International Journal of Electrical and Computer Engineering 5:6 2010

Adaptive Sliding Mode Observer for a Class of

Systems
D.Elleuch, T.Damak

technique applied to synthesize the observer.

A systematic approach to synthesize the adaptive observer
is developed by [3] for Lipchitz nonlinear systems .Following
that, an adaptive sliding mode observer for a system as in [3]
will be studied. Consider the nonlinear systems.

AbstractIn this paper, the performance of two adaptive

observers applied to interconnected systems is studied. The
nonlinearity of systems can be written in a fractional form. The first
adaptive observer is an adaptive sliding mode observer for a Lipchitz
nonlinear system and the second one is an adaptive sliding mode
observer having a filtered error as a sliding surface. After comparing
their performances throughout the inverted pendulum mounted on a
car system, it was shown that the second one is more robust to
estimate the state.

x& = Ax + ( x, u ) + T f ( x, u )

y = Cx

KeywordsAdaptive observer, Lipchitz system, Interconnected

fractional nonlinear system, sliding mode.

With: x

; n ; f = diag ( f1 , f 2 L , f n ); A is

a constant matrix; parameter vector; y

n

I. INTRODUCTION

The nonlinear functions

such as:

IFFERENT forms of nonlinear systems with varying

parameters are presented in literature. For this reason
many adaptation laws are developed for each system to
estimate the unknown parameters. To have appropriate
adaptation laws, some techniques are used such as an adaptive
controller ([6], [7], [16]) and an adaptive observer ([2], [5],
[8], [9], [17]), where an augmentation of the vectors of state
including the state and the unknown parameters is used [10].
An observer is called adaptive observer if it uses an adaptation
law to estimate the unknown parameters [10].
In literature ([1], [2], [3], [4]), most of the adaptive
observers are developed for a class of lipchitz nonlinear
systems throughout a classical observer. But the adaptation
laws proposed for a class of systems are not usually robust for
all systems of the same class.
In this work, we have shown the modest performance of an
adaptation law proposed for a lipchitz nonlinear system which
can be written in a fractional form. Then, an adaptive sliding
mode observer designed to a fractional nonlinear system is
applied. These adaptation laws have been compared through
the inverted pendulum mounting on a cart system in the case
of constant parameters.

and

f are two lipchitz matrices

f ( x, u ) f ( x , u ) 1 x x

( x, u ) ( x, u ) 2 x x
1 , 2 n .
Considering the following lyapunov function

1 T
1 ~ ~
e e + T
2
2
~
Where = and e = x x

V =

(1)

This lyapunov function satisfies the condition of stability,

the derivative of V is negative, if:
The architecture of the adaptive sliding mode
observer is:

x& = Ax + ( x, u ) + T f ( x , u ) + L( y Cx )

sign( y Cx )
& 1
= f T ( x , u )( y Cx )

II. ADAPTIVE SLIDING MODE OBSERVER FOR A CLASS OF

LIPCHTIZ SYSTEM

(2)

This adaptive observer is stable and converges to

the desired state if:
(3)
k >
In [14], it is shown that the adaptive sliding mode observer
can give a good performance for a Lipchitz nonlinear system
when the parameters are constant or varied.

In literature, the adaptive observers are more studied in the

case of classical approaches. It is proved that the adaptation
law depends on the class of nonlinear systems and on the
D.Elleuch is with Industrial Processes Control Unit Sfax University, ENIS,
BP W, 3038, Sfax Tunisia e-mail :( dorsaf _ell@hotmail.fr).
T.Damak is with Industrial Processes Control Unit Sfax University, ENIS,
BP W, 3038, Sfax Tunisia e-mail: (tarak.damak@enis.rnu.tn)}.

368

International Journal of Electrical and Computer Engineering 5:6 2010

III. ADAPTIVE SLIDING MODE OBSERVER HAVING A FILTERED

ERROR AS A SLIDING SURFACE
Some nonlinear systems are written in the general structure
(4) like the robot manipulator ([11], [12]) and the inverted
pendulum mounting on a cart [13].

