IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 7, NO.

2, APRIL 1999 133

Stabilizing Controller Design for Uncertain

Nonlinear Systems Using Fuzzy Models
Marcelo C. M. Teixeira and Stanislaw H. Żak, Member, IEEE

Abstract—A Lyapunov-based stabilizing control design method where the state vector , the control input
for uncertain nonlinear dynamical systems using fuzzy models , and the matrices and are of appropriate dimen-
is proposed. The controller is constructed using a design model sions. The above information is then fused with the available
of the dynamical process to be controlled. The design model is
obtained from the truth model using a fuzzy modeling approach. IF–THEN rules where the th rule can have the form
The truth model represents a detailed description of the process Rule IF is AND AND is
dynamics. The truth model is used in a simulation experiment
to evaluate the performance of the controller design. A method THEN
for generating local models that constitute the design model is
proposed. Sufficient conditions for stability and stabilizability of where is the th fuzzy set of the th
fuzzy models using fuzzy state-feedback controllers are given. rule. Let be the membership function of the fuzzy set
The results obtained are illustrated with a numerical example and let
involving a four-dimensional nonlinear model of a stick balancer.
Index Terms— Fuzzy modeling, Lyapunov’s method, stability,
uncertain systems.

Then, given a pair the resulting fuzzy system

I. INTRODUCTION
model is inferred as the weighted average of the local models

T HE first step in the controller design procedure is the

construction of a “truth model” of the dynamics of the
process to be controlled. The truth model is a simulation model
and has the form

that includes all the relevant characteristics of the process.

The truth model is too complicated for use in the controller
design. Thus, we need to develop a simplified model that
can be used to design a controller. Such a simplified model
is labeled by Friedland [1] as the design model. The design
model should capture the essential features of the process. In
many situations, there may be human experts who can provide
a linguistic description of the process in terms of IF–THEN
rules. Combining the available mathematical description of the
process with its linguistic description results in a fuzzy system (1)
model. Such an approach to modeling was proposed by Takagi
and Sugeno [2] and further developed by Sugeno and Kang where for
[3]. This type of model is called the Takagi–Sugeno (T-S) or
Takagi–Sugeno–Kang (TSK) fuzzy model. The discrete-time
version of this model was analyzed by Tanaka and Sugeno [4],
and Tanaka [5]. Wang et al. [6], Cao et al. [7], and Feng et al.
[8] used TSK fuzzy models to design stabilizing controllers for Note that for
nonlinear systems. Other approaches to fuzzy modeling were
proposed in [9] and [10]. One can construct a TSK model if
local description of the dynamical system to be controlled is
available in terms of local linear models
and
(2)
Manuscript received January 10, 1997; revised October 8, 1998. M. C. M.
Teixeira was supported by Fundação de Amparo à Pesquisa do Estado de São In this paper, we combine the truth model of the process
Paulo (FAPESP) of Brazil. with its linguistic description to form a fuzzy design model.
M. C. M. Teixeira is with the Department of Electrical Engineering,
FEIS/UNESP, 15385-000 Ilha Solteira-SP, Brazil. The class of the truth models that we consider in this paper
S. H. Żak is with the School of Electrical and Computer Engineering, is given by
Purdue University, West Lafayette, IN 47907 USA.
Publisher Item Identifier S 1063-6706(99)02799-X. (3)
1063–6706/99$10.00  1999 IEEE
134 IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 7, NO. 2, APRIL 1999

