Vous êtes sur la page 1sur 10

1

Abstract The magnetic levitation system is a non


linear mechatronic system accepted both for the
specific mechatronic area and for other engineering
fields , as mentioned in numerous references[1], [2],[3].
The current period is characterized by multiple
requirements of achieving high precision mobile
mechanical systems, managed by an intelligent control
system and appointed to various industrial
technologies: transportation, magnetic bearings, kinetic
energy store systems, special actuators, haptic
magnetic levitation,etc,[4],[5].Magnetic levitation
system is a highly nonlinear process. This working
model is used to verify the design concepts of
Lyapunov based stabilizing controller by T-S Model
Adaptation Scheme.A Lyapunov-based stabilizer and a
Takagi-Sugeno (T-S) model based Adaptive stabilizing
controller using fuzzy plant model have been proposed
here. Here a two-part controller has been developed in
order to overcome the modeling uncertainties, The first
part stabilizes the nominal plant; without modelling
uncertainties. The second part of the controller is to
reject modelling uncertainties. In both cases, fuzzy
state-feedback control technique has been applied.
In case of Adaptive stabilizing controller, some adaptive
nature to the control generation itself has been
introduced and we consider one part instead of dividing
the control law in two parts. Thus introducing separate
control law for external disturbance and noise rejection
is taken care off automatically by the adaptive nature. A
real time nonlinear magnetic levitation system has
been simulated with these controller to show the
effectiveness, merits and the design procedure of the
proposed state feedback controller in MATLAB
Toolbox.

Prof. Alok Kole is with the Department of Electrical and
Elecctronics Engineering,Mody Institute Of Technology &
Science,Lakshmangarh-332311,Rajasthan,India.
Ms. is with the Department of Electrical and Elecctronics
Engineering,Mody Institute Of Technology & Science


I. Introduction
First we shall develop the controller by using fuzzy
design model, which combines the mathematical
description of the on process in the form of a working
model and the linguistic description of the process.
The fuzzy design model of particular interest is of the
TSK fuzzy model .

II. Development of the Physical model
The continuous fuzzy dynamic model is expressed by
fuzzy IF-THEN rules, which represent local linear input-
output relations of nonlinear systems. The i
th
rule of the
fuzzy model can be written as the following form:
Plant Rule i :
IF ) t ( x
1
is
i
1
M AND ) t ( x
n
is
i
n
M
THEN ) ( ) ( ) ( t u B t x A t x
i i
+ = (1)
where
i
j
M , n ., . . . . . . 2, , 1 = j is the
th
j fuzzy set of
the
th
i rule, ) t ( x and ) t ( u are the state vector and the
control input and
i
A and
i
B are the matrices of
appropriate dimensions. Let ) x (
j
i
j
be the
membership function of the fuzzy set
i
j
M

and let



Given a pair of )) t ( u ), t ( x ( , the final output of the
resulting T-S fuzzy model can be derived as the
weighted average of local models as follows:
Adaptive Stabilizing controller design by T-S Model for
uncertain Nonlinear Systems

Alok Kole
1
,
2

1
EEE Department ,Mody Institute Of Technology & Science,Lakshmangarh-332311,Rajasthan,India,.
alokkole123@yahoo.co.in
1
EEE Department ,Mody Institute Of Technology & Science,Lakshmangarh-332311,Rajasthan,India.

) ( )) ( (
1
j
m
j
i
j
i
x t x w
=
= o


2


r
1 = i
i
i i
r
1 = i
i
)) t ( x ( w
) u B + x A ))( t ( x ( w
= x


= ) u B + x A (
i i
r
1 = i
i


= u ) B ( + x ) A (
i
r
1 = i
i i
r
1 = i
i


= u ) ( B + x ) ( A (2)
A fuzzy design model has been formed by combining
the working model of the dynamical system with its
linguistic description. The truth model is given by
)) x , t ( u )( x ( G ) x ( f x + + =

(3)
Where
n n
: f 9 9 ,
m n n
: G

9 9 , and
) x , t ( represents modelling uncertainties. It is
assumed that the uncertainty modelling function has
an upper limit and is bounded by a nonnegative
function x) (t, = so that
) x , t ( ) x , t (
p
(4)
where
p
. denotes the vector p norm.
Finally T-S fuzzy design model is represented in the
following form:
)) x , t ( u )( ( B x ) ( A x + + =

