LMI Approach To Output Feedback Control For Linear Uncertain Systems With D-Stability Constraints

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

JOURNAL OF OPTIMIZATION THEORY AND APPLICATIONS: Vol. 113, No. 2, pp.

357–372, May 2002 (2002)

LMI Approach to Output Feedback Control


for Linear Uncertain Systems with
D-Stability Constraints1
Z. WANG2 AND K. J. BURNHAM3

Communicated by E. Polak

Abstract. This paper deals with the problem of designing output feed-
back controllers for linear uncertain continuous-time and discrete-time
systems with circular pole constraints. The uncertainty is assumed to be
norm bounded and enters into both the system state and input matrices.
We focus on the design of a dynamic output feedback controller that,
for all admissible parameter uncertainties, assigns all the closed-loop
poles inside a specified disk. It is shown that the problem addressed can
be recast as a convex optimization problem characterized by linear
matrix inequalities (LMI); therefore, an LMI approach is developed to
derive the necessary and sufficient conditions for the existence of all
desired dynamic output feedback controllers that achieve the specified
circular pole constraints. An effective design procedure for the expected
controllers is also presented. Finally, a numerical example is provided
to show the usefulness and applicability of the present approach.

Key Words. Linear systems, dynamic output feedback, norm-bounded


uncertainty, robust control, regional pole placement, LMI approach.

1. Introduction

As is well known, pole location is associated directly with the dynam-


ical characteristics of linear time-invariant systems, and thus one of the most
popular approaches to defining performance specifications is to put some

1
The work of the first author was supported in part by the Alexander von Humboldt Foun-
dation of Germany and the University of Kaiserslautern, Kaiserslautern, Germany.
2
Staff Member, Control Theory and Applications Centre, School of Mathematical and Infor-
mation Sciences, Coventry University, Coventry, United Kingdom.
3
Professor, Control Theory and Applications Centre, School of Mathematical and Information
Sciences, Coventry University, Coventry, United Kingdom.
357
0022-3239兾02兾0700-0357兾0  2002 Plenum Publishing Corporation
358 JOTA: VOL. 113, NO. 2, MAY 2002

constraints on the pole location for the closed-loop transfer matrix. In


designing practical control systems, it may be satisfactory that the closed-
loop poles be in a specified region. More specifically, the exact pole assign-
ment is not required in practice and it is usually sufficient to place the poles
in a prescribed region (often a disk) in the left-half complex plane for
continuous-time systems or in the unit disk for discrete-time systems.
In the past decade, the problem of controller design for assigning all
closed-loop poles within a disk has been an active area of research; see e.g.
Refs. 1–6. Moreover, the problem of robust circular pole assignment sub-
jected to plant parameter perturbations has been studied extensively. The
robust regional (circular) stability of a system subjected to uncertainties was
analyzed in Refs. 7–9. Also, Ref. 10–11 dealt with the controller design for
robust regional pole assignment, but the poles of the nominal system were
required to be located in the specified region, and thus these design methods
were not suitable for the unstable nominal systems. Recently, Refs. 12–14
established the parametrization of robust state feedback or output feedback
controllers for regional pole placement for linear uncertain systems, and
sufficient conditions for the existence of desired controllers were given based
on manipulations of some algebraic matrix equations; Ref. 15 investigated
the HS control with regional stability constraints, but the system uncer-
tainty was not presented; Ref. 16 focused on the robust circular pole assign-
ment for parametric uncertain systems without the assumption of nominal
stability. There, a so-called quadratic d-stabilizability concept was proposed
and then the necessary and sufficient conditions were derived for the quad-
ratic d-stabilizability in terms of a parameter dependent augmented Riccati
equation. It should be pointed out that an appropriate parameter has to be
initialized (turned) when solving such a parameter-dependent Riccati equa-
tion. Subsequently, the theoretical results obtained in Ref. 16 were applied
successfully in the control of a flexible robot arm; see Ref. 17. Unfortu-
nately, only the state feedback case was studied in Refs. 16–17. Also, Ref.
18 tackled the problem of quadratic guaranteed cost control with robust
pole placement in a disk, but the output feedback case was still not inten-
sively included. This situation motivates us to investigate the problem of
designing dynamic output feedback controllers that meet the expected circu-
lar-pole constraints for admissible parameter uncertainty, and if possible,
obtain the necessary and sufficient conditions for the existence of the desired
controllers without any conservativeness.
On the other hand, in the past few years, linear matrix inequalities
(LMIs) have gained much attention for their computational tractability and
usefulness in control engineering (see e.g. Ref. 19) as the so-called interior-
point method (see Ref. 20) has been proved to be numerically very efficient
JOTA: VOL. 113, NO. 2, MAY 2002 359