where , , and the vector-valued different from what the designer had in his or her mind. For
function represents modeling uncertainties. The only example, suppose that the truth model of a plant has the form
information that is available to the designer about uncertainties
(6)
is their bounds. Specifically, we assume that the uncertainty
modeling function is bounded by a nonnegative function For simplicity of notation, let . Then,
; that is we can represent model (6) as
(4) (7)
where denotes the vector norm. A detailed discussion of Expanding by means of a Taylor series around
condition (4) is included in Section IV. Observe that affects yields
the system dynamics via the input matrix in a similar
fashion as the control input does. For this reason, we say that
the uncertainty satisfies the matching condition.
The resulting fuzzy design model has the form higher order terms (8)
(5) where
Our goal in this paper is to give methods for constructing (9)
stabilizing state-feedback controllers for nonlinear systems
modeled by (3). We construct a stabilizing controller using To write expressions for the second and the third terms of (8)
the fuzzy design model of the form (5) and a Lyapunov based as functions of and , let be the th element of the
approach. The fuzzy design model is obtained using the truth matrix . Then,
model and available linguistic description of the process. The
(10)
controller performance is tested on the truth model.
The paper is organized as follows. In the next section, we
use the truth model—the stick balancer; that is, an inverted where the th element of the matrix has the form
pendulum mounted on motor-driven cart to illustrate our
method for constructing fuzzy design models. The derivation
of the truth model of the stick balancer is found in the
Appendix. In Section III, we analyze the stability and stabiliz-
Finally
ability of the fuzzy design models. In Section IV, we present
a method for designing stabilizing controllers for a class
(11)
of uncertain nonlinear dynamical systems. In the controller
design, we employ fuzzy design model. Our design is tested
by means of simulations based on the truth model. Concluding A point is an equilibrium point of (7)
remarks are offered in Section V. if ; that is, if at we have . Let
and . Note that
II. CONSTRUCTING THE DESIGN MODEL
In the stabilization problem of a nonlinear plant, we are
concerned with constructing a controller so that starting from Then, the linearized model about the equilibrium is
an arbitrary point in some neighborhood of the operating obtained by neglecting higher order terms and observing that
point, the controller forces the closed-loop system trajectory for the equilibrium point . The linearized model
to converge to the operating point. On the other hand, if the has the form
starting point coincides with the operating point, the closed-
loop system trajectory is expected to stay at this point for all (12)
subsequent time. It is clear from the above description that
an operating point satisfying the above requirements must be where
an asymptotically stable equilibrium state of the closed-loop
system. In constructing a controller, we use a design model that and
is constructed from the truth model. In this paper, we combine
the mathematical description of the process in terms of the Suppose now that we wish to construct a stabilizing con-
truth model and a linguistic description of the process to obtain troller for the truth model , where
a fuzzy design model. The fuzzy design model is of the form of so that is an asymptotically stable
the TSK fuzzy model. Essential components of the TSK model equilibrium point of the closed-loop system. Our first step is to
are local linear models. These local linear models describe construct a fuzzy design model. We start with generating linear
the plant dynamical behavior at its different operating points. local models describing plant’s behavior at selected operating
We now show that using a common linearization approach to points. It is natural to first construct a local linear model that
constructing local models may result in a fuzzy design model describes plant’s behavior about the equilibrium state .
TEIXEIRA AND ŻAK: STABILIZING CONTROLLER DESIGN FOR UNCERTAIN NONLINEAR SYSTEMS 135

We can obtain this first local linear model by using the above the offset term . The designer must then be
described linearization technique. The resulting model is given careful when constructing local models using Taylor series
by . Next, we construct local linear models linearization because this approach, in general, will not yield
describing plant’s behavior at the remaining operating points. a local model that is linear. We note that Taylor’s linearization
Suppose that is the next operating state of interest. The will yield a linear system in and if the equilibrium point
result of Taylor’s linearization of a nonlinear model about an .
operating nonequilibrium point is an affine rather than linear We now describe our method for constructing local models
model. Even if the operating point is an equilibrium point, using the truth model of the system to be controlled. We
Taylor series linearization, in general, will not yield a local assume that the truth model has the form given by (6). Suppose
model that is linear in and . Indeed, suppose that the that we are given an operating state that does not have to
operating point is an equilibrium point; that is be an equilibrium state of (6). Our goal is to construct a linear
model in and that approximates the behavior of (6) in
(13) the vicinity of the operating state ; that is, we wish to find
The resulting linearized model is constant matrices and such that in a neighborhood of
for any (20)
and
(14) for any (21)
(15)
Since is arbitrary, we have to have
We can represent model (15) in the form
(22)
(16)
Thus, we are left with finding a constant matrix such that
The term does not have to be equal to zero in a neighborhood of
and, hence, model (16) is not a linear model, but rather an
(23)
affine model. We illustrate on a numerical example that the
term is indeed in general not equal to zero. and
Example 1: We consider a subsystem of model (80), de- (24)
scribed in the Appendix, that has the form
Let denote the th row of the matrix . Then, for the
purpose of further analysis, we represent condition (23) as
(25)