(5)
Using a Lyapunov based approach, we are interested
in the construction of state-feedback controllers for the
fuzzy design model of the form (5) derived from
nonlinear systems. The fuzzy design model is
developed from the linguistic description of the
process and the working model.
Also a common Lyapunov matrix is found using an
iterative algorithm and the robust stability analysis is
performed on the influence of parameter uncertainty.
Finally the performance of the stabilizing controller on
the working model is investigated.
III.Design of the Working model of the Magnetic
Levitation System
We introduce the most basic magnetic suspension
system whose motion is constrained in the vertical
plane[6] The schematic diagram of the electromagnetic
suspension system [7] is shown in figure 1. This
magnetic levitation system consists of a ferromagnetic
ball suspended in a voltage-controlled magnetic field.
The objective of the system is to control the vertical
position of the ball at a prescribed reference level by
adjusting the current in the electromagnet through the
input voltage E. The ball position is read by an optic
transducer which can detect the distance between the
iron ball and the electromagnet. Note that this simple
electromagnet system is unstable without feedback
control. We derive an ideal mathematical model of the
system based on physical laws and several
assumptions. The dynamic model of the system can be
written as the following three equations which show
the motion of the iron ball, electromagnetic force and
equation of an electric circuit of the electromagnet,
respectively.
) t ( f Mg
dt
) t ( x d
M
2
2
= (6)
2
0
)
x + ) t ( x + X
) t ( i + I
( k = ) t ( f (7)
) t ( e + E = )) t ( i + I ( R +
dt
) t ( di
L (8)
where M is the mass of the iron ball, X is the steady
gap between the electromagnet and the iron ball, x(t) is
the deviation from X, I is the steady current and i(t) is
the deviation from I, E is the steady voltage, e(t) is the
deviation from E, f(t) is the electromagnetic force, k and
x
0
are coefficients of f(t) which are determined by
experiments, L is an inductance of the electromagnet
( EM ) and R is the resistance of the EM.
From (7) we can write
)
X
) t ( x
2 1 ( X )) t ( i ) t ( i I 2 I ( k )
X
) t ( x
1 ( X ) ) t ( i I ( k ) t ( f
0
2 -
0
2 2 2 -
0
2
0
2
+ + = + + =

( 9)
where
0 0
x + X = X .
Let say and x = x , x = x
2 1

3
x = x

. Substituting these
values in (9), we can simplify (9) as
k x x M x M x M M ) t ( f
3 1 4 3 3 1 2 1
+ + = (10)
where
2
3 1
-3
0
2
3
-2
0
x x x 2 x X =
2 2
0 1
I kX M

= ;
3
0
2
2
kX I 2 M

= ;
2
0 3
kIX 2 M

= ;
3
0 4
kIX 4 M

=


3




Substituting the value of ) t ( f from (10) in (6), we can
simplify (6) in terms of state variables as


M M x x M x M x M x
9 8 3 1 7 3 6 1 5 2
+ + =

(11)
where
M
M - Mg
M ,
M
M
M ,
M
M
M ,
M
M
M
1
8
4
7
3
6
2
5
= = = = ,
M
k
= M
9

Similarly from (8), we can write
) I
L
R
L
E
( u )
L
1
( x )
L
R
( x
3 3
+ + =

(12)
By combining (11) and (12), we obtain a state-space
model of the magnetic levitation system of the form
(
(
(

3
2
1
x
x
x

=
(
(
(
(
(


L
R
0 0
M x M 0 M
0 1 0
6 1 7 5
(
(
(

3
2
1
x
x
x
+
(
(
(
(

L
1
0
0
) u (
1
+

(13)
where ) M M ( L
9 8 1
= and
2
3 1
3
0
2
3
2
0
x x x 2 x X

=

IV. Lyapunov based Stability problem formulation
and Stability analysis
We first write sufficient conditions for asymptotic
stability in the large of unforced continuous-time fuzzy
model
x

