Robust Fuzzy Sliding Mode Controller for

Discrete Nonlinear Systems

Hafedh Abid1, Mohamed Chtourou2, Ahmed Toumi1,

1 Unité de Commande et de Procédés Industriels (UCPI), ENIS BP. W,
3038 Sfax, Tunisie
[email protected] , [email protected]
2 Unité de Commande Intelligente, design et optimisation des Systèmes complexes,

(ICOS) ENIS,B.P. W, 3038 sfax, Tunisie

[email protected]

Abstruct: In this work we are interested to discrete robust fuzzy sliding mode
control. The discrete SISO nonlinear uncertain system is presented by the
Takgi-Sugeno type fuzzy model state. We recall the principle of the sliding
mode control theory then we combine the fuzzy systems with the sliding mode
control technique to compute at each sampling time the control law. The con-
trol law comports two terms: equivalent control law and switching control law
which has a high frequency. The uncertainty is replaced by its upper bound.
Mass spring dumper is used to check performances of the proposed fuzzy robust
sliding mode control scheme.

1. Introduction

Most of the industrial plants contain nonlinearities or/and uncertainties. Several tech-
niques have been used to identify the nonlinearities to reach wanted performances.
The robust control is one of the techniques that permits to reach the desired perform-
ances in presence of external or/ and internal disturbances. Besides the stability, the
state variables of the system must follow the reference one independently of uncer-
tainties. In literature, many methods have been developed in continuous or in discrete
time to solve the tracking problem for nonlinear systems.
In the last few years many researches combine sliding mode techniques with intelli-
gent one, such as Neuronal systems or fuzzy systems [3, 4], to lead to better results.
The sliding mode control (SMC) was originally developed for variable structure sys-
tems in continuous domain. Utkin [14] gives a thorough description of the sliding
mode theory in continuous time. Also, Slotine and Li [17] describe in detail continu-
ous sliding mode controllers.
At the end of the twentieth century, the research of discrete time SMC has been at-
tracted more attention, such as [7,9,11, 12], as for the implementation of the control-
ler on a digital computer requires a sampling time and the assumption of an infinite
switching time does not hold any more.

STA'2007-ACS-185, pages 1 à 15
Sciences et Techniques de l'Automatique
© Centre Universitaire de Publication 2007
2 STA′2006 – Topics …, pages 1 à X

The fuzzy systems have been combined with classical sliding mode control to provide
robust stability to the fuzzy controller. The combination of the two control principles,
is called fuzzy sliding mode control (FSMC), it provides an alternative to design a
robust controller for nonlinear systems with uncertainties [15, 16]. Our contribution
in this work consists of presenting a new robust controller based on the integration of
the dynamic Takagi-Sugeno type fuzzy state model for discrete nonlinear systems in
the classical sliding mode control theories. Actually the integration of the fuzzy logic
in the sliding mode is due to the fact that the stability is difficult to illustrate in the
case of a fuzzy controller contrary to a controller by sliding mode where the stability
is inherent. This interest is due to the fact, that, unlike the controller by sliding mode
where the stability is inherent it is difficult to illustrate the stability of fuzzy control-
ler. This paper is organised as follow. In Section two, we recall the concept of dis-
crete Takagi-Sugeno type fuzzy state model for nonlinear systems. Then, we describe
in the first part of the third section the sufficient and necessary reaching conditions of
the sliding mode control for the discrete nonlinear systems. The fuzzy sliding mode
controller for discrete time of nonlinear systems is developed in the second part of the
third section. In Section 4, the proposed discrete fuzzy sliding mode control
“DFSMC” scheme is used to control mechanical nonlinear system. Finally, commen-
taries and concluding are given in the final section.

2. Problem Statement and Fuzzy Systems

2.1. Problem Statement

Consider a class of discrete nonlinear systems SISO described by the following

equations:
⎧ x 1 ( k + 1) = x 2 ( k )
⎪
⎪. (1)
⎪⎪ .
⎨
⎪ x i ( k + 1) = x i +1 ( k )
⎪ x (k + 1 ) = f ( X ( k )) + g (X (k ))u (k )
⎪ n
⎪⎩ y ( k ) = C X ( k )

T
where, C=[1,0,…,0], X ( k ) = ⎡⎣ x 1 ( k ) . . . x n ( k ) ⎦⎤ ∈ ℜ n is a
state variable vector, f(X(k)) and g(X(k)) are two discrete bounded nonlinear func-
tions,
f ( X ( k )) = f o ( X ( k )) + ∆ f , g ( X ( k )) = g o ( X ( k )) + ∆ g
 Paper title 3

In order for (1) to be controllable, it is required that g(X(k))≠0.

The nominal control law can be expressed as:

u (k ) = 1
T e g (X (k ))
−K ( T
E (k ) + x nd ( k + 1) − x nd (k ) − T e f ( X ( k )) ) (2)

In general case, the functions f(X(k)) and g(X(k)) are badly known so, the control law
(2) cannot be implanted. To overcome these difficulties, many approaches are used
such as adaptive control, linearization around operating points, fuzzy control,… etc.

2.2. Takagi-Sugeno Fuzzy Model

The advantage of the T-S type fuzzy models is that their description permits the utili-
zation of the state representation, and by consequence to exploit the maximum of the
potential relative to this representation. The Takagi-Sugeno (T-S) type fuzzy model
can be viewed as a natural expansion of piecewise linear partition for nonlinear sys-
tems. The nonlinear system is represented as a collection of the fuzzy IF-THEN rules,
where each rule describes the local dynamics by a linear system model. The general
fuzzy model is achieved by fuzzy amalgamation of the linear systems models [20,
29]. The ith rule of the discrete fuzzy model has the following form:
ith plant rule :

⎧⎪ X (k + 1) = Adi X (k) + B di u(k) , (3)

IF z 1 is µi 1 and z 2 is µi 2 and .. and z n is µin THEN ⎨
⎪⎩ y(k) = C i X (k) ; i = 1, 2...c

{ }
where µ ij are the fuzzy sets, Ci = [1, 0, 0, …, 0], X(k) = x1(k) . . . xn (k) ∈ℜ [ ] n

is the state vector, u (k) is the input vector, Adi ∈ Rnxn is the state matrix, Bdi ∈ Rnxm is
the input matrix, c is the number of the IF-THEN fuzzy rules;
z (k ) = [z1 (k ) . . . z n (k )] are the premise variables they represent some meas-
urable system variables, they can be chosen as a state variables. It is assumed in this
paper that the premise variables do not depend on the input variables u(k);
The discrete Takagi-Sugeno type fuzzy model of the certain system is inferred as
follows:
c (4)
∑ w i ( z (k ))[ A d i X ( k ) + B d i u ( k )]
i =1
X (k ) = c
∑ w i ( z ( k ))
i =1
4 STA′2006 – Topics …, pages 1 à X

For each rule Ri is attributed a weight wi(z(k)) which depends on the membership
function’s grade of the premise variables zj(t) in the fuzzy sets µij , which is pre-
sented by the following expression:
n (5)
wi ( z ( k )) = ∏ µij ( z j ( k )) ;
j =1

where, µij(zj (k)) is the membership grade of z j(k) of the fuzzy set µij and wi(z(k)) >0 ;
for all i = 1, … c ;
The normalized weight is defined as h j ( z ( k )) which is presented as follow:

w j ( z ( k )) (6)
h j ( z ( k )) = c
;.
∑ wi ( z ( k ))
i =1

c
0< hi(z(k)) <1; i = 1, … c ; ∑ h (z(k))
i =1
i =1

The polytopic form of the state equation is expressed as:

c (7)
X ( k + 1) = ∑ h i ( z ( k ))[ A di X ( k ) + B di u ( k )]
i =1

Where, Adi and Bdi are the discrete state matrices of the ith sub-system. They will be
written as follow:
⎡ 0 1 0 ... 0 ⎤ (8)
⎢ 0 0 1 ... 0 ⎥
⎢ ⎥
A di = ⎢ ⎥; B di = [ 0 ,0 ,...,b in ]T
⎢ ⎥
⎢ 0 0 0 ... 1 ⎥
⎢⎣ − a i 1 − ai2 − a i3 ... − a in ⎥⎦

The output of the discrete Takagi-Sugeno type fuzzy model for the uncertain nonlin-
ear systems can be inferred as:
(9)
⎧
( )
c
⎪ X (k + 1 ) = ∑ h i ( z ( k )) ( A d i + ∆ A d i ) X ( k ) + ( B d i + ∆ B d i )u (k )
⎪ i =1
⎨ c
⎪ y (k ) = ∑ h ( z ( k ))C X (k ) ;
⎪⎩ i =1
i i

where, ∆Adi and ∆Bdi are the uncertainty discrete state matrices of the ith sub-system.
They can be written as:
 Paper title 5

⎡ 0 0 0 ... 0 ⎤
⎢ 0 0 0 ... 0 ⎥
⎢ ⎥
∆ A di = ⎢ ⎥; ∆ B di = [ 0 ,0 ,..., ∆ b in ] T
⎢ ⎥
⎢ 0 0 0 ... 0 ⎥
⎢⎣ − ∆ a i1 − ∆ ai2 − ∆ a i3 ... − ∆ a in ⎥⎦

It is required that all bin is different from zero to assure the controllability of (7).

3. Sliding mode Control law and reaching conditions

The sliding mode control, first appeared in the Soviet literature [14], it has been
widely recognized as a potential approach to uncertain dynamical non-linear systems
that are subject to external disturbances and parameter variations. In sliding mode
control (SMC), the control action forces the system trajectories to cross the state
space’s manifold which is called the sliding surface [14]. The system trajectories are
then constrained to the sliding surface for all subsequent time via the use of high
speed switching control. The most significant advantage of the sliding mode is ro-
bustness against the system parameters changes or disturbances. The chattering phe-
nomena, is the major disadvantage associated with the sliding mode control because it
can excite the high frequency dynamics of unmodelled nonlinearity.
Many researches are interested to continuous sliding mode control, whereas only a
few researches are interested in discrete-time sliding mode control. A discrete version
of SMC has a great importance as the control implementation is realized by digital
components which need a sampling period to compute the appropriate controller. It
must be pointed out that the discrete version of SMC cannot be obtained from their
continuous counterpart by means of simple equivalence. Dote and Hoft [6 ] are the
first among those who are interested in the discrete SMC problem and used an
equivalent form of the continuous reaching condition to give the following discrete
reaching condition:
[S ( k + 1 ) - S ( k ) ]S ( k ) p 0 (20)

Milosavljevic [7 ] suggested the concept of the quasi-sliding mode and pointed out
that the condition (10) is not sufficient for such a discrete sliding mode control.
Sarpturk, et al. [8], proposed the following reaching condition:
S ( k + 1) p S ( k ) (13)

Furuta [ 9] used the equivalent form of a Lyapunov-type of continuous reaching con-

1
dition V& p 0, where V = S 2 , to give the discrete version.
2
1 (42)
V ( k + 1) − V ( k ) p 0 with V ( k + 1) = (S ( k ) )2
2
6 STA′2006 – Topics …, pages 1 à X

Weibing Gao et al. [11] pointed out that all these forms of reaching conditions are
incomplete to have a satisfactory characterization of a discrete-time sliding mode
control. They suggest that the state trajectory of a discrete sliding mode control sys-
tem must have some attributes which form the basis of the discrete sliding mode con-
trol [11].

3.1 Discrete fuzzy sliding mode control law

For the discrete SMC we use the following reaching law:

S (k + 1) = S (k ) − qTe S (k ) − εTe S (k ) sgn((S (k )) , with: 1-qT e f 0, ε f 0,q f 0 (53)

The sliding surface is defined as:

S (k ) = G T X (k ) (64)

where, GT is a constant row vector GT= [g1, …, gn-1, 1] such as all the roots of the
following polynomial are situated inside the unit complex disc:
h(s) = s n −1 + g n −1 s n − 2 + ... + g 1
The control’s objective is to lead the states toward the sliding surface S and maintain
them at each sampling time.
Like in continuous time sliding mode control, the control law in the discrete time
sliding mode control has two terms [3, 6, 8, 14]:
- equivalent control term, u e , is designed to maintain the states on the de-
sired sliding surface once the states reach it,
- switching control term, u s , is designed to compel the states to move to-
wards the desired sliding surface
u = ue + us (75)

We would like to verify the sufficient conditions proposed by \Sarpturk, et al., [11]
1987\ , \Furuta 1990 \ and \Sira-Ramirez 1991\
We recall the expression of the discrete time reaching law is expressed as:

(S ( k + 1) ) = ((1 − qT )S ( k ) − ε T S ( k ) sgn( S ( k )) ) (86)

Let's verify the condition proposed by Sarpturk which is described as:

S ( k + 1) p S ( k )

S ( k + 1) = (1 − qT )S ( k ) − ε T S ( k ) sgn( S ( k )) (97)
 Paper title 7

S ( k + 1) = (1 − qT )S ( k ) − ε T S ( k ) (108)

S ( k + 1 ) = (1 − ( q + ε ) T )S ( k ) (119)

S ( k + 1) p S ( k ) if and only if (1 − ( q + ε )T ) p 1

So,
0 p ( q + ε )T p 2 (20)

Let's verify the condition proposed by by Furuta which is described as:

(S ( k + 1) )2 p (S ( k ) )2 .

(S ( k + 1) ) =
2
((1 − qT )S ( k ) − ε T S ( k ) sgn( S ( k )) )
2
(212)

(S ( k + 1) ) =
2
((1 − qT )2 S ( k ) 2 − 2 (1 − qT )ε TS 2
( k ) + (ε TS ( k ) )
2
) (22)

(S ( k + 1) ) =
2
((1 + (( q + ε ) T ) 2
)
− 2 (ε + q ) T S ( k ) 2 ) (23)

(S ( k + 1 ) ) = (1 − ( q + ε ) T
2
)2 S ( k ) 2 (24)

(S ( k + 1 ) ) p S ( k ) 2 if and only if (1 − ( q + ε ) T )2 p1
2

So,
0 p ( q + ε )T p 2 (25)

Let's verify the condition proposed by \Sira-Ramirez 1991\ which is described as:
S (k) S (k + 1) p S 2 (k)

S ( k + 1) S ( k ) = (1 − qT )S 2 ( k ) − ε TS ( k ) S ( k ) sgn( S ( k )) (26)
8 STA′2006 – Topics …, pages 1 à X

S (k +1)S (k ) = (1 − qT )S 2 (k ) − ε T S (k ) = (1 − qT − ε T )S 2 (k )) ≤ (1 − qT − ε T ) S 2 (k ) (27)
2

(1 − qT − ε T ) p 1

0 p ( q + ε )T p 2 (28)

So, S ( k + 1) S ( k ) p S 2 ( k )
On conclusion the proposed reaching law satisfies all sufficient conditions proposed
by \Sarpturk, et al., [11] 1987\ , \Furuta 1990 \ and \Sira-Ramirez 1991\ if and only
if the following inequality is verified: 0 p ( q + ε ) T p 2
a) The equivalent control law
In this stage we consider the certain dynamics systems, this means that:
∆Ai= 0nxn and ∆Bi=[0, . . . ,0]T .
The equivalent control term, u e , is designed to keep the system on the sliding sur-
face in the sliding stage. The ideal quasi sliding mode should satisfy the following
equations:
⎧ S (k ) = 0 (29)
⎨
⎩ S ( k + 1) = 0

We define the discrete-time equivalent control term by solving S ( k + 1) = 0 so,

r
(30)
GT ∑ hi (z(k )) ( A i (k + 1)X (k ) + B iu e (k ) ) = 0, k = 0, 1,...
i =1

The equivalent control term is given by the following expression:

( ) (313)
r −1 ⎡ r ⎤
u e (k ) = − ∑ hi (z (k ))G T B i
⎢
⎢ G T ∑ hi (z (k ))Ai (k + 1)X (k ) ⎥⎥
i =1 ⎢ ⎥
⎣ i =1 ⎦

We assume that the following condition is hold:

( r
∑ h i ( z ( k ))G T B i ≠ 0
i =1
) (32)

b) The switching control law:

From the reaching law we can write:

 Paper title 9

S ( k + 1) = S ( k ) − q T e S ( k ) − ε T e S ( k ) sg n (( S ( k )) (33)

We know that: S ( k ) = G T
X ( k ) and S ( k + 1) = G T X ( k + 1) So,

( ) (34)
r
S(k + 1) - S(k) = ∑ hi ( z ( k )) G T Ai X (k ) + G T Bi u (k ) −G T X (k )
i =1

If we compare the equations (23) and (24) we deduct the global control law:

−1
⎛r ⎞ ⎛r ⎞ (35)
( )
u(k) = −⎜ ∑hi (z (k ))GT Bi ⎟ ⎜ ∑hi (z (k )) GT Ai X (k ) − (1−qT )S (k ) + εT S (k ) sgn(S (k )) ⎟
⎝ i =1 ⎠ ⎝ i =1 ⎠

From the equations (21) and (25) we obtain the switching control term:

−1
⎛ r ⎞ (36)
u s (k) = − ⎜ ∑ h i ( z ( k ))G T B i ⎟ ( −(1 − qT )S (k ) + εT S ( k ) sgn(S ( k )) )
⎝ i =1 ⎠

3.2. Robust fuzzy sliding mode Control law

Consider the discrete nonlinear system (1) in the perturbed conditions. It can be de-
scribed by the T-S type fuzzy model (4). Where, ∆Adi and ∆Bdi represent the parame-
ters variations and the external disturbance respectively for each sub-model of the
system.
We assume that the uncertainty, which represent parameters variations and the exter-
nal disturbance, verify the matching conditions. So the matrices ∆Adi and ∆Bdi can be
expressed as:
( (
∆ A di = B di A i and ∆ B di = B di B i (37)
( (
where, Ai is a row vector and Bi is a scalar . They should be written as:

(
[
Ai = ∆ a i1 ∆ ai2 ∆ a i3 ... ∆ a in ] (
; B i = ∆ bi (38)

Then the system of equations (9) becomes:

10 STA′2006 – Topics …, pages 1 à X

(39)
⎧
( )
c ( (
⎪ X (k + 1) = ∑ h i ( z ( k )) A i X ( k ) + B i (u( k )+ A i X ( k ) + B i )
⎪ i =1
⎨ c
⎪ y (k) = ∑ h ( z ( k ))C X ( k ) ;
⎪⎩ i =1
i i

However, the equivalent control term for uncertain model becomes:

−1
⎛ r ⎞ ⎛ r ( ( ⎞ (40)
⎝ i =1 ⎠ ⎝ i =1
(
ue (k) = − ⎜ ∑ hi (z (k ))G T Bi ⎟ ⎜ ∑ hi (z (k )) G T Ai X (k ) + G T Bi (Ai X (k ) + Bi ⎟
⎠
)
So, the global control law (36) becomes:

−1
⎛r ⎞ ⎛r ( ( (414)
( )
u (k) =−⎜ ∑hi (z (k ))GT Bi ⎟ ⎜ ∑hi (z (k )) GT Ai X (k ) +GT Bi (Ai X (k ) + Bi −(1−qTe )S (k ) +εTe S (k ) sgn(S (k ))
⎝ i =1 ⎠ ⎝ i =1

( (
In general case A i and B i are unknown, but their upper bound are known, so the
global control law can not be implemented. However, to overcome this difficulty we
( (
replace respectively the unknown terms A i and B i by the following expressions:

⎧ T (42)
⎛ (T (
(
⎪ A i = ⎜ eig ( A iup A iup ) ⎞
⎟
⎨ ⎝ ⎠
⎪B = ∆b
⎩ i i max

Where,
(
[
Aiup = − ∆ a i1 max − ∆ a i 2 max − ∆ a i 3 max ... − ∆ a in max ] (
; Biup = ∆ bi max
We define a new set of perturbations and control parameters as follow:
⎧⎪ S ig = G T B i A i X(k ) (43)
⎨
⎪⎩ F i g = G T B i B i

The choice of Sig and Fig is done to ensure that the sign of the incremental S (k) is
opposite to the sign of S (k).
The global control law will be expressed as:
−1
⎛r ⎞ ⎛r ⎞
⎝ i =1 ⎠ ⎝ i =1
( )
u (k ) =−⎜ ∑hi (z (k ))GT Bi ⎟ ⎜ ∑hi (z (k )) GT Ai X (k ) +Sig + Fig + (Sig + Fig )sgn(S (k ) − (1−qTe )S (k ) +εTe S (k ) sgn(S (k ))⎟
⎠
(44)

3.3. Tracking robust fuzzy sliding mode Control law

 Paper title 11

If we are interested to the tracking problem the sliding surface, which is expressed in
the last section in continuous time by: S (t ) = G T X (t ) , will be expressed in the error
space as follow Slotine [17] :
(n −1)
⎛ d ⎞ ~ ~
S ( x;t ) = ⎜ + λ ⎟ X = a0 ~
x (n -1) + a 1 ~
x (n - 2) + ... + a n -1 ~
x = G X (t )
⎝ dt ⎠
Where, λ is a positive integer, G is a row vector defined as: G = [ g n -1 , ..., g 1 , 1] ,
~
X (t ) = X − X , X d = [ X 1 d . . . X nd ]T represents the desired vector.
d

The tracking problem will be transformed into stability problem. Indeed, S ( x;t ) = 0
~
represents an equation whose unique stationary solution is X = 0 . The tracking prob-
lem of the desired vector Xd comes back to locate to zero the sliding surface for all
time t f 0 .
Like the continuous version, the discrete form of the sliding surface will be expressed
as :
~ (45)
S ( x;k ) = G X ( k )

where, X~ ( k ) = X ( k ) − X d ( k )
Thus, the dynamics of the sliding mode is defined by:
~
n −1
d n −1 − k X (46)
∑
k =0
gk
dt n −1 − k
= 0

Consequently, the system in sliding mode is stable if all roots of the following poly-
nomial h(s) are situated inside the unit complex disc:
n −1
h(s) = ∑g
k =0
k s n −1− k = g 0 s n −1 + g n −1 s n − 2 + ... + g n −1 = 0

Where, s is the Laplace operator.

The control law will be expressed as:
−1
⎛r ⎞ ⎛r ⎞ (47)
⎝ i =1 ⎠ ⎝ i =1
( )
u (k) =−⎜∑hi (z (k ))GT Bi ⎟ ⎜∑hi (z (k )) GT Ai X% (k )+Sig +Fig +(Sig +Fig )sgn(S(k ) −(1−qTe )S(k )+εTe S(k )sgn(S(k ))
⎠

4. Illustration:

To illustrate the performance of the presented approach, we choose the Mass spring
damper as example which is widely used in the control literature of nonlinear system.
It is described in continuous version by the following equation [19]:

&& + h1 ( x(t), x(t)

Mx(t) & ) + h2 ( x(t) ) = φ ( x(t)
& ) u(t ) (48)
12 STA′2006 – Topics …, pages 1 à X

where M is the mass and u(t) is the force; h1(x(t), x&(t)) is the nonlinear or uncertain
term with respect to the damper, h2(x(t)) is the nonlinear or uncertain term with re-
spect to the spring, and ϕ (x(t))
& is the nonlinear term with respect to the input.
We have :
h 1 (x (t), x&(t)) = c 1 x (t) + c 2 x&(t) (49)
2
h 2 (x (t)) = c 3 x (t) + c 4 x (t)
φ ( x&(t) ) = 1 + c 5 x& 3 ( t )

⎧ x&1 (t) = x 2 (t) (50)

⎪
⎨ x& 2 (t) = f(x1 , x 2 ) + g(x1 , x 2 )u + d (t)
⎪ y(t) = x (t )
⎩ 1

where
1 (515)
f(x1 , x2 ) = − (h (x(t), x(t))
& + h2 (x(t)))
M 1
1
g(x1 , x 2 ) = φ(x(t))
&
M
and d(t) denotes a disturbance on the state of the system.
The domain of state variables variation is [-1.5; 1.5]. The parameters of system are:
M=1.0, c1=0, c2=1, c3=0.01, c4=0.1 and c5= 0.13.
The system becomes:
x ( t ) = − x& ( t) − 0 . 0 1 x ( t ) − 0 . 1 x ( t ) 3
&& + ( 1 + 0 . 1 3 x& 3 ( t) ) u ( t) (52)

⎡ x1(t) ⎤
The premise variables are x(t) = z(t) = ⎢ ⎥
⎣ x 2(t)⎦
The membership functions are:

x12 ( t) (53)
µ 1 1 ( x1 ( t ) ) = µ 2 1 ( x1 (t) ) = 1 −
2 .2 5
x12 (t)
µ 3 1 ( x1 ( t) ) = µ 4 1 ( x1 ( t) ) =
2 .2 5
x 23 ( t)
µ 1 2 ( x 2 ( t) ) = µ 2 2 ( x 2 ( t ) ) = 0 . 5 +
6 .7 5
x 23 (t)
µ 3 2 ( x 2 ( t )) = µ 4 2 ( x 2 ( t) ) = 0 . 5 -
6 .7 5

The discrete T-S Fuzzy model of system described in continuous by equation (52) is
defined by the following four fuzzy rules:
Plant rules
 Paper title 13

Rule 1: IF x1 is µ11 and x2 is µ12 THEN X (k + 1) = A1 X (k ) + B1u

Rule 2: IF x1 is µ21 and x2 is µ22 THEN X (k + 1) = A2 X (k ) + B2u
Rule 3: IF x1 is µ31 and x2 is µ32 THEN X (k + 1) = A3 X (k ) + B3u
Rule 4: IF x1 is µ41 and x2 is µ42 THEN X (k + 1) = A4 X (k ) + B4u
For nominal values of M, c2, c3 and c4, matrices Ai and Bi are given by:

⎡ 1 0 . 01 ⎤ ⎡ 0 ⎤
Ad 1 = Ad 2 =
⎢⎣ − 0 .0001 0 . 99 ⎥⎦ ;B d 1 = B d 3 = ⎢⎣ 0 . 0143 ⎥⎦ ;

⎡ 1 0 . 01 ⎤ ⎡ 0 ⎤
Ad 3 = Ad 4 =
⎢⎣ − 0 .0023 0 . 99 ⎥⎦ ;B d 2 = B d 4 = ⎢⎣ 0 .0056 ⎥⎦

⎡ 0 0 ⎤
∆ A 1up = ∆ A = ⎢ ∆ B 1up = ∆ B 3up = 0 . 043 ;
0.003 ⎥⎦
2up
⎣ 0.0000

⎡ 0 0 ⎤
∆ A 1up = ∆ A = ⎢ ∆B = ∆B = 0 . 0017 ;
0.003 ⎥⎦
2up 2up 4up
⎣ 0.0007

We present in the figures 1, 2, 3, 4, 5 and 6 the simulation results of the behavior of

the state variables x1(k), x2(k), the position and velocity error ε1(k) and ε2(k), the
sliding surface and the control law respectively of the inverted pendulum with pa-
rameters vary of an uncertain way in time.
x (0) = [- π 60 ; 0 ] , G =[15 1], Te=0.01; q=70, ε =0.015;
T

Fig. 1. Evolution of x1and Fig.2. Evolution of x2 and de Fig. 3. Evolution of the

x1d. x2d. position error
14 STA′2006 – Topics …, pages 1 à X

Fig. 4. Evolution of the Fig. 5. Evolution the sliding Fig. 6 Evolution of the con-
speed error surface trol law

Commentary
We notice that the state variables of the system follow appropriately the reference
signal and that the tracking error converges toward zero. Although the uncertainties
reach 30% of the nominal parameters of the system the state variables track the refer-
ence one and the errors converge toward zero.

5.Conclusion

In this paper we present a robust fuzzy sliding mode controller for discrete nonlinear
systems. First, we recall the discrete Takagi-Sugeno type fuzzy model, then the
principe of the sliding mode control in discrete time. The uncertainty are assumed to
be verifie the matching conditions. We develop a robust controller based on the
sliding mode and the dynamic T-S fuzzy state model. The uncertainties are replaced
bye the bigger eigen-value of the upper bound matrices of uncertainties. The
expressions in discrete time of both equivalent control term and hitting term are
developed. The tracking control law is developed. Simulation results for mass spring
damper with parameters variation show the performance of the proposed control law.
Bibliographie
[1] Michio Sugeno “On Stability of fuzzy Systems expressed by rules with singleton
consequents”. IEEE Trans on Fuzzy Syst,Vol 7 N°2 Feb 1999.
[2] T.Takagi and M.Sugeno. "Fuzzy identification of systems and its applications to
modeling and control", IEEE trans syst. Man,Cybern. Vol15. 116-132 Jan/Feb1985.
[3]W.Chang, J.Bae Park, Y.Hoan Joob and G.chen "Design of robust fuzzy model-
based controller with sliding mode control for SISO non-linear systems" Fuzzy Sets
and Systems 125 (2002) pp 1-22.
[4] Hafedh Abid Mohamed Chtourou et Ahmed Toumi, ”A Sliding Mode Based Ro-
bust Fuzzy Controller For a Class of Uncertain Systems” SSD05, Mrach 2005
Sousse Tunisia.
[6] Y. dote and R.G..hoft, "Microprocessor based sliding mode controller for dc
motor drives" presented at the Industrial Application. SOC. Annu. Meeting, Cincin-
nati, OH, 1980.
 Paper title 15

[7] D. Milosavljevic, "General conditions for the existence of a quasi-sliding mode on

the switching hyperplane in Discrete Variable Systems," Automat. Remote Contr.,
vol. 46, pp(307-314) 1985.
[8]. S. Z. Sarpturk, Y Istefanopulos, and O. Kaynak, "On the stability of Discrete-
time sliding mode control systems," IEEE Trans. Automat. Contr., vol. 32, N° 10,
pp(930-932) 1987.
[9] K. Furuta, "Sliding Mode Control of a discrete system”, Systems and Control
Letter, Volume 14, pp145-152 1990.
[11] Weibing Gao, Yufu Wang and Abdollah Homaifa, "Discrete-Time Variable
Structure Control Systems" IEEE Transactions on Industrial, Electronics, volume 42,
N°2, April 1995, pp(117-122).
[12] Weibing Gao., and Hung, J.C. , “Variable structure control of nonlinear sys-
tems: a new approach”, IEEE Transactions on Industrial Electronics, 40, 45, 1993
[13] H. Lee, E. Kim, H.J. Kang et M. Park, “A new sliding-mode control with fuzzy
boundary Layer”, Fuzzy Sets and Systems 120 (2001), pp. 135-143
[14] Utkin, V. I, “Sliding Modes and their Application in Variable Structure Sys-
tems” (Moscow: Nauka) 1974 (in Russian, and also, 1978, Mir, in English).
[15]. Yu, X. H., Man, Z. H. and Wu, B. L “Design of fuzzy sliding-mode control
systems”, Fuzzy Sets and Systems, 95, pp.295-306. ., (1998),“
[16]. Ting, C. S., Li, T. H. S, and Kung, F. C., ,“An approach to systematic design of
the fuzzy control system”, Fuzzy Sets and Systems, 77, pp. 151-166. (1996)
[17] Slotine, J. J. E. and Li. W., (1991), “Applied nonlinear control”, Prentice Hall,
Englewood Cliffs, NJ.
[18] Utkin, V. I, “Variable structure systems with sliding mode”, IEEE Transactions
on Automatic Control, Vol. AC-22, No. 2, pp 212-222, April., (1977)
[19]. Kazuo Tanaka, Takayuki Ikeda and Hua O.Wang. "Robust stabilization of a
class of uncertain nonlinear systems via fuzzy control: Quadratic stability, H∞ Con-
trol Theory, and linear Matrix Inequalities". IEEE Trans on Fuzzy System, Vol 4 N°1
Feb 1996.
[20] Mehrdad Hojati and Saeed Gazor “Hybrid Adaptive Fuzzy Identification and
Control nonlinear Systems” IEEE Transactions ON Fuzzy Systems vol 10, N°2. April
2002 pp 198-210.
[21]. H. Sira-Ramirez. ”structure at infinity, zero dynamics and normal forms of sys-
tems undergoing sliding motion”. Int. J. systems Sci., 21(4):665–674, 1988