(17) and we represent (24) as

(26)
where m/s , kg, kg, , where the th component of is . Expanding
m. We linearize model (17) about . For the left-hand side of (25) about and neglecting second- and
to be a component of an equilibrium state, we have higher order terms yields
to have . Hence, the equilibrium state is
(27)
where is the gradient, a column vector, of
We next compute so that (13) holds. Performing simple evaluated at . Now, using (26), we represent (27) as
calculations we get . Then, using (10) we obtain
(28)
(18) where is arbitrary but “close” to . Our task now is to
determine a constant vector ; that is as “close as possible”
Next, using (11) we obtain to and satisfies the constraint . Let

(19)
Then, we can formulate our task as a constrained optimization
Note that problem of the form
minimize
(29)
subject to
The above example illustrates the fact that the system being
linear in and is in general only We note that (29) is a convex constrained optimization prob-
affine, and not linear in and because of the presence of lem. This means that the first-order necessary condition for a
136 IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 7, NO. 2, APRIL 1999

minimum of is also sufficient (see, for example, Chong We again use (35) to compute , the second row of the matrix
and Żak [11, ch. 21]). The first-order conditions for the of the second local model. We get
optimization problem (29) are

(30)

(31) The above obtained local models are the same as the ones that
were proposed by Wang et al. [6] using a heuristic approach.
where is the Lagrange multiplier and the subscript in
indicates that the gradient is computed with respect to . III. STABILITY AND STABILIZABILITY OF FUZZY MODELS
Performing the required differentiation in (30) yields
In this section, we first present sufficient conditions for
(32) asymptotic stability in the large of unforced continuous-time
fuzzy model
(33)
(36)
Recall, that we consider now the case when . Premul-
tiplying (32) by and substituting (33) into the resulting where for and
equation yields . We next discuss the stabilizability problem of forced fuzzy
models using fuzzy state-feedback control strategies.
(34) It is straightforward to show that a sufficient condition for
asymptotic stability in the large of the equilibrium state
Substituting given by (34) into (32), we obtain of fuzzy system (36) is that there exists a symmetric positive
definite matrix such that for
(37)
It is obvious that a necessary condition for the existence of a
(35) common symmetric positive definite satisfying (37) is that
Example 2: We illustrate the above procedure for construct- each be asymptotically stable; that is, the eigenvalues of
ing local linear models on the nonlinear model (17) that we each be in the open left-hand complex plane. We now give
analyzed in Example 1. This model was also considered by another necessary condition for the existence of a common
Wang et al. [6, p. 20]. Suppose that, as in Wang et al. [6], that satisfies (37).
we wish to construct two local models; one describing the Theorem 1: Suppose that each , in
plant operation about and the other model when fuzzy model (36) is asymptotically stable and there exists a
is about . In both models, we assume that and symmetric positive definite such that conditions (37) hold.
is arbitrary. Note that the nonlinear model (17) is in the Then the matrices
format given by (6). We first construct a local model about
the operating state . The local model for this
case can be obtained using Taylor’s linearization about the
equilibrium point . where and are asymptot-
We next construct the second local model that corresponds ically stable.
to and . Let . Then, the Proof: Suppose conditions (37) hold. Let
matrix can be obtained from (17) using (22). We have
(38)

where for . Summing both

sides of (38) yields
We next compute the matrix of the second local model
using (35). We first compute the first row . We note that (39)
and hence
By assumption , and since for
, the symmetric matrices
The operating state is . We compute
, the first row of , using (35) to get

also positive definite. Hence, by the Lyapunov theorem the ma-

trices A , are asymptotically stable.
TEIXEIRA AND ŻAK: STABILIZING CONTROLLER DESIGN FOR UNCERTAIN NONLINEAR SYSTEMS 137

