A Comparative Study of Sliding Mode Observers

Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169

Volume: 3 Issue: 5 2923 - 2929


_______________________________________________________________________________________________

A Comparative Study of Sliding Mode Observers

A.Charles1, M.Nandhini2, A.Sakthivel3


1
Assistant Professor, Karpagam College of Engineering, Coimbatore, India
2
Assistant Professor, Karpagam College of Engineering, Coimbatore-India
3
Associate Professor, Karpagam College of Engineering, Coimbatore-India

Abstract—In state space design, state feedback control is a powerful control technique. In state feedback control the Feedback of complete state
vector gives the designer to have total control over the closed loop poles. Many times in state feedback control, all states are not available to feed
back. To measure all unmeasurable states of the system the observer is needed. In real time system the system uncertainty will occur. So
observer should estimate the state of the system in the presence of uncertainty. It is needed to control the system for which observer should be
very robust to estimate the states of the system correctly. So here Utkin observer and Walcott zak observer`s are taken here. Their performances
are compared to find the robustness against the system uncertainty. Utkin sliding mode observer has only the switching gain to minimize the
observer error. But the Walcott-zak observer has static and nonlinear observer gain to minimize the observer error.

Index Terms—Utkin observer, state feedback control, Walcott zak observer.


__________________________________________________*****_________________________________________________

cantilever beam model is taken and the evaluation of the


I. INTRODUCTION
algorithms for smart cantilever beam is present.
In state feedback control the Feedback of complete state vector
gives the designer to have total control over the closed loop II. UTKIN OBSERVER
poles. Many times in state feedback control, all states are not Consider a continuous time linear system described by
available to feed back. To measure all unmeasurable states of x (t )  Ax(t )  Bu(t ) (1)
the system the observer is needed. A key feature in the Utkin y (t )  Cx (t ) (2)
observer [5] is the introduction of a switching function in the
nn nm
observer to achieve a sliding mode and also stable error Where A  R ,BR and p  m .Assume that the
dynamics. This sliding mode characteristic which is a matrices B and C are of full rank and pair ( A, C ) is
consequence of the switching function is claimed to result in observable.
system performance which includes insensitivity to parameter As the outputs are to be considered, it is logical to effect a
variations, and complete rejection of disturbances. Similar to change of coordinates so that the outputs appear as components
the Utkin observer, the sliding mode functional observer also of the states. One possibility is to consider the transformation
utilizes a switching function in its design and invariably will
inherit the benefits of insensitivity to noise rejection as
x  Tc x
reported for the Utkin observer. The Utkin Sliding Mode N c 
Observer does not have a static observer gain in its structure Where, Tc  0  (3)
and instead, the switching gain plays the role of stabilizing the  
n( n  p )
error dynamics. However, the disadvantage of this sliding Where the columns of N c  R span the null space of C.
observer structure reveals it- self when there exist uncertainties This transformation is nonsingular, and with respect to this
and disturbances. In this case, the observer can only estimate new coordinate system, the new output distribution matrix is
the states with a bounded error and not asymptotically. The
Walcott zak sliding mode Observer [4] has static and CTc
1

 0I p  (4)
nonlinear observer gain it reduces error due to system Where
uncertainty the Walcottzak observer algorithm explained in p is the number of output from the system.
section III. n is the order of the system.
If the other system matrices are written as
In section II discusses about Utkin observer. Section III
1  A11 A12 
discusses about walcottzak observer. In section IV the smart Tc ATc    (5)
 A21 A22 

2923
IJRITCC | May 2015, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 3 Issue: 5 2923 - 2929
_______________________________________________________________________________________________
 B1  e Ty e y   e y (17)
Tc B  B  (6)
 2
Then the nominal system can be written as is satisfied. Consequently, an ideal sliding motion will take
x1 (t )  A11 x1 (t )  A12 y (t )  B1u (t ) (7) place on the surface
S o  {(e1 , e y ) : e y  0} (18)
y (t )  A21 x1 (t )  A22 y (t )  B2 u (t ) (8)

