Sei sulla pagina 1di 14

79

CHAPTER 6

DESIGN OF STATE FEEDBACK CONTROLLER FOR


INTERVAL MODEL

6.1 INTRODUCTION

The problem of an interval analysis approach for the designing state


feedback controllers for linear time-invariant systems is extensively treated in
the control system literature by Marcia (2005). Linear models of real systems
sometimes include parameters whose values are unknown but bounded in
compact sets, often described in the form of closed intervals. The control
problem consists in finding a state feedback gain so as to place all closed loop
poles in the left-hand side of the complex plane( robust stabilization) or in
some prescribed region of it(robust performance) for every possible set of
system parameters. The basic goal of any practical motivated feedback design
is that the feedback system should preserve stability and closed–loop
performance requirements in the face of plant parameter uncertainties i.e.,
stability robustness and performance robustness respectively. Kharitonov’s
theorem states that the Hurwitz stability of a real (or complex) interval
polynomial family can be guaranteed by the Hurwitz stability of four
(or eight) prescribed critical vertex polynomials in this family. This result is
significant since it reduces checking stability of infinitely many polynomials
to checking stability of finitely many polynomials, and the number of critical
vertex polynomials need to be checked is independent of the order of the
polynomial family.
80

This chapter is organized as follows: In section 6.2, the


preliminaries about interval polynomial is presented. The state feedback
controller is designed using interval polynomial and simplex method in
section 6.3 and 6.4. Conclusion is given in section 6.5.

6.2 INTERVAL POLYNOMIAL PRELIMINARIES

An important approach to the subject is via expressing the


characteristic polynomial by an interval polynomial, i.e., a polynomial whose
each coefficient vary independently in a prescribed interval. The stability
analysis of polynomials subjected to parameter uncertainty have received
considerable attention after the celebrated theorem of Kharitonov (1979) ,
which assures robust stability under the condition that four specially
constructed “extreme polynomials”, called Kharitonov’s polynomials are
called Hurwitz. The problem of robust stability of interval polynomial is
discussed in Barmish (1989), Bartlett et al (1988), Battacharya (1987),
Keel et al (1999). Based on interval polynomial theory, Soh et al (1987)
proposed a technique for the design of a robust pole placement controller.
Rotstein (1990) proposed a computational method to design a robust
characteristic polynomial placement controller. Several results have appeared
in the literature which aims at reducing test of Hurwitz stability of entire
family to a small subset of entire family.

6.2.1 Interval Polynomials and Kharitonov’s Theorem

Consider a family F of real rational polynomials

 (s) = 1 s   2 s 2  ....... n 1s n 1   n s n (6.1)

where,  i  [ x i , x i ], (i  1, 2,..., n) , such a family of polynomial is called family


of uncertain or interval polynomial.
81

According to Kharitonov’s theorem every polynomial in the family


 (s) is Hurwitz if and only if the following four extreme polynomials are
Hurwitz and Kharitonov (1979).

K1P ( s )  x0  x1s  x 2 s 2  x3 s3  x 4 s 4  x5 s5  ...,

2 ( s )  x  x s  x s 2  x s3  x s 4  x s5  ...,
KP 0 1 2 3 4 5
(6.2)
3 ( s )  x  x s  x s 2  x s3  x s 4  x s5  ...,
KP 0 1 2 3 4 5

4 ( s )  x  x s  x s 2  x s3  x s 4  x s5  ...,
KP 0 1 2 3 4 5

6.2.2 Stability of Interval Matrices

The structured parametric uncertainties are represented at variations


in elements of system matrix A. Let
.
x = Ax + Bu (6.3)
be the state space representation of a system then the system is an interval
system if
A = {A = [a ] : a  [a ij , a ij ]}
ij ij
(6.4)
= {A = [a ] : a ij  a  a ij}
ij ij

where x n , u m , A   nxn


, B nxm

An interval matrix A is a real matrix in which all the elements are


known to the extent that each belongs to a specific closed interval. In precise
terms, an n x n interval matrix A is a set of real matrices as defined in
equation (6.4). The stability of interval matrices can be tested using various
82

condition Bialas (1983), Heinen (1984) and Lin et al (1988). Lin (1988)have
given comparatively simple conditions to determine stability of interval
matrices.

6.2.3 Bilas Criterion

A necessary and sufficient condition for stability of interval


matrices is given by Bialas (1983). The condition involves testing of stability
of various matrices formed from lower and upper bounds of given interval
matrix. The implementation of this test requires determining characteristic
2
polynomials of 2n matrices of dimension n x n. For example for 2 x 2
matrices, it involves checking stability of 16 characteristic polynomials. For
3 x 3 matrices, 512 characteristic polynomials are required to checking the
stability of those matrices. Therefore this test involves much computational
complexity.

6.2.4 Gershgorin’s Theorem

Various relationships between the elements of a matrix and the location of its
eigen values have been determined .One such relationship is Gershgorin’s
theorem by Meyer, C.D(2004) which states that for an n x n matrix A, every
eigenvalue λ satisfies at least one of the conditions
n
│λ - aii│  ∑ │aji│i,j=1, 2, 3………n. (6.5)
j 1
j i

n
│λ-aii│  ∑│aji│i,j=1,2,3………n (6.6)
j 1
j i
83

That is the eigenvalues of a matrix lie in the closed circular regions


centered at the diagonal elements of the matrix .as seen the theorem provides
actual numerical bounds on the distances to these diagonal elements. Thus
this theorem gives radius of circle around diagonal elements in which
eigenvalues lie.

6.2.5 Heinen’s Criterion

It is possible to study location of the eigenvalues of the matrices


belonging to an interval matrix based on Gershgorin’s theorem. This criterion
is discussed by Heinen (1984). The basic idea lies in determination of a worst
case matrix A€A and if this A is stable then every A€A is stable. The worst
case A is clearly such that bounds on circular regions are as far to the right as
possible in the complex plane. With a little thought, it can be seen that the
‘worst case’ A is thus given by
W= [wij]
where
wij =cii

i, j=1, 2, 3 ………n (6.7)

Wij= max {|b ij|, |cij|}

i, j=1, 2, 3………n
i≠j

Applying Gershgorin’s theorem to w, it is seen that all the Eigen values of W


must satisfy at least one of the conditions
n
│λ – wii │≤  wij i, j=1, 2, 3……….n (6.8)
j 1
j i
84

Thus the eigenvalues are bounded on the right by the quantities,


n
Mi = wii + w
j1
ij i, j=1, 2, 3……....n (6.9)
ji

These must each be negative to guarantee that each of the λ’s lies
strictly in the left half plane ,thus leading to the stability of W, and hence A.
This result is summarized as follows

The interval system matrix A is stable if following condition is


satisfied,
n
a ii   max(| a ij |,| a ij |)  0 ( i=1, 2, 3….n ) (6.10)
j1
j i

An alternative form of this result gives alternate expression of


Heinen criterion as follows

The interval matrix A is stable if ,


n
a ii   max(| a ji |,| a ji |)  0 (i,j=1,2,3……….n) (6.11)
j1
j i

Stability conditions given above are sufficient and hence if these


conditions are not satisfied, we can not comment on stability. It is also
observed that the above conditions have no chance of being satisfied unless
each of the diagonal elements of W (namely the a ii ) is negative.
85

6.3 CONTROLLER DESIGN USING INTERVAL


POLYNOMIALS

The controller is designed using interval polynomial for the model


given in equation (2.9) to reduce the amplitude of vibration at first resonance.
The closed loop system with controller is shown in Figure 6.1. Let G(s) be the
transfer function of model in equation (2.9). The coefficients of the numerator
and denominator of the transfer function are varying independently in
prescribed intervals. The transfer function G(s) in interval form is

+ C(s) G(s)
Controller Transfer
__ function

Figure 6.1 Closed Loop System

n 1s  n 0
G(s):= { G ( s )  2
(6.12)
s  d 1s  d 0

n 0  [0.5497, 1.0464]
n1  [0.2768,  0.2748]
d 0  [2095.8, 2096.8]
d1  [0.4, 1]

The characteristic polynomial of the closed loop system is shown in


figure 6.1 is
2
(k , s)  s  ( d1  k n1 ) s  ( d 0  kn 0 )

 
(6.13)
1 ( k ) 0 (k )
86

The controller is a constant gain, C(s) = k that is to be adjusted, if


possible, to robustly stabilize the closed loop system. It is possible to
determining the range of values of the gain k  [,  ] for which the closed
loop system is robustly stable, i.e. stable for all G(s) G(s).

Since the parameters d i , n j , i = 0,1,2, j = 0,1,2 vary independently it

follows that for each fixed k, (k,s) is an interval polynomial. Using the
bounds given in equation (6.13) the following coefficient bounds for positive
k is obtained.

δ1 (k)  [0.4 - 0.2768k, 1 - 0.2748k]

δ 0 (k)  [2095.8 + 0.5497k, 2096.8 + 1.0464k]

Since the leading coefficient is +1 the remaining coefficients must


be all positive for the polynomial to be Hurwitz. This leads to the constraints:

0.4 - 0.2768k > 0


(6.14)
1 - 0.2748k > 0

From the kharitonov’s theorem applied to second order interval


polynomials it can be easily shown that to ascertain the Hurwitz stability of
the entire family it suffices to check in addition to positivity of the
coefficients, the Hurwitz stability of only the second kharitonov polynomial
K 2 (s). In this problem, the entire system is Hurwitz if and only if in addition to

the above constraints (6.14).

inf{δ1 (s) • δ 0 (s)}  sup{δ0 (s)} (6.15)


87

Based on the Cauchy’s inequality method (6.15), the following


inequality equation is obtained:

(0.4 - 0.2768k)(2095.8 + 0.5497k)  (2096.8 + 1.0464k)

It follows that the closed loop system is stabilized if and only if


K[-15.7818,0.00569]

The open and close loop step response obtained in simulation for
lower and upper bound systems are shown in Figures 6.2 and 6.3.

Open Loop
Closed Loop

Figure 6.2 Open and Closed Loop Response of Lower Bound System
88

Open Loop
Closed Loop

Figure 6.3 Open and Closed Loop Response of Upper Bound System

6.4 CONTROLLER DESIGN USING SIMPLEX METHOD

The controller is designed using simplex method for the model


given in equation (2.9) to reduce the amplitude of vibration at first resonance.
Let the linear state feedback control u = - kx be applied to the system, where
k is a constant state feedback matrix. The closed loop system becomes

x  ( A I  BI k)x (6.16)

The state feedback gain can be found so that the closed loop system
matrix ( A I  B I k) is stable. The interval matrix ( ( A I  B I k) is stable if
n
d i i   m a x (| c ij |, | d ij |)  0 fo r ( i  1, 2 ...n )
j1 (6.17)
j i
89

Or
n
d ii   m ax (| c ji |, | d ji |)  0 fo r (i  1, 2 ...n )
j1 (6.18)
j i

where dii is the diagonal upper bounds of ( A I  B I k) and Cij , dij are the off

diagonal lower and upper bounds respectively of ( A I  B I k) . This will lead to


inequalities involving the unknown elements of k. These inequalities can be
solved to obtain the state feedback gain. Let
- + - + - +
[a11 , a11 ] [a12 , a12 ] ... [a1n , a1n ] 
[a-21 , a+21 ] - +
[a 22 , a 22 ]
- +
... [a 2n , a2n ]
AI =  
. . . . (6.19)
[a- , a+ ] - + - + 
 n1 n1 [a n2 , a n2 ] ... [a nn , ann ]

- + - + - +
[b11, b11] [b12 , b12 ] ... [b1m, b1m]
[b-21, b+21] - +
[b22 , b22 ]
- +
... [b2m, b2m]
BI =  
. . . . (6.20)
[b- , b+ ] - + - + 
 n1 n1 [bn2 , bn2 ] ... [bnm, bnm]

be the system and input matrices of a given interval system.

Let k

 k11 k12 ... k1n 


 k 21 k 22 ... k 2n 
k 
 . . . .  (6.21)
 k km2 ... kmn 
m1
90

be the state feedback gain matrix. The sign of K ij for ( i=1..m, j=1..n) is not
known in this problem. Every real numbers can be expressed as the difference
of two nonnegative numbers. Hence the state feedback gain is expressed as

 k11  k11 k12  k12  k1n  k1n 


       
   k 21  k 21 k 22  k 22  k 2n  k 2n 
k k k (6.22)
     
       
 k m1  k m1 k m2  k m2  k mn  k mn 

where k ij  0, k ij  0for(i  1,...m, j  1,...n)

The following steps are required to design a state feedback gain ;

1. Apply Heinen’s condition to closed-loop system matrix


( A I  B I k) to get the constraints on k for ( i=1..m, j=1..n),
i,j

2. Convert the minimization of the objective function


m n


i 1 j1
kij with the above constraint into linear programming

problem in standard form,

3. Solve the linear programming problem by simplex method to


get the optimum state feedback gain

The state feedback gain obtained in this way is


K   -0.322 -0.432 .

The open and close loop step response obtained in simulation for
lower and upper bound systems are shown in Figures 6.4 and 6.5.
91

Open Loop
Closed Loop

Figure 6.4 Open and Closed Loop Step Response of Nominal System

Open Loop
Closed Loop

Figure 6.5 Open and Closed Loop Step Response of Lower Bound System
92

Open Loop
Closed Loop

Figure 6.6 Open and Closed Loop Step Response for Upper Bound
System

6.5 CONCLUSION

The state feedback controller to suppress the first vibration mode of


a piezo actuated cantilever beam is designed using interval polynomial and
simplex method. Heinen’s condition has been invoked to the closed-loop
system matrix. The minimization of certain objective function subject to
constraints has been converted into linear-programming problem. It has been
solved to obtain the state feedback gain using simplex method.

The controller is designed for the second order internal model using
interval polynomial approach, in order to reduce the amplitude of vibration at
first resonance. The transfer function in interval form is constructed and the
characteristic polynomial of the closed loop system is identified to robustly
stabilize the closed loop system using interval polynomial. The performance
of the controller has been evaluated through simulation.

Potrebbero piacerti anche