The following corollary can be viewed as a continuous- stable in the large is that for some symmetric positive definite
time counterpart of Tanaka and Sugeno [4, theorem 4.3, pp. the following conditions are satisfied:
146–147].
Corollary 1: Suppose that each , in
(42)
fuzzy model (36) is asymptotically stable and there exists a
and
symmetric positive definite such that conditions (37) hold.
Then, the matrices (43)
We now give a sufficient condition for the stability of system
(41) for the case when conditions (43) are not satisfied. To
are asymptotically stable. proceed, we need the following notation. Suppose that there
We mention that Tanaka and Sugeno’s Theorem 4.3 can is a positive definite such that conditions (42) are satisfied.
be generalized to correspond to Theorem 1 rather than to the Represent conditions (42) as
above corollary.
The above corollary can be restated in an equivalent way
as follows. If there exist and such that a matrix (44)
is not asymptotically stable, then there is no positive definite where each is symmetric positive definite. Let denote
matrix such that for . The the minimum eigenvalue of . Since , we have
following example illustrates this fact.
(45)
Example 3: Given the unforced fuzzy system model de-
scribed by (36), where and Now, let
(46)
where now we do not require to be positive definite. Note
Obviously, each , , , is asymptotically stable. that and, hence, the eigenvalues of are real.
However, there is no common positive definite such that Let denote the minimum eigenvalue of . We are now
for because the matrix ready to state and prove the following theorem.
Theorem 2: Suppose that each
in (41) is asymptotically stable and
there exists a symmetric positive definite such that
is unstable since its eigenvalues are conditions (42) hold. Then, the closed-loop fuzzy model (41)
is asymptotically stable in the large if the matrix

Remark 1: Theorem 1 can be used to check if there does

not exist a positive definite matrix satisfying conditions .. .. .. (47)
(37) because this theorem gives us a sufficient condition for . . .
nonexistence of a common that satisfies (37).
We now present a sufficiency condition for asymptotic is positive definite.
stabilizability in the large of fuzzy model given by (1) and Proof: Let
(2) using fuzzy state feedback control law

(40) be a Lyapunov function candidate for the closed-loop system

modeled by (41). We then evaluate the time derivative of
The closed-loop system has the form on the trajectories of (41) to get

(41)

Suppose that the gain matrices where chosen so that the

matrices

have their eigenvalues in the open left-hand complex plane; (48)

that is, they are asymptotically stable. Let
for We now use that fact that for any symmetric matrix ,
we have
Wang et al. [6] observed that a sufficient condition for the
closed-loop system modeled by (41) to be asymptotically
138 IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 7, NO. 2, APRIL 1999

where is the smallest eigenvalue of . Using this family of fuzzy controllers that reject the uncertainty .
fact in (48) yields These controllers are obtained by fuzzifying the controllers
proposed by Hui and Żak [13]. To describe this family we
need the following notation. Let

(53)

.. .. .. Let denote the norm of a vector; that is, if a vector

. . . , then its norm is defined as

A well-known result concerning norms is the Hölder in-

.. equality
.

(49)
All norms on are equivalent in the sense that if and
The Lyapunov derivative is negative definite if the matrix are norms on , then there exist positive constants
is positive definite, which completes the proof. and such that
for all (54)
IV. LYAPUNOV-BASED STABILIZERS
Finally, the gradient of , where it is defined, has the form
In this section, we apply a Lyapunov control approach to the
fuzzy control design for uncertain dynamical systems. Some
of the Lyapunov-based control strategies, using nonfuzzy con- (55)
trollers for uncertain systems, are documented in Zinober [12]. Applying the chain rule, we compute
The method of constructing stabilizing controllers presented
in this section requires that each of the matrices
is already asymptotically stable and that there
exists a common symmetric positive definite .. (56)
.
such that for
(50)
Let for some . We will show that when
If this is not the case, we break the design of stabilizing control
law into two parts . We first design of the and (57)
form
then the closed-loop system driven by is
(51) asymptotically stable in the large for any uncertainty such
that .
Theorem 3: Given fuzzy system model (5), suppose that
so that the system [given by (51)] is such that there exists a common satisfying
conditions (42) and (43) or conditions of Theorem 2. Then,
the closed-loop system driven by , where is
given by (57), is asymptotically stable in the large for any
is asymptotically stable in the large and there exists a common uncertainty such that .
such that conditions (42) and (43) or conditions of Theorem Proof: Suppose that we already constructed , given by
2 are satisfied. Note that the system modeled by (5) and driven
(51), so that (42) and (43) or conditions of Theorem 2 hold.
by (51) has the form
System (5) driven by takes the form (52); that is