for solving the LMIs. The number of control problems that can be formu-
lated as LMI problems is large and continues to grow. For example, the
continuous-time and discrete-time HS control problem was thoroughly
solved in Ref. 21 based on LMIs. In Ref. 22, a lot of elegant results were
obtained for robust quadratic dissipative control of linear uncertain systems
in terms of LMIs. Chilali and Gahinet (Ref. 23) studied in detail the design
of state-feedback or output-feedback HS controllers that satisfy additional
constraints on the regional pole location. An LMI approach was utilized in
Ref. 23, and an extension to the uncertain case was also included. However,
the uncertain systems investigated in Ref. 23 were described by a polytopic
state-space model; that is, the system matrices take values in the matrix
polytope, and thus the uncertainties were structured. Up to now, there has
been few papers dealing with robust pole assignment problems subjected
to unstructured norm-bounded parameter uncertainties using the LMI
approach, especially for the discrete-time case.
In this paper, an LMI approach developed in Ref. 21 is exploited to
study the problem of the robust output feedback controller design for un-
certain continuous-time and discrete-time systems with circular pole con-
straints. The parameter uncertainty is allowed to be unstructured but norm
bounded, and it exists in both the system state and the system input
matrices. The aim of this problem is the design of a dynamic output feed-
back controller such that, for all admissible uncertainties, the closed-loop
poles are placed within a specified disk. A unified LMI-based approach,
which is suitable for both linear uncertain continuous-time and discrete-
time systems, is developed to derive the necessary and sufficient conditions
for the existence of all desired dynamic output feedback controllers that
achieve the specified circular pole constraints in terms of LMIs. The solu-
tions of the LMIs, if they exist, can be exploited to construct effectively the
expected controllers. It is worthwhile to mention that the LMI toolbox (see
Ref. 24) is now very powerful to solve LMIs. We provide a numerical
example to illustrate the obtained theory. Finally, we point out that:
(i) the main results developed in this paper introduce no
conservativeness;
(ii) although the uncertainty in this paper is assumed to be norm
bounded, it is not difficult to consider a more general description
of the uncertainty (such as the generalized positive real un-
certainty, Ref. 25) within the same framework;
(iii) no stability assumption is needed in this paper for the nominal
system.
The paper is organized as follows. In Section 2, we formulate the
dynamic output feedback controller design problem for both linear
360 JOTA: VOL. 113, NO. 2, MAY 2002

uncertain continuous-time and discrete-time systems with circular pole con-


straints. In Section 3, first we recall some useful results, then derive the
necessary and sufficient conditions for the existence of the expected control-
lers in terms of three LMIs, and next present an effective approach to con-
structing these desired controllers. An illustrative example is provided in
Section 4 to show the usefulness of the LMI approach developed. Finally,
some concluding remarks are given in Section 5.
The notation is standard. Throughout this paper, ⺢n and ⺢nBm denote
respectively the n-dimensional Euclidean space and the set of all nBm real
matrices. The superscript T denotes matrix transposition and the notation
X¤Y [respectively, XHY], where X and Y are symmetric matrices, means
that XAY is positive semidefinite [respectively, positive definite]. In is the
nBn identity matrix, diag{. . .} stands for a block diagonal matrix, and
Ker(X) refers to the null space of the linear operator X.

2. Problem Formulation and Assumptions

Consider both the uncertain linear continuous-time system

ẋp (t)G(Ap C∆Ap)xp (t)C(Bp C∆Bp)u(t), (1a)


yp (t)GCp xp (t)CDp u(t), (1b)

and the uncertain linear discrete-time system

xp (kC1)G(Ap C∆Ap)xp (k)C(Bp C∆Bp)u(k), (2a)


yp (k)GCp x(k)CDp u(k), (2b)

where xp ∈⺢nx is the state vector, u∈⺢nu is the control input vector,
yp ∈⺢ny is the measured output vector, and Ap , Bp , Cp are known constant
matrices. It is assumed that Dp G0 and that the triple (Ap , Bp , Cp) is stabiliz-
able and detectable. It should be pointed out that the assumption Dp G0
causes no loss of generality, while considerably simplifying the algebraic
manipulations; see Ref. 21. ∆Ap and ∆Bp represent parametric perturbations
in the system state matrix and control input matrix, and are assumed to be
of the following form (see e.g. Refs. 26–27):