= x A
i
r
1 = i
i

(14)
where 0 )) t ( x ( =
i i
for r ., . . . . 2,. , 1 = i and
1
r
1 i
i
=
=
We next discuss the stabilization problem of
forced fuzzy models using fuzzy state-feedback control
strategies. The necessary condition for the existence of
a common symmetric positive definite P satisfying (14)
is that each
i
A be asymptotically stable; that is, the
eigen values of each
i
A be in the open left-hand
complex plane. Tanaka and Sugeno prescribed an
important criterion for stability of T-S fuzzy model,
which is stated below in the form of a theorem.
Tanaka and Sugeno prescribed an important criterion
for stability of T-S fuzzy model, which is stated below in
the form of a theorem [8]
Theorem 1: The equilibrium state 0 x = of fuzzy
system (14) is globally asymptotically stable if there
exits a common symmetric positive definite matrix P
such that
0 < PA + P A
i
T
i
(15)
where r} ., . . . . 2, {1, i
k

Theorem 2 : Suppose that each ) F B A (
i i i
,
r ., . . . . 2, 1, = i in (17) is asymptotically stable and
there exists a symmetric positive definite P such that
condition (16) and (17) hold, then the closed-loop
fuzzy model (14) is asymptotically stable in the large if
the matrix
(
(
(
(




= A
r
2
r 2
2
r 1
2
r 2 2
2
12
2
r 1
2
12 1
... / /
. .. .. .
/ .. /
/ .... /

is positive definite.
0 ) ( ) ( ( +
i i i
T
i i i
F B A P P F B A , r , . . . . . 2, 1, i = (16)
and
ij
T
ij
PG P G + < 0 , r j i <

(17)
where G
ij
is given by,
) F B A ( ) F B A ( G
i j j j i i ij
+ = for r j i <


Next we present a sufficient condition for asymptotic
stabilizability in the large of fuzzy model given by (1)
and (3) using the state feedback control law

r
1 j
j j
x F u
=
= (18)


4


Then the closed-loop system has the form
x ) F B A ( x
j i i j
r
1 i
r
1 j
i
=
= =

(19)
The gain matrices
i
F

is chosen in a way such that
the matrices

i i i
F B A ,

r ., . . . 2,. 1, = i
have their eigen values in the open left-hand complex
plane, that is, they are asymptotically stable.
V. Lyapunov Based Stabilizers Design for
Magnetic Levitation System
The following are the conditions for construction of
stabilizing controllers as documented documented in
Zinober [9];:
- Each of the matrices
i
A , r ., . . . . 2, 1, = i is
already asymptotically stable.
- Theorem 1 as mentioned in (15) needs to be
satisfied.
If this is not the case, we break the design of stabilizing
control law u into two parts u = u
1
+u
2
. We first compute
u
1
as
x F u
j
r
1 j
j 1

=
= (20)
so that the system x

= x ) F B A (
j i i j
r
1 i
r
1 j
i

= =

asymptotically stable in the large and there exists a
common P such that conditions (16) and (17) or
conditions of theorem 2 are satisfied. Note that the
system modeled by (5) and driven by (20) has the from
x

=
=
x A
r
1 i
i i i
r
1 i
i
B
=
) x F (
j
r
1 j
j

=
+ ) u ( B
1 2 i
r
1 i
i
+
=

= x ) F B A (
j i
r
1 i
i j
r
1 j
i

= =
+ ) u ( B
1 2 i
r
1 i
i
+
=
(21)
Now we proceed with the design of
2
u to reject
uncertainty .Then
2 1
u u u + = will asymptotically
stabilize in the large the uncertain system (5). These
controllers are obtained by fuzzifying the conditions
proposed by Hui and Zak [10]. To describe the family,
we need the following notation. Let
[ ] Px B = s ...... s s = s
T
i
r
1 = i
i
T
m 2 1


Finally the gradient of
p
s ,i.e.,
p
s V is defined as,
(
(
(
(
(
(

=
(
(

c
c
c
c
c
c
= V

) (
) (
) (
s
1
s s s
s
1
2
1
2
1
1
1
1
2 1
m
p
m
p
p
p
p
T
m
p p p
p
s sign s
s sign s
s sign s
s s s