We now show that

(58)
(52) is a Lyapunov function for the closed-loop system where
satisfies (42) and (43) or conditions of
Next, we proceed with the design of . The role of is to Theorem 2. For this, we evaluate the Lyapunov derivative of
reject the uncertainty . Then will asymptotically ; that is, we evaluate on the trajectories of the closed-
stabilize in the large the uncertain system (5). We propose a loop system, and show that the Lyapunov derivative of is
TEIXEIRA AND ŻAK: STABILIZING CONTROLLER DESIGN FOR UNCERTAIN NONLINEAR SYSTEMS 139

negative definite. We have The parameter values are the same as in Example 1. We use
membership functions of the form

The corresponding fuzzy system model is

(59)
sign (61)
Applying Hölder’s inequality to the above yields
where because for all
and the friction coefficient . We then proceed to
construct a stabilizing controller. We first construct
(60)
By Lyapunov’s theorem, the closed-loop system is asymptot- such that the system
ically stable in the large.
Example 4: Using system model (81), we propose the fol-
lowing two rules describing the plant dynamics: (62)

Rule 1: IF is about
is asymptotically stable in the large. To better test our stabi-
THEN
lization algorithm, we randomly chose the set of desired poles
and for to be in the interval [ 4, 1], and we got
Rule 2: IF is about
THEN
where . Linear local models are The desired poles for were also chosen in a
constructed using the procedure described in the previous random fashion to be in the interval [ 4, 1]. In this case,
section and, in particular, relations (22) and (35). We have we got the set

For this choice of the closed-loop poles of the local models,

we obtain the following feedback gain matrices:

and

Let

The matrices and of the local model corresponding to and

the second rule are

Then, it is easy to check that sufficient conditions given by

(42) and (43) for asymptotic stability in the large of system
(62) are satisfied. This completes the design of .
We now construct . We use (53) to construct

In this example, the number of inputs is and, hence,

(57) reduces to

sign
140 IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 7, NO. 2, APRIL 1999

a simplified version of the truth model and usually neglects

high-frequency dynamics of the plant. Thus, the designer has
to cope with an incomplete mathematical model of the process
to be controlled. To account for modeling uncertainties, we
propose using a two-part controller. The first part stabilizes the
nominal plant; that is, the plant model that does not include
modeling uncertainties. The role of the second part of the
controller is to reject modeling uncertainties. In both cases,
fuzzy state-feedback control strategies were used. In order to
be able to apply the proposed controllers in real-life systems,
further research is needed to account for limited authority of
actuators, unmatched uncertainties, as well as inaccessibility
of state variables. We mention that algorithms for constructing
fuzzy state estimators using TSK fuzzy models were proposed
in [14] and [15].