[∆Ap , ∆Bp ]GMp F[Np1 , Np2 ], (3)

where F∈⺢iBj is an uncertain matrix bounded by

F TF⁄I (4)
JOTA: VOL. 113, NO. 2, MAY 2002 361

and Mp , Np1 , Np2 are known constant matrices of appropriate dimensions


which specify how the elements of the nominal matrices Ap and Bp are affec-
ted by the uncertain parameters in F. ∆Ap and ∆Bp are said to be admissible
if both (3) and (4) hold.
In this paper, we consider the dynamic output feedback controller with
respect to the continuous-time system (1) described by
ẋc (t)GAc xc (t)CBc yp (t), (5a)
u(t)GCc xc (t)CDc yp (t), (5b)
and the dynamic output feedback controller with respect to the discrete-
time system (2) described by
xc (kC1)GAc xc (k)CBc yp (k), (6a)
u(k)GCc xc (k)CDc yp (k), (6b)
where xc ∈⺢nc, 0⁄nc ⁄nx , and the extreme case nc G0 represents the static
gain state feedback.
Applying the output feedback controllers (5) and (6) respectively to the
systems (1) and (2), we obtain the closed-loop augmented continuous-time
and discrete-time systems as follows:
ẋ(t)G[(AC∆A)C(BC∆B)KC]x(t), (7a)
y(t)GCx(t), (7b)
x(kC1)G[(AC∆A)C(BC∆B)KC]x(k), (8a)
y(k)GCx(k), (8b)
where

冤x 冥 , 冤x 冥 , 冤0 冥 冤0 冥
xp yp Ap 0 Bp 0
x_ y_ A_ , B_ , (9)
c c 0 Inc

冤0 冥 冤B 冥
Cp 0 Dc Cc
C_ , K_ , [∆A, ∆B]GMF [N1 , N2], (10)
Inc c Ac

冤 0 冥,
Mp
M_ N1 _[Np1 , 0], N2 _[Np2 , 0]. (11)

We now consider a stable disk region D(q, r) in the complex plane with
the center at qCj0 and the radius r; that is, D(q, r) belongs to the left-half
complex plane for continuous-time systems and to the unit disk for discrete-
time systems. To this end, we are in a position to state that: the robust
circular pole-constrained design (RCPD) problem under study is to design
362 JOTA: VOL. 113, NO. 2, MAY 2002

dynamic output feedback controllers of the forms (5), (6), or equivalently, to


determine the parameter matrix K such that, for all admissible uncertainties
∆A and ∆B, the poles of the closed-loop systems are assigned inside a pre-
specified disk D(q, r); i.e., the roots of
det[sIA(AC∆A)C(BC∆B)KC]G0
are within the disk D(q, r). In this case, we shall say that the closed-loop
systems (7), (8) are robustly D-stable or possess robust D-stability. The
main purpose of the next section is to characterize the necessary and suf-
ficient conditions under which the uncertain systems (7), (8) are robustly D-
stabilizable via dynamic output feedback control, and then give a practical
design procedure of the corresponding output feedback controllers.

3. Main Results and Proofs

We begin with offering some important lemmas. First, the following


result provides a key for solving Problem RCPD. It is shown that the the
robust D-stability of linear uncertain systems is closely related to the exist-
ence of a positive-definite solution to a quadratic matrix inequality.

Lemma 3.1. Let the disk D(q, r) and the controller parameter matrix
K be given. Consider the quadratic matrix inequality
Aq[(AC∆A)C(BC∆B)KC]TPAqP[(AC∆A)C(BC∆B)KC]
C[(AC∆A)C(BC∆B)KC]TP[(AC∆A)
C(BC∆B)KC]C(q2Ar2)PF0. (12)
Then, the closed-loop systems (7), (8) are robustly D-stable if and only if
there exists a positive definite solution P to (12).

