3

International Journal of Control and International Journal of Control and International Journal of Control and International Journal of Control
rol and Automation Automation Automation Automation

Vol. 3, No. Vol. 3, No. Vol. 3, No. Vol. 3, No. 2 22 2, , , , June June June June, 2010 , 2010 , 2010 , 2010

21

LMI-based tracking control for Takagi-Sugeno fuzzy model

Afef Abdelkrim, Chekib Ghorbel, Mohamed Benrejeb

UR LA.R.A Automatique, Ecole Nationale dIngnieurs de Tunis
BP 37 le Belvdre 1002 Tunis, Tunisie
Tlphone : +216 71 874 700, Fax : +216 71 872 729
afef.abdelkrim@esti.rnu.tn, Chekib.Ghorbel@enit.rnu.tn,
mohamed.benrejeb@enit.rnu.tn

Abstract
This paper deals with the problem of tracking control for Takagi-Sugeno fuzzy model. An
LMI (Linear Matrix Inequality) formulation is suggested to make possible the convergence of
the state vector of the continuous-time system to a desired one using a new approach, called
MultiQuadratic Fuzzy Lyapunov (MQFL). A fourth order unstable nonlinear system is
studied to illustrate the efficiency of this formulation and to compare the corresponding
simulation results with the obtained stability conditions based on the quadratic Lyapunov
function.

Keywords Takagi-Sugeno (TS) fuzzy model, stability, Linear Matrix Inequalities (LMI),
quadratic Lyapunov function

1. Introduction
Takagi-Sugeno (TS) fuzzy models have been of great importance in the academic research
and the industrial applications [1]-[3]. The main idea is based on the use of sector
nonlinearity concept, also known as, the polytopic transformation method, which decomposes
a complex nonlinear system into a set of linear subsystems using fuzzy IF-THEN rules. These
rules locally represent a linear input-output relation. Then a global nonlinear model is
constructed by assembling all the linear subsystems with associated fuzzy membership
functions. The stability of these models is, most of the time, studied using the quadratic
Lyapunov approach [4]-[10]. The obtained conditions are given in terms of Linear Matrix
Inequalities (LMI) and can be efficiently solved by convex programming techniques [11],
[12].
The stability conditions based on the use of the quadratic Lyapunov function are
conservative as a single common symmetric positive definite matrix verifying all Lyapunov
inequalities is required. It is also rejected by certain systems such as the saturated systems, the
International Journal of Control and Automation International Journal of Control and Automation International Journal of Control and Automation International Journal of Control and Automation

22

piecewise linear systems, etc. Some works show the contribution of the polyquadratic and the
piecewise quadratic Lyapunov functions, [13], [14].
Our new approach, called MultiQuadratic Fuzzy Lyapunov functions (MQFL), permits to
locally find a symmetric positive definite matrix to each model of the basis. Therefore, the
obtained stability conditions for the global fuzzy model are necessary and not sufficient.
In this paper, some basic notions of the TS fuzzy model in the continuous-time case are
presented in the first section. Based on the quadratic Lyapunov function, the second section
introduces the LMI formulation which allows the convergence of the state vector of the
closed-loop TS fuzzy model to the desired vector. In the third section, the same work is
redone using the MQFL function. Finally, the case of a fourth order unstable continuous
nonlinear system is considered to illustrate the proposed approach.
Notations
The symbol ( ) denotes the transpose elements in the symmetric positions, for example,
( ) 0 X + < stands for 0
T
X X + < and
( )
0
A B
C
| |
<
|
\
stands for 0
T
A B
B C
| |
<
|
|
\