( x)

V is positive for an error e

sj

V is radically unbounded with respect e i.e

sj
V when esj
V& 0 e sj 0

x& = F ( x, x& , u )

(4)
This form can be transformed into the fractional form .If the
parameter vector of the system is unknown, the joint
estimation of the parameter and that of the state becomes more
complicated because the parameter vector is designed in the
numerator and in the denominator. The form of system (4) is
written as a set of interconnected fractional systems having the
following form:

x&1 j = x 2 j

f j ( x, u )

S j x& 2 j =
( x, u )

y = x1 j

Where

The function

x&1 j = x 2 j s1 j

T W fj ( x , u )
s2 j
S j x& 2 j = T
W ( x , u )

sup(W fj ( x1 j , esj + l j )WT ( x )

&
1

= esj
T

W ( x1 j , l j ) sup(W fj ( x) +

j = 1L m

For each subsystem the output is y = x1 j and the output

and the state

vector is x = x11 x 21 x12 x 22 L x 2 m .

The nonlinearity f j ( x, u ) and

j are

and

( x, u ) = 0 ( x ) + T W ( x, u )
With x ,
n

i =1

p , f j ( x, u ) and ( x, u ) C 1 ,

e sj = e&1 j + e1 j : is the filtered error.

The error vector of each subsystem is

Assumption:

With

e j = x1 j x1 j

W fj ( x, u ) > W ( x, u )

esj

V = ( x1 j , + l j , x 2 j )d

W fj ( x, u ) and W ( x, u ) have the same sign

(8)

Where

(x) will be ( x 2 j ) ,

( x1 j , + l j , x 2 j ) = ( x1 j , + l j ) ( x 2 j ) and

B.
Definition: Weighting Control Lyapunov Function
(WCLF)
For a state vector bounded. We define the following
function:

l j = x 2 j [ 0]e j

esj

The state vector of each subsystem becomes:

(6)

x j = ( x1 j , esj + l j )

Where

x 2 j x 2 j

To design the observer, the expression (6) becomes:

W fj ( x, u ) <

V = ( x, + l j )d

nxn

sij = k ij esj + ij sign(e sj ) , i = 1,2 ; j = 1, LL, m

following assumption.

i =1

sij is the sliding surface which is written as

The function W fj ( x, u ) and W ( x, u ) must satisfy the

ij and > 1

ij

With

W fj ( x, u ) and W ( x, u ) p .

A.

(7)

The stability of the adaptive observer is guaranteed for all

gains verifying this relation:

expressed such as

f j ( x, u ) = f 0 j ( x) + T W fj ( x, u )

( x) is a weighting function.

Theorem:
If the system (5) satisfies the assumption and if the function
WCLF (6) verifies the condition as [6], the adaptive sliding
mode observer for the form of the system (5) is defined as
follows:

(5)

of all the system is y = x11 x12 x 22 L x1m

is a smooth function

The lyapunov function (8) satisfies the same condition as

(6).
These two architectures of adaptive observer are applied to
an inverted pendulum mounted on a cart to test their

( x , + l j ) = ( x ) ( x, + l j )

The function V is a Weighting Control Lyapunov Function

(WCLF) if:

369

International Journal of Electrical and Computer Engineering 5:6 2010

x&1 = x 2 + L1 ( x1 x1 ) 1 sign( x1 x1 )

2
x& 2 = 1 Fx x 2 + 2 ( x& 4 cos( x 3 ) x 4 sin( x 3 )) +

L2 ( x 2 x 2 ) 2 sign( x 2 x 2 )
&
x 3 = x 4 + L3 ( x3 x 3 ) 3 sign( x3 x 3 )
&
1
&
x 4 = j 3 ( g sin( x 3 ) x 2 cos( x 3 )) + F x 4 +
L4 ( xi x i ) 4 sign( xi x i ); i = 1,2 or 3

performance.

IV. EXAMPLE
The studied adaptive observers are applied to the inverted
pendulum mounting on a cart [15] having a Lipchitz
nonlinearity which can be transformed into a fractional form.
The equations of motion of the systems are:

( M + m) &x& + Fx x& + ml (&& cos( ) & 2 sin( )) = u

&&
j + F & m lg sin( ) + ml&x& cos( ) = 0
With x is the linear displacement and is the angular
displacement; m the pendulum mass; M the car mass; l the
link length and Fx and F are the forces.