APPENDIX
Fig. 1. Plots of x1 versus time for two different initial values of x1 :
CONSTRUCTING THE TRUTH MODEL OF THE STICK BALANCER
x1 (0) = 1:2 and x1 (0) = 0:6. In both cases, x2 (0) = x3 (0) = x4 (0) = 0. Here we develop a nonlinear four-dimensional truth model
of the stick balancer, also known as the cart with an inverted
pendulum. In our derivation of this truth model, we use
Newton’s laws in a similar fashion to Kwakernaak and Sivan
[16] and Ogata [17], where linearized models of the stick
balancer were developed. An alternative derivation, using
the method of D’Alembert, can be found in Cannon [18,
Sec. 22.4]. This truth model is used to construct fuzzy design
model and to test the performance of the stabilizing controller
design.
A free-body diagram of a cart on which a stick (an inverted
pendulum) is mounted is shown in Fig. 3. We denote by
and —the horizontal and vertical
reaction forces, respectively, and by and —the coordinates
of the fixed nonrotating coordinate frame . The angular
displacement of the stick from the vertical position is denoted
by . The mass of the cart is denoted by , while the
mass of the stick by . The length of the stick is 2l, and its
center of gravity is at its geometric center. The control force
Fig. 2. Plots of x3 versus time for two different initial values of x1 : applied to the cart is labeled by . We assume that the wheels
x1 (0) = 1:2 and x1 (0) = 0:6. In both cases, x2 (0) = x3 (0) = x4 (0) = 0.
of the cart do not slip. We model the frictional force of the
cart wheels on the track by
where we can take . We combine and to obtain
the final form of stabilizing controller. This controller has the (64)
form
where is the cart friction coefficient. Friedland [1, p. 201]
refers to this model of friction as the classical Coulomb friction
(63) model. Let be the coordinates of the center of gravity
The above controller globally stabilizes the design model. In of the stick. Then
our simulations we connected the controller to the truth model.
(65)
Typical responses are shown in Figs. 1 and 2. The controller
locally stabilizes the truth model. We observed that when
rads and , then the We are now ready to write down equations modeling the
equilibrium state of the closed-loop system consisting system. The equation that describes the rotational motion of
of the truth model driven by (63) becomes unstable. the stick about its center of gravity is obtained by applying
the rotational version of Newton’s second law. Summing the
moments about the center of gravity of the stick we obtain
V. CONCLUSIONS
A common approach to the control system design is to use
(66)
a design model to construct a controller. The design model is
TEIXEIRA AND ŻAK: STABILIZING CONTROLLER DESIGN FOR UNCERTAIN NONLINEAR SYSTEMS 141

Substituting now (71) and (72) into (66) yields

(74)

We next substitute from (73) into (74) and perform

manipulations to get

(75)

Let

Then, we can represent (73) as

(76)
Fig. 3. Free-body diagram of a cart with an inverted pendulum.
We substitute (76) into (75) to obtain as shown in (77), shown
at the bottom of the page. Let and . Using
where
the expression for given by (67), we represent (77) in
state–space format as follows:
ml (67)

is the moment of inertia of the stick with respect to its center

of gravity.
Next, we write the equation that describes the horizontal
motion of the center of gravity of the stick. Applying Newton’s (78)
second law along the axis yields

(68) We next substitute , obtained from (75), into (76) to get

Performing the required differentiation, we get

(69) (79)

The equation that describes the vertical motion of the center of Let and . Then, combining (78) and (79) we
gravity of the stick is obtained by applying Newton’s second obtain a state–space model of the inverted pendulum on a cart
law along the axis. We have of the form

(70)

Performing the needed differentiation yields

(71)

Finally, we apply Newton’s second law to the cart to get

(72)
(80)
Substituting (72) into (69) gives

(73)

(77)
142 IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 7, NO. 2, APRIL 1999

For the purpose of the controller design, we represent (80) in [12] A. S. I. Zinober, Ed., Deterministic Control of Uncertain Systems.
an equivalent form London, U.K.: Peter Peregrinus, 1990.
[13] S. Hui and S. H. Żak, “Robust output feedback stabilization of uncertain
dynamic systems with bounded controllers,” Int. J. Robust Nonlinear
Contr., vol. 3, pp. 115–132, Aug. 1993.
[14] X. Ma, Z. Sun, and Y. He, “Analysis and design of fuzzy controller and
fuzzy observer,” IEEE Trans. Fuzzy Syst., vol. 6, pp. 41–51, Feb. 1998.
[15] K. Tanaka, T. Ikeda, and H. O. Wang, “Fuzzy regulators and fuzzy
observers: Relaxed stability conditions and LMI-based design,” IEEE
Trans. Fuzzy Syst., vol. 6, pp. 250–265, May 1998.
[16] H. Kwakernaak and R. Sivan, Linear Optimal Control Systems. New
York: Wiley, 1972.
[17] K. Ogata, Modern Control Engineering, 3rd ed. Englewood Cliffs, NJ:
Prentice-Hall, 1997.
[18] J. R. H. Cannon, Dynamics of Physical Systems. New York: McGraw-
Hill, 1967.

