m o r d i n w ot thi

bsnrtlun Contml Conlenno

TP14 = 5:20 Ikltlmon,Mayland June 1994
1994 ACC
STABILIZATION OF UNCERTAIN SYSTEMS
USING UNITY RANK OUTPUT FEEDBACK AND THE EDGE THEOREM

M.B. Argoun and M.M. Bayoumi

Dept. of Aerospace Engineering Dept. of Electrical Engineering
Cairo University, Giza,Egypt Queen's Univ. ,Kingston,Ont.,Canada.

ABSTRACT of the available freedom but this relationship is made

A method is presented for robust stabilization of uncertain linear. Specifically let G (s,q) be the uncertain p r r open
systems using unity-rank output feedback. The coefficients loop transfer function: O

of the characteristic polynomial are assumed to depend Go(s,q) =l/Ao(s.q)[ T (s,q) I (4)
linearly on interval parameters. The feedback is restricted
to allow linear dependence of t h e closed-loop characteristic and let the output feedback be given by the unity-rank
polynomial on t h e elements of t h e feedback matrix. T h i s matrix F E Rrxp, where:
r e s u l t s in a bilinear polynomial dependence problem which is F = H K (5)
solved using t h e Edge Theorem. and H E R ' ~ ~ , K E R ' ~respectively.
~, Then the closed loop char-
1. Introduction acteristic polynomial Ac(s.q) is related t o the open loop
Most of the work on robustness of uncertain polynomials has characteristic polynomial A (s,q) by, [SI:
addressed the analysis aspects of the problem, with Ac(s,q) = Ao(s,q) + Klip[T (s,q)lpxrHrxl (6)
relatively few papers discussing synthesis. In terms of
analysis, Kharitonov's theorem 111, gives conditions f o r The relationship (6) is bilinear in the elements of the
robust stability of interval polynomials, while the Edge compensator. To simplify matters, we choose the elements of
Theorem, I21, provides the necessary tool f o r robustness H (or K ) t o be fixed. This obviously causes some loss of t h e
analysis of polynomials with linearly dependent coeffic- available freedom in selecting t h e elements of t h e feedback
ients. More recently, attention has been directed towards matrix F. On the other hand, we get a simple relationship
developing synthesis procedure f o r incorporating these and f o r Ac(s,q), which depends linearly on the elements of K,
other robustness results in the design of uncertain control and bilinearly on K and q. From ( 6 ) . if H is chosen t o have
systems I3.41. -
unitv elements (or any other apriori assigned elements), t h e
In this paper we make use of t h e result in [21 to design closed loop polynomial Ac(s,q) can be written as:
output feedback robust compensators f o r a general MIMO un-
certain system with perturbations given on the elements of
p,(s.q)l
the s t a t e matrix A, and occurring linearly in the character- (7)
istic polynomial. We r e s t r i c t the compensator itself t o be a
unity-rank proportional output feedback. This preserves the where the coefficients of .-
g,@,q,,i=l,2 ,...p, depend
4 linearly
linear nature of the dependence of t h e closed loop poly-
nomial coefficients on the elements of the feedback on the perturbation parameters q.
compensator. A procedure is then given t o obtain a class of 3. S t a b i l i z a t i o n p r o c e d u r e
compensators which will stabilize the uncertain system. The
remaining freedom can then be used t o optimize the According t o t h e Edge Theorem [21, t o stabilize an entire
Compensator or t o Utilize i t t o achieve an additional family of polynomials whose coefficients depend linearly on
objective such as pole placement. A detailed example is interval parameters, i t is necessary and sufficient t o
given t o illustrate the procedure. stabilize the exposed edges of t h e polytope of polynomials
formed by t h e interval parameters taking their extreme
2. Problem f o r m u l a t i o n a n d development values. Our basic idea in t h i s p a p e r is as follows:
Consider the state space system If we find the s e t of feedback compensators K i which
stabilizes each e d g e of the polytope of polynomials
kt)=Ax(t)+Bu(t), yCt)=Cx(t) (1)
Ac(s,q), then t h e allowable compensators will lie in t h e
where x is a s t a t e vector, x(t)ERn,AERnX", BERnXrand CdRPxn.
Let some of the elements of t h e matrix A, be subject t o intersection of all t h e edge-stabilizing sets. From t h a t
unknown but bounded perturbations q; i.e. allowable set we choose our compensator t o satisfy some
criterion such as pote placement of t h e nominal system.
A:=A(q), qeQ ,{q:q1,q2...q,, g,~q1~?4,. i=I.Z,..m} (2)
To check f o r stability of a n edge we can use one of several
where, zi and 7,. ,...
i=1,2 m, a r e known lower and upper methods I2,61. Our purpose in this paper i s t o develop a
constructive synthesis procedure f o r robust stabilization
bounds on the perturbations. I t can be easily verified t h a t based on these results. Note t h a t if we treat the elements
if the perturbations occur in elements in t h e same row or of K as constants in (71, then f o r a specific choice of the
column of t h e m a t r i x A, t h e coefficients of the compensator K, the stability of t h e uncertain polynomial is