1 1
r r
i j i j
i j i j
j i
X X X X
< = =
>
=

3
12 13 23 ij
i j
a a a a
<
= + +

BMI : Bilinear Matrix Inequalities
LMI : Linear Matrix Inequalities

2. Takagi-Sugeno fuzzy model representation
Consider a continuous nonlinear system described by the following form:
( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( )
x t f x t g x t u t
y t h x t
= +
&
(1)
where ( ) f , ( ) g and ( ) h are the nonlinear functions with appropriate dimensions,
| |
1 2
T
n
x x x x = K and
m
u are respectively the state and the input vectors.
Based on the sector nonlinearity concept, the system (1) can be represented with the TS
fuzzy model (2). Its i-th rule is given by [1]:
<Plant rule i>
International Journal of Control and International Journal of Control and International Journal of Control and International Journal of Control and Automation Automation Automation Automation

23

If ( ( )
1
z t is
1i
F ) and and ( ( )
p
z t is
pi
F ) Then
( ) ( ) ( )
( ) ( )
i i
i i
x t A x t B u t
y t C x t
= +
&
(2)

where r is the number of IF-THEN rules.
( )
1
,...,
p
z z are the premise variables, in general,
they are the state variables
i
x of the system.
ji
F ( ) 1, 2, ..., j p = are the fuzzy sets.
n n
i
A

,
n m
i
B

and
1 n
i
C

.
The defuzzification process of the TS fuzzy model (2) can be represented as:
( ) ( ) ( ) ( ) ( )
( ) ( ) ( )
1
1
r
i i i
i
r
i i
i
x t t A x t B u t
y t t C x t
=
=
= +
&
(3)
In the literature, the commonly used validities
i
are calculated as:
( ) ( ) ( )
1
/
r
i i i
i
t w t w t
=
=

(4)
where the weights
i
w given to every fuzzy rule (2) are:
( ) ( ) ( )
1
p
i ji j
j
w t F z t
=
=
(5)
t ,
( )
ji j
F z is the grade of membership of
j
z in
ji
F .

Note that:
( )
( )
1
0 1 for 1, 2, ,
1
i
r
i
i
t i r
t
=
=
K
(6)
In this paper, our objective is to achieve ( ) ( ) 0
d
x t x t as t + , where
d
x is the
desired state vector given by:
( ) ( ) ( ) ( )
1 2
T
d d d nd
x t x t x t x t = (

K (7)
We have choosen to apply the geometric method [10] calculating the distance
i
d between
the state variables
i
x of the TS fuzzy model (3) and the desired state variables
id
x as:

24

( ) ( ) ( ) for 1, 2, ,
i i id
d t x t x t i r = = K (8)
The normalized distance
norm
i
d is given by:
( ) ( ) ( )
1
/
r
norm
i i j
j
d t d t d t
=
=

(9)

The validities
i
are given by:
( ) ( ) ( )
1
/
r
i i j
j
t t t
=
=

(10)
with:
( ) ( )
( )
( )
2
1
1 1 exp
norm
r
j norm
i i
j
j i
d t
t d t
=
| | | |
| |
| |
| =
|
|
|
|
|
\
\ \
(11)
represents a variable regulating parameter between 0 and 0.99.

All
i
defined in (10) verify the same convexity conditions as (6).

Remark 1: If n r then the distance
i
d can be calculated between the partial output
i
y
of the TS fuzzy model (2) and the desired partial output
id
y as ( ) ( ) ( )
i i id
d t y t y t = where
( ) ( ) ( )
1
r
d i id
i
y t t y t
=
=
.
3. Stability conditions based on the quadratic Lyapunov function
Based on the quadratic Lyapunov function, stability analysis of a TS fuzzy model is to
seek a common positive definite matrix P verifying 0 V > and 0 V <
&
for all the models of
the basis. Its basic results can be found in [4]-[9], [15]-[17].
To achieve ( ) ( ) 0
d
x t x t as t + , an elementary fuzzy control law
i
u is applied:
<Controller rule i>
If ( ( )
1
z t is
1i
F ) and and ( ( )
p
z t is
pi
F ) Then ( ) ( ) ( ) for 1, 2, ...,
i i i d
u t K x t N x t i r = + =
(12)
where
i
K and
i
N are the feedback and the tracking gains, respectively.


25

The global fuzzy control law u is done by:
( ) ( ) ( ) ( ) ( )
1
r
i i i d
i
u t t K x t N x t
=
= +
(13)
Then, substituting (13) in (3) leads to:
( ) ( ) ( )
( )
( )
( ) ( ) ( )
1 1
1
r r
i j ij ij
d i j
r
i i
i
x t
x t t t G H
x t
y t t C x t

= =
=
| | (
( = |
(

|
&
(14)
with
ij i i j
G A B K = and
ij i j
H B N = .

Note that e is the difference the state vector x of the continuous TS fuzzy model (14)
and the desired state vector
d
x , supposed constant, as:
( ) ( ) ( )
d
e t x t x t = (15)
The time derivative ( )
( ) d e t
e t
dt
= & is given by:
( ) ( ) ( )
( )
( )
1 1
r r
i j ij ij
d i j
e t
e t t t G
x t

= =
| | (
( = |
(

|
\
& (16)
with
ij ij ij
G H = + .

The expression of e& contains the dominant terms (for 1, 2,..., i r = ) and the coupled terms
(for 1 i j r < ). It can be rewritten as:
( ) ( ) ( ) | |
( )
( )
( ) ( )
( )
( )
2
1
2
2 2
r
ij ji ij ji
i ii ii i j
d d i i j
G G e t e t
e t t G t t
x t x t

= <
| | | | + + ( ( ( | | | |
= + | | ( | | ( (
| |
( \ \ \ \

&
(17)
Theorems 1 and 2 are developed in terms of the BMI and LMI, respectively, to achieve
( ) ( ) 0
d
x t x t as t + .

Theorem 1 The closed-loop continuous system (17) is quadratically globally stable if there
exists a common symmetric positive definite matrix P satisfying the following BMI
formulation:

26

( )
( )
( ) ( )
* 0
* 0 for 1, 2, ,
* *
ii ii
PG P
P P i r
P
+ | |
|
< =
|
|
\
K (18a)
( )
( )
( ) ( )
* 0
2 2
* 0 for 1
* *
ij ji ij ji
G G
P P
P P i j r
P
| + + | | | | |
+ | | |
| \ \
|
<
|
|
|
|
\
(18b)
Proof : Based on the quadratic approach of Lyapunov verifying:
( ) ( ) ( ) ( ) 0
T
V e t e t P e t = > (19a)
( ) ( ) ( ) ( ) ( ) ( ) 0
T T
V e t e t P e t e t P e t = + <
&
& & (19b)

the time derivative ( ) V e
&
is given by:
( ) ( )
( )
( )
( ) ( ) ( ) ( )
( )
( )
2
1
2 0
2
T
r
ij ji
i ii i j
d d i i j
e t e t
V e t t t t
x t x t

= <
| + | ( ( | |
= + < | | ( (
|
\ \

&
(20)
with:
( )
( )
*
for 1, 2, ,
* 0
ii ii
ii
PG P
i r
| + |
=
|
\
K (21a)
( )
( )
*
2 2 for 1
2
* 0
ij ji ij ji
ij ji
G G
P P
i j r
| | + + | | | |
+ + | | |
<
| \ \
|
\
(21b)
As all 0 1
i
, for 1, 2, , i r = K , and
( )
( )
2
0
d
e t
x t
(
(

, (20) is equivalent to:
0 for 1, 2, ,
ii
i r < = K (22a)
0 for 1
2
ij ji
i j r
+
< (22b)
The matrix inequalities (22a) and (22b) can be rewritten as:
( )
( )
| |
1
* 0
0 0 for 1, 2, ,
*
ii ii
PG P
P P i r
P P

| + | | | (
+ < =
| |
(
\ \
K (23a)

27

( )
( )
| |
1
* 0
2 2 0 0 for 1
*
ij ji ij ji
G G
P P
P P i j r
P
P
| | + + | | | |
+ | | | ( | |
+ <
| | \ \ (
\
|
\
(23b)
While applying to (23a) et (23b) the Schurs Lemma , see Annexe, we get the results of
Theorem 1.

