You are on page 1of 6

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/256430541

Pole placement and compensator design of generalized systems

Article  in  Systems & Control Letters · January 1987


DOI: 10.1016/0167-6911(87)90028-4

CITATIONS READS

48 483

3 authors, including:

Yue-Yun Wang
General Motors Company
88 PUBLICATIONS   934 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Yue-Yun Wang on 02 January 2019.

The user has requested enhancement of the downloaded file.


Systems & Control Letters 8 (1987) 205-209 205
North-Holland

Pole placement and compensator design


of generalized systems
Yue-Yun WANG, Song-Jiao SHI, Zhong-Jun ZHANG
The Department of Automatic Control, Shanghai Jiao Tong University, People’s Republic of China

Received 22 April 1986


Revised 13 August 1986

Abstract: The system considered is a generalized system E.? = Ax + Bu, y = Cx with matrix E singular. When the impulsive modes
of the system are both controllable and observable, it is shown in this paper that these modes can be eliminated by almost any
constant output feedback. For a strongly controllable and observable generalized system, after its impulsive modes are eliminated, the
poles of the system can be assigned arbitrarily by state feedback. Even if the states are not available, the paper adopts an efficient
method to design a compensator for the system.

Keywords: Generalized system, Output feedback, Strong controllability, Strong observability, Pole placement, Compensator.

1. Introduction

Consider the time-invariant generalized system


E2=Ax+ Bu, y=cx, 0)
where the matrices E, A, B, C are respectively r X r, I: X r, r X m, IX r dimensional and E is singular.
Assume invertibility of SE -A so that (1) has unique solutions for all Ex(O-) and u(t). Let n = degree of
1SE - A 1, k = rank E with n G k; then in system (1) there exist n exponential modes and k - n impulsive
modes. Correspondingly, the free-state responses of system (1) will exhibit exponential motions as well as
k - n impulsive motions [5].
According to [5], if both exponential modes and impulsive modes of (1) are controllable,(or observable),
the system is termed strongly controllable (or observable). For a strongly controllable system (l), imI&ive
modes may be transformed into exponential modes by state feedback, and the poles of the closed-loop
system can be assigned arbitrarily [l]. However, in many practical cases, the states are not available and
the design must be further investigated.
In [4], an observer was constructed, which could measure the states of a strongly observable system (1)
in an asymptotic sense, while impulsive motions only appear at the instant t = 0. Any perturbation such as
disturbances may cause a change of the initial state value Z(OJ of the observer, resulting in an error
A =2(0-)-x(0-) at I = O-.
In this case, although a control u = Fx + u may eliminate impulsive motions,. the control u = FZ -I- u
does not, where 2 is the states of the observer.
For this reason, a compensator design procedure is expounded in this paper. This compensator not only
eliminates impulsive modes, but also assign the closed-loop poles to the prespecified points.

2. Pole placement

Throughout this paper, system (1) is assumed to be strongly controllable and observable, so that. it
should pass the following two tests for strong controllability.

0167-6911/87/$3.50 0 1987, Elsevier Science Publishers B.V. (North-Holland)


206 Yue-Yun Wang et al. / Pole placement and compensator design

Test 1. All the exponential modes of (1) are controllable if and only if
rank[sE-A B]=r forallfinites. (2)

Test 2. Given system (l), use an allowed transformation defined by [5] to bring the pencil [SE - A, B] to
the form
[SE, -A, A, B] with E, of fullcolumnrank. (3)
Then all the impulsive modes are controllable if and only if [5]
[E, A, B] has full row rank. (4

From the principle of duality, strong observability can be checked by the duals of Test 1 and 2.
Since rank E = k < t, without loss of generality, it is assumed that the matrices E, A, B, C in (1) have
the special form

(5)
as they can always be brought to this form by an allowed transformation of the kind (PEQ, PAQ, PB, CQ)
with P, Q nonsingular.
In terms of (5), we rewrite (1) as
k1 = A,,x, + A+, + B,u, 0 = A,lx, + A+, + B2u,
y = c,x, + c,x,. (6)

In (6) if A, is nonsingular, the system is called normal. In this case the generalized characteristic
polynomial is

“k - A,, - A,2
= I-A,,1 1% - A,, + A,,&% 1 (7)
-41 -42

where the polynomial degree n = k, so there exists no impulsive mode in the system. When A,, is singular,
the system has k - n impulsive modes. This unnormal case will be investigated in the following
discussions.

2.1. The effect of output feedback on impulsive behavior

Since (1) is both strongly controllable and observable, from impulsive controllability Test 2 and its
duality test, it follows

which means that [A,,, B2] is of full row rank, and [A&, CzlT is of full column rank. When

[sy ;] -
(9)

is vietied as a system matrix in regular state-space form, it is seen that the zero eigenvalues of singular A,,
are both controllable and observable modes of this system. There are two possibilities for (9). For the first,
the-following theorem was proved in [2].
Yue-Yun Wang et til. ( Pole placement and compensator design 207

Theorem 1. If the system X(A,,, B,, C,) is completely controllable and observable, then for any finite
number of points p p ( pl,. . . , pi) on the complex plane C, the set
F*={FIFER~~/ and a(A,, + B,FC*) np #a} (10)
is either an empty set or a hypersurface in Rmxr.

For the second possibility, system (9) may not be completely controllable and ‘observable. Using
Kalman’s canonical decomposition, we can separate system (9) into four parts with the controllable and
observable part denoted by (AzXc,,, BzcO,C,,). It is well known that output feedback u = Fy merely
affects the controllable and observable modes of system (9), while the set of uncontrollable and/or
unobservable modes of (9), represented by a( A&, remains invariant under output feedback. Thus we have
44, + B,FCd = (~(&co + BzcoFCzco) W u(A,), from which a corollary of Theorem 1 is deduced.

Corollary 1. Take any finite number of points p A ( pl,. . . , pj) on the complex plane C with p n u(A,) = J3.
The set
F*={FIFER~XI and u(A,+ B,FCz) np#fl} WV
is either an empty set or a hypersurface in Rmx’.

Now substitute u = Fy + v into (6); one gets


A
1, = A,,x, +,&x2 + B1v, 0 =&xl +,&x2 + B1v, (11)
Y=c,xl+qx,, (11)’
where ail = A,, + B,FCl, & = A,, + B,FC,, dzl = A,, + B, FC,, dz2 = A,, + B, FCz. As the zero eigen-
values of A,, @ u(A,,), take p = 0, p n u(A,,) =fl. From Theorem 1 and its corollary, it follows:

Theorem 2. For almost any F, p n a(A, + B, FC,) = fl, where p = 0.


The above theorem shows that impulsive modes of the system can be eliminated by almost any F, for
AI, is now nonsingular.

Remark. If u = Fx, + v, instead of output fetdback, is substituted into (6), one has AZ = A,, -t B2F. From
Theorem 2 where C, = 1, it is known that A,, is nonsingular for almost all F. Therefore, state feedback
can also eliminate impulsive modes.

2.2. Pole placement

After eliminating impulsive modes, the system can be decomposed into two parts, namely a regular
state-space subsystem and a static subsystem. Let

02)
Substitute x = Nz into (11) and (ll)‘, and premultiply both sides of (11) by M (this kind of transformation
does not alter the structural properties of the system). (11) and (11)’ transform into
i, = Aozl + BOv, 0 = z2 + Bzv, (1%
y = C,z, + DOv, (14).
where A,, --ill - &d,;lA^,,, B, = B, - &d&lBz, C, = Cl - C,&li&l, D, = -C-&&. The vector z1
represents the dynamic state variables, and the vector z2 represents the static state variables.
208 Yue-Yun Wang et al. / Pole placement and compensator design

Since the system is strongly controllable and observable, it must satisfy Test 1 and its duality test, that
is

rank d--A,
0
0

-I
B,
B,
1 1
=rank d-A,
0 B,
0
0
I=rsI rank
d-A,
0
[ 1 0 -I =r. 05)
co 0

Equality (15) shows that the state-space subsystem Z(A,, B,,, C,) is controllable and observable. Now the
eigenvalues of A, consist of two sets. One set is obtained by transforming k - n impulsive modes into
exponential modes with output feedback u = Fy + u, and another set is the original n exponential modes
of system (1).
Since (A,, B,) is controllable, there exists KE Rmxk, such that a(A, + B,K) can be assigned to the
specified points on the complex plane C. Let Fl = [K, O]W’ and substitute u = F,x into (11). It follows

Ei= (A + BFC+ BF,)x (16)

and

1SE - (A + BFC + BF,) 1 = t,b ‘I -_“;-KBo” “I =\ClI.SI-A,-B,KI (17)


2

where J/ is a constant. Thus, the poles of the generalized system can be assigned to the specified points by
state feedback.

3. Compensator design

When the states of the system are not available, which implies that it is not possible to uniquely
determine zi from equation (14), one may design an observer as

~~=(A,+LC,)i,+(B,+LD,)o-Ly and i,= -B2u, 08)

where a(A, + X0) can be assigned arbitrarily because (A,, Co) is observable. The error zi - z^i satisfies

~(zl-~l)=(Ao+LCo)(zl-il). 09)

Let the measured states be 2 = Ni. If we substitute u = FIX instead of u = F,x into (ll), the overall system

Fig. 1.
Yue- Yun Wung et al. / Pole placement and compensator design 209

will be represented by Figure 1 and the closed-loop system is described by the following equivalent
equation:

I[1
0 21
0 i1 (20)
I -72

I[ 0I[A,
and the similarity transformation

-I I -LC,,
BOK
A,+LC,+B,,K I[ 1
I
I
0
I =
A,+ B,K
0
B,,K
A,, + LC,
shows that the poles of the closed-loop system correspond to the eigenvalues of A, + B,K and A, + LC,,.
(21)

4. Conclusion

For a strongly controllable system (l), it was shown in [l] that, under state feedback, the poles of the
system can be assigned arbitrarily. However in that paper, the computational approach to pole placement
was not the most efficient, because it related to decomposing system (1) into the following two subsystems:
R, = A,x, + Blu, Ji, =x2 + B2u
where J is rhlpotent.
As we know, this decomposition is more difficult to carry out, but the special form (6) is easier to
obtain. For this reason, our pole-placement approach is simpler than that of [l].
Even though the states are not available, if system (1) is both strongly controllable and observable, we
may design a compensator for the system. This compensator not only eliminates impulsive modes, but also
assigns the closed-loop poles to the specified points on the complex plane. Finally, we conclude this
section by summing up the compensator design procedure as follows.
(1) Choose F at random and compute
1
A,, = A,, + B, FC, ,' Al2 = A,, + B,FC,, A^21=A,, + B,FC,, 222=A22+~2~c.
(2) Compute
A, = A;, - A;2&1A21, B,= B, -zi12&1B2, co = Cl - c2a;1d421, D,,= - C2&'B2.
Choose K and L to assign the eigenvalues of A, + B,,K and A, + LC, to the specified points on the
complex plane C.
(3) Construct the compensator
B,=(A,+Lc,)~~+(B,+LD,)u-my, u = Ki; + Fy:

[l] D. Cobb, Feedback and pole placement in descriptor variable systems, Internar. J. Control 33 (1981) 1135-1146.
[2] E.J. Davison, Decentralized stabilization and regulation in linear multivariable system, in: T. Ho and S. Mitter, Eds., Directions in
Decentralized Control (Pergamon Press, Oxford-New York, 1976).
[3] T. Kailath, Linear Systems (Prentice-Hall, Englewood Cliffs, NJ, 1980).
[4] M. El-Tohami, V. Lovass-Nagy and R. Mukundan, On the design of observers for generalized state-space systems using singular
value decomposition, Internat. J. Control 38 (1983) 673-683.
[5] G.C. Verghese, B.C. Levy, T. Kailath, A generalized state-space for singular systems, IEEE Trans. Automat. Control 26 (1981)
811-831.
[6] A.J.J. Van Der Weiden and O.H. Bosgra, The determination of structural properties of a linear multivariable system by
operations of system singularity 2. Non-proper systems in generalized state-space form, Internat. J. Control 32 (1980) 489-537.

View publication stats

You might also like