(22)
where .
p
denote the p norm of a vector
Let
q
q
) x , t ( for 0
q
> . If
p
q 2
s u = ,
1 p and 1 =
q
1
+
p
1
(23)
then the closed system driven by
2 1
u u u + = is
asymptotically stable in the large for any uncertainty
such that
q
q
s .
Theorem 3 : Given fuzzy system model (5), suppose
that u
1
given by (20) is such that there exists a
common P satisfying conditions (16) and (17) or
conditions of theorem 2. Then the closed-loop system
driven by , u u u
2 1
+ = where
2
u is given by (23), is
asymptotically stable in the large for any uncertainty c
such that
q
q
s .
Proof: From (21) , we can write
x ) F B A ( x
j i
r
1 i
i j
r
1 j
i
=
= =

+ ) u ( B
1 2 i
r
1 i
i
+
=
(24)
We now prove that Px x V
T
= is a Lyapunov function
for the closed-loop system where P= P
T
>0 satisfies
(16) and (17) or conditions of Theorem 2; For this, we
evaluate the total time derivative of V; that is, we
evaluate
dt
dv
on the trajectories of the closed loop
system, and show that the Lyapunov derivative of V is
negative definite. We have
dt
dv
=
= =
r
1 i
r
1 j
j i i j i
T
x ) F B A ( ( P x 2 + ) u ( B
1 2 i
r
1 i
i
+
=
)
<
1
T
p
T
q
s 2 s s 2 +



5




1
T
p
q
s 2 s 2 +
The eq. of
1
is in the form as
M
M
g
1
1
= ) x x X 2 x X (
M
k
2
3 1
3
0
2
3
2
0

.
As the electromagnetic forces must be greater than the
gravitational force
g ) x x X 2 x X (
M
k
2
3 1
3
0
2
3
2
0
>

so 0
1
< .Hence
0
dt
dv
<
So by Lyapunovs theorem, the closed-loop system is
asymptotically stable in the large.
VI. Adaptive Controller design by T-S Fuzzy Model
Here we will briefly discuss about the design of T-S
model based adaptive fuzzy controller proposed by
Park and Cho and then we will extend this concept for
the adaptation of the Magnetic Levitation system.

If the unknown parameters in a plant vary under
the influence of external disturbances and or noise, the
state must first be estimated before an adaptive control
law can be synthesized. The estimated state (t) x
is
compared with the measured state x(t) of the plant and
the parameters are varied by the adaptive law in such a
way that the estimated state (t) x approaches the
measured state x(t) as t .
We consider the process described by the fuzzy IF-
THEN rules of equation (1). Parameter estimator for
the T-S fuzzy model is designed using the following
parameterized plant equation.
( ) ( )
r
1 i
i s i i s
u B x A A x A x
=
+ + =

(25)
where A
s
is an arbitrary stable matrix ,i.e., all its eigen
values are on the left half of the s-plane.
The estimation model may be written as,
( ) ( )
r
1 i
i s i i s
u B x A A x A x
=
+ + =

(26)
where ) t ( A
i

and ) t ( B
i

are the estimates of the


matrices ) t ( A
i
and ) t ( B
i
respectively at time t and
n
) t ( R x e

is the estimate of the state vector x(t). The


matrices ) t ( A
i
and ) t ( B
i
are to be varied by the
adaptive law.
The error in estimating the state vector can be
computed as
x x e


= u
w
B
~
w
x
w
A
~
w
e A
i
i i
i
i i
s

= (27)
where
i i i
A A A
~
=

,
i i i
B B B
~
=


The adaptive law has the general structure of
( ) u , e , x x, M = A
i i

and
( ) u , e , x x, E = B
i i

(28)
The guideline for selecting M
i
and E
i
are such that at
the equilibrium

i ie
A A =

,
i ie
B B =

and 0 e
e
= ( 29)
where
ie
A

and
ie
B

are the estimated values of A


i
and
B
i
respectively under equilibrium condition.
Now to find out the actual structure of the adaptation
law Lyapunov function is chosen as given below,
( )
|
|
.
|

\
|
+
|
|
.
|

\
|
+ =
= =
r
1 i
i 2
i
T
i
r
1 i
i 1
i
T
i T
i i
r
B
~
P B
~
tr
r
A
~
P A
~
tr Pe e B
~
, A
~
, e V
(30)
where tr(.) stands for the trace of a matrix ,
0 > P = P
T
is the solution of I PA P A
s
T
s
= + and
0 > r , r
i 2 i 1
are constant scalars. Since the matrix A
s
is
stable, solution of P will always exist
Now the total time derivative of the Lyapunov function
V is computed as,
|
|
|
|
|
|
.
|

\
|

+
+ =
T
i
T
i i
i
i
T
i
T
i
T
i i
i
i
T
i
T
Pe
w
B w
r
PE B
Pe
w
A w
r
PM A
tr e e V
u
~

~
x
~

~
2

2
1




6


(31)
Now V

in equation (31) will be negative definite if the


following two conditions are satisfied simultaneously.
T
i
T
i i
i 1
i
T
i
Pe
w
A
~
w
r
PM A
~
x

=
,
T
i
T
i i
i 2
i
T
i
u Pe
w
B
~
w
=
r
PE B
~

(32)
Using relations (32), the adaptive control law becomes,

T
i
i
i 1 i
x e
w
w
r = A

and
T
i
i
i 2 i
u e
w
w
r = B


VII. Design of Lyapunov based Adaptive stabilizing
controller by T-S Fuzzy Model
The adaptation of uncertain parameters of system
matrix A
i
and input matrix B
i
can be combined with the
design of the stabilizing controller. In the previous
section , we designed the control law u containing two
components, u
1
and u
2
. The first component was
synthesized based on state feedback control for a
stable linearized system whereas the other component
was intended to compensate the uncertain parameter
variation provided the absolute magnitude was
constrained with an upper bound, i.e.
) x , t ( ) x , t (
p
.
By making adaptation inherently present in the
generation process itself for the controller, the
uncertain variation of parameters can be taken care of
for closed loop system stabilization. It will be found
that, this can be achieved without using two distinct
components of controllers. Besides, it will improve the
performance of the closed loop system even in
presence external disturbances.
Now in the new scheme the control law generated in
the previous section becomes

r
1 j
j j 1
x F u u
=
= =

(33)
where, F

is the estimated feedback matrix obtained


based on the adapted system matrix
i
A

and input
matrix
i
B


For any system we can express the closed loop system
without adaptation as using the control law
=
=
r
1 j
j j 1
x F u similar to equation (33) as

r
1 i
1
r
1 j
j j i i
r
1 i
i i
) x F ( B x A x
= = =
|
|
.
|

\
|
+ =


= ( )
r
1 i
1 i i
r
1 i
r
1 j
j i i j i
B x F B A
= = =
+ (34)
If the effect of noise
1
is ignored, the closed loop
system will appear as:
( )
r
1 i
r
1 j
j i i j i
x F B A x
= =
=

(35)
If parameters of the matrices ) t (
i
A

and ) t (
i
B

are
adapted with time t by an adaptive control law , then the
coefficients of feed back matrix ) t (
j
F

will also be
varying in such a way that the estimated value ) t ( x

will
approach the measured value x(t) as t approaches
infinity.
In terms of estimation model the closed loop system
using control law of equation (33) becomes
( )
( ) ( )
( )

1
1
1 1
1
B B
x F F B B
- x A A x
r
i
i s i i
r
i
r
j
j s j j i s i i
r
i
i s i i
T
T T
T
=
= =
=
+ +
|
|
.
|

\
|
+ +
+ =
c


=
( )
1
r
1 i
i s i i
r
1 i
j s j i s i i s i j i
B T B x F T F B T B A T A -
= =
|
.
|

\
|
+ +
|
.
|

\
|
+ |
.
|

\
|
+ +


(36)
Since, the effect of uncertainties is taken care of by
adaptation of the feedback matrix, the term involving
1
is dropped from equation (36) to get the following
simplified closed loop system
( )
r
1 i
j s j i s i i s i j i
x F T F B T B - A T A x
=
|
.
|

\
|
+ |
.
|

\
|
+ + =

(37)
From equation (35) and (37) we deduce the constraint
of choosing F

V in such a way that




7


i) ) t ( x ) t ( x

and ii)
( ) ( )
r
1 i
j s j i s i i s i j i
r
1 i
r
1 j
j i i j i
F T F B T B A T A F B A
= = =
|
.
|

\
|
+ |
.
|

\
|
+ +


(38)
The outline of the Adaptation scheme is shown in
details in the following diagram Fig.2
VIII. Testing and Application of Lyapunov based
stabilizing Controller for Magnetic Levitation
system Model without adaption











Figure 2 : Schematic structure of T-S model based
To better test our stabilization algorithm, we randomly
chose the set of desired poles for ) F B A (
1 1 1
to be in
the interval [-40,-90], and we got
80] 40 48 [
1
= ,

The desired poles for ) F B A (
2 2 2
were also chosen
in a random fashion to be the interval [-40, -90], In this
case, we got the set

85] 42 50 [
2
=
For this choice of the closed-loop poles of the local
models, we obtain the following feedback gain
matrices:
| | 552 . 119 9511 . 374 9099 . 29664 F
1
= ,
And | | 127.283 199.3104 5752 . 15832 F
2
= ,
Typical responses of stabilizing controller without
adaption are shown in figures 3 and 4 with different
initial conditions.
0 100 200 300
-1
0
1
2
3
x 10
-3
t, msec
x
1
0 100 200 300
-0.05
0
0.05
t,msec
x
2
0 100 200 300
-0.6
-0.4
-0.2
0
0.2
t, msec
x
3
0 100 200 300
-60
-40
-20
0
20
t,msec
u

Figure 3 : Plots of X(t) and u(t) with time for
initial values of x(0)= [0.002 0.000 0.000 ] without
adaptive control
0 100 200 300
-2
0
2
4
6
x 10
-3
t,msec
x
1
0 100 200 300
-0.1
-0.05
0
0.05
0.1
t, msec
x
2
0 100 200 300
-0.6
-0.4
-0.2
0
0.2
t, msec
x
3
0 100 200 300
-80
-60
-40
-20
0
20
t, msec
u

Figure 4 : Plots of X(t) and u(t) with time for
initial values of x(0)= [0.004 0.000 0.000 ] without
adaptive control
( ) ( ) + + =
=
r
1 i
i s i i s
u B x A A x A x


x
x


) u B x A ( x
i i
r
1 i
i
+ =

=


Control
Law
Process
Model
MMmmM
odel
Computation of
Feedback
Coefficient
i
F


T
i i 2 i
u e r B =


T
i i i
e r A x
1
=


x F

u
j
r
1 j
i

=
=



8


IX. Testing and Application Lyapunov based
Stabilizing Controller for Magnetic Levitation
system Model with Adaptation
Lyapunov based Adaptive Stabilizing controller has
been applied and tested for magnetic levitation system.
The responses of the controller are shown in the
subsequent figures. The fuzzy model of the Magnetic
Levitation system as given by (5) is,
( ) ( ) ) + u ( B + B + x A + A = x
1 2 2 1 1 2 2 1 1


where u is the control signal and ) M M ( L
9 8 1
= .

The control u is given by

( ) sign(s) * x F F u
q 2 2 1 1
+ = for the system
( ) =
= =
2
1 i
2
1 j
j i i j i
x F B A x


The desired closed loop poles are arbitrarily chosen as,

[-60 -40 -50]; for ( )
1 1 1
F B A

[-62 -42.5 -52]; for ( )
2 2 2
F B A
We adapt A and B matrices and in turn adapt the
feedback matrices and the final set of matrices is given
as,
F
1
= [26063.8543 336.4879 104.09];
F
2
= [13742.3603 176.4913 109.6735];

The P matrix is computed and the final controller is
given as,
( ) sign(s) * - x F + F = u
q 2 2 1 1
.
The initial conditions are taken as, x(0) = [0.002 0 0]
and x(0) = [0.004 0 0]. fig 5 and fig 6 represent the
corresponding responses of the Adaptive stabilizing
controller for magnetic levitation system. Fig 5 gives
the plots of x(t) = [ x
1
(t) x
2
(t) x
3
(t) ] along with control
action u(t) with time for initial condition x(0) = [0.002
0 0] whereas fig 6 shows the corresponding plot for
initial conditions x(0) = [0.004 0 0]. It is observed that
in case of adaptive controller the position of the
suspended ball returns to to the stable equilibrium
within 100 mili seconds for both the initial conditions,
but in case of non-adaptive it is 200 mili seconds.

0 50 100 150 200
-4
-2
0
2
4
x 10
-3
t, msec
X
1
0 50 100 150 200
-0.2
-0.1
0
0.1
0.2
t, mse
x
2
0 50 100 150 200
-1
-0.5
0
0.5
1
t, msec
x
3
0 50 100 150 200
-40
-20
0
20
40
t, msec
u


Figure 5 : Plots of X(t) and u(t) with time for initial
values of x(0)= [0.002 0.000 0.000] with fuzzy
adaptive control

Also the adaptation of the system matrix parameters
along with the input matrices are shown in figure 7,8 ,9
and 10 respectively. We also observe that system
state matrices A

and input matrices B

are
converging to stable values ( vide figures 7,.8, 9 and
10) .


9



0 50 100 150 200
2
3
4
5
6
x 10
-3
t, msec
x
1
0 50 100 150 200
-0.1
-0.05
0
0.05
0.1
0.15
t, msec
x
2
0 50 100 150 200
-0.8
-0.6
-0.4
-0.2
0
tm msec
x
3
0 50 100 150 200
-80
-60
-40
-20
0
t, msec
u


Figure 6 : Plots of X(t) and u(t) with time for
initial values of x(0)= [0.004 0.000 0.000] with
fuzzy adaptive control
0 50 100
0.058
0.059
0.060
t, msec
a
1
1
(
t
)
0 50 100
1.059
1.060
1.061
t, msec
a
1
2
(
t
)
0 50 100
0.0588
0.0589
0.0590
t, msec
a
1
3
(
t
)
0 50 100
6247.4233
6247.4234
6247.4235
t,msec
a
2
1
(
t
)
0 50 100
0.0596
0.0597
0.0598
t, msec
a
2
2
(
t
)
0 50 100
34.8985
34.8986
t, msec
a
2
3
(
t
)
0 50 100
0.0596
0.0597
0.0598
t, msec
a
3
1
(
t
)
0 50 100
0.0590
0.0591
0.0592
t ,msec
a
3
2
(
t
)
0 50 100
-28.7653
-28.7652
-28.7651
t, msec
a
3
3
(
t
)


Figure 7 : Adaptation of the state matrix A


( 3 3 ) for the system. Plot of three rows of the
component of A

matrices for 1
st
rule with time
with initial conditions x(0) = [0.002 0 0 ]


0 50
-0.0457
-0.0458
-0.0459
-0.0460
t, msec
a
1
1
(
t
)
0 50
0.9543
0.9544
0.9545
0.9546
t, msec
a
1
2
(
t
)
0 50
-0.0457
-0.0456
-0.0455
-0.0454
t, msec
a
1
3
(
t
)
0 50
6247.3187
6247.3188
6247.3189
6247.3190
t, msec
a
2
1
(
t
)
0 50
-0.0457
-0.0456
-0.0455
t, msec
a
2
2
(
t
)
0 50
69.6334
69.6333
69.6332
t, msec
a
2
3
(
t
)
0 50
-0.0457
-0.0456
-0.0455
-0.0454
t, msec
a
3
1
(
t
)
0 50
-0.0457
-0.0456
-0.0455
t, msec
a
3
2
(
t
)
0 50
-28.8699
-28.8698
-28.8697
-28.8696
t, msec
a
3
3
(
t
)

Figure 8 : Adaptation of the state matrix A

( 3 3 )
for the system. Plot of three rows of the component of
A

matrices for 2
nd
rule with time with initial
conditions x(0) = [0.002 0 0 ]

0 10 20 30 40 50
0.0780
0.0785
0.0790
t, msec
b
1
1
(
t
)
0 10 20 30 40 50
0.0780
0.0785
0.0790
t, msec
b
1
2
(
t
)
0 10 20 30 40 50
1.2426
1.2428
1.2430
t, msec
b
1
3
(
t
)

Figure 9: Adaptation of the input matrix B

( 1 3 )
for the system. Plot of three rows of the component of
B

matrices for 1
st
rule with time with initial
conditions x(0) = [0.002 0 0 ]


10


0 5 10 15 20 25 30 35 40 45 50
0.05830
0.05835
0.05840
t , msec
b
1
1
(
t
)
0 5 10 15 20 25 30 35 40 45 50
0.05830
0.05835
0.05840
t, msec
b
1
2
(
t
)
0 5 10 15 20 25 30 35 40 45 50
1.2224
1.2225
1.2226
t, msec
b
1
3
(
t
)

Figure 10 : Adaptation of the input matrix B

( 1 3 )
for the system. Plot of three rows of the component of
B

matrices for 2
nd
rule with time with initial
conditions x(0) = [0.002 0 0 ]
X. CONCLUSION
A Lyapunov-based stabilizing control design technique
and a T-S model based Adaptive stabilizing controller
for uncertain, nonlinear dynamical systems using fuzzy
plant model has been presented. The design model is
the simplified version of the working model of the
physical process and usually neglects high-frequency
dynamics of the plant. Thus the incomplete
mathematical model of the process to be controlled
need to be studied thoroughly based upon its system
dynamics and then the working model is to be
developed. To overcome the modelling uncertainties,
we used a two-part controller. The first part stabilizes
the nominal plant; that is, the plant model that does not
include modelling uncertainties. The function of the
second part of the controller is to reject modelling
uncertainties. In both the cases, fuzzy state-feedback
control technique has been employed. In case of
adaptive stabilizing controller, some adaptive nature to
the control generation itself has been introduced and
we consider u
1
instead of dividing the control law in two
parts. Thus the objective of introducing separate
control law u
2
for external disturbance and noise
rejection is taken care off automatically by the adaptive
nature. Here also u
1
can cope up with sudden change
in system parameter due to various uncertainties. From
the figures 7,8, 9 and 10, we found that if we adapt
state matrix A

and input matrixB

at time t, then the


feed back gain matrix F

will also adapt and


consequently estimated state track the measured state
as t .So the convergence of the adaptive
parameters of matrices A

and B

for Adaptive control


are shown in these plots. A real time nonlinear
magnetic levitation system has been simulated with
these controller to show the effectiveness, merits and
the design procedure of the proposed state feedback
controller. In order to apply the proposed controller in
real-life physical systems, further research is needed
to account for limited authority of actuators,
unmatched uncertainties, as well as inaccessibility of
state variables.

REFERENCES
[1] Lilienkamp, L. A., Lundberg, K., Low-cost magnetic
levitation project kits for teaching feedback system
design,
http://web.mit.edu/klund/www/papers/ACC04_maglev.p
df
[2] Shiao, Y. S., Design and Implementation of a
Controller for a Magnetic Levitation
System, Proc. Natl. Sci. Counc., vol.11, no.2, 2001,
pp.88-94, 2001
[3] Guess, T. M., Alciatore, D. G., Model Development
and Control Implementation for a
Magnetic Levitation Apparatus, Proc. Of !995 ASME
Comp. in Eng. Conf., Boston,sept.1995

[4] Li, J. H., DSP- Based Control of a PWM-driven
Magnetic Levitation System, IEEE ICSS2005 Intern.
Conf. on Systems & Signals
[5] Kemin, K., Tekkouk, O., Constrained generalised
predictive control with estimation by genetic algorithm
for a magnetic levitation system, Intrn. J. of Innovative
Comp., Inform.Adn Control, v.2,no.3, 2006, p.543
[6] C.R.Knsope and E.G.Collins(Editor):Special Issue
on Magnetic Bearing Control, IEEE Trans. On Control
System Technology vol.4, no.5, pp.481-608,1996.
[7] M. Fujita,T.Namerikawn,et al.,-Synthesis of an
Electromagnetic Suspension System,, IEEE Trans.
Automatic Control, vol.40., no.3, pp.530-536, 1995.
[8] Marcelo C.M.Teixeira and Stanislaw
H.Zak,,Stabilizing Controller Design for Uncertain
Nonlinear Systems Using Fuzzy ModelsIEEE
Transactions on Fuzzy Systems , vol.7,No.2 April
1999.
[9] A.S.I.Zinober,Ed, Deterministics control of uncertain
Systems London, U.K, Peter Peregrinusa,1990.
[10] S.Hui and S.H.Zak,robust output feedback
stabilization of uncertain dynamic systems with
bounded controllers, Int.J. Robustn Nonlinear Contr,
vol.3, pp.115-132, Aug 1993.

Vous aimerez peut-être aussi