The obtained stability conditions (18a) and (18b) are not linear. However, by making
variable changes and some matrix transformations such as pre- and post-multiplying them by
the diagonal matrix
( )
1 1 1
, , P P P

(

, replacing
ij i i j
G A B K = ,
ij i j
H B N = ,
ij ij ij
G H = + ,
1
S P
= ,
i i
U K S = and
i i
V N S = , (18a) and (18b) can become linear, see Theorem 2.

Theorem 2 The closed-loop continuous system (17) is quadratically globally stable if there
exists a common symmetric positive definite matrix
1
S P
= and vectors
i
U and
i
V of
appropriate dimensions satisfying the following LMI formulation:

( )
( )
( ) ( )
* 0
* 0 for 1, 2, ,
* *
i i i i i i i i
A S B U A S B U B V
S S i r
S
+ + | |
|
< =
|
|
\
K (24a)
( )
( )
( )
( )
( ) ( )
,
* 0
2 2
* 0 for 1
* *
i j i j j i
A A S B U B U
i j
S S i j r
S
| |
+

|
+
|
|
<
|
|
|
\
(24b)

with:
( ) ( )
,
i j i j j i i j j i
i j A A S B U B U B V B V = + + +
The gains
i
K and
i
N are given by:
and for 1, 2,...,
i i i i
K U P N V P i r = = = (25)
4. Stability conditions based on the MQFL function
If the polyquadratic Lyapunov function, which is given by:
( ) ( ) ( ) ( ) ( ) ( )
1
r
T
i i
i
V e t e t z t P e t
=
=

(26)

28

where
i
are defined in (10), then, the continuous-time derivate of V is:
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( )
( )
1 1 1
0
r r r
i T T T
i i i i i
i i i
t
V e t e t t P e t e t t P e t e t P e t
t

= = =
| | | | | |
= + + <
| | |

\ \ \

&
& & & (27)
Is used, we can get no information about the monotony of the quantity
( )
i
t
t

, and therefore
the matrix inequalities (27) cannot be given in terms of LMI. To surpass this problem, we
suggest to use the MQFL function which locally finds a symmetric positive definite matrix
i
P for each model of the basis. The proposed MQFL function is given by:
If ( ( )
1
z t is
1i
F ) and and ( ( )
p
z t is
pi
F ) Then ( ) ( ) ( ) ( ) for 1, 2, ,
T
i i
V e t e t P e t i r = = K (28)
The obtained stability conditions, given in Theorem 3, are necessary and not sufficient, ie, if
the stability is verified to every model it is not necessarily true for the global model.

Theorem 3 The closed-loop continuous system (17) is globally stable if there exist r
symmetric positive definite matrices
1
q q
S P
= , for 1, 2, , q r = K , and vectors

iq
U and
iq
V of
appropriate dimensions satisfying the following LMI formulation:

( )
( )
( ) ( )
* 0
* 0 for , 1, 2,...,
* *
i q i iq i q i iq i iq
q q
q
A S B U A S B U B V
S S i q r
S
| | + +
|
< = |
|
|
\
(29a)
( )
( )
( )
( ) ( )
, ,
* 0
2 2
* 0 for 1 and 1, 2, ,
* *
ij q i jq j iq
q q
q
A S B U B U i j q
S S i j r q r
S
| |
+
|
|
< = |
|
|
|
\
K (29b)
with:
( ) , ,
ij q i jq j iq i jq j iq
i j q A S B U B U B V B V = + +
The gains
i
K and
i
N are given by:
and for 1, 2,...,
i ii i i ii i
K U P N V P i r = = = (30)

29

Proof: Consider the MQFL function defined in (28), its time derivate ( ) ( )
( ) ( )
q
q
dV e t
V e t
d t
=
&

is given by:
( ) ( ) ( ) ( ) ( ) ( ) 0
T T
q q q
V e t e t P e t e t P e t = + <
&
& & (31)
Then, substituting (17) in (31) leads to:
( ) ( )
( )
( )
( ) ( ) ( ) ( )
( )
( )
2
1
2 0
2
T
q q
r
ij ji
q
q i ii i j
d d i i j
e t e t
V e t t t t
x t x t

