Proceedings of the 36th FM13 2:30
Conference on Decision & Control
San Diego, Califomia USA December 1997
Robust Controllability and Observability for MIMO in LTI
Systems with Unidirectional Perturbation
Cksar Elizondo' , Joaquin Collado
FIME UANL
Doctorado en Ingenieria Elkctrica
Apdo. postal 139-F, C.P. 66450
San Nicolh de 10s Garza, N.L.
MEXICO
Email: celizond43gama.fime.uanl.m
j
[email protected] Abstract 2 Preliminaries
Abstract. Robust controllability and observability are The controllability of a linear time invariant sys-
studied for multiple input multiple output linear time tem is determined if and only if the controllabil-
+
invariant system, with matrices A = A0 rA1 , B = ity Kalman matrix U 15 Rnxnp is of rank n , U =
+ +
Bo rB1 , C = CO rC1 , where for T = 0 the system [B AB A2B ...A"-lBI , it is equivalent to the ma-
is controllable and observable. In this paper we present trix UU8 be non-singular.
the limits of the open interval of the real parameter r
, in which controllability or observability are preserved For our purpose it is useful the mathematical relation
( all proofs are omited). shown in [2, 31, the proof is omitted but it is based in
partitioned matrix properties shown in [4].
Keywords: Robust controllability, robust observability,
unidirectional perturbation.
-
1 Introduction 0 I 0 0 0 -
0 0 I 0 0
Controllability and observability conditions for LTI 0 0 0 I 0
systems had being reviewed in several cases: in [l]for
single input and single output was obtained the limits 0 0 0 0 I
in physical parameters which controllability or observ- - -
" I-"- -ulN-2 -"-3 '" -m1
ability is lost, the distance from a controllable pair to
the nearest uncontrollable system was obtained in [7]
for multiple input linear systems, uncertainty bounds
for matrix A using B and C fixed matrix are obtained
in [6]for SI systems in controllability and SO systems in The controllability Kalrnan matrix is a function of the
observability. The purpose of this paper is to study the parameter r , for simp!licity we will write U and the
robust controllability and observability in LTI systems, matrix product UU* has in its elements: constants and
the parametric uncertainty is an unidirectional pertur- terms in r , r2 , ... , rZn, in such form that it can be
bation: A=AO+rAl,B=Bo+rBl,C=Co+rCl in expressed as so as matrix polynomial in r of degree 2n
a real parameter r. Given a LTI system with fixed m& and order n .
trix: Ao,Al E P X nBo,& , E R n x p , C0,Ci E %qxn
and a (Ao,Bo) controllable pair and a (Ao,CO)observ-
Proposition 2 (Matriz polynomial of the UU8 prod-
able pair, the problem is to find the limits {rmin,r m a }
where the individual properties are preserved. One ap-
uct) Let: AD, AI E IRnxn Bo, BI E Rnxp, A =
A. + TA1, B = B o + T B ~r, E 92. Let U E IfZnxnp be
proach to solve this problem could be obtained using
the controllability Kalman matrix corresponding to pair
the results shown in [5] but it can only be applied for
single input systems. + +
( A , @ , U = [(Bo rBl)i(Ao rAl)(Bo rBl)i(Ao + +
lPartially sponsored by Conacyt. rA1)2(Bo+ rB1);.. . i(& + rAl)"-'(Bo + rB1)],then
0-7803-3970-8197 $10.00 0 1997 IEEE 4369
the matrix product UU* can be expressed b y a matrix 4 Example
polynomial whose degree is 2n and order n: UU* =
+ + + +
MO rM1 r2M2 . . . T ~ ~ M ~Mi, ,E X n x n . Given the matrices A0 = [0,1,0,0;0,0,1,0;0,0,0,1;-2,-
4,-3,-1], Bo = [1,0;0,1;0,1;-1,1], A1 = [-1,1,0,0;0-
The calculation of matrix coefficients is tedious 1,0,0;0,0,0,1;-2,0,-3,0],B1 = [l,-1;0,1;-1,1;-3,1]and the
by multiplication and factorization methods, then it (Ao,Bo) controllable pair. What is the open interval in
is proposed the next procedure. r preserving controllability in the pair ( A ,B ) , where
A = A0 +rA1 and B = Bo +rB1 ?. Solution: With the
given matrices and procedures of this paper it is com-
Theorem 3 Let UU* be the matrix polynomial above puted the matrix M and its eigenvalues: X i , = -1
described, then the matrices coefficients MI, are ob-
and A i m = 0 then -1 < r < 0;) .
tained b y derivative procedures as follows. Mk =
k!
5 Conclusions
The problem of Controllability and Observability for
Where:
MIMO in LTI Systems with Unidirectional Perturba-
tion is solved by means of real eigenvalues of a par-
titioned companion matrix M , conformed by matrix
coefficients of the matrix polynomial UU* . The limits
minimum and maximum of open interval in real para-
meter r for controllability and observability, are deter-
mined by the inverse value of the minimum negative
3 Main Result and maximum positive real eigenvalues of the matrix
M respectively.
The problem is to find the positive and negative limits
for r where determinant of UU*vanish, it is translated
to find the eigenvalues of a matrix M E X N n x N n . Mo- 6 References
tivated by the result in Robust Stability of M. F‘u and
B. R. Barmish we present the following theorem. [l]J. Ackermann, Robust Control Systems with Uncer-
tain Physical Parameters, Springer-Verlag 1993.
Theorem 4 Let A , B be the matrices used in matrix
[2] B. R. Barmish, New Tools for Robustness of Linear
polynomial of the UU* product and let (A0,Bo) be a
controllable pair. Then the open interval in r for con- Systems. Macmillan Publishing Company, 1994.
trollability of the pair ( A , B ) is described as follows: [3] M. Fu, B. R. Barmish, “Maximal Unidirectional
1
rmin < r < rmax, rmin = rm, = -.-L- Perturbation bounds for stability of polynomials and
Xiin(M) &L(M)
matrices“, Systems and Control Letters, Vol. 11, pp.
Where M is generated in according to Characteris- 173-178, 1988.
tic polynomial of block partitioned companion ma-
trix and Matrix polynomial of the UU* product, us- [4] F. R. Gantmacher, “The Theory of Matrices“,
ing mi = MOlMi-1 . The eigenvalues A,, , :A
and Chelsea Publishing Company, 1990.
means negative and positive real eigenvalues of ( M )
[5] T. Tsujino, T. Fujii, K. Wei, ‘‘ On the Connection
respectively. In case that one of them does not exist,
Between Controllability and Stabilizability of Linear
then the nonexistent eigenvalue will be taken as zero.
Systems with Structural Uncertain Parameters “, Au-
In multiple input cases the interval goes whole reals tomatica, Vol 29, No. 1, 1993.
for almost all directions because the set of controllable
[6]K.Wang, A. Michel, “Necessary and Sufficient Con-
systems is simply connected.
ditions for the Controllability and Observability of a
Class of Linear Time-Invariant Systems with Interval
Corollary 5 Let to use A* , C* instead of A , B re- Plant “, IEEE Transactions on Automatic Control, Vol.
spectively in Matrix polynomial of the UU* product and 39, No. 7, July 1994.
let (A0,Co) be an observable pair. Then the open in-
terval in r for observability of the pair ( A , C ) is de- [7] M. Wicks, R. A. DeCarlo, “Computing The D i s
scribed as follows: r,in < r < r, rmin = l
tance to an Uncontrollable System“, IEEE Transac-
1
min(M) ’ tions on Automatic Control, Vol. 36, No. 1, January
rmax =
Gm=x(M) 1991.
4370