where It follows that after some finite time t s , for all subsequent
 x1  time, e y  0 and e y  0
Tc x    (9)
y 
The observer proposed by Utkin [1] has the form Equation (15) then reduces to
~
e~ (t )  A e~ (t )
1 11 1 (19)
x1 (t )  A11 x1 (t )  A12 y (t )  B1u (t )  Lv (10) Which, by choice of L , represents a stable system and
y (t )  A21 x1 (t )  A22 y (t )  B 2 u (t )  v (11) e~1  0 consequently, xˆ1  x1 as t   .
Where ( x1 , y1 ) represent the state estimates for
III. WALCOTT ZAK OBSERVER
( n p ) p
( x1 , y1 ), L  R is a constant feedback gain matrix and
SYNTHESIS OF A DISCONTINUOUS OBSERVER
the discontinuous vector v is defined component wise by Consider the dynamical system
vi  M sgn( yˆ i  yi ) x (t )  Ax(t )  Bu(t )  B (t , x, u ) (20)
nn nm pn
Where M  R If the errors between the estimates and the Where A  R , BR ,C  R and P  m in addition the
true states are written as a e1  xˆ1  x1 and e y  yˆ  y then matrixes B and Care assumed to be of full rank. The function

from equations (7)to (11) the following error system is f : R  R n  R m  R n is unknown and represents the system
obtained uncertainty. A natural problem to consider initially is the
e1 (t )  A11e1 (t )  A12 e y (t )  Lv (12) special case when the uncertainty is matched: suppose
f (t , x, u )  B (t , x, u ) (21)
e y (t )  A21e1 (t )  A22 e y (t )  v ` (13)
Where the function  : R  R n  R m  R m is unknown, but
Since the pair ( A, C ) is observable the pair ( A11 , A21 ) is also bounded, so that
observable. As a consequence , L can be chosen to make the  (t, x, u)  r1 u   (t, y) (22)
spectrum of A11  LA21 lie in C  . Define a further change of
Where r1 is a known scalar and  : R  R p  R is a known
coordinates
function.
~  I n p L  Suppose that there exists a linear change of coordinates so
T   (14)
0 I p  that the system can be written as
x1 (t )  A11 x1 (t )  A12 y (t )  B1u (t )
~  e  Le The error system with respect to the
and let e x1 (t )  A21 x1 (t )  A22 y (t )  B2 u (t )  D2 (23)
1 1 y.
n p
new coordinate can be written as Where x1  R , yR p
and the matrix A11 has stable
~ ~
e~1 (t )  A11e~1 (t )  A12 e y (t ) (15) eigenvalues. Consider an observer of the form
~
e (t )  A e~ (t )  A e (t )  v (16) xˆ (t )  A xˆ (t )  A yˆ (t )  B u (t )  A e (t )
1 11 1 12 1 12 y (24)
y 21 1 22 y

Where yˆ 1 (t )  A21 xˆ1 (t )  A22 yˆ (t )  B 2 u (t )  ( A12  A22


s
)e y (t )  v (25)
~ ~ ~
A11  A11  LA21 , A12  A12  LA22  A11 L and s
~ Where A22 is a stable design matrix and ey  yˆ  y (26)
A12  A22  A21 L .
 P2 e y
It follows from (3.12) that in the domain   (t , y, u ) D2 if ey  0
v P2 e y (27)
  {(e1 , e y ) : A23 e1  0.5 max ( A22  ey  M 
T
A22 ) 
 0 else
Where M is some small positive scalar, the reachability
 (t , y, u )  r1 u   (t , y)   o
condition
e1 (t )  xˆ1 (t )  x1 (t )

2924
IJRITCC | May 2015, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 3 Issue: 5 2923 - 2929
_______________________________________________________________________________________________
e1 (t )  A11e1 (t ) (28) If represents the state estimate for and then the
robust observer can conventiently be written as
e y (t )  A21e1 (t )  s
A22 e y (t )  v  D2 (29)
xˆ(t )  Axˆ(t )  Bu(t )  G Ce(t )  G v
1 n (37)
There exists a family of symmetric positive definite matrices
P2 such that the uncertain dynamical error system above is
quadratically stable. x
Where the linear gain
Let Q1  R ( n  p )( n  p ) and Q2  R p p be symmetric positive
 A12 
definite design matrices and define P2  R p p to be the unique Gl  To1  s 
(38)
 A22  A22 
symmetric positive definite solution to the Lyapnouv equation.
T
s
P2 A22  A22
s
P2  Q 2 (30) and the nonlinear gain
Using the computed value of P2 define 0 
G n  D2 To1   (39)
Qˆ  A22
T
P2 Q21 P2T A22  Q1 (31) I p 
and
and notice that Qˆ  Qˆ T  0 .
 P2 Ce
Let P1  R( n  p )( n  p ) be unique symmetric positive definite   (t , y, u ) D2 if FCe  0
v P2 Ce
solution to the Lyapunov equation  0 else
AT P  P T A  Qˆ
11 1 1 11 (32)
(40)
Consider the quadratic form given by
Even in the special case when D  B the observer
V (e1 , e y )  e1T P1e1  e Ty P2 e y (33) formulation (37) to (40) is different from that of
As a candidate Lyapunov function. The derivative along the for the case when p  m since their results
system trajectory guarantee sliding will take place on the surface in the error
V  e1T Qˆ 1e1  e1T A21
T
P2 e y  e Ty A21
T
P2 e1
(34)
 