= <
| | | |
+ ( (
| | = + <
( (
| |

\ \

&
(32)
with:
( )
( )
*
for , 1, 2, ,
* 0
q ii q ii
q
ii
P G P Z
i q r
+ | |
= |
|
\
K (33a)
( )
( )
*
2 2 for 1 and 1, 2, ,
2
* 0
ij ji ij ji q q
ij ji q q
G G Z Z
P P
i j r q r
| | + + | | | |
+ + | | |
< =
| \ \
|
\
K (33b)
As all 0 1
i
and
( )
( )
2
0
d
e t
x t
(
(

, (32) is equivalent to:

0 for , 1, 2, ,
q
ii
i q r < = K (34a)
0 for 1 and 1, 2, ,
2
q q
ij ji
i j r q r
+
< = K (34b)
The obtained matrix inequalities (34a) and (34b) can be rewriten as follows:
( )
( )
1
* 0
0 0 for , 1, 2, ,
*
q ii q ii
q q
q
q
P G P Z
P P i q r
P
P

+ | | | | (
( + < = | |
(

| |
\ \
K (35a)
( )
( )
1
* 0
2 2 0 0 for 1 and 1, 2, ,
*
ij ji ij ji
q q
q q
q
q
G G Z Z
P P
P P i j r q r
P
P
| | + + | | | |
+ | | | ( | |
( + < = |
| ( \ \

|
| \
\
K
(35b)
While applying to (35a) and (35b) the Schurs Lemma, we get:

30

( )
( )
( ) ( )
* 0
* 0 for , 1, 2, ,
* *
q ii q ii
q q
q
P G P Z
P P i q r
P
| | +
|
< = |
|
|
\
K (36a)
( )
( )
( ) ( )
* 0
2 2
* 0 for 1 and 1, 2, ,
* *
ij ji ij ji
q q
q q
q
G G Z Z
P P
P P i j r q r
P
| + + | | | | |
+ | | |
| \ \
|
< =
|
|
|
|
\
K (36b)
While pre- and post-multiplying the obtained matrix inequalities (36a) and (36b) by a
diagonal matrix
( )
1 1 1
, ,
q q q
P P P

(

, replacing
ij i i j
G A B K = ,
ij i j
H B N = ,
ij ij ij
Z G H = +
and making the following variable changes
1
q q
S P
= ,
iq i q
U K S = and
iq i q
V N S = , we get
the results of Theorem 3.
5. Numerical example
Consider a nonlinear unstable system as follows [16], [17]:
( ) ( ) ( ) ( ) ( ) ( ) ( )
( )
( )
2
1 1 2 3 4 1
sin 0.1 1 x t x t x t x t x t x t u t = + + + + & (37a)
( ) ( ) ( )
2 1 2
2 x t x t x t = & (37b)
( ) ( ) ( ) ( ) ( )
2
3 1 1 2 3
0.3 x t x t x t x t x t = + & (37c)
( ) ( ) ( ) ( )
4 3 4
sin x t x t x t = & (37d)

Assume that ( ) | |
1
, x t a a and ( ) | |
3
, x t b b where a and b are positive numbers.
Using the sector nonlinearity concept, the nonlinear continuous system (37) is represented by
the global TS fuzzy model (14) with ( ) ( )
1 1
z t x t = , ( ) ( )
2 3
z t x t = and 4 r = .

The premise membership functions are given by:
2 2
1 11 13
0 x F a F = + and ( )
( )
3 31 3 32 3
sin
sin
b
x F x F x
b
= +
The fuzzy sets
12 11
F F = ,
14 13
F F = ,
33 31
F F = and
34 32
F F = are in | | 0, 1 and verify
11 13
1 F F + = and
31 32
1 F F + = with:

31

2
1
11
2
x
F
a
= and
( ) ( )
( ) ( )
3 3
3
3
31
3
sin sin
for 0
sin
1 for 0
b x x b
x
x b b
F
x

The consequent matrices are given by:
1
2
1 1 1 0.1
1 2 0 0
1 0.3 0
0 0 1 1
A
a
| |
|
|
=
|
|
|
\
,
( ) ( )
( ) ( )
2
2
1 1 sin / 0.1
1 2 0 0
1 0.3 0
0 0 sin / 1
b b
A
a
b b
| |
|
|
= |
|
|
|
\
,
3
1 1 1 0.1
1 2 0 0
1 0 0.3 0
0 0 1 1
A
| |
|
|
=
|
\
,
( ) ( )
( ) ( )
4
1 1 sin / 0.1
1 2 0 0
1 0 0.3 0
0 0 sin / 1
b b
A
b b
| |
|
|
=
|
|
|
\
,
2
1 2
1
0
0
0
a
B B
| |
+
|
|
= =
|
|
|
\
and
3 4
1
0
0
0
B B
| |
|
|
= =
|
|
\
.

In the simulation, assume that 1.4 a = , 0.7 b = and choose the initial states
| | 0.3 0.7 0.6 0.4
T
.

Remark 2: Each of the fourth models of the basis contains an unstable pole.

While solving the LMI (24a), (24b), (29a) and (29b), we find the following quadratic
Lyapunov matrix P and the MQFL matrices
i
P , for 1, 2, , 4 i = K :

( )
0.1152 0.0156 0.0471 0.0021
0.1058 0.0070 0.0071
0.0644 0.0170
* 0.0376
P
| |
|
|
=
|
|
|
\

( )
1
0.6141 0.1384 0.2832 0.0088
0.4173 0.0766 0.0183
0.3028 0.0705
* 0.1358
P
| |
|
|
=
|
|
|
\

( )
2
0.6277 0.1424 0.2907 0.0096
0.4179 0.0790 0.0180
0.3069 0.0708
* 0.1357
P
| |
|
|
=
|
|
|
\


32

( )
3
0.6762 0.2604 0.3343 0.0128
0.5637 0.1673 0.0134
0.3655 0.0805
* 0.1516
P
| |
|
|
=
|
|
|
\

( )
4
0.6277 0.1424 0.2907 0.0096
0.4179 0.0790 0.0180
0.3069 0.0708
* 0.1357
P
| |
|
|
=
|
|
|
\

The gains
i
K and
i
N defined in (25) in the quadratic case and (30) in the MQFL case are
given in Table 1 and Table 2, respectively.

Table1.Gainsvalues
i
K
and
i
N
inthequadraticcase

| |
| |
| |
| |
1
2
3
4
1.3805 0.9864 1.0174 0.1366
1.3809 0.9890 0.9825 0.1244
3.1822 2.1119 2.3276 0.3124
3.1875 2.1278 2.2815 0.2953
K
K
K
K
=
=
=
=

| |
| |
| |
| |
1
2
3
4
1.0427 0.6486 0.6795 0.1028
1.0431 0.6511 0.6716 0.0906
2.1934 1.1264 1.7241 0.2167
2.1182 1.1278 1.3611 0.8201
N
N
N
N
=
=
=
=

Table2.Gainsvalues
i
K
and
i
N
intheMQFLcase

| |
| |
| |
| |
1
2
3
4
1.9367 1.1984 1.3103 0.1803
1.9713 1.2092 1.2927 0.1697
4.8771 3.7605 3.4949 0.4634
4.4795 2.6120 2.9872 0.4062
K
K
K
K
=
=
=
=

| |
| |
| |
| |
1
2
3
4
1.5988 0.8606 0.9724 0.1465
1.6177 0.8911 0.9613 0.1359
3.9941 2.6733 2.4798 0.3774
3.9325 1.7612 2.3167 0.3401
N
N
N
N
=
=
=
=

Figure 1 illustrates the trajectories of the state variables
1
x ,
2
x ,
3
x and
4
x of the
continuous system (37) described by the closed-loop TS fuzzy model (14) and the chosen
constant desired state variable
d
x given by a step of amplitude 0.5 .