0 1 0
1 = M + m
0 0 0

ml

Where = 2 =
, A = 0 0 0

m+M

=
ml

0 0 0
3

( x, u ) = 0 ,
0
0

x& 4 cos( x3 )

Fx x 2 2
0
x 4 sin( x3 )

f ( x, u ) = 0

0
0

g sin( x3 )
0
0
x& cos( x )

3
2

(10)

The adaptation law:

&
1
1 = Fx x 2 e2
1

&
1 &
( x 4 cos( x 3 ) x 42 sin( x 3 ))e2
2 =

&
1 1
j ( g sin( x 3 ) x& 2 cos( x 3 ))e4
3 =
3

In the state space, the system is written as:

x&1 = x 2

x& 2 = 1 Fx x 2 + 2 ( x& 4 cos( x3 ) x 42 sin( x3 ))

x& 3 = x 4
&
1
x 4 = j [ 3 ( g sin( x3 ) x& 2 cos( x3 )) + F x 4 ]

1 0 0 0

y = 0 1 0 0

0 0 1 0

(11)

The nominal parameters used in simulations are:

M = 3.2 Kg ; m = 0.535 Kg ; j = 0.062 Kgm 2 ;

(9)

l = 0.365m; Fx = 6.2 Kg / s; F = 0.009 Kg / m 2 and

g = 9.807m / s 2
And the simulation conditions are:

L1 = 40; L2 = 25; L3 = 20; L4 = 35; 1 = 2 = 3 = 10 3 ,

= 8; (0) = (0) = (0) = (0) = 0; x (0) = 1;
4

x 2 (0) = 3; x 3 (0) = 10 3 ; x 4 (0) = 1; x1 (0) = 1;

0
0
1 ,
F

x 2 (0) = 3; x3 (0) = 10 3 ; x 4 (0) = 1;

1

= 2 *10 6 ;

= 7 * 10 6 ;

= 2 * 10 6 ;

In figure (1), it is shown that the adaptive sliding mode

observer designed for a Lipchtiz nonlinear system has a
modest performance to estimate the angular speed and the
parameters.

Applying the adaptive observer (2) to this system, the

following expressions are obtained:
The sliding mode observer:

370

International Journal of Electrical and Computer Engineering 5:6 2010

Estimation of the state x4

Estimate the 3rd parameter

120

0.3

100

0.25

desired state
estimated state

80

0.2

0.15
60

0.1
desired parameter
estimated parameter

40

0.05
20

0
0

50

100

150

-0.05

time(s)

50

100

150

time(s)

(a) Estimation of the state

x4

(d) Estimation the parameter

Estimate the 1st parameter

Fig.1: Estimation of the state and the parameters by an adaptive

sliding mode observer

0.35

0.3

To apply the second adaptive observer the system should be

put on fractional form.
After development and simplification, the equation of the
system (9) becomes:

0.25

0.2

0.15

0.1

x&11 = x 21

[ Fx x 21 + j 1 (ml ) 2 g sin( x12 ) cos( x12 )

s1
1
2
x& = (ml )( j F x 22 cos( x12 ) x 22 sin( x12 )]
21
(m + M ) j 1 (ml ) 2 ((cos( x12 )) 2 )

desired parameter
estimated parameter

0.05

50

100

150

time(s)

(b) Estimation of the parameter

(12)

x&12 = x 22

( M + m) F x 22

2
2

(ml ) x 22 sin( x12 ) cos( x12 ) +

s2
j 1
(ml ) Fx x 21 cos( x12 ) + (m +

M )(ml ) g sin( x12 ))

x& 21 =
1
2
2

(m + M ) j (ml ) ((cos( x12 )) )

Estimate the 2nd parameter

0.02

0
desired parameter
estimated parameter

-0.02

-0.04

-0.06

The output vector is y = x11

vector is

-0.08

x12 ] and the parameter

= [ (ml ) 2 (ml ) (m + M ) (m + M )(ml )]

2
With: 1 = (ml ) ; 2 = ( ml ) ; 3 = ( m + M )
and

-0.1

-0.12

(13)

50

100

150

time(s)

(c) Estimation the parameter

4 = (m + M )(ml )

Applying the architecture (9), it yields to the following

systems:
The sliding mode observer having a filtered error is:

x&11 = x 21 s11

[ Fx x 21 + j 1 (ml ) 2 g sin( x12 ) cos( x12 )

s1
2
( ml )( j 1 F x 22 cos( x12 ) x 22
sin( x12 ))]
&

=
x
s 21
21
1
2

( m + M ) j (ml ) ((cos( x12 )) 2 )

371

International Journal of Electrical and Computer Engineering 5:6 2010

j 2 (cos( x12 )) 2 ((e s 2 + l 2 ) 2 x 22

)1

2
+ ( j 2 F x (cos( x )) 2
&

j
F
x

x
21
2
22
12

1
1 = 1 e s 2 1

+ j (es 2 + l 2 ) )3

2
4 j g (cos( x12 )) 2 + 21000

s 22

2
2
1

& = 1e j Fx x 21 (cos( x12 )) 1 + j Fx x 21 3

(21)
2
2
s2

80

With:

2
2
s11 = k11 e s1 + 11sat(e s1 ) ; s 21 = k 21 e s1 + 21sat(e s1 )
( j F (cos( x12 )) (es 2 + l 2 )

1 2
; s12 = k12 e s2 + 12 sat(e s2 ) ; s 22 = k 22 e s2 + 22 sat(e s2 );
1

j x 22 )1 + j Fx x 212 +

e s1 = e&11 + e11 ; es 2 = e&12 + e12 ;

& = 1e ( j 2 F x (cos( x )) 2 +
3
3
22
s2
12

e11 = x11 x11 ;e12 = x12 x12 ; and x 22 = es 2 + l 2

1

j ( F ((es 2 + l 2 ) x 22 )) 3

The adaptation law is expressed to the subsystems s 2

1000

j
g
+

because it contains all the parameters. Since we have

1 2
j x 22

&4 = 41es 2 j 2 g (cos( x12 )) 2 1 + j 1 g3 + 1600

1
j F x x 21
(22)
sup (W f 22 ( x12 , e s 2 + l 2 ) ) = 1
;

The
simulation
parameters
are:

j F x 22
1 (0) = 0.01;2 (0) = 0 ; 3 (0) = 21; 4 (0) = 0.6 ; k11 = 0.85 ;
j 1 g

k 21 = 1; k12 = 0.1; k 22 = 0 .1; 11 = 12 = 21 = 10 5

2
1
j (cos( x12 ))
22 = 30 ; 11 = 6 * 10 9 ; 21 = 10 5 ; 31 = 10 5 ;

41 = 10 6 ; = 5.x11 (0) = 2 ; x 21 (0) = 1; x12 (0) = 10 5 ;

W ( x) =
1

x 22 (0) = 0 ; x11 (0) = 2 ; x 21 (0) = 3; x12 (0) = 10 4 ; x 22 (0) = 1

According to the simulation results, the adaptive sliding mode

(20)

x&12 = x 22 s12

2
( M + m) F x 22 (ml ) 2 x 22
sin( x12 ) cos( x12 )

1
s2
j + (ml ) Fx x 21 cos( x12 ) +

( m + M )(ml ) g sin( x12 ))

x& =
1
2
2
21

(
)
(
)
((cos(
))
)
+

m
M
j
ml
x
12

observer having a filtered error as a sliding surface shows the

good performances in estimating the angular speed x 22
(Figure 2.a), whereas the parameters do not converge (Figure
2.b.c.d.e). This result is the same as in [7].

2
j 1 x 22

1
j F x x 21

; sup (W f 22 ( x 22 ) ) = 1

j F x 22
j 1 g

Estimate the angular speed x22 with an adaptive

sliding mode observer having a filtered error as a surface
40

The adaptation law is the following:

30
desired state
estimated state
20

10

-10

-20

(a) Estimation of the state

372

x22

10
time(s)

12

14

16

18

20

International Journal of Electrical and Computer Engineering 5:6 2010

Estimate the 1st parameter

Estimate the 4th parameter

1.4

-0.01

1.3
1.2

-0.02
desired parameter
estimated parameter

-0.03

1.1

desired parameter
estimated parameter

1
-0.04
0.9
-0.05
0.8
-0.06

0.7

-0.07
-0.08

0.6

(b) Estimation of the parameter

10
time(s)

12

14

16

18

0.5

20

10
time(s)

12

14

16

18

20

Fig. 2 Estimation of the state and the parameters by an adaptive

observer having a filtered error as sliding surface and constant
parameters

Estimate the 2nd parameter

desired parameter
estimated parameter

-0.1

(e) Estimation of the parameter

0.05

-0.05

By comparing this result to the first one, we have shown

that the adaptive sliding mode observer having a filtered error
is more robust than the adaptive sliding mode observer
designed to a lipchitz nonlinear systems to estimate the
angular speed of the inverted pendulum. But both are not
efficient to estimate all the parameters.

-0.15
-0.2
-0.25
-0.3
-0.35
-0.4

V. CONCLUSION
0

(c) Estimation the parameter

10
time(s)

12

14

16

18

20

In this work, the performances of the two adaptive sliding

mode observers are studied through the example of inverted
pendulum mounting on a cart. It was shown that the
performance of the adaptive sliding mode observer attributed
to a lipchitz nonlinear system is modest when the nonlinearity
is written in a fractional nonlinear form. An adaptive sliding
observer having a filtered error as a sliding surface is applied
to a fractional nonlinear system; it gives better performance to
estimate the states than the first one. But both architectures of
the adaptive observer are not efficient to estimate all the
parameters.

Estimate the 3rd parameter

8
7
6
desired parameter
estimated parameter

5
4
3

REFERENCES
2

[1]
1
0

(d) Estimation the parameter

10
time(s)

12

14

16

18

[2]

20

[3]

[4]

[5]

[6]

373

R. Rajamani Adaptive observers for active automotive suspensions:

theory and experiment IEEE transaction on control systems technology.
Vol 32, N1 ,mars 1995
F. Zhu The design of full order and reduced-order adaptive observers
for nonlinear systems IEEE international conference on control
automation, China, 30 May to june 1, 2007.
Y. M. Cho , R.Rajamani A systematic approach to adaptive synthesis
for nonlinear systems IEEE ,1995 ,pp 478-482.
C.Cheng ``Design of adaptive observation schemes for lipschtiz
nonlinear systems'' JSME international journal, series C, Vol 44, N 3,
2001.
H.Saadaoui, De leon, M.djemai, J.P Barbo High order sliding mode and
adaptive observers for a class of switched systems with unknown
parameter: A comparative studyProceeding of the 45th IEEE
conference on decision and control, san diego ,CA , USA. December 1315 2006.
S.S.Ge , C.C.Hang, T-Zhang A direct adaptive controller for dynamic
systems with a class of nonlinear parametrizations Automatica, vol
35,1999, pp 741-747.

International Journal of Electrical and Computer Engineering 5:6 2010

[7]

[8]
[9]

[10]
[11]

[12]

[13]

[14]

[15]

[16]

[17]

F.Y.Zeng , B.Dahhou Adaptive control of nonlinear fermentation

process via MRAC technique Appl.math.modelling, vol 17 February,
1993,PP 58-69.
R. Marino Adaptive observers for single output nonlinear systems
IEEE transaction on automatic control, vol 35, N9, September, 1990 v.
N.M.Iyer, A. E. Forell Design of stable adaptive nonlinear observer for
an exothermic stirred Tank reactor Computers Chem.Engng, Vol 20,
N9,1996, pp 1141-1147.
G. Besanon nonlinear observers and applicationsSpringer-verlag
Berlin Heidelberg, 2007.
Y-P Chen; J-L Chang Sliding mode force control of manipulators
Proceedings of the national science council, Republic of China ,part
A,physical science and engineering, vol 32, N2 ,1999, pp 281-288.
Paul C-P. Chao, Chien-Yu shen Design and implementation of a sliding
mode controller and high gain observer for output tracking of a three
axis pickup Sensors and actuators A 135, 2007, pp 713-730.
K.Hakiki, B. Mazari, ,A. Liazid ,S. Djaber Fault reconstruction using
sliding mode observers American journal of applied sciences, vol 3,
N1 ,2006, pp: 1669-1674.
D.Elleuch, T.Damak Adaptive sliding mode observer for Lipchitz
nonlinear systems with varying parameters 1 0th International
conference on Sciences and Techniques of Automatic control \&
computer engineering, Hammamet, Tunisia ,December 20-22,2009,
pp1821-1833.
K.hakiki, B.Mazariari, A.Liazid , S.Djaber Fault reconstruction using
sliding mode observer American journal of applied sciencez, vol 3,N1,
2006, pp: 1669-1674.
Ibrahim F. Jasim Stable Robust Adaptive Controller and Observer
Design for a Class of SISO Nonlinear Systems with Unknown Dead
Zone World Academy of Science, Engineering and Technology 59,
2009.
Grouz Faten, Sbita Lassad Speed Sensorless IFOC of PMSM Based
On Adaptive Luenberger Observer International Journal of Electrical
and Electronics Engineering 2:1 2009.

374