space given by e  R n : FCe  0 . In the above formulation
 e y Q2 e y  2e y P2 v  2e y P2 D2
T T T this is guaranteed.
It is easy to verify that Let ( A, D, C ) represent the linear part of the uncertain
system in (20) which represents the propagation of the
(e y  Q21 P2 A21e1 ) T Q2 (e y  Q21 P2 A21e1 ) T 
uncertainty through to the output. Consider the problem of
e Ty Q2 e y  e1T A21
T
P2 e y  e Ty A21
T
P2 e1 constructing an observer for the uncertain system of the form
(35)
 e1T A21
T
P2 Q21 P2 A21e1 z (t )  Az(t )  Bu(t )  G1Ce (t )  G n v (41)
where e  z  x , v is discontinuous about the hyperplane
Substituting the identity (35) into equation (34) and writing for  
S o  e  R n : Ce  0 and Gl , Gn  R ( n p ) are appropriate gain
notational convenience (e  Q 1 P A e )
y as
2 2 e~21 1 y matrices. The purpose of this section is determining the class
then of systems for which the observer (41) provides quadratic
stability of the error system despite the presence of bounded
V  e1T Qˆ T e1  e1T A21
T
P2Q21 P2 A21e1  e~yT Q2 e~y  2eTy P2 v  2eTy P2 D2
matched uncertainty. The canonical form from the section will
 e1T Q1e1  e~yT Q2 e~y  2eTy P2 v  2eTy P2 D2 provide an intermediate step for establishing the form in
section from which the observer was designed.
 e1T Q1e1  e~yT Q2 e~y  2eTy P2 D2  2 (t , y, u) D2 P2 e y Let G l and G n be appropriate gain matrices so that
Using uncertainty bound and the bound for  () from Ao  A  Gl C is stable, and assume an ideal sliding mode
equation (25) in the inequality above insensitive to uncertainty exists on the hyperplane in the error
V  e1T Q1 e1  e~yT Q2 e~y  2  (t , y, u ) D 2 P2 e y space given by S o . The error system satisfies

 2r1 u   (t , y )  D 2 P2 e y
e(t ) Ao e(t )  D (t , x, u )  G n v (42)
For a unique equivalent control to exist, det (CG n )  0 .
 e1T Q1e1  e~yT Q2 e~y  2 o D2 P2 e y
e(t )  ( I  Gn (CG n ) 1 C )e(t )
 0 for (e1 , e y )  0 (43)
 ( I  Gn (CG n ) 1 C ) D (t , x, u )
And hence the error system is quadratically stable.
Consider the hyper plane in the error space given by To be insensitive to the uncertainty it follows that


S o  e  R n : Ce  0  (36) ( I  G n (CG n ) 1 C ) D =0

2925
IJRITCC | May 2015, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 3 Issue: 5 2923 - 2929
_______________________________________________________________________________________________
or equivalently 
c  CTL1  0 I p 
D  G n (CG n ) 1 C (44) From the definition of and
Since by assumption rank ( D )  q . Therefore it c ab be
assumed without loss of generality that the system ( A, D, C ) is
and so the uncertainty distribution matrix is given by
in the canonical form. If the nonlinear gain matrix is
partitioned so that
G n  Finally, if , it can be shown by direct evaluation
Gn    (45)
G 2  that
Then CG n  TG 2 and so det det(G 2 )  0 . From equation (45).
It follows that the poles of the (linear) reduced order motion This is stable by choice of L. The system triple ( A, D, C ) is
are given by now in the canonical form (24) a robust observer exists.

 
Ao 11  G1G21 Ao 21  (46) In the special case where an observer of the form (45)
which is insensitive to the uncertainty in (20) exists if one only
where  Ao 11 and  Ao 21 represent the top left and bottom left
if
sub-blocks of the closed-loop matrix Ao partitioned in a . (49)
compatible way to the canonical form. By definition the The invariant zeros of ( A, D, C ) lie in .(49).That is, the
matrix Ao  A  Gl C , so triple ( A, D, C ) is minimum phase and relative degree 1. In
( Ao )11  A11  Gl C 11 this case the restriction that guarantees the
Where Gl C 11 represents the top left sub-block of the square existence of exactly invariant zeros and therefore the
matrix G l C . However, it is easy to check that Gl C 11  0 for
reduced order sliding motion is totally determined by these
zeros.
all Gl  R rr and so  Ao 11  A11 . Similarly it can be shown
that  Ao 21  A21 and consequently. IV. SIMULATION RESULTS

  
  Ao 11  G1G21 Ao 21   A11  G1G21 A21  (47)
State space model of Smart cantilever beam with uncertainty
is considered in this work
From equation (44) it follows that
G1G 21 D21  0  x1 (t )   92.1084 64.5070 - 39.8911 65.1749 
  
Which after considering the structure of implies  x 2 (t )   - 159.5286 14.3813 112.5734 - 118.4229
G1G 21  
 G 0  x 3 (t )  116.4182
  
- 111.6173 - 15.247 160.9807 

Where G1  R( n  p )( p  q ) and therefore from the definition of  x 4 (t )  - 63.1027 39.0227 - 63.7560 - 93.4438 

A11 it follows that  x1 (t )  - 0.5220 - 0.0141


     
A11  G1G 21 A21  A11  G A211  x 2 (t )  0.2457 u (t )  - 0.0387 (10 * sin(190t ))
 x 3 (t )  - 0.3766 0.0421 
By construction the pair ( A11 , A211 ) is such that      
 x 4 (t ) 0.7240  0.0058 
zeros of ( A, D, C)  A11o   ( A11  G A211 ) for all y(t )  1 0 0 0x1 (t ) x 2 (t ) x3 (t ) x 4 (t )T
G  R ( n  p )( p  q ) Assume initial conditions of the actual system is
Let ( A, D, C ) represent the system and suppose rank (CD )  q
x(0)  1.5 1.5 0 0T
and any invariant zeros lie in C  . Without loss of generality it
Assume initial conditions of the estimator system is
xˆ(0)  0 0 0 0T
can be assumed that the system is already in the canonical
o
where the matrix A11 is stable. As a consequence there exists a
System uncertainty is  ( x, t , y )   sin(190t ) ;
( n  p )( p q )
matrix LR such that ( A11  LA211 ) is stable. Assume that input to the actual system and observer system as
Define nonsingular transformation as same. That is u=0 or known.
I L Observe the system states using Utkin observer
TL   n  p  (48)
The Utkin observer is explained in II.
 0 T
Where L I  0(n p)q  Applying the Utkin observer for the above system. The results
obtained as
After changing coordinates with respect to TL , the new output Assume M=1 obtains L as 0.1 0.1 0.1
distribution matrix becomes
2926
IJRITCC | May 2015, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 3 Issue: 5 2923 - 2929
_______________________________________________________________________________________________

Amplitude
Figure.1 System uncertainty Figure.6 Actual state x (t ) vs estimated state xˆ (t )
3 3

Figure.2 Actual state vs estimated state

Figure.7 Estimation error between actual state x (t ) vs estimated state xˆ (t )


3 3

Figure.3 Estimation error between actual state vs estimated state

Figure.8 Actual state x 4 (t ) vs estimated state xˆ 4 (t )

Amplitude

Figure.4 Actual state vs estimated state

Figure.9 Estimation error between actual state x 4 (t ) vs estimated state xˆ 4 (t )

The observer starts at 0 sec but it response to track the actual


state after 2 sec only. It tries to approach the actual state and it
continuously minimizes the error but due to system uncertainty
present in actual system it cannot able to track.
Observe The System States Using Walcott Zak Observer
FiFigure.5 Estimation error between actual state vs estimated state

The Walcott Zak observer is expressed as.


xˆ(t )  Axˆ(t )  Bu(t )  G Ce(t )  G v
1 n
Assumptions
Stable design matrix s
A22  10
 (t , y, u )  15

2927
IJRITCC | May 2015, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 3 Issue: 5 2923 - 2929
_______________________________________________________________________________________________
From the algorithm Linear gain
Gl  376 - 366 255 266T
Nonlinear gain Gn  0.0141 0.0387  0.0421  0.0058T

Figure.14 Estimation error between actual state vs estimated state

Figure.10 Actual state vs estimated state

Figure.15 Estimation error between actual state x (t ) vs estimated state xˆ (t )


3 3

Figure.11 Actual state vs estimated state

Figure.16 Estimation error between actual state x 4 (t ) vs estimated state xˆ 4 (t )

Figure.12 Actual state x (t ) vs estimated state xˆ (t )


3 3

Fig.17 Estimation error between actual state vs estimated state.

The observer starts at 0 sec but it response to track the actual


state after 6 sec. It tries to approach the actual state and it
continuously minimizes the error and it eliminates the error
due to system uncertainty.
Figure.13 Actual state x 4 (t ) vs estimated state xˆ 4 (t )

V. CONCLUSION
From the results obtained, the Utkin observer starts to track
the actual system states after 1 sec and it is unable to approach
the actual system states in presence of uncertainty. This can be
seen from the simulation results. Because the Utkin Sliding
mode observer does not have a static observer gain in its
structure and instead, the switching gain Lν plays the role of
stabilizing the error dynamics. So it is unable to minimize the
error due to system uncertainty.
2928
IJRITCC | May 2015, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 3 Issue: 5 2923 - 2929
_______________________________________________________________________________________________
From the results obtained, the Walcott Zak observer starts to
track the actual system states after 1 sec and it is able to
approach the actual system states after 6 sec. Because the
Walcott Zak observer has two error stabilizing components one
is feedback the estimation error and another one is the
switching function. The switching function has the range of
upper bound value of the system uncertainty. The estimation
error is feedback to the observer with linear gain and the
switching function is feedback with the non linear gain
these two feedback error gains play an important role to
minimize the error due to system uncertainty very effectively
and observe the system, for this the system uncertainty is
assumed to be an unknown function but with bounded range.
So the Walcott Zak observer is robust against system
uncertainty when compared to the Utkin observer.
Both Utkin observer and Walcott Zak observer are having
one disadvantage. The problem of observing the states of a
system, some of whose inputs are not available for
measurement. Utkin observer and Walcott Zak observer cannot
observe the system states, under such conditions. Both are
suitable for only the input to the system is available and zero.
In future the unknown input observer can be designed with
controller.

VI. REFERENCES
[1] Drakunov, S. Utkin,V. “Sliding mode observers Tutorial,” Decision and
control 34th IEEE Conference on volume. 4 pages 3376-3378, 1995.
[2] V. I. Utkin, “Variable structure systems with sliding modes,” IEEE
Trans. Automatic. Control vol. AC-22, pp. 212–222, 1977.
[3] B. L. Walcott, S. H. Zak., “Combined Observer Controller synthesis for
Uncertain Dynamical Systems with Applications,” IEEE Transactions on
Systems, Man and Cybernetics, vol. 18,No. 1, pp.88-104, 1988.
[4] Edwards C, Spurgeon SK. “Sliding mode control: theory and
applications,” vol. 7, Systems and Control Book Series. Taylor &
Francis, London, pg no 128-147, 1998.
[5] Utkin VI. Sliding modes in control and optimization,
Communications and Control Engineering Series. Springer-Verlag,
Berlin, 1992.
[6] Edwards, C., and Spurgeon, S.K., On the Development of
Discontinuous Observers,Department of Engineering Report 92-41,
December 1992, Leicester University, Int. J.Control, Vol. 59, No. 5,
1211-1229, 1994.
[7] J. J. Slotine, “Sliding controller design for nonlinear systems,” Int.J.
Contr., vol. 40, no. 2, pp. 421-434, 1984.
[8] R. G. Morgan and U. Ozguner, “Decentralized variable structurc control
algorithm for robotic manipulators.” IEEE 1. Rohot ,1 iitomat., vol. RA-
1, no. 1, pp. 57-65, 1985.
[9] Walcott, B.L., Codess, M.J. and Zak, S.H., Comparative Study of
Nonlinear State Observation Techniques, Int,J.Control, V.45, no. 6, pp.
2109-2132,1987.
[10] Drakunov, S.V. Sliding-Mode Observers Based on Equivalent Control
Method, Proceedings of the 31stIEEE Conference on Decision and
Control (CDC),Tucson, Arizona, December 16-18, 1992, pp. 2368-2369.

2929
IJRITCC | May 2015, Available @ http://www.ijritcc.org
_______________________________________________________________________________________

You might also like