33

0 1 2 3 4 5 6 7 8
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
(a)
x
1
(
t
)
0 1 2 3 4 5 6 7 8
0.45
0.5
0.55
0.6
0.65
0.7
(b)
x
2
(
t
)
0 1 2 3 4 5 6 7 8
0.3
0.35
0.4
0.45
0.5
0.55
0.6
(c)
x
3
(
t
)
0 1 2 3 4 5 6 7 8
-0.4
-0.2
0
0.2
0.4
0.6
(d)
x
4
(
t
)
xd
MQFL approach
Quadratic approach

Fig.1.Trajectoriesofthestatevariables
1
x
,
2
x
,
3
x
and
4
x
andthedesiredstate
vector
d
x
in the cases of the MQFL function (solid line) and the quadratic function (dotted line)
In the Figure 2, are given the trajectories of the global fuzzy control law u in the two
cases.

0 1 2 3 4 5 6 7 8
-2.5
-2
-1.5
-1
-0.5
0
0.5
time (s)
u (MQFL approach)
u (Quadratic approach)

Fig.2.Trajectoriesof
u


34

When the MQFL or the quadratic Lyapunov functions are used, the feedback
i
K and the
tracking
i
N gains guarantee the stability of the continuous system (37), initially unstable,
described by the fuzzy closed-loop (17). Thus, all the trajectories of the state variables
i
x
can track the desired state vector
d
x very well. The global fuzzy model response is better in
the MQFL case than in the quadratic Lyapunov one.
6. Conclusion
In this paper, a contribution in tracking control for the Takagi-Sugeno fuzzy model is
considered. Using the MultiQuadratic Fuzzy Lyapunov (MQFL) approach, a new LMI
formulation is suggested in order to allow the convergence of the state vector of the system to
a desired one. This approach guarantees the local stability of each model of the basis;
however, the stability of the global model cannot be concluded. When comparing the MQFL
and the quadratic Lyapunov approaches, the results of simulation, in the case of a fourth order
continuous nonlinear system, are satisfactory. Our works will focus mainly on seeking
sufficient LMI-based stability conditions for the global model using the introduced
polyquadratic Lyapunov approach.
Appendix [Lemma of Schur]
Let the matrices M , L and Q of appropriate dimensions where
T
M M = and
0
T
Q Q = > [12], [13]:
1
0 0
T
T
M L
M L QL
L Q
| |
+ < < |
|
\

7. Refferences
[1] K. Tanaka and H. O. Wang, Fuzzy control systems design and analysis. A linear matrix
inequality approach, (John Wiley and Sons, New York, 2001).
[2] T. Takagi and M. Sugeno, Fuzzy identification of systems and its applications to modeling and
control, IEEE Trans. Syst. Man., Cybern, vol. SMC-15, no. 1, pp. 116-132, Jan. 1985.
[3] M. Vidyasagar, Nonlinear System Analysis, Prentice-Hall, Englewood Cliffs, 1993.
[4] H. O. Wang, K. Tanaka and M. F. Griffin, An approach to fuzzy control of nonlinear systems:
stability and design issues, IEEE Trans. on Fuzzy Syst., vol. 4, no. 4, pp. 14-23, Feb. 1996.
[5] K. Tanaka and M. Sugeno, Stability and design of fuzzy control systems, Fuzzy Set and
Systems, vol. 45, no. 2, pp. 135-156, 1992.
[6] C. H. Fang, Y. S. Liu, L. Hong and C. H. Lee, A new LMI-based approach to relaxed quadratic
stabilization of TS fuzzy control systems, IEEE Trans. on Fuzzy Syst., vol. 14, no. 3, pp. 386-397,
June 2006.

35

[7] K. Tanaka, T. Ikeda and H. O. Wang, Fuzzy regulators and fuzzy observers: relaxed stability
conditions and LMI-based designs, IEEE Trans. on Fuzzy Syst., vol. 6, no. 2, pp. 250-265, May
1998.
[8] E. Kim and H. Lee, New approaches to relaxed quadratic stability condition of fuzzy control
systems, IEEE Trans. on Fuzzy Syst., vol. 8, no. 5, pp. 523-534, Oct. 2000.
[9] J. Joh, Y. H. Chen and R. Langari, On the stability issues of linear Takagi-Sugeno fuzzy
models, IEEE Trans. on Fuzzy Syst., vol. 6, no. 3, pp. 402-410, Aug. 1998.

[10] Y. Morre, Design and implementation of control law for Takagi-Sugeno fuzzy models, PhD.
Thesis, LAMIH, University of Valenciennes and Hainant-Cambrsis, Jan. 2001 (in French).
[11] S. Boyd, L. El Ghaoui, E. Feron and V. Balakrishann, Linear matrix inequalities in system and
control theory (SIAM, Philadelphia, PA, 1994).
[12] P. Gahinet, A. Nemi rovski, A. J. Laub and M. Chilali, LMI control toolbox, the Math Works
(Natick, MA, 1995).
[13] M. Chadli, D. Maquin and J. Ragot, On the stability analysis of multiple model systems, ECC,
Porto, pp.1894-1899, 2001.
[14] M. Johansson, A. Rantzer and K. Arzen, Piecewise quadratic stability for affine Sugeno systems,
Fuzzy IEEE98, Anchorage, Alaska, 1998.
[15] C. Ghorbel, A. Abdelkrim and M. Benrejeb, Observers for continuous nonlinear systems
containing unknown parameters and described by Takagi-Sugeno fuzzy model, Int. Journal of
Control and Intelligent Systems, vol. 38, n. 2, pp. 103-109, April 2010.
[16] C. Ghorbel and A. Abdelkrim, Contrle de poursuite dun systme complexe dcrit par
lapproche algbrique de Kharitonov : Approche LMI , Revue des Sciences et Technologies de
lAutomatique (e-sta), vol. 7, n. 1, pp. 42-47, March 2010.
[17] C. Ghorbel, A. Abdelkrim and M. Benrejeb, An adaptive fuzzy control of continuous nonlinear
systems, IEEE, 6th International Multi-Conference on Systems, Signals and Devices, SSD09,
March 23-26, 2009, Djerba, Tunisia.


36

Authors

Afef Abdelkrim has obtained a Diploma of Engineer in 2000 and the
PhD. in Electrical Engineering at the National School of Engineers of
Tunis (ENIT) in 2005; she is now an Assistant Professor at the High
School of Technology and Computer Science (ESTI). Her research
interests are in handwriting process modelling and synthesis and in
the artificial neuronal network domains.

Chekib Ghorbel received the Engineer Diploma degree in
Electrical Engineering and the Master in Automatic Control from
Ecole Nationale dIngnieurs de Sfax (Tunisia) in 2005 and 2006,
respectively. He is currently preparing the PhD. degree in Electrical
Engineering within the framework of LARA (ENIT). His research is
related to stability and stabilization of the complex continuous-time
systems described by Takagi-Sugeno fuzzy model.

Mohamed Benrejeb has obtained the Diploma of "Ingnieur IDN"
(French "Grande Ecole") in 1973, the Master degree of Automatic
Control in 1974, the PhD. in Automatic Control of the University
(USTL) of Lille in 1976 and the DSc of the same University in 1980.
He is currently a full Professor at the Ecole Nationale dIngnieurs
de Tunis (Tunisia) and an Invited Professor at the Ecole Centrale de
Lille (France). His research interests are in the area of analysis and
synthesis of complex systems based on classical and non conventional approaches and
recently in discrete event system domain.

3

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

3

Uploaded by

Copyright:

Available Formats

International Journal of Control and International Journal of Control and International Journal of Control and International Journal of Control

rol and Automation Automation Automation Automation

= , for 1, 2, , q r = K , and vectors

You might also like