characteristic polynomial and of the numerator polynomials guaranteed if the edges are Hurwitz. The objective of t h e
will be linearly dependent on t h e parameters 9,. Let this be synthesis procedure thus becomes t h a t of selecting X such
the case, and let in addition, t h e characteristic polynomial t h a t this is satisfied. Substituting t h e extreme values of
be nonrobust, i.e., unstable f o r some values of t h e the parameters in ( 7 ) we obtain t h e vertex polynomials of
perturbations q E Q. Let u s use t h e output feedback law: Ac(s,q);Yl(s,K),Yz(s,K),.., Vl(s,K), which are themselves
u(t) = -F y(t) (3) polynomials with linear depen- dence on the elements of K.
This allows us t o use the Edge Theorem to find t h e range of
Then, we address the following problem: K which preservs t h e stability of t h e vertices. This,
Find a computationally tractable way to obtain the feed-
however, is not sufficient f o r the stability of the edges.
back matrix F s u c h that t h e closed loop characteristic Let us express an edge polynomial Y connecting the two
polynomial is robust for all values of q E Q. 1J
One of t h e most simple and yet powerful methods of design vertex polynomials Y ( s , K ) and Y(s.K) as:
of output feedback compensators is the unity-rank feedback J
method [SI. The main advantage of the method is t h a t the V i J( s . K . h ~ = ( J - h ) Y i ( s , K ) + h Y J ~ s , K ~h, E t0,ll. (8)
relationship between the closed loop characteristic poly- Ideally, we would like t o find K t h a t satisfies ( 8 ) f o r
nomial and t h e open loop characteristic polynomial is h E [0,11.However, the condition on the elements of K t o
bilinear in the elements of the feedback matrix F. By satisfy ( 8 ) would be algebraically difficult t o solve. W e
restricting the elements of F in a certain way, we lose some opt instead f o r the following computer assisted procedure.
2348
i ) Using the Edge theorem f i n d the sets K t which stabilize Now, f o r K to stabilize the closed loop system, it has to
stabilize the exposed edges of the polytope formed by (16).
the vertex polynomials Vl,i=1,2 ,...,1. This is done f o r each To determine K which achieves this objective, we f i n d K
vertex polynomial by finding the extreme values of the which simultaneously stabilizes the vertex polynomials
elements of K f o r which the polynomial has roots on the V1(~1,~2), V2(g1,q2), V3(ql.q2) and V4(ql.q2). Since these
imaginary axis. Since the vertex polynomials are polytopic polynomials are linearly dependent on k and k
in K , we can easily plot the regions of K f o r stability o f 2' we can use
the vertices. the Edge theorem to map the regions in which K should l i e in
i i ) Find the admissible set Xu f o r the stability of all the order t o preserve the stability of the vertex polynomials.
These regions as well as their intersection are shown in
vertices f r o m the intersection of the sets Ki,i=1,2,..,1; Figure (1). The compensator K would be chosen to s a t i s f y
i.e.: X = IK: K E n XI, i=1,2,.,,11 (9) some specific criterion such a s pole placement of the
nominal system. In this example, K i s chosen arbitrarily to
i i i ) select the design values Xde Xu, and check f o r the
be K=C-2,41. Checking the stability of the edges f o r this
stability of all edges. compensator, they are all found stable.Thus this compensator
iv) i f an edge has one or more unstable points, Select an stabilizes the uncertain system f o r all values of the
unstable point on the edge and call it the new vertex Vli1. uncertainty q E Q.
Find K 1 + l that stabilizes this vertex. 5. Conclusion
v ) Find the new admissible set A method is presented f o r the design of a unity-rank output
X u + , = IK: K E n XI, i=1,2,..,1+1) (10) feedback compensator t h a t robustly stabilizes an uncertain
. _
system with linear polynomial dependence on interval
iv) Select a new compensator in this set and repeat the parameters. The method exploits the special structure of the
steps. unity-rank compensator t o produce bilinear closed-loop
Remarks: 1) The original set Ku is the s e t of compensators dependence on t h e uncertain parameters and t h e elements of
t h a t guarantees t h e necessary condition ,i.e., the stability the compensator. This allows the use of the Edge theorem t o
o f all the vertices. So i t is guaranteed t o be t h e largest characterize the s e t of all possible robust stabilizing
possible set f o r K . compensators.
2) Every time an unstable point i s added, i t cuts a bad
portion of t h a t admissible region. The portion cut is gum-
anteed t o be unstable f o r some point in the uncertainty
domain. So, as we proceed, t h e admissible region g e t s
smaller, with more likelihood of stabilizing the edges. In
this sense, the method i s guaranteed t o find a stabilizing
compensator, if one exists. The next example illustrates t h e
synthesis procedure.
4. Example
The uncertain system i s given by:

q, E [-LZI and q, E [-1,ZI (12)

The open loop system i s non-robust in this range. The un-
certain open-loop t r a n s f e r function i s given by (41, where:
the elements of T(s,q) are:
t (s,q)=-s-3+q1-0.5q 2.t 12(s,q)=-(s2+(3-q )S+(0.25-o.5ql-q2)
K1
Ao(s,q)=s3+(3.5-ql)s2+ (4- 1.5 4,-q )s +(6.5-3 q i0.25 4,) Figur? (1) : Regions of robusi stability
Let the unity rank feedback gain be given by :
REFERENCES:
F=KH=lkl k J [ l j (13)
I l l Kharitonov, V. L., "Asymptotic stability of an equilibr-
ium position of a family of systems of linear differential
Let hl= h2= I , then Ac(s,q) is polytopic in kl and k2. write equations," Dif ferentsial'nye Uravneniya, 14. 2086-2088.
[21 Bartlett, A.C., Hollot, C . V . , and Lin,H.,"Root locations
this polynomial as, of an entire polytope of polynomials: i t suffices t o check
Ac(s,q)=f o(s.q)+klf I(s,q)+ k 2 f 2(s.q) (14)
the edges", Math. Cont. Sign. Systems,vol 1, 1988,pp. 61-71.
Note that f Js'q), f I(s,q) and f 2(s,q) are polynomials which 131 Bernstein,D. and Haddad, W.,"Robust controller synthesis
using Kharitonov's theorem," I E E E Trans. Automat. Contr.,
depend linearly on q, and 4,. These are given by: Vol. AC-37, no. 1, pp. 129-132, Jan., 1992.
fo~~,q~=(s3+3.5s2+4s+6.5)-q,(S2+1.5s+3)-qz(~ [41 Wei, K., and Barmish, B.R., "Making a polynomial Hurwitz
invariant by choice of feedback gains," Int.J.Contr., vol.
50, n0.4, pp. 1025-1038, 1989.
(15) I51 Seraji, H., "Pole assignment techniques f o r multivar-
iable systems using unity-rank output feedback," Int. J .
Substituting the extreme values f o r q, and q, we obtain the
Contr., vol. 21, no. 6, pp. 945-954, 1975.
following f o u r vertex polynomials: [61 Fu, A. and Barmish, E. R., "Stability of convex and
V1(-I,-I~=(s3+4.5~2+6.5s+9.3~-k~2+5~+2.Z5~+k2~~2+2.5S+~.~~ linear combinations of polynomials arising in robustness
2 problems," in Proc. Conf. Inform. Sci. Syst., John Hopkins
V (-1, 2)=(s3+4.5s2+3.5s+9.9)-kl(s2+5s+0.75~+k2(s +5.5s-3) univ., Baltimore, MD, 1987.
S il.5)
V*(Z,-l) =(~~+1.5s~+2~+0.3)-kl(s~+Zs+O.75)+k~(~~+Z.5
k,(sZ+2s+5.25)+k2(s2+5.5s-3)
V~~Z,Z~=(s3+2.5sZ-s+0.9)- (16)

2349