Marcelo C. M. Teixeira was born in Campo

(81) Grande-MS, Brazil, in 1957. He received the B.Sc.
degree from Escola de Engenharia de Lins, EEL,
Brazil, in 1979, the M.Sc. degree from Universidade
ACKNOWLEDGMENT Federal do Rio de Janeiro, COPPE-UFRJ, Brazil,
in 1982, and the D.Sc. degree from Pontificia
The authors would like to thank the reviewers for their Universidade Catolica do Rio de Janeiro, PUC-
comments on the manuscript improvement. RJ, Brazil, in 1989, all in electrical engineering.
In 1982, he joined the Department of Electrical
Engineering of Universidade Estadual Paulista,
REFERENCES UNESP, at Ilha Solteira-SP, Brazil, where he is
currently an Associate Professor. In 1996 and 1997 he was a Visiting Scholar
[1] B. Friedland, Advanced Control System Design. Englewood Cliffs, NJ:
in the School of Electrical and Computer Engineering at Purdue University,
Prentice-Hall, 1996.
[2] T. Takagi and M. Sugeno, “Fuzzy identification of systems and its in West Lafayette, IN. His interests include control theory and applications,
applications to modeling and control,” IEEE Trans. Syst., Man, Cybern., neural networks, variable structure systems, and fuzzy systems.
vol. SMC-15, pp. 116–132, Jan./Feb. 1985.
[3] M. Sugeno and G. T. Kang, “Structure identification of fuzzy model,”
Fuzzy Sets Syst., vol. 28, pp. 15–33, Oct. 1988.
[4] K. Tanaka and M. Sugeno, “Stability analysis and design of fuzzy control
Stanislaw H. Żak (M’81) received the Ph.D. degree
systems,” Fuzzy Sets Syst., vol. 45, pp. 135–156, Jan. 1992.
[5] K. Tanaka, “Stability and stabilizability of fuzzy-neural-linear control from Warsaw University of Technology, Poland, in
systems,” IEEE Trans. Fuzzy Syst., vol. 3, pp. 438–447, Nov. 1995. 1977.
[6] H. O. Wang, K. Tanaka, and M. F. Griffin, “An approach to fuzzy control From 1977 to 1980, he was an Assistant Professor
of nonlinear systems: Stability and design issues,” IEEE Trans. Fuzzy at the Institute of Control and Industrial Electronics
Syst., vol. 4, pp. 14–23, Feb. 1996. (ISEP), Warsaw University of Technology. From
[7] S. G. Cao, N. W. Rees, and G. Feng, “Quadratic stability analysis and 1980 to 1983 he was a Visiting Assistant Professor
design of continuous-time fuzzy control systems,” Int. J. Syst. Sci., vol. in the Department of Electrical Engineering, Univer-
27, no. 2, pp. 193–203, 1996. sity of Minnesota, Minneapolis. In 1983 he joined
[8] G. Feng, S. G. Cao, N. W. Rees, and C. K. Chak, “Design of fuzzy the School of Electrical and Computer Engineering
control systems with guaranteed stability,” Fuzzy Sets Syst., vol. 85, pp. at Purdue University, West Lafayette, IN. He is a
1–10, Jan. 1997. coauthor of Topics in the Analysis of Linear Dynamical Systems (Warsaw,
[9] E. Kim, M. Park, S. Ji, and M. Park, “A new approach to fuzzy Poland: PWN, 1984). He is also a coauthor of An Introduction to Optimization
modeling,” IEEE Trans. Fuzzy Syst., vol. 5, pp. 328–337, Aug. 1997. (New York: Wiley, 1996). He is a past Associate Editor of Dynamics and
[10] J. Lygeros, “A formal approach to fuzzy modeling,” IEEE Trans. Fuzzy Control. He has worked in various areas of control, optimization, and neural
Syst., vol. 5, pp. 317–327, Aug. 1997. networks.
[11] E. K. P. Chong and S. H. Żak, An Introduction to Optimization. New Dr. Żak served as Associate Editor of IEEE TRANSACTIONS ON NEURAL
York: Wiley, 1996. NETWORKS from 1993 to 1997.