Proof. It is easy to see that the specified D-stability constraint is


equivalent to the Schur stability of the matrix
ϒ_(1兾r) [(AC∆A)C(BC∆B)KCAqI]; (13)
i.e., the eigenvalues of ϒ are all located inside the unit circle D(0, 1). From
the discrete-time Lyapunov stability theory, we know that ϒ is a Schur
matrix if and only if there exists a positive-definite matrix P satisfying the
inequality
ϒ TPϒAPF0,
which is the same as (12). This proves Lemma 3.1. 䊐
JOTA: VOL. 113, NO. 2, MAY 2002 363

Next, we will recall some important matrix inequalities to be used


extensively in the sequel.

Lemma 3.2. Schur Complement Lemma (Ref. 28). Given the constant
matrices M, L, Q of appropriate dimensions, where M and Q are symmetric
and QH0, then MCLTQLF0 if and only if

AQ −1 L
冤 冥 冤 冥
M LT
F0 or F0.
L −Q −1 LT M

Lemma 3.3. Projection Lemma (Ref. 21). Given the matrices


GGGT ∈⺢ mBm, Y∈⺢rBm, and Z∈⺢sBm, there exists a matrix Θ∈⺢rBs
satisfying GCYTΘZCZ TΘTYF0 if and only if the matrix G is positive
definite on ker(Y) and ker(Z), i.e., N TY GN Y F0, and N TZ GN Z F0, where
N Y and N Z are any matrices whose columns form bases of the null spaces
of Y and Z respectively.

The next lemma reveals that the robust stability of a discrete-time


system subjected to norm-bounded uncertainty can be guaranteed by the
existence of the positive-definite solution to a matrix inequality.

Lemma 3.4. See Ref. 27. Let A∈⺢nBn, M∈⺢nBi, N∈⺢ jBn,
QGQ T ∈⺢nBn be given matrices, and let F∈⺢iBj be any matrix satisfying
FTF⁄I. Then, there exists a positive-definite matrix PH0 meeting
(ACMFN )TP(ACMFN )CQF0 if and only if there exists a scalar (H0
such that

AP −1C(MM T A
冤A T
( −1N TNCQ
F0. 冥
For the sake of simplicity, we make the following definitions:
Acq G(AAqInxCnc)CBKC, (14)

冤 冥
−P (1兾r)P(AAqInxCnc) (1兾r)PM 0(nxCnc)Bj
(1兾r)(AAqInxCnc ) P −P
T
0(nxCnc )Bi N T1
GG , (15)
(1兾r)M P T
0iB(nxCnc ) −( −1Ii 0iBj
0 jB(nxCnc ) N1 0 jBi −(I jBj
YG[(1兾r)B TP, 0(nuCnc)B(nxCnc), 0(nuCnc)Bi, N T2 ], (16)
ZG[0(nyCnc)B(nxCnc), C, 0(nyCnc)Bi, 0(nyCnc)Bj ], (17)
T T
XG[(1兾r)B , 0(nuCnc)B(nxCnc), 0(nuCnc)Bi, N ].2 (18)
364 JOTA: VOL. 113, NO. 2, MAY 2002

The following theorem shows that the existence conditions of the


expected robust D-stability output feedback controllers can be expressed in
terms of two matrix inequalities.

Theorem 3.1. Consider the linear uncertain continuous-time or


discrete-time system (1) or (2). Assume that N X and N Z represent the
matrices whose columns form bases of the null spaces of X and Z, respect-
ively. Then, there exists a dynamic output feedback controller being of the
form (5) or (6) such that the closed-loop system (7) or (8) is robust D-stable
if and only if there exists a positive scalar (H0 such that the matrix
inequalities

冤 冥
AP (1兾r)P(AAqInxCnc) (1兾r)PM 0(nxCnc)Bj
T
T
(1兾r)(AAqInxCnc) P −P 0(nxCnc)Bi N T1
N Z −1
N Z F0, (19)
(1兾r)M P T
0iB(nxCnc) −( Ii 0iBj
0 jB(nxCnc) N1 0 jBi −(I jBj
−1

冤 冥
AP (1兾r)(AAqInxCnc) (1兾r)M 0(nxCnc)Bj
T (1兾r)(AAqInxCnc)T −P 0(nxCnc)Bi N T1
N X N X F0 (20)
(1兾r)M T 0iB(nxCnc) −( −1Ii 0iBj
0 jB(nxCnc) N1 0 jBi −(I jBj

have a positive-definite solution PH0 of dimension (nx Cnc)B(nx Cnc).

Proof. The proof technique is along lines similar to those used in Ref.
21. Note that ϒ can be rewritten as

ϒG(1兾r)[(AC∆A)C(BC∆B)KCAqI]
G(1兾r)[(AAqI )CBKC]C(1兾r)MF(N1 CN2 KC). (21)

Lemma 3.1 means that the closed-loop systems (7), (8) are robustly D-stable
if and only if there exists a positive-definite solution PH0 to the inequality
(12) or

ϒTPϒAPF0.

Then, by considering the relationship (21) and Lemma 3.4, we know that
the above inequality holds if and only if there exists a scalar (H0 satisfying

AP −1C((兾r2) MM T
冤 冥
(1兾r)[(AAqI )CBKC]
F0, (22)
(1兾r)[(AAqI )CBKC] ( −1 (N1 CN2 KC)T (N1 CN2 KC)AP
T
JOTA: VOL. 113, NO. 2, MAY 2002 365

which can be rearranged as

AP −1
冤 冥
(1兾r)[(AAqI )CBKC]
(1兾r)[(AAqI )CBKC] −P
T

(1兾r)M T 0 T
(1兾r)M T 0
冤 冥冤 冥冤 冥
(I 0
C F0. (23)
0 N1 CN2 KC 0 ( −1I 0 N1 CN2 KC

It follows from Lemma 3.2 that (23) holds if and only if the following
inequality is true:
AP −1

冤 冥
(1兾r)Acq (1兾r)M 0(nxCnc)Bj
(1兾r)A T
cq −P 0(nxCnc)Bi (N1 CN2 KC)T
F0, (24)
(1兾r)M T
0iB(nxCnc) −( −1Ii 0iBj
0 jB(nxCnc) N1 CN2 KC 0 jBi −(I jBj
where Acq is defined in (14).
Premultiplying and postmultiplying the inequality (24) by
diag{P, I, I, I} respectively yield

冤 冥
AP (1兾r)PAcq (1兾r)PM 0(nxCnc)Bj
(1兾r)A P −P
T
cq 0(nxCnc)Bi (N1 CN2 KC)T
F0, (25)
(1兾r)M TP 0iB(nxCnc) −( −1Ii 0iBj
0 jB(nxCnc) N1 CN2 KC 0 jBi −(I jBj
which, by using the definitions (15)–(17), can be expressed as
GCYTKZCZ TK TYF0. (26)

Let N Y , N Z , N X be respectively the matrices whose columns span the null


spaces of Y, Z, X. We now use Lemma 3.3 to derive the solvability con-
ditions for K meeting (26). It follows directly from Lemma 3.3 that there
exists a solution K to the inequality (26) if and only if
T T
N Y GN Y F0, N Z G N Z F0. (27)

Note that the relationship between NY and NX is


P −1

冤 冥
0(nxCnc)B(nxCnc) 0(nxCnc)Bi 0(nxCnc)Bj
0(nxCnc)B(nxCnc) InxCnc 0(nxCnc)Bi 0(nxCnc)Bj
N YG N X. (28)
0iB(nxCnc) 0iB(nxCnc) Ii 0iBj
0 jB(nxCnc) 0 jB(nxCnc) 0 I jBi
366 JOTA: VOL. 113, NO. 2, MAY 2002

Then, substitute (28) into N TY G N Y F0 to give (20). Also, it is obvious that


N TZ G N Z F0 is just (19). This completes the proof of Theorem 3.1. 䊐

Remark 3.1 It is noticeable that the inequality (19) is linear on P, but


nonlinear w.r.t ( or ( −1, while the inequality (20) involves both P and P −1.
This brings considerable difficulties in practical application. However, we
can derive simpler forms by partitioning P. Specifically, we shall reduce the
conditions to a pair of linear matrix inequalities (LMIs) which can be solved
very efficiently by the Matlab LMI toolbox (Ref. 19).
For simplifying the presentation, we define a linear function of the
positive-definite matrix J∈⺢nxBnx as

f (J )_(1兾r2)(Ap J ATp AqAp JAqJ ATp Cq2InxC(Mp M Tp). (29)

The following theorem provides the necessary and sufficient conditions for
the solvability of the robust D-stability output feedback controller design
problem in terms of three linear matrix inequalities (LMIs).

Theorem 3.2. Consider the linear uncertain continuous-time or


discrete-time system (1) or (2). Let W1 , W2 , W3 denote respectively the bases
of the null spaces of B Tp , N Tp2 , Cp . Then, there exists a dynamic output feed-
back controller of the form (5) or (6) such that the closed-loop system (7)
or (8) is robust D-stable, if and only if there exist a positive scalar (H0 and
two positive-definite matrices J, T∈⺢nxBnx such that the following three
LMIs hold:

W T1 AJCf (J ) Ap J N Tp1
冤 冥冤 冥 冤0 冥
0 W1 0
F0, (30)
0 W T2 Np1 J ATp Np1 J N Tp1 A(I j W2
W T3 (−TC( −1N Tp1 Np1) W3 W T3 ATp T 0

冤T Ap W 3
0
−T
M Tp T
T Mp F0,
−( −1I
冥 (31)

冤I T冥H0.
J I
(32)

Proof. First, partition P and P −1 conformably to A as follows:

冤U 冥 冤V 冥
T U J V
PG T , P −1 G T . (33)
T̂ Ĵ
JOTA: VOL. 113, NO. 2, MAY 2002 367

Next, noticing that


Z_[0(nyCnc)B(nxCnc) , C, 0(nyCnc)Bi , 0(nyCnc)Bj ]

冤0 冥
0nyB(nxCnc) Cp 0nyBnc 0nyB(iCj )
G ,
ncB(nxCnc) 0ncBnx Inc 0ncB(iCj )
X_[(1兾r)B T, 0(nuCnc)B(nxCnc ) , 0(nuCnc)Bi , N T2 ]
(1兾r)B Tp 0nuB(nxCncCi) N Tp2
冤0 冥,
0nuBnc
G
ncBnx (1兾r)Inc 0ncB(nxCncCi) 0ncBj

it is easy to check that NX and NZ are respectively of the form

冤 冥 冤 冥
W1 0 0 InxCnc 0 0
0 0 0 0 W3 0
N XG , N ZG . (34)
0 InxCncCi 0 0 0 0
0 0 W2 0 0 IiCj

Note that the second row of N X and third row of N Z are identically zero.
Now, by using Lemma 3.2, it is immediate to verify the fact that
W T3 (−TC( −1N Tp1 Np1)W3 W T3 ATp T
冤T Ap W 3 −TC(T Mp M Tp T 冥
W T3 (−TC( −1N Tp1 Np1)W3 W T3 ATp T T

冤 冥 冤 冥 冤 冥
0 0
G C (I F0 (35)
TAp W3 −T TMp TMp
is equivalent to the inequality (31). Considering this fact and substituting
both the partition of P in (33) and the second expression of (34) into the
inequality (19), it can be shown that, through some straightforward matrix
manipulations, the inequality (19) is equivalent to the inequality (31). On
the other hand, substituting the partition of P −1 in (33) and the first
expression of (34) into the inequality (20), it can be readily established also
that the condition (20) is equivalent to (30). Hence, P satisfies (19)–(20)
if and only if J, T satisfy (30)–(31). Finally, PH0 and P∈⺢(nxCnc)B(nxCnc)
if and only if J, T satisfy inequality (32); see Ref. 29. This proves the
theorem. 䊐

Theorem 3.2 gives the existence conditions for the robust output feed-
back D-stabilization problem. Assume that the LMIs in (30)–(32) admit
feasible solutions (not necessarily unique) J, T, (. By considering the defi-
nitions in (33), we are now ready to propose the corresponding design pro-
cedure for the expected robust output feedback controllers.
368 JOTA: VOL. 113, NO. 2, MAY 2002

Design Procedure.
Step 1. Compute the full-column rank nx Bnc real matrices V and U
such that VU T GIAJT.
Step 2. Find the unique solution PH0 of the linear equation

冤U 冥 冤 冥
T I I J
T GP ,
0 0 VT

which is always solvable for P (see Lemma 7.5 of Ref. 29)


because TH0 and V are of full-column rank.
Step 3. With the above matrix P, calculate the controller parameters
Ac , Bc , Cc , Dc by solving the LMI (26) for K defined in (10).

Remark 3.2. It is obvious that the inequality (30) in linear on J and


(, while the inequality (31) in linear on T and ( −1, but (30) and (31) are not
jointly linear on ( or ( −1. A possible approach to dealing with the three
LMIs in Theorem 3.2 is to first solve the LMI (30) for ( and J, and then
use the obtained ( and J to solve the LMIs (31)–(32) for T.

Remark 3.3. When we consider the dynamic state feedback case (i.e.,
Cp GI), it is easy to test that, in Theorem 3.2, only the LMI (30), which is
linear on both J and (, is needed. The corresponding design procedure can
also be simplified. Furthermore, if we consider the static state feedback case
(i.e., nc G0), the results obtained in this paper reduce to those of Ref. 16,
since the Riccati matrix equation conditions derived in Ref. 16 can be con-
verted readily into the associated LMIs. If we consider the static state feed-
back case when the uncertainty disappears, actually the present results
recover those of Ref. 1. On the other hand, since the existence conditions
for the robust D-stability controllers proposed in Refs. 12–13 are merely
sufficient, the results obtained in this paper generalize those of Refs. 12–13.

Remark 3.4. It is worth pointing out that, although the uncertainty


in this paper is assumed to be norm bounded, it is not difficult to consider
a more general description of the uncertainty within the same framework
developed in this paper. For example, it has been respectively shown in Ref.
25 and Ref. 22 that the generalized positive real uncertainty systems and
the dissipative uncertainty of linear uncertain systems, although they are
more involved, can be taken into account by introducing some LMIs, and
the corresponding robustness can be guaranteed when the solutions to cer-
tain LMIs are known to exist. Thus, following along the line of this paper,
we will be able to enforce further the desired D-stability constraint in terms
JOTA: VOL. 113, NO. 2, MAY 2002 369

of an LMI which has a form similar to (26), and then the present techniques
can all be applied.

4. Numerical Example

In this section, a simple numerical example is provided to show the


usefulness of the present approach. Without loss of generality, we consider
the static output feedback case for the linear uncertain continuous-time
system (7); i.e.,

nc G0, KGDc .

Assume that the system parameters of (7) are

冤 冥 冤 冥
A2 0 0 0 0
Ap G 0 A3 1 , Bp G 1 0 ,
0 0 1 0 1

冤 冥
0.5 0
冤 冥
0 0.5 0
Cp G , Np2 G 0 0 ,
0 0 0.5
0 0.1

and

Mp Gdiag{0.2, 0.01, 0}, Np1 Gdiag{1, 0, 0}, FG(sin σ )I3 ,

where σ is a scalar uncertain parameter.


It is clear that the open-loop system is unstable. We assume that the
robust D-stability constraint is that, for all the admissible uncertainties, the
poles of the closed-loop matrix

Ac _(AC∆A)C(BC∆B)KC,

or the closed-loop matrix

Acp _(Ap C∆Ap)C(Bp C∆Bp)Dc Cp

for the static output feedback case, are all assigned inside the prespecified
disk D(3, 2).
By employing the controller design procedure provided in the previous
section, with respect to four different choices of (U, V) satisfying
370 JOTA: VOL. 113, NO. 2, MAY 2002

VU T GIAJT, we arrive at four static output feedback gains as follows:


1.240 −1.924 1.208 −2.008
K1 G 冤0.127 −4.489
, 冥 K2 G 冤−0.004 −10.459
, 冥
−1.239 −2.116 −1.229 −2.127
K3 G 冤 0.093 −4.569
, 冥 K4 G 冤−0.034−9.998
. 冥
It is not difficult to test that, with the four feedback gains provided above,
the robust circular pole constraint is met for all these four cases. We can
see from this example that the approach proposed in the present paper
possesses both effectiveness and flexibility.

5. Conclusions

This paper has introduced an LMI approach to the robust output feed-
back control for linear uncertain continuous-time and discrete-time systems
with circular pole constraints. We have derived some linear matrix inequal-
ities whose solutions indicate the achievability of the desired robust D-
stability; i.e., we have shown that the robust D-stabilization problem can
be recast as a convex optimization problem under constraints of LMIs
which can be solved effectively using the recently developed LMI tool. Then,
based on these LMIs, we have given a simple procedure to construct the
required output feedback stabilizing controllers. It has been shown that the
main results obtained in this papers include those of most recent papers
concerning robust D-stability synthesis and design as special cases. A
numerical example has demonstrated the applicability of the proposed
approach.

References

1. FURUTA, K., and KIM, S. B., Pole Assignment in a Specified Disk, IEEE Trans-
actions on Automatic Control, Vol. 32, pp. 423–427, 1987.
2. HADDAD, W. M., and BERNSTEIN, D. S., Controller Design with Regional Pole
Constraints, IEEE Transactions on Automatic Control, Vol. 37, pp. 54–69,
1992.
3. KAWASAKI, N., and SHIMEMURA, E., Pole Placement in a Specified Region
Based on a Linear Quadratic Regulator, International Journal of Control, Vol.
48, pp. 225–240, 1988.
4. LEE, T. T., and LEE, S. H., Discrete Optimal Control with Eigenûalues inside a
Circular Region, IEEE Transactions on Automatic Control, Vol. 31, pp. 958–
962, 1986.
JOTA: VOL. 113, NO. 2, MAY 2002 371

5. WANG, Z., Parametric Circular Eigenûalue Assignment for Descriptor Systems


ûia State Feedback, IMA Journal of Mathematical Control and Information,
Vol. 17, pp. 57–66, 2000.
6. WANG, Z., CHEN, X., and GUO, Z., Controller Design for Continuous Systems
with Variance and Circular Pole Constraints, International Journal of Systems
Science, Vol. 26, pp. 1249–1256, 1995.
7. CHOU, J. H., Pole-Assignment Robustness in a Specified Disk, Systems and Con-
trol Letters, Vol. 16, pp. 41–44, 1991.
8. HORNG, H. Y., CHOU, J. H., and HORNG, I. R., Robustness of Eigenûalue
Clustering in Various Regions of the Complex Plane for Perturbed Systems,
International Journal of Control, Vol. 57, pp. 1469–1483, 1993.
9. RACHID, A., Robustness of Pole Assignment in a Specified Region for Perturbed
Systems, International Journal of Control, Vol. 21, pp. 579–585, 1990.
10. JUANG, Y. T., and CHEN, K. H., Robust Pole-Assignment of Linear Dynamic
Systems, Control Theory and Advanced Technology, Vol. 5, pp. 67–74, 1989.
11. TSAY, S. C., FONG, I. K., and KUO, T. S., D-Stability Analysis for Discrete
Optimal Regulator, Control Theory and Advanced Technology, Vol. 6, pp. 237–
246, 1990.
12. WANG, Z., Robust HS Control for Uncertain Linear Continuous Systems with
Circular Pole Assignment ûia Output Feedback, Systems Science, Vol. 23, pp. 39–
54, 1997.
13. WANG, Z., TANG, G., and CHEN, X., Robust Controller Design for Uncertain
Linear Systems with Circular Pole Constraints, International Journal of Control,
Vol. 65, pp. 1045–1054, 1996.
14. YAZ, E., SKELTON, R. E., and GRIGORIADIS, E., Robust Regional Pole Assign-
ment with Output Feedback, Proceedings of IEEE Conference on Decision and
Control, pp. 3009–3013, 1993.
15. YEDAVALLI, R. K., and LIU, Y., HS Control with Regional Stability Constraints,
Automatica, Vol. 31, pp. 611–615, 1995.
16. GARCIA, G., and BERNUSSOU, J., Pole Assignment for Uncertain Systems in a
Specified Disk by State Feedback, IEEE Transactions on Automatic Control,
Vol. 40, pp. 184–190, 1995.
17. DAAFOUZ, J., GARCIA, G., and BERNUSSOU, J., Robust Control of a Flexible
Robot Arm Using the Quadratic d-Stability Approach, IEEE Transactions on
Control Systems Technology, Vol. 6, pp. 524–533, 1998.
18. REZA MOHEIMANI, S. O., and PETERSEN, I. R., Quadratic Guaranteed Cost
Control with Robust Pole Placement in a Disk, IEE Proceedings on Control The-
ory and Applications, Vol. 143, pp. 37–43, 1996.
19. BOYD, S., GHAOUI, L. EL, FERON, E., and BALAKRISHNAN, V., Linear Matrix
Inequalities in System and Control Theory, Studies in Applied Mathematics,
SIAM, Philadelphia, Pennsylvania, Vol. 15, 1994.
20. NESTEROV, Y., and NEMIROVSKY, A., Interior-Point Polynomