensional case, the more general than its linear counterpart.
output can be expressed as y = ax 1 + bx 2 + cx 1 x 2 + d, for some The intention of this paper is to provide a concise overview
scalars a, b, c and d, see [5], [8]. on the fuzzy control research since the pioneering works per-
Mamdani-type fuzzy systems are linguistically understand- formed by Mamdani’s group at Queen Mary College. Here, no
able since fuzzy variables are used in both the premises and the attempt is made to comprehensively review the literature which

FEbruary 2019 | IEEE Computational intelligence magazine 57

 between different control objectives can be per-
The design of Mamdani fuzzy control remains formed by fuzzy reasoning. For these reasons,
model-free and essentially heuristic. Therefore, we until now MFC has been successfully applied to
a large number of industrial processes, see [12],
still lack a systematic framework as well as analytical [23], [25], [26] for constructive surveys on
tools to study rigorously the stability of Mamdani prominent applications of MFC.
fuzzy systems. Unfortunately, except for a few exceptions,
the design of MFC remains model-free and
essentially heuristic. As a direct consequence, we
includes several excellent books [14], [17]–[20] and thousands still lack at present a systematic framework as well as analytical
of technical articles, see for instance [5], [10], [12], [21]–[23] tools to study rigorously the stability of Mamdani fuzzy systems
and references therein. Instead, only a selective and exemplary [22]. Establishing such a theoretical stability framework for Mam-
list of some breakthrough results is given to tracing back the dani fuzzy systems is expected to be particularly challenging due
evolution of fuzzy control systems. Our primary aim is to pro- to the novelty of the fuzzy mathematics and language [7]. The
vide a historical rationale for each type of fuzzy systems and its mainstream idea for stability analysis has been to consider the
current research status without excessive mathematical com- Mamdani fuzzy controller as a nonlinear controller, then the fuzzy
plexity. The emphasis is put more on the fuzzy model-based sta- control design is recast as a nonlinear control approach using
bility analysis than model-free fuzzy one. In particular, for the absolute stability theory, sliding mode control, adaptive fuzzy con-
sake of simplicity and illustration, only approaches based on trol, etc. Excellent reviews on the stability issues of MFC systems
Lyapunov stability theory for continuous-time dynamical sys- can be found, for instance, in [5], [12], [22].
tems without any specific performance issue are focused in the To overcome the “model-free” major drawback of MFC,
paper. However, it is stressed that theoretical results on stability fuzzy model-based control approaches were originally pro-
analysis and control design with or without various perfor- posed by Sugeno’s research group at Tokyo Institute of Tech-
mance specifications, for instance H 3 and H 2 with respect to nology [6], [13]. Such control approaches enable a systematic
external disturbances, robustness with respect to time delay or framework to deal with the stability analysis and control design
modeling uncertainty, etc., and their discrete-time counterparts of nonlinear dynamical systems in the following general form:
have also been widely reported in the fuzzy control literature
[14], [22]. Finally, we present our viewpoint on the perspectives xo (t) = f (x (t)) + g (x (t)) u (t)(1)
and challenges of the future fuzzy control research.
Notation: X N denotes the set " 1, 2, f, N , . R is the field where x (t) ! R n is the state vector and u (t) ! R m is the con-
of real numbers. For a vector x ! R n and i ! X n, x i denotes trol input. Without loss of generality, the nonlinear system (1)
the ith entry of x. I denotes the identity matrix of appropriate satisfies the following assumption.
dimension. For a matrix X, X < indicates its transpose and X (ij ) Assumption 1. The origin x 0 = 0 ! R n is an equilibrium
denotes its element of the ith row and jth column. For any of system (1) such that f (0) = 0. Moreover, the vector fields
square matrix X, X 2 0 indicates a symmetric positive definite f (·) and g (·) are sufficiently smooth, i.e., f ! C 2 and g ! C 2.
matrix, and He X = X + X <. The symbol * stands for matrix Fuzzy model-based approches to study theoretically the
blocks that can be deduced by symmetry. The time dependency nonlinear system (1) are discussed in the subsequent sections.
of the variables is omitted when convenient.
III. Takagi-Sugeno Fuzzy Systems
II. Mamdani Fuzzy Systems After a brief description of T-S fuzzy models, this section
The basic idea of Mamdani fuzzy control (MFC) is to represent reviews some key points on the stability analysis and control
the process states by means of linguistic variables and to exploit design of this type of fuzzy systems.
these variables as inputs to control rules [18]. Hence, MFC enables
to incorporate the expert’s skills and experience through a set of A. System Description
fuzzy IF-THEN rules [17], [19]. Thanks to this particular feature, A T-S fuzzy model of the continuous-time nonlinear system
the effectiveness of MFC has been clearly proved in the following (1) is described in the following form [14]:
situations [12]. First, no acceptable mathematical model is available
for the controlled plant. Second, human operators play a crucial Rule R i: IF z 1 (t) is M i1 and f and z p (t) is M ip
role in the control process and can provide qualitative control rules 
THEN xo (t) = A i x (t) + B i u (t) (2)
in terms of fuzzy logic sentences. Moreover, as pointed out by
Zadeh, MFC is task-oriented control which is in contrast to set- where R i denotes the ith fuzzy inference rule, r is the number
point-oriented feature of conventional control approaches [24]. This of inference rules, M ij, with i ! X r and j ! X p, are the fuzzy
allows MFC to achieve easily multi-objective goals by simply set- sets, and (A i, B i ) the state-space matrices of appropriate dimen-
ting some fuzzy control rules under one criterion and others sions of the ith local model. The vector of premise variables is
under a different performance criterion. The coordination defined as z (t) = [z 1 (t) fz p (t)]. Using the center-of-gravity

58 IEEE Computational intelligence magazine | FEbruary 2019

method for defuzzification, the T-S fuzzy model (2) can be rep- Concerning the first remark, the possibility to reformulate
resented in the following compact form: the stability analysis of T-S fuzzy systems as a convex optimiza-
r tion problem has undeniably sparked the growing interest in this
xo (t) = / h i (z) (A i x (t) + B i u (t))(3) type of fuzzy systems [14]. This enables systematic and effective
i =1
frameworks for stability analysis and control design of general
where the normalized MF h i (z) is defined as nonlinear systems, see detailed discussions on prominent results
in the 90s and in the early of 2010s in [11], [22]. Another very
p
~ i ( z) recent survey on this research topic can be found in [29].
h i ( z) = , ~ i (z) =% i
n j (z j), i ! Xr .
r The two other remarks are concerned with the conservative-
/ ~ i (z) j =1
ness issue of T-S fuzzy control theory [15]. Currently, most of
i =1
the research effort has been focused on this issue. There are two
The grades of membership of the premise variables in the mainstreams to reduce the conservatism of T-S fuzzy model-
respective fuzzy sets M ij are given as n ij (z j). Note that the based approaches, which are briefly discussed hereafter. The first
normalized MFs satisfy the following convex sum property: one is based on the choice of different families of Lyapunov
function candidates eventually combined with slack variables
r r
0 # h i (z) # 1, / h i (z) = 1, / ho i (z) = 0 (4) introduced via robust control tools such as Finsler lemma, S—
i =1 i =1 procedure, etc. [28]. The second mainstream consists in finding
ways to exploit more efficiently the information on the MFs
Remark 1. Although the premise variables can represent any for stability analysis.
type of variables in nonlinear systems, most of the works in the
T-S fuzzy literature deal with the case for which the premise 1) Relaxations with Different Choices of Lyapunov Functions
variables are composed of a subset of the system state x (t). This The conservatism of the results due to the use of a single qua-
assumption is also adopted along this paper. dratic Lyapunov function (6) was emphasized in [30], [31]. To
Remark 2. Using the sector nonlinearity approach in [14], overcome this drawback, more general classes of Lyapunov
the T-S fuzzy model (3) can be directly derived from (1). In this functions have been suggested in the T-S fuzzy literature,
case, both representations of the affine nonlinear system are including piecewise Lyapunov functions (PLFs) [30], [32], [33],
strictly equivalent in a compact set of the state space. fuzzy Lyapunov functions (FLFs) depending on the MFs [31],
[34], [35], line integral Lyapunov functions (LILFs) [34], [36],
B. Lyapunov-Based Stability Analysis [37], polynomial Lyapunov functions depending on the MFs
Consider the T-S fuzzy system (3) with u (t) / 0 as follows: with arbitrary degree [38]–[40], multidimensional fuzzy Lyapu-
r nov functions [41], [42], and so on.
xo (t) = / h i (z) A i x (t)(5) An interesting remark when using FLFs for stability analysis
i =1
of the continuous-time T-S fuzzy system (5) is that the stability
The first results on the stability analysis were proposed in [27] for conditions depend on the time-derivatives of the MFs, thus the
the T-S fuzzy system (5), and in [13] for its discrete-time counter- derivatives of the system state. This implies more numerical and
part. Consider a quadratic Lyapunov function (QLF) of the form theoretical complexities and requires a much larger effort to
obtain convex formulations than a quadratic Lyapunov based
V (x) = x < Px, P 2 0 (6) framework. To overcome this difficulty, different alternatives have
been proposed to consider the upper bounds of the time-deriva-
The following result is readily obtained. tives of the MFs. However, this usually leads to a local analysis
Theorem 1. [27] The equilibrium of the T-S fuzzy system setting [41]–[44]. Another alternative is to make use of PLFs or
(5) is globally asymptotically stable if there exists a common LILFs to avoid the presence of the time-derivatives of the MFs
positive definite matrix P such that in the stability conditions. However, in these cases some special
structures should be imposed on the Lyapunov matrices which
A <i P + PA i 1 0, i ! X r (7) introduce some conservatism. In addition, using PLFs is only
suitable for T-S fuzzy systems with triangular or trapezoidal MFs
The three following remarks deserve particular attention. inducing particular state-space partitions [22], [30], [32]. More-
❏❏ Conditions (7) are expressed in terms of linear matrix over, LILFs-based approaches usually require that the premise
inequalities (LMIs). Hence, the stability analysis can be easily variables are the system state, i.e., z / x. Nevertheless, this
checked with available numerical solvers [28]. assumption can be recently avoided in [37]. Applying PLFs for
❏❏ A common Lyapunov matrix variable P has to exist for all the stability analysis of T-S fuzzy systems was already discussed in
local linear subsystems. detail in [22]. Hence, only the two other cases, namely FLFs and
❏❏ The MFs of the T-S fuzzy system (5), i.e., its nonlinearities, LILFs, are given below to illustrate the above discussion.
are considered as uncertainty. Then, the stability analysis is Consider a simple example of FLFs depending explicitly on
embedded in the conventional robust control theory. MFs as follows [31]:

FEbruary 2019 | IEEE Computational intelligence magazine 59

 r
is also to maximize the domain of attraction included inside
V (x) = x < c / h i (z) Pi m x, Pi 2 0 (8)
i =1 the state region defined by the above assumption [41]. Note
that exploiting different types of properties of fuzzy MFs and
whose time-derivative is given by relaxing tools from robust control theory, several local stability
R r V conditions have been proposed in the FLFs literature [44]. It is
S / ho i (z) Pi *W important to note also that unlike the continuous-time case,
Vo (x) = ; o E S r W;xo E .
x < Si = 1 Wx
the stability analysis for discrete-time T-S fuzzy systems does
x
SS / h i (z) Pi 0WW not suffer from this kind of difficulty in dealing with the time-
Ti = 1 X derivatives of MFs when using FLFs [22], [46].
Remark 3. When the positive definite matrices are Remark 5. In Theorem 2, matrices X, M, N, are consid-
imposed as P1 = g = Pr = P, then the QLF (6) is straightfor- ered as slack variables, used for relaxation purposes. By impos-
wardly recovered from (8). Therefore, the QLFs are only a spe- ing X = 0, Pi = P 2 0, for 6i ! X r , and M =-P, it is easy
cial case of the FLFs. Observe also that the terms ho i (z), i ! X r , to prove that the result of Theorem 2 includes that of Theo-
appear explicitly in the expression of Vo (x). This makes the sta- rem 1. This theoretically confirms that compared to quadratic
bility analysis much more involved with this choice. It should Lyapunov functions, FLFs enable less conservative stability anal-
be stressed that although the fuzzy Lyapunov function (8) ysis, see also Remark 3.
belongs to a more general class of Lyapunov function candi- For T-S fuzzy systems with z k = x k, k ! X n, i.e., the prem-
dates compared to the quadratic one, there is no guarantee that ise variables are explicitly the state variables, line integral Lyapu-
FLFs-based results are less conservative than those derived from nov functions can be exploited to avoid dealing with the
QLFs, see [45] for a counterexample. This is primarily due to time-derivatives of the MFs [36]. Consider now a Lyapunov
the presence of a priori unknown time-derivatives of the MFs function of the following form:
in the theoretical developments when using FLFs [35]. Hence, r
the stability conservatism strongly depends on the way how V ( x) = 2 #
C [0, x]
/ h i (g) Pi g, dg (12)
i =1
such unknown time-derivatives are handled.
For simplicity, one could exploit the ideas related to proper- where C [0, x] is a path from the origin to the present state, dg
ties of the MFs to deal with their time-derivatives. For instance, is an infinitesimal displacement, G ·, · H denotes the inner prod-
from the convex sum property (4), it follows that uct. As shown in [36], if the matrices Pi are written as
R ri = 1 ho i (z) X = 0, for any matrix X. This implies that
r r Pi = D 0 + D i (13)
/ ho i (z) Pi = / ho i (z) (X + Pi) .
i =1 i =1
where
Assume there exist upper bounds of the time-derivatives of the
MFs such that ho i (z) # z i, for some positive scalars z i, i ! X r . R0 d 12 g d 1nVW R a i1 V
S Sd 11 0 g 0 W
Then, it follows that S 0 d a22i2 g 0 W
Sd 12 0 g d 2nW
D0 = S ,
r
h W D i = SS h h j h WW,
x < c / ho i (z) (Pi + X ) m x # x < Pz x (9)
h h j
SS W
i =1 d
T 1n d 2n g 0 WX S 0 0 g d anninW
T X
r
where Pz = R (Pi + X ). Using a null-term approach or
i =1 zi
Finsler lemma [28], inequality (9) can be exploited for stability and a ik indicating which M k fuzzy set is used for the ith rule.
analysis as summarized in the following theorem. Then, it follows from (12) and (13) that
Theorem 2. [34] Given a T-S fuzzy system (5), and upper r
bounds on the time-derivatives of the MFs as ho i (z) # z i, i ! X r . dV (x) = 2 / h i (x) Pi x.
If there exist positive definite matrices Pi ! R n # n, with i ! X r , i =1

and matrices X ! R n # n, M ! R n # n, N ! R n # n, satisfying the Exploiting this special structure of the Lyapunov function, the
following linear matrix inequalities: following theorem can be stated.
Theorem 3. [34] Given a T-S fuzzy system (5) with z / x.
Pi + X 2 0, i ! X r (10) If there exist symmetric matrices Pi ! R n # n, with i ! X r , and
matrices M ! R n # n, N ! R n # n such that
; E 1 0, i ! X r
Pz - MA i - A <i M < *
(11)
Pi + M < - NA i N + N <
Pi 2 0, i ! X r (14)
with Pz given by (9). Then, the T-S fuzzy system (5) is asymp-
; E 1 0, i ! X r (15)
-MA i - A <i M < *
totically stable.
Pi + M < - NA i N + N <
Remark 4. Differently from Theorem 1, Theorem 2 only
allows for local stability analysis of system (5) since the system with Pi having the structure in (13). Then, the T-S fuzzy system
state has to satisfy ho i (z) # z i, i ! X r . Hence, an implicit goal (5) is globally asymptotically stable.

60 IEEE Computational intelligence magazine | FEbruary 2019

 Remark 6. A great advantage of Theorem 3 compared to with Pi having the structure in (13), b ij = S (ij) + w i + v and
Theorem 2 is that the assumption on the upper bounds of the

Qi = ; E.
time-derivatives of the MFs is not required anymore for LMI- -MA i - A i< M < *
based stability analysis. Such an assumption is not always veri- Pi + M < - NA i N + N <
fied, especially in T-S fuzzy control context [35]. However, the
special structure of the matrices Pi as shown in (13) may Then, the T-S fuzzy system (5) is globally asymptotically stable.
induce some conservatism. In fact, up to now it is still hard to Remark 7. The key difference between Theorems 3 and 4
get a definitive answer on which approach leads to less conser- lies on the extra decision variable Z. Via this slack variable, the
vative stability conditions. shape information of the MFs (16), also represented by the sca-
lars b ij, can be exploited to reduce the stability conservatism in
2) Exploiting the Knowledge of the Membership Theorem 4. This shape information can be used to deal with
Functions for Stability Relaxations relatively complex MF shapes, i.e., with some minor modifica-
The MFs used to “blend” the local linear submodels of the T-S tions, multiple constraints on the MF shapes are easily handled.
fuzzy system (3) represent the nonlinearity of system (1). Howev- However, it usually requires a preliminary optimization step to
er, these MFs have been widely considered as system uncertainty, find the tightest constraint verifying (16) that could be applied
and only their convex sum property (4) has been exploited in to the MFs of the studied system.
most of the existing works based on quadratic, fuzzy and poly- Besides the two above mainstreams, research efforts have
nomial Lyapunov functions [15]. Whenever the shape of the been also devoted to reduce the stability conservatism caused
MFs and its intrinsic time-varying characteristic are not explicitly by the sufficiency of fuzzy summations, i.e., the ways how MFs
taken into account in the stability analysis, the conservativeness are dropped out to obtain a finite set of LMI conditions.
issue still remains, see further details in [29], [47]–[49]. Although numerous results have been proposed to deal with
Several approaches have been proposed to consider explicitly this source of conservativeness [55]–[61], the most prominent
the shape of the MFs in the stability analysis. These approaches approach relies on Pólya’s theorem. Based on checking the
can be classified into two following categories [50]. First, the positiveness of multidimensional matrices, the authors in [58]
membership-function-approximation approaches exploit the applied Pólya’s theorem to derive asymptotically necessary and
MF information via alternative similar functions such as staircase sufficient LMI-based conditions for the stability and perfor-
MFs [51], piecewise linear MFs [52]. Second, the membership- mance of T-S fuzzy systems. It should be stressed that as the
bound-dependent approaches exploit the bound information of homogeneous degree of a multiple summation fuzzy Lyapu-
MFs for stability analysis [48]. In addition, the MFs image space nov function increases, the conservatism of the stability con-
and the order relations among the MFs have been also exploit- ditions decreases thanks to the introduction of more degrees
ed, see for instance [50], [53], [54]. Note that membership- of freedom. However, Pólya’s theorem based approaches are
function-approximation and image-space approaches generally conceptual rather than implementable since the computa-
lead to a higher number of convex stability constraints. Con- tional burden swiftly increases in a way that most numerical
cerning membership-bound-dependent approaches, slack vari- solvers crash.
ables are usually introduced into the stability conditions for
relaxation purposes. As an illustrative example, the following C. Takagi-Sugeno Fuzzy Control Design
theorem presents the idea in [48] on using the MFs shape infor- An important application of stability theorems is to design sta-
mation to reduce further the stability conservatism of Theorem bilizing fuzzy controllers based on T-S fuzzy models of nonlin-
3 by introducing new relaxation variables. More discussions on ear plants. The control design of T-S fuzzy models are usually
the membership-function-dependent stability analysis can be addressed with the following steps: (1) choose a specific form
found in the recent survey [29]. of the control law, (2) find the respective closed-loop T-S fuzzy
Theorem 4. (adapted from [48] and [34]) Given a T-S representation, (3) apply a set of stability analysis conditions to
fuzzy system (5) with z / x. Consider the MFs vector the closed-loop representation, and (4) transform these condi-
h (x) = 6h 1 (x) h 2 (x) f h r (x)@< such that tions into LMI-based formulations. The challenge usually lies
on the last step since without any transformation at this step,
h (x) < Sh (x) + h (x) < w + v # 0 (16) the design conditions are expressed in terms of bilinear
matrix inequalities (BMIs) instead of LMIs, leading to
where S ! R r # r , w ! R r and v ! R are given. If there exist numerical difficulties.
symmetr ic matr ices Pi ! R n # n, i ! X r , and matr ices Different (state or output) feedback control schemes can be
Z ! R 2n # 2n, M ! R n # n, N ! R n # n such that applied to T-S fuzzy models [14], [22], [46]. The most com-
monly used control law is based on the so-called parallel dis-
Z 2 0, Pi 2 0, i ! X r (17) tributed compensation (PDC) concept, for which the fuzzy
Q i - b ii Z 1 0, i ! X r (18) controller shares the same fuzzy rules and sets as the T-S fuzzy
model. As a result, a PDC controller is obtained from a convex
Q i + Q j - (b ij + b ji) Z 1 0, i, j ! X r , j 2 i (19) blending of the linear local feedback gains and the MFs of the

FEbruary 2019 | IEEE Computational intelligence magazine 61

T-S fuzzy model. Within LMI-based control framework, this Remark 10. It should be stressed that various alternative
control approach was originally proposed in [62] and was control schemes with different degrees of design conservatism
named “PDC” in [55]. have been proposed in the T-S fuzzy literature, including non-
For illustration, consider a PDC control law of the form PDC control [35], [46], [64], [65], fuzzy observer-based
control with measurable premises [57], [66]–[68], with
r
u (t) = / h j (z) K j x (t)(20) unmeasurable premises [69]–[72], fuzzy static output feedback
j =1 control [73]–[77], etc., and their numerous extensions. Since
full state-information is generally not available in practice,
Then, the closed-loop T-S fuzzy system can be obtained from observers are usually designed to reconstruct the system state
(3) and (20) as follows: for state-feedback control. T-S fuzzy observer designs with or
without performance specifications have been largely devel-
r r
xo (t) = / / h i (z) h j (z)^A i + B i K jhx (t)(21) oped using Lyapunov stability arguments and LMI techniques,
i =1 j =1 see for instance [22], [78]–[84] and related references. Note
that in the framework of T-S fuzzy systems whose premises are
The basic result is based on the quadratic Lyapunov function state variables, two following cases are distinguished. If the
(6). For the closed-loop stability analysis, its time-derivative premise variables are measurable, then the approaches in [57],
along the trajectory of system (21) is required to be negative, [66], [67], [78], [79] can be directly applied. The second case,
i.e., Vo (x) = xo < Px + x < Pxo 1 0, which results in in which the premise variables are not measurable, is much
more challenging. This is due to a mismatch between the T-S
/ / h i (z) h j (z) He 6(A i + B i K j) < P@ 1 0 (22)
r r
fuzzy systems and the corresponding T-S fuzzy observers. In

i =1 j =1 this situation, it is necessary to estimate the premise variables as
shown in [69]–[72], [83].
Note that (22) is expressed in terms of BMI due to the cou-
pling between the Lyapunov matrix P and the feedback gains IV. Piecewise Multiaffine Systems
K j, j ! X r . However, by applying a simple congruence trans- This section discusses the use of PMA modeling for theoretical
formation [28] to (22), with X = P -1 2 0, it is easy to show studies of the nonlinear system (1).
that if the following LMI constraint holds:
A. Description of PMA Systems
r r
/ / h i (z) h j (z) He [A i X + B i Y j] 1 0 (23) To introduce the PMA modeling, let us first consider the non-
i =1 j =1 linear system (1) with u / 0 for simplicity, which results in the
following autonomous system:
Then, the PDC control law (20) asymptotically stabilizes the
xo = f (x)(24)
T-S fuzzy system (3) with K j = Y j X -1, for j ! X r .
Remark 8. A direct consequence of sharing the same MFs Remark 11. It is important to stress that T-S fuzzy model-
for the PDC control and the T-S fuzzy system is that double based approaches require rewriting system (24) in the form
convex sums appear, see for example (23). Similar to the stabili- xo = A (z) x, where the premise vector z regroups all nonlinear
ty analysis discussed above, several approaches are available to terms involved in the matrix A (·). Up to now, no systematic
reduce the design conservatism at the expense of a higher procedure is available to select, among the infinite possibilities,
computational cost. The key difference between stability analy- the “best” parameterization f (x) = A (z) x guaranteeing to
sis and control design is that the latter usually requires addition- achieve stability and other performance specifications. There-
al matrix transformations and/or matrix inversions of decision fore, the issue on the non-uniqueness of T-S fuzzy representa-
variables to derive LMI-based formulations. Apart from some tion still remains open [15]. As shown below, we do not have
special cases, this makes the use of some classes of Lyapunov this source of conservatism with PMA model-based approach-
functions with special structures on their parameter matrices, es since the form (24) is directly dealt with. However, PMA
for instance LILFs and PLFs, very challenging for control modeling leads to appropriation errors. At the present stage of
design purposes. progress, taking into account such errors for stability analysis
Remark 9. Compared to LILFs and PLFs, FLFs as given in of the nonlinear system (24) still remains open in PMA mod-
(8) can be applied more easily to the synthesis problem [35], el-based framework.
[42], [63]. However, in this case, for T-S fuzzy systems (5) with Without loss of generality, we assume that x i #
z / x, the time-derivatives of the MFs involve the system x i # xr i, i ! X n, where x i and xr i denote respectively the upper
dynamics, i.e., ho i (x) = ^2h i /2x h< xo , which can include the con- and lower bounds of the ith entry of x. As a consequence, the
trol action u (t) to be designed. Hence, it is not obvious to state x belongs to the set R = 6x 1, xr 1@ # f # 6x-n, xr n@ . Let us
check a posteriori the validity region of the designed controller partition the state-space of system (24) as follows:
[35]. This is the biggest gap between stability analysis and con-
trol design when using fuzzy Lyapunov functions. x- j = | [j1] 1 | [j2] 1 f 1 | [jN j +1] = xr j, j ! X n (25)

62 IEEE Computational intelligence magazine | FEbruary 2019

 Let K v = X N1 +1 # f # X Nn +1 be the set of multi-indexes by PMA model (29) with arbitrary accuracy on R by increas-
corresponding to all the vertices induced by the partition (25) ing the number of piecewise regions with (25).
and K r = X N1 # f # X Nn the set of multi-indexes corre- Remark 13. PMA systems are characterized by the trian-
sponding to the regions. For i = (i 1, f, i n) ! K r , the region gular MFs defined in (26). Other types of MFs for PMA sys-
6| 1[i1], | 1[i1 +1]@ # f # 6| n[in], | n[in +1]@ is denoted by Ri and tems, for instance Gaussian MFs or trapezoidal MFs, could be
K i = " i 1, i 1 + 1 , # f # " i n, i n + 1 , is the set of multi-indexes considered. However, the parametric expression (28) of x (t) is
corresponding to all vertices of Ri. For k ! K i, | k is the ver- crucial for the stability analysis of PMA systems, which is
tex of Ri whose jth component is defined as | [jk j], for j ! X n . directly derived from triangular MFs. Moreover, note also that
For a hyper-rectangle region Ri , for i ! K r , we consider the triangular MFs are largely employed in fuzzy control systems
following set of fuzzy rules: and in practice [85].
Consider a particular case with n = 2. Then, the resulting
IF x 1 (t) is h 1[k1] (x 1) and f and x n (t) is [k n]
hn (x n), second-order PMA model (29) with the MFs (26) corresponds
THEN xo (t) is Fk, k ! K i, to the parametric expressions of the singleton-type fuzzy sys-
tems studied in [5], [16]. In this case, by eliminating one of the
where Fk = f (| k) is the singleton vector, and the triangular triangular MFs from the parametric expression (29), the fol-
membership function h [jk j] (x j) with respect to x j (t), for j ! X n lowing state-space representation of PMA systems can be read-
and k j ! X N j +1 , is defined as ily obtained:
Z xo = A i (x) x + n i, x ! Ri, i ! K r
* ni =
[k j - 1 ]
] x j -| j 6 [k j -1], | [jk j]@ and j $ 2 / h k ( 0) F k
] | [jk j] - | [jk j -1] , if x j ! | j (30)
] k ! Ki
[k j]
h j (x j ) = [ | j
[k j + 1 ]
(26)
[k j] , if x j ! 6| j , | j @ and j # N j
-x j [k j] [k j +1]
] [k j + 1 ] where n i = 0 , for i ! K Z , and h k (0) denotes the value of
]] | j -| j
0, otherwise. h k (x) for x = 0. The state-dependent matrix is of the form
\
A i ^ x h = h61k1@ ^x 1h S ^k 1 $ h + h61k1 ^x 1 h S ^k1 + 1, $ h (31)
+1@
Note that the MFs (26) satisfy the following properties [5]:
i j +1
hj
[k j ]
( x j ) $ 0, / [k j ]
hj ( x j ) = 1, j ! X n . for k ! K i and i ! K r . The explicit expressions of S (k 1, $) and
k j =i j
S (k 1 + 1, $), their derivations, and alternative equivalent state-
Given x(t) with these membership functions, xo (t) can be space representations of (30)-(31) can be found in [5].
inferred by taking the weighted average of Fk as follows: Remark 14. For second-order PMA systems, both represen-
n tations (29) and (30) are strictly equivalent. Note also that system
xo = / h k (x) Fk, h k (x) = % h [jk j] (x j)(27) (30) is found to be a biaffine system. Moreover, with expression
k ! Ki j =1
(31), the PMA system (30) can be viewed as a piecewise polytopic
The following expression of x(t) can be derived from that of affine system with two vertex systems: xo = S (k 1, $) x + n i and
the triangular MFs [5]: xo = S (k 1 + 1, $) x + n i .

x= / h k (x) | k (28) B. Literature Review on PMA Systems

k ! Ki
Many advantages of using PMA systems for control purposes
Based on (27) and (28), the parametric expression of PMA sys- were highlighted in [5]. First, a PMA model (29) is easily
tem on R = , i ! Kr Ri can be expressed as follows [7]: obtained from the mathematical expression (24) of a nonlinear
system as shown above. Second, PMA models have a general
xo = / h k (x) Fk, x= / h k (x) | k (29) approximation capacity of any smooth nonlinear systems. Third,
k ! Kv k ! Kv
PMA models are simply implemented by means of look-up
For stability analysis, we assume that the equilibrium x / 0 of tables (LUT) which are one of the most widely used practical
system (29) corresponds to the vertex | k0 of the state-space tools in the industry for model approximation and control
partition, for a given k 0 ! K v . Let K Z be the set of multi- implementation, especially in automotive and aerospace engi-
indexes for regions containing the origin which is called zero- neering. Finally, PMA systems are fully parametric, i.e., both the
regions, and K NZ = K r \K Z (called non-zero regions). We denote state vector and its rate can be expressed by parametric expres-
also K )i = K i \ {k 0}, for i ! K Z . sions in (29). As shown later, this enables a systematic frame-
Remark 12. Observe in (29) that the weights with respect work to study PMA systems.
to x j, j ! X n, in the premises are computed by the multiplica- Despite these practical and theoretical advantages, up to
tion of h [jk j] (x j), k j ! X N j +1. Therefore, the fuzzy reasoning 1999 there was no rigorous stability framework on PMA sys-
used here is characterized by normalized membership func- tems. Almost all papers had been focused on the practical appli-
tions, multiplicative weights calculation, and weighted average cations of PMA systems rather than their theoretical studies [5].
aggregation [5]. Note also that system (24) can be approximated By fully taking into account the information of the triangular

FEbruary 2019 | IEEE Computational intelligence magazine 63

MFs as functions of state variables, Sugeno first set the theoreti- simple and effective stability framework. The key idea relies on
cal foundations on a quadratic Lyapunov stability framework for the fact that within a piecewise region, each of these expressions
both continuous-time and discrete-time PMA systems [5]. is uniquely determined by its vertex values and its restriction to
These results were then improved in [16] to achieve necessary and the region is a convex combination of these values. Moreover, the
sufficient stability conditions with respect to a common quadratic piecewise feature of PMA systems is fully exploited via a piece-
Lyapunov function (6). The basic idea of the stability analysis in wise Lyapunov function. This led to promising results on stabili-
[5], [16] consists in using both parametric expressions (29) and ty analysis of PMA systems of any order, and has opened new
the state-space representation (30) to analytically set V (x) > 0, research perspectives as discussed in the subsequent sections.
Vo (x) 1 0, for x ! 0, as the stability conditions for each piece-
wise region. Following the same line, Taniguchi and Sugeno C. LMI-Based Stability Analysis for General PMA Systems
proposed in [86] necessary and sufficient stability conditions with Consider a piecewise quadratic Lyapunov function candidate
respect to a piecewise Lyapunov function of the form parameterized as follows [30]:

V (x) = 1 x < Pi x + q i< x + ri, i ! K r (32)

V ( x) = ) t < t t
x < Pi x, for x ! Ri, i ! K Z
2 (33)
x P j x, for x ! R j, j ! K NZ
where the explicit expressions of Pi, q i and ri, for i ! K r , can
be found in [86]. In addition, the conditions guaranteeing that where xt = 6x < 1@< and
function (32) is a proper Lyapunov function candidate are also
)t
discussed therein. Note that the Lyapunov function (32) can be Pi = M <i TM i, for i ! K Z
t j, for j ! K NZ (34)
t <j TM
considered as a piecewise approximation of an arbitrary Lyapu- Pj = M
nov function by second-order functions. These preliminary
results of Sugeno’s group clearly show the great potential of The constraint matrices are constructed as follows [30]:
PMA model-based approaches to get rid of the conservative-
Lt i = 6L i l i@, M
t i = 6M i m i@ (35)
ness issue of T-S fuzzy theory, for which MFs are usually
regarded as unknown parameters [15]. such that
In parallel, without considering the knowledge on the MFs, Lt i xt $ 0, x ! Ri, i ! K r ,
numerical approaches were developed for stability analysis and t i xt = Mt j xt , x ! Ri + R j, i, j ! K r ,
M
control design of PMA control systems in [8], [87]. Also based
on Lyapunov stability arguments, there is however a fundamental where l i = 0, m i = 0, i ! K Z .
difference compared to the results in [5], [16], [86]. That is, Remark 15. Since the Lyapunov function (33) combines
these works essentially focused on the characteristics of PMA the power of quadratic functions near an equilibrium point
controllers performed on T-S fuzzy objective systems. As such, with the flexibility of piecewise functions in the large, it can
the stability analysis of PMA control systems is embedded into lead to less conservative results compared to those based on a
the conventional T-S fuzzy theory, which is not compatible common quadratic Lyapunov function [7], [30]. Indeed, the
with the original motivation of PMA control systems. Indeed, latter can be regarded as a special case of (33) by considering
the final goal in PMA control theory is to provide a stability M i = I and L i = 0.
analysis framework of nonlinear systems by embedding them in Note that all free parameters of the Lyapunov function V (x)
PMA systems to avoid the inherent drawbacks of T-S fuzzy defined in (33) are collected in the symmetric matrix T and
control systems [5]. the expression of Pi is linear in T. This feature allows for an
It should be stressed that all the above-mentioned results on LMI-based framework to study the stability of PMA systems as
PMA control systems suffer from two major drawbacks [7]. First, shown in the following theorem.
the proposed stability conditions are expressed in terms of non- Theorem 5. [7, Theorem 1] Given a PMA system in (29).
linear matrix inequalities, which obviously induce numerical If there exist a symmetric matrix T ! R N # N , symmetric matri-
difficulties. Second and more importantly, these conditions are ces with nonnegative entries U q ! R 2n # 2n and W q ! R 2n # 2n (for
only applicable to second-order PMA systems. Note that state- q ! K r ), matr ices Y1i ! R n # n, Y2i ! R 1 # n (for i ! K Z ),
space representation (30) of PMA systems was exclusively Yt1j ! R (n +1) # (n +1) and Yt2j ! R 1 # (n +1) (for j ! K NZ ), satisfying
employed to study PMA systems in [5], [8], [16], [86], [87]. the LMI constraints (36), (37), (38) and (39).Then, the PMA sys-
However, such a representation is very challenging for theoreti- tem (29) is asymptotically stable. Moreover, the piecewise qua-
cal study due to its piecewise polytopic affine feature [5], see also dratic function V (x) defined in (33) is a Lyapunov function of
Remark 14. As a consequence, the extensions of these results to this system.
high-dimensional PMA systems are not obvious. Due to these
restrictive drawbacks, PMA control systems have not yet Pi - L <i U i L i 2 0, i ! K Z (36)

He = G 1 0, i ! K Z , k ! K i)
received much attention from the (fuzzy) control community. Y1i + L <i W i L i /2 -Y1i | k + Pi Fk
Only very recently, it is demonstrated in [7] that the specific rep- Y2i -Y2i | k
resentation of PMA via parametric expressions (29) enables a (37)

64 IEEE Computational intelligence magazine | FEbruary 2019

Pt j - Lt <j U j Lt j 2 0, j ! K NZ (38) implemented with look-up-tables which is an outstanding fea-
ture for real-time control applications [5].
Yt1j + Lt <j W j Lt j /2 -Yt1j |t k + Pt j Ft k
He = G 1 0, j ! K NZ , k ! K j
Using similar arguments as for Theorem 5, the following
Yt2j -Yt2j |t k theorem provides sufficient conditions to design LUT control-
(39) ler (42) for nonlinear systems.
Theorem 6. Given a PMA system in (40) and a stabilizing
with Pi and Pt j defined in (34) and feedback gain H of the linearized system (A, B) defined in
(41). Assume there exist a symmetric matrix T ! R N # N , sym-
Ft k metric matrices with nonnegative entries U q ! R 2n # 2n and
Ft k = = G, |t k = ; E, k ! K j, j ! K NZ .
|k
0 1 W q ! R 2n # 2n (for q ! K r ), matrices Y1i ! R n # n, Y2i ! R m # n,
Y3i ! R 1 # n (for i ! K Z ), Yt1j ! R (n +1) # (n +1), Yt2j ! R m # (n +1),
Remark 16. LMI conditions (36) and (38) are imposed for Yt3j ! R 1 # (n +1) (for j ! K NZ ), and the control input vertices
m#1
each region to guarantee that (33) is a proper Lyapunov func- yk ! R (for k ! K v = {0}), satisfying the following matrix
tion candidate. The set of LMIs (37) and (39) ensures that the inequalities (43), (44), (45) and (46). Then, the control law (42)
value of the Lyapunov function decreases along any trajectory asymptotically stabilizes the PMA system (40). Furthermore,
of the PMA system (29). the piecewise quadratic function (33) is a Lyapunov function of
Remark 17. For the stability results in Theorem 5, the para- the closed-loop system.
metric expressions of both x and xo in (28) can be fully exploit-
Pi - L <i U i L i 2 0, i ! K Z (43)
ed via Finsler lemma [28], see [7] for details. Since expression
Ci Pi Gq -Y1i | k + Pi Fk + Hy k
He >Y2i + H -I/2 H 1 0,
(28) is equivalent to that of the triangular MFs (26), the informa-
tion on the MFs can be thus easily taken into account in the y k - Y2i | k
Y3i 0 -y k< y l /2 - Y3i | k
theoretical studies of PMA systems to reduce the conservatism.
Note that this is not the case of T-S fuzzy systems whose MFs i ! K Z , q ! K i, k ! K i), l ! K i) (44)
are often considered as uncertainty [15].
Pt j - Lt <j U j Lt j 2 0, j ! K NZ (45)
R t t k -Yt1j |t k + Pt j Ft k + Ht y kVW
D. Lyapunov-Based LUT Control Design S Cj Pt j G
Following the same modeling procedure as in Section IV-A, He SYt2j + Ht -I/2 y k - Yt 2j |t k W 1 0,
S W
the following PMA model of system (1) is straightforwardly S Y3j t 0 -y k y l /2 - Y3j |t k W
< t
obtained on R : T X
j ! K NZ , k ! K j, l ! K j  (46)

xo = / h k (x) (Fk + Gk u), x= / h k (x) | k (40) where Ht = 6H 0@ and

k ! Kv k ! Kv

Gt k = ;GkE, k ! K j, j ! K NZ ,
where Gk = g ( | k). Assume that the linearized system (A, B) of 0
(1) around x = 0 is stabilizable. This means that there exists a C i = Y1i + ^L <i W i L i + H < H h /2, i ! K Z ,
linear feedback gain H such that (A + BH ) is Hurwitz with Ct j = Yt1j + ^Lt <j W j Lt j + Ht < Ht h /2, j ! K NZ .

A =; E , B = ; E (41)
2f 2g
Remark 19. The slack matrices U q and W q, for q ! K r , are
2x x = 0 2x x = 0
introduced into the conditions of Theorems 5 and 6 through the
This control gain H can be designed in advance by any lin- S-procedure [28]. This contributes to reduce the conservatism
ear control technique to guarantee some local closed-loop since the piecewise-region feature of PMA systems can be fully
properties of (1). To stabilize the PMA system (40), let us con- exploited via the constraint matrices defined in (35).
sider the state-feedback control law of the form Remark 20. The design conditions in Theorem 6 are
expressed in terms of BMIs due to the product y <k y l involved
u (x) = / h k (x) y k + Hx (42) in (44) and (46). Note that for each vertex, there are 2 n associ-
k ! Kv
ated BMI-based conditions to be verified. Hence, the value of
n
where the control vertex at the origin must be assigned as y k, for k ! K v = {k 0}, should be imposed identical for 2 con-
y 0 = 0 and other input vertices y k, with k ! K v = {0}, have to ditions concerning the same vertex to avoid sliding modes and
be designed. chattering phenomena.
Remark 18. The control law (42) is composed of two
parts. The parametric expression part R k ! K h k (x) y k is defined in
v V. Future Perspectives of Fuzzy Control Systems
the same way as x and xo , see (40). The incorporation of the This paper provides a concise discussion on the evolution of
linear feedback part Hx in (42) is obviously crucial to guaran- fuzzy control systems. Through a selective list of references, we
tee the closed-loop stability in the case where system (1) is present the historical motivations and the current research
open-loop unstable around x = 0. Note that (42) can be easily progress of three types of fuzzy systems: Mamdani-type fuzzy

FEbruary 2019 | IEEE Computational intelligence magazine 65

systems, T-S fuzzy systems and PMA (or singleton-type fuzzy) induced delay, the information transmitted over the network to
systems. Great advances on both fundamental and application the controller is usually delayed, causing a mismatch between
aspects of fuzzy control have been made with a huge number the premise variables. This mismatch can be seen as extra con-
of available publications on the topic. However, many interest- straints in the design of NCSs [95]–[99]. Moreover, it can be
ing and important issues still remain challenging, which pro- very restrictive to assume that the premise variables of the fuzzy
vide fantastic opportunities for the research on fuzzy control systems and the fuzzy controllers/observers are synchronous if
systems in the future. Since we believe that within fuzzy con- bilateral networks (controller-to-actuator and sensor-to-con-
trol context, only fuzzy model-based approaches enable sys- troller) have to be designed [99].Then, the asynchronous trans-
tematic frameworks for stability analysis and control design of mission issue should be considered.
nonlinear systems, below our personal perspectives are focused ❏❏ The second example is related to the fact that in any con-
on T-S fuzzy systems and PMA systems. trol system, fault detection, diagnosis and recovery is deci-
sive to have a robust and resilient system operation. In this
A. Open Issues on Takagi-Sugeno Fuzzy Systems context, control reconfiguration plays a crucial role to
A large research effort has been focused on reducing the con- obtain a fault tolerant control (FTC) that can effectively
servativeness of stability analysis and control design. As dis- handle severe actuator/sensor faults while still guarantee-
cussed above, this has been done by searching for more general ing a desired closed-loop performance [100]. To achieve
Lyapunov function candidates and clever manipulations to this goal, it is crucial to have effective fault detection and
bring the problem to an LMI framework. isolation (FDI) schemes to support the control reconfigu-
Although there are conditions considering the information ration in a proper way [101]. One idea behind the control
on the shape of the MFs, see for instance [29], [47], we believe reconfiguration is to redesign the problem for the faulty
that more powerful methodologies could arise with the use of system by choosing possibly a new structure and adapting
interval type-2 T-S fuzzy modeling techniques [88]–[91]. These the control law to this new scenario. A possible solution
fuzzy models allow to deal with uncertain grades of member- for this problem relies on the use of two-step design pro-
ship. Hence, type-2 T-S fuzzy systems are very useful in the cedures for which FDI and FTC schemes are separately
cases where exact MFs are difficult to be chosen and/or there dealt with. To this end, virtual actuators/sensors can be
is a need to cope with large amounts of uncertainties [89], [90]. used to mask the actuator/sensor faults [102]. More spe-
However, for stability analysis and control design, the lower and cifically, an FDI scheme via T-S fuzzy virtual actuator/
upper MFs have to be simultaneously considered leading to sensor models have to be designed to achieve the recon-
more elaborated manipulations and computational burden [92], figuration goal. This requires the design of a T-S fuzzy
[93]. Possible contributions go toward building less conservative observer-based controller, in which the T-S fuzzy control-
LMI-based conditions for interval type-2 T-S fuzzy models ler and the T-S fuzzy observer possibly do not share the
depending on MFs, reducing the gap between type-1 and same MFs or it is necessary to estimate their unmeasur-
type-2 T-S fuzzy modeling methodologies. For instance, mem- able premise variables.
bership-dependent stability conditions for both type-1 and
type-2 T-S fuzzy systems were recently proposed in [50]. To this B. Open Issues on Piecewise Multiaffine Systems
end, it is considered that the MFs belong to a unified space. Lyapunov-based studies of PMA systems are much more limit-
Then, an extrema-based method is proposed to construct a ed compared to those of T-S fuzzy systems. As mentioned pre-
polyhedron convex hull to enclose the membership distribu- viously, this is due to theoretical challenges when dealing with
tion in this space. Despite the interests of the proposed results, this type of fuzzy systems, especially for high-order systems.
much research effort should be still devoted to the stabilization Although the outstanding contributions in [5], [7], [16] throw
problem, and even to the stability analysis to reduce further the new light on the research of PMA control systems, several
gap between both types of fuzzy systems. important issues need to be solved in the future.
In our opinion, the design of T-S fuzzy controllers/observ- ❏❏ How to fully take into account the knowledge of triangular
ers that do not share the same MFs or premise variables with MFs to derive tractable necessary and sufficient stability conditions
T-S fuzzy models is another important topic. This topic should for high-order PMA systems? Two of the authors have provid-
deserve more thorough consideration since it can greatly ed a discussion on this issue in [16], [86]. However, the stability
improve the flexibility of the control/observer design in many conditions are only applicable to second-order PMA systems
cases [69]. Also, such a design scheme is involved in current and expressed in terms of nonlinear matrix inequalities.
research mainstreams on T-S fuzzy model-based approaches as ❏❏ How to develop an effective framework to design LUT
shown in the two following illustrative examples. controllers from the PMA stability results? BMI-based
❏❏ The first example is concerned with the design of nonlinear design conditions in Theorem 6 induce numerical diffi-
networked control systems (NCSs) in which network-induced culties and may be conservative, especially for high-order
imperfections, for instance network-induced delay and sam- systems. PMA systems can be also used for LUT-based
pling issues, are intrinsic [70], [94]. One of the major difficulties control implementation of feedback error learning (FEL)
in this case is the fact that, when considering the network- scheme [103]. In FEL control, it is crucial to build an

66 IEEE Computational intelligence magazine | FEbruary 2019

 online pseudo-inverse model of the general nonlinear [13] K. Tanaka and M. Sugeno, “Stability analysis and design of fuzzy control systems,”
Fuzzy Sets Syst., vol. 45, no. 2, pp. 135–156, 1992.
plant for feedforward control purposes. Up to now, neural [14] K. Tanaka and H. Wang, Fuzzy Control Systems Design and Analysis: A Linear Matrix
networks have been intensively used for this aim. Howev- Inequality Approach. New York, NY, USA: Wiley, 2004.
[15] A. Sala, “On the conservativeness of fuzzy and fuzzy-polynomial control of nonlinear
er, due to computational and overfitting reasons, PMA sys- systems,” Annu. Rev. Control, vol. 33, no. 1, pp. 48–58, 2009.
tems are considered as a powerful alternative to neural [16] M. Sugeno and T. Taniguchi, “On improvement of stability conditions for con-
tinuous Mamdani-like fuzzy systems,” IEEE Trans. Syst. Man, Cybern. B, Cybern. (1995–
networks. Despite the practical effectiveness of LUT-based 2012), vol. 34, no. 1, pp. 120–131, 2004.
FEL control, more rigorous stability analysis is required in [17] W. Pedrycz, Fuzzy Control and Fuzzy Systems. Research Studies, 1993.
[18] R. R. Yager and D. P. Filev, Essentials of Fuzzy Modeling and Control. New York, NY,
the future. USA: Wiley, 1994.
❏❏ How to extend the theoretical results concerning the “clas- [19] K. Passino and S. Yurkovich, Fuzzy Control. Reading, MA, USA: Addison-Wesley,
1998.
sical” PMA system (40) to a wider class of systems such as [20] R. Babuška, Fuzzy Modeling for Control. Springer Science, 2012.
time-delay PMA systems, descriptor PMA systems, switch- [21] J. M. Mendel, “Fuzzy logic systems for engineering: A tutorial,” Proc. IEEE, vol. 83,
no. 3, pp. 345–377, 1995.
ing PMA systems and so on? The research on this issue [22] G. Feng, “A survey on analysis and design of model-based fuzzy control systems,”
depends obviously on the progress of the two previous ones. IEEE Trans. Fuzzy Syst., vol. 14, no. 5, pp. 676–697, 2006.
[23] R.-E. Precup and H. Hellendoorn, “A survey on industrial applications of fuzzy
❏❏ Whether does the stability of PMA system (40) imply that control,” Compu. Ind., vol. 62, no. 3, pp. 213–226, 2011.
of the nonlinear system (1)? To solve this issue, the approxi- [24] L. Zadeh, “The evolution of systems analysis and control: A personal perspective,”
IEEE Control Syst., vol. 16, no. 3, pp. 95–98, 1996.
mation errors should be characterized, and explicitly consid- [25] P. P. Bonissone, V. Badami, K. H. Chiang, P. S. Khedkar, K. W. Marcelle, and M. J.
ered in the stability analysis as highlighted in [7]. Schutten, “Industrial applications of fuzzy logic at general electric,” Proc. IEEE, vol. 83,
no. 3, pp. 450–465, 1995.
❏❏ For a given nonlinear system, how to obtain its “best” PMA [26] A. Van der Wal, “Application of fuzzy logic control in industry,” Fuzzy Sets Syst., vol.
model in terms of facilitating both accurate modeling and 74, no. 1, pp. 33–41, 1995.
[27] K. Tanaka and M. Sugeno, “Stability analysis of fuzzy systems using Lyapunov’s di-
effective stability/performance analysis, and reducing the rect method,” in Proc. North America Fuzzy Information Processing Society, Toronto, Canada,
numerical complexity (namely the number of piecewise 1990, vol. 1, pp. 133–136.
[28] S. Boyd, L. El Ghaoui, E. Feron, and V. Balakrishnan, Linear Matrix Inequalities in
regions)? The results in [104] may bring some interesting System and Control Theory, ser. Studies in Applied Mathematics, vol. 15. Philadelphia, PA,
ideas to this open issue. USA: SIAM, 1994.
[29] H.-K. Lam, “A review on stability analysis of continuous-time fuzzy-model-based
A part from the above open issues, many other challenges and control systems: From membership-function-independent to membership-function-de-
perspectives concerning fuzzy control systems have been raised pendent analysis,” Eng. Appl. Artif. Intell., vol. 67, pp. 390–408, 2018.
[30] M. Johansson, A. Rantzer, and K.-E. Årzén, “Piecewise quadratic stability of fuzzy
in numerous publications, for instance [10], [11], [22], [23] and systems,” IEEE Trans. Fuzzy Syst., vol. 7, no. 6, pp. 713–722, 1999.
references therein. To conclude the paper, we strongly believe [31] K. Tanaka, T. Hori, and H. O. Wang, “A multiple Lyapunov function approach
to stabilization of fuzzy control systems,” IEEE Trans. Fuzzy Syst., vol. 11, no. 4, pp.
that PMA model-based approaches could provide a promising 582–589, 2003.
future for fuzzy control, which would have great impacts on [32] G. Feng, “Stability analysis of discrete-time fuzzy dynamic systems based on piece-
wise Lyapunov functions,” IEEE Trans. Fuzzy Syst., vol. 12, no. 1, pp. 22–28, 2004.
the control community. From the theoretical viewpoint, they [33] V. Campos, F. Souza, A. Torres, and R. Palhares, “New stability conditions based
allow overcoming the conservativeness issue of T-S fuzzy mod- on piecewise fuzzy Lyapunov functions and tensor product transformations,” IEEE Trans.
Fuzzy Syst., vol. 21, no. 4, pp. 748–760, 2013.
el-based approaches [7], [16]. From the application viewpoint, a [34] L. Mozelli, R. Palhares, and G. Avellar, “A systematic approach to improve multiple
systematic LUT-based control approach is of special interest for Lyapunov function stability and stabilization conditions for fuzzy systems,” Inf. Sci., vol.
179, no. 8, pp. 1149–1162, 2009.
real-time applications [5]. Therefore, PMA based approaches [35] T.-M. Guerra, M. Bernal, K. Guelton, and S. Labiod, “Non-quadratic local stabiliza-
could provide a viable answer to the issue: “Only straightforward tion for continuous-time Takagi-Sugeno models,” Fuzzy Sets Syst., vol. 201, pp. 40–54,
2012.
in theory plus straightforward in practice are great solutions to [36] B.-J. Rhee and S. Won, “A new fuzzy Lyapunov function approach for a Takagi-Sug-
change a discipline”, recently raised in [11]. eno fuzzy control system design,” Fuzzy Sets Syst., vol. 157, no. 9, pp. 1211–1228, 2006.
[37] T. González, A. Sala, and M. Bernal, “A generalised integral polynomial Lyapunov
function for nonlinear systems,” Fuzzy Sets Syst., vol. 356, pp. 77–91, 2019.
References [38] E. Tognetti, R. Oliveira, and P. Peres, “Selective H 2 and H 3 stabilization of Tak-
[1] E. H. Mamdani, “Application of fuzzy algorithms for control of simple dynamic plant,” agi-Sugeno fuzzy systems,” IEEE Trans. Fuzzy Syst., vol. 19, no. 5, pp. 890–900, 2011.
Proc. IEEE, vol. 121, no. 12, pp. 1585–1588, 1974. [39] H. Zhang and X. Xie, “Relaxed stability conditions for continuous-time T-S fuzzy-
[2] L. Zadeh, “Fuzzy algorithms,” Inf. Control, vol. 12, no. 2, pp. 94–102, 1968. control systems via augmented multi-indexed matrix approach,” IEEE Trans. Fuzzy Syst.,
[3] L. A. Zadeh, “Outline of a new approach to the analysis of complex systems and deci- vol. 19, no. 3, pp. 478–492, 2011.
sion processes,” IEEE Trans. Syst. Man, Cybern., vol. SMC-3, no. 1, pp. 28–44, 1973. [40] Y.-J. Chen, M. Tanaka, K. Tanaka, and H. Wang, “Stability analysis and region-of-
[4] E. Mamdani and S. Assilian, “An experiment in linguistic synthesis with a fuzzy logic attraction estimation using piecewise polynomial Lyapunov functions: Polynomial fuzzy
controller,” Int. J. Man. Mach. Stud., vol. 7, no. 1, pp. 1–13, 1975. model approach,” IEEE Trans. Fuzzy Syst., vol. 23, no. 4, pp. 1314–1322, 2015.
[5] M. Sugeno, “On stability of fuzzy systems expressed by fuzzy rules with singleton [41] D.-H. Lee and D.-W. Kim, “Relaxed LMI conditions for local stability and local
consequents,” IEEE Trans. Fuzzy Syst., vol. 7, no. 2, pp. 201–224, 1999. stabilization of continuous-time Takagi-Sugeno fuzzy systems,” IEEE Trans. Cybern., vol.
[6] T. Takagi and M. Sugeno, “Fuzzy identification of systems and its applications to mod- 44, no. 3, pp. 394–405, 2014.
eling and control,” IEEE Trans. Syst. Man, Cybern. B, Cybern. (1995–2012), vol. SMC-15, [42] D.-H. Lee, Y.-H. Joo, and M.-H. Tak, “Local stability analysis of continuous-time
no. 1, pp. 116–132, 1985. Takagi-Sugeno fuzzy systems: A fuzzy Lyapunov function approach,” Inf. Sci., vol. 257,
[7] A.-T. Nguyen, M. Sugeno, V. Campos, and M. Dambrine, “LMI-based stability analysis pp. 163–175, 2014.
for piecewise multi-affine systems,” IEEE Trans. Fuzzy Syst., vol. 25, no. 3, pp. 707–714, 2017. [43] L. Mozelli, R. Palhares, F. Souza, and E. Mendes, “Reducing conservativeness in recent
[8] E. Kim, “A new computational approach to stability analysis and synthesis of linguistic stability conditions of T-S fuzzy systems,” Automatica, vol. 45, no. 6, pp. 1580–1583, 2009.
fuzzy control system,” IEEE Trans. Fuzzy Syst., vol. 12, no. 3, pp. 379–388, 2004. [44] V. Campos, A.-T. Nguyen, and R. Palhares, “A comparison of different upper-bound
[9] B. Kosko, “Fuzzy systems as universal approximators,” IEEE Trans. Comput., vol. 43, inequalities for the membership functions derivative,” IFAC-PapersOnLine, vol. 50, no.
no. 11, pp. 1329–1333, 1994. 1, pp. 3001–3006, 2017.
[10] A. Sala, T.-M. Guerra, and R. Babuška, “Perspectives of fuzzy systems and control,” [45] A.-T. Nguyen, R. Márquez, and A. Dequidt, “An augmented system approach for
Fuzzy Sets Syst., vol. 156, no. 3, pp. 432–444, 2005. LMI-based control design of constrained Takagi-Sugeno fuzzy systems,” Eng. Appl. Artif.
[11] T.-M. Guerra, A. Sala, and K. Tanaka, “Fuzzy control turns 50: 10 years later,” Fuzzy Intell., vol. 61, pp. 96–102, 2017.
Sets Syst., vol. 281, pp. 168–182, 2015. [46] T.-M. Guerra and L. Vermeiren, “LMI-based relaxed non-quadratic stabilization
[12] M. Sugeno, “An introductory survey of fuzzy control,” Inf. Sci., vol. 36, no. 1–2, conditions for nonlinear systems in the Takagi-Sugeno’s form,” Automatica, vol. 40, no.
pp. 59–83, 1985. 5, pp. 823–829, 2004.

FEbruary 2019 | IEEE Computational intelligence magazine 67

[47] A. Sala and C. Ariño, “Relaxed stability and performance conditions for Takagi- [75] J. Qiu, S. X. Ding, H. Gao, and S. Yin, “Fuzzy-model-based reliable static output
Sugeno fuzzy systems with knowledge on membership function overlap,” IEEE Trans. feedback H 3 control of nonlinear hyperbolic PDE systems,” IEEE Trans. Fuzzy Syst.,
Syst. Man, Cybern. B, vol. 37, no. 3, pp. 727–732, 2007. vol. 24, no. 2, pp. 388–400, 2016.
[48] A. Sala and C. Ariño, “Relaxed stability and performance LMI conditions for Tak- [76] A.-T. Nguyen, K. Tanaka, A. Dequidt, and M. Dambrine, “Static output feedback
agi-Sugeno fuzzy systems with polynomial constraints on membership function shapes,” design for a class of constrained Takagi-Sugeno fuzzy systems,” J. Franklin Inst., vol. 354,
IEEE Trans. Fuzzy Syst., vol. 16, no. 5, pp. 1328–1336, 2008. no. 7, pp. 2856–2870, 2017.
[49] M. Narimani and H.-K. Lam, “Relaxed LMI-based stability conditions for Takagi- [77] Y. Wei, J. Qiu, and H. R. Karimi, “Reliable output feedback control of discrete-time
Sugeno fuzzy control systems using regional-membership-function-shape-dependent fuzzy affine systems with actuator faults,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 64,
analysis approach,” IEEE Trans. Fuzzy Syst., vol. 17, no. 5, pp. 1221–1228, 2009. no. 1, pp. 170–181, 2017.
[50] X. Yang, H.-K. Lam, and L. Wu, “Membership-dependent stability conditions for [78] P. Bergsten, R. Palm, and D. Driankov, “Observers for takagi-sugeno fuzzy systems,”
type-1 and interval type-2 T-S fuzzy systems,” Fuzzy Sets Syst., 2018. DOI: 10.1016/j. IEEE Trans. Syst.Man, Cybern. B, Cybern. (1995–2012), vol. 32, no. 1, pp. 114–121, 2002.
fss.2018.01.018. [79] M. C. Teixeira, E. Assuncao, and R. G. Avellar, “On relaxed LMI-based designs
[51] H.-K. Lam and M. Narimani, “Quadratic-stability analysis of fuzzy-model-based for fuzzy regulators and fuzzy observers,” IEEE Trans. Fuzzy Syst., vol. 11, no. 5, pp.
control systems using staircase membership functions,” IEEE Trans. Fuzzy Syst., vol. 18, 613–623, 2003.
no. 1, pp. 125–137, Feb. 2010. [80] Z. Gao, X. Shi, and S. X. Ding, “Fuzzy state/disturbance observer design for T-S
[52] H.-K. Lam, “Polynomial fuzzy-model-based control systems: Stability analysis via fuzzy systems with application to sensor fault estimation,” IEEE Trans. Syst. Man, Cybern.
piecewise-linear membership functions,” IEEE Trans. Fuzzy Syst., vol. 19, no. 3, pp. B, Cybern. (1995–2012), vol. 38, no. 3, pp. 875–880, 2008.
588–593, 2011. [81] G. B. Koo, J. B. Park, and Y. H. Joo, “Decentralized sampled-data fuzzy observer
[53] M. Bernal, T.-M. Guerra, and A. Kruszewski, “A membership-function-dependent design for nonlinear interconnected systems,” IEEE Trans. Fuzzy Syst., vol. 24, no. 3, pp.
approach for stability analysis and controller synthesis of Takagi-Sugeno models,” Fuzzy 661–674, 2016.
Sets Syst., vol. 160, no. 19, pp. 2776–2795, 2009. [82] X. Xie, D. Yue, and C. Peng, “Multi-instant observer design of discrete-time fuzzy
[54] V. Campos, L. Torres, and R. Palhares, “Using information on membership function systems: a ranking-based switching approach,” IEEE Trans. Fuzzy Syst., vol. 25, no. 5,
shapes in asymptotically exact triangulation approaches,” in Proc. 51st IEEE Conf. Decision pp. 1281–1292, 2017.
and Control, 2012, pp. 6205–6210. [83] L. Li, M. Chadli, S. X. Ding, J. Qiu, and Y. Yang, “Diagnostic observer design for
[55] H. Wang, K. Tanaka, and M. Griffin, “An approach to fuzzy control of nonlinear sys- T-S fuzzy systems: Application to real-time-weighted fault-detection approach,” IEEE
tems: Stability and design issues,” IEEE Trans. Fuzzy Syst., vol. 4, no. 1, pp. 14–23, 1996. Trans. Fuzzy Syst., vol. 26, no. 2, pp. 805–816, 2018.
[56] H. Tuan, P. Apkarian, T. Narikiyo, and Y. Yamamoto, “Parameterized linear matrix [84] T.-M. Guerra, R. Márquez, A. Kruszewski, and M. Bernal, “ H 3 LMI-based ob-
inequality techniques in fuzzy control system design,” IEEE Trans. Fuzzy Syst., vol. 9, server design for nonlinear systems via Takagi-Sugeno models with unmeasured premise
no. 2, pp. 324–332, 2001. variables,” IEEE Trans. Fuzzy Syst., vol. 26, no. 3, pp. 1498–1509, 2018.
[57] X. Liu and Q. Zhang, “New approaches to H 3 controller designs based on fuzzy [85] W. Pedrycz, “Why triangular membership functions?” Fuzzy Sets Syst., vol. 64, no.
observers for T-S fuzzy systems via LMI,” Automatica, vol. 39, no. 9, pp. 1571–1582, 2003. 1, pp. 21–30, 1994.
[58] A. Sala Ariño, “Asymptotically necessary and sufficient conditions for stability and [86] T. Taniguchi and M. Sugeno, “Stabilization of nonlinear systems based on piecewise
performance in fuzzy control: Applications of Polya’s theorem,” Fuzzy Sets Syst., vol. 158, Lyapunov functions,” in Proc. IEEE Int. Conf. Fuzzy System, 2004, vol. 3, pp. 1607–1612.
no. 24, pp. 2671–2686, 2007. [87] E. Kim, “A new approach to numerical stability analysis of fuzzy control systems,”
[59] A. Kruszewski, A. Sala, T.-M. Guerra, and C. Ariño, “A triangulation approach to IEEE Trans. Syst. Man, Cybern. C, Appl. Rev., vol. 31, no. 1, pp. 107–113, 2001.
asymptotically exact conditions for fuzzy summations,” IEEE Trans. Fuzzy Syst., vol. 15, [88] J. M. Mendel, “Type-2 fuzzy sets and systems: An overview,” IEEE Comput. Intell.
no. 5, pp. 985–994, 2009. Mag., vol. 2, no. 1, pp. 20–29, 2007.
[60] V. Montagner, R. Oliveira, and P. Peres, “Convergent LMI relaxations for quadratic [89] H. Hagras, “Type-2 FLCs: A new generation of fuzzy controllers,” IEEE Comput.
stabilizability and control of Takagi-Sugeno fuzzy systems,” IEEE Trans. Fuzzy Syst., vol. Intell. Mag., vol. 2, no. 1, pp. 30–43, 2007.
17, no. 4, pp. 863–873, 2009. [90] H.-K. Lam and L. D. Seneviratne, “Stability analysis of interval type-2 fuzzy-model-
[61] R. Márquez, T.-M. Guerra, M. Bernal, and A. Kruszewski, “Asymptotically neces- based control systems,” IEEE Trans. Syst. Man, Cybern. B, Cybern. (1995–2012), vol. 38,
sary and sufficient conditions for Takagi-Sugeno models using generalized non-quadratic no. 3, pp. 617–628, 2008.
parameter-dependent controller design,” Fuzzy Sets Syst., vol. 306, pp. 48–62, 2017. [91] M. Biglarbegian, W. W. Melek, and J. M. Mendel, “On the stability of interval type-2
[62] K. Tanaka and M. Sano, “A robust stabilization problem of fuzzy control systems and TSK fuzzy logic control systems,” IEEE Trans. Syst. Man, Cybern. B, Cybern. (1995–2012),
its application to backing up control of a truck-trailer,” IEEE Trans. Fuzzy Syst., vol. 24, vol. 40, no. 3, pp. 798–818, 2010.
no. 2, pp. 119–134, 1994. [92] H.-K. Lam, H. Li, C. Deters, E. L. Secco, H. A. Wurdemann, and K. Althoefer,
[63] M. Bernal and T.-M. Guerra, “Generalized nonquadratic stability of continuous- “Control design for interval type-2 fuzzy systems under imperfect premise matching,”
time Takagi-Sugeno models,” IEEE Trans. Fuzzy Syst., vol. 18, no. 4, pp. 815–822, IEEE Trans. Ind. Electron., vol. 61, no. 2, pp. 956–968, 2014.
2010. [93] B. Xiao, H.-K. Lam, and H. Li, “Stabilization of interval type-2 polynomial-
[64] D. H. Lee, J. B. Park, and Y. H. Joo, “Approaches to extended non-quadratic stability fuzzy-model-based control systems,” IEEE Trans. Fuzzy Syst., vol. 25, no. 1, pp.
and stabilization conditions for discrete-time Takagi-Sugeno fuzzy systems,” Automatica, 205–217, 2017.
vol. 47, no. 3, pp. 534–538, 2011. [94] L. Zhang, H. Gao, and O. Kaynak, “Network-induced constraints in networked
[65] X. Xie, H. Ma, Y. Zhao, D. W. Ding, and Y. Wang, “Control synthesis of discrete- control systems: A survey,” IEEE Trans. Ind. Electron., vol. 9, no. 1, pp. 403–416, 2013.
time T-S fuzzy systems based on a novel non-PDC control scheme,” IEEE Trans. Fuzzy [95] T. Oliveira, R. Palhares, V. Campos, P. Queiroz, and E. Goncalves, “Improved Tak-
Syst., vol. 21, no. 1, pp. 147–157, 2013. agi-Sugeno fuzzy output tracking control for nonlinear networked control systems,” J.
[66] K. Tanaka, T. Ikeda, and H. Wang, “Fuzzy regulators and fuzzy observers: Relaxed Franklin Inst., vol. 354, no. 16, pp. 7280–7305, 2017.
stability conditions and LMI-based designs,” IEEE Trans. Fuzzy Syst., vol. 6, no. 2, pp. [96] Z. Gu, D. Yue, and E. Tian, “On designing of an adaptive event-triggered commu-
250–265, 1998. nication scheme for nonlinear networked interconnected control systems,” Inf. Sci., vol.
[67] B.-S. Chen, C.-S. Tseng, and H.-J. Uang, “Mixed H 2 /H 3 fuzzy output feedback 422, pp. 257–270, 2018.
control design for nonlinear dynamic systems: An LMI approach,” IEEE Trans. Fuzzy [97] C. Peng, D. Yue, and M. Fei, “Relaxed stability and stabilization conditions of net-
Syst., vol. 8, no. 3, pp. 249–265, 2000. worked fuzzy control systems subject to asynchronous grades of membership,” IEEE
[68] J. Yoneyama, M. Nishikawa, H. Katayama, and A. Ichikawa, “Design of output feed- Trans. Fuzzy Syst., vol. 22, no. 3, pp. 1101–1112, 2014.
back controllers for Takagi-Sugeno fuzzy systems,” Fuzzy Sets Syst., vol. 121, no. 1, pp. [98] C. Peng, M. Yang, J. Zhang, M. Fei, and S. Hu, “Network-based H 3 control for
127–148, 2001. T-S fuzzy systems with an adaptive event-triggered communication scheme,” Fuzzy Sets
[69] S.-K. Nguang and P. Shi, H 3 fuzzy output feedback control design for nonlin- Syst., vol. 329, pp. 61–76, 2017.
ear systems: an LMI approach,” IEEE Trans. Fuzzy Syst., vol. 11, no. 3, pp. 331–340, [99] D. Zhang, Z. Zhou, and X. Jia, “Networked fuzzy output feedback control for dis-
2003. crete-time Takagi-Sugeno fuzzy systems with sensor saturation and measurement noise,”
[70] H. Li, C. Wu, S. Yin, and H.-K. Lam, “Observer-based fuzzy control for nonlinear Inf. Sci., vol. 457–458, pp. 182–194, 2017.
networked systems under unmeasurable premise variables,” IEEE Trans. Fuzzy Syst., vol. [100] J. Lan and R. Patton, “Integrated design of fault-tolerant control for nonlinear
24, no. 5, pp. 1233–1245, 2016. systems based on fault estimation and T-S fuzzy modeling,” IEEE Trans. Fuzzy Syst., vol.
[71] L. K. Wang, H. G. Zhang, and X. D. Liu, “ H 3 observer design for continuous-time 25, no. 5, pp. 1141–1154, 2017.
Takagi-Sugeno fuzzy model with unknown premise variables via nonquadratic lyapunov [101] Y. Zhang and J. Jiang, “Bibliographical review on reconfigurable fault-tolerant con-
function,” IEEE Trans. Cybern., vol. 46, no. 9, pp. 1986–1996, 2016. trol systems,” Annu. Rev. Control, vol. 32, no. 2, pp. 229–252, 2008.
[72] J. Dong and G.-H. Yang, “Observer-based output feedback control for discrete-time [102] P. Odgaard and J. Stoustrup, “A benchmark evaluation of fault tolerant wind turbine
T-S fuzzy systems with partly immeasurable premise variables,” IEEE Trans. Syst. Man, control concepts,” IEEE Trans. Control Syst. Technol., vol. 23, no. 3, pp. 1221–1228, 2015.
Cybern. Syst., vol. 47, no. 1, pp. 98–110, 2017. [103] L. Eciolaza, T. Taniguchi, M. Sugeno, D. Filev, and Y. Wang, “Piecewise bilinear
[73] S.-S. Chen, Y.-C. Chang, S.-F. Su, S.-L. Chung, and T.-T. Lee, “Robust static out- models for feedback error learning: Online feedforward controller design,” in Proc. IEEE
put-feedback stabilization for nonlinear discrete-time systems with time delay via fuzzy Int. Conf. Fuzzy System, 2013, pp. 1–8.
control approach,” IEEE Trans. Fuzzy Syst., vol. 13, no. 2, pp. 263–272, 2005. [104] X.-J. Zeng and M.-G. Singh, “Approximation accuracy analysis of fuzzy systems as
[74] S.-W. Kau, H.-J. Lee, C.-M. Yang, C. H. Lee, L. Hong, and C.-H. Fang, “Robust function approximators,” IEEE Trans. Fuzzy Syst., vol. 4, no. 1, pp. 44–63, 1996.
H 3 fuzzy static output feedback control of T-S fuzzy systems with parametric uncertain-
ties,” Fuzzy Sets Syst., vol. 158, no. 2, pp. 135–146, 2007. 

68 IEEE Computational intelligence magazine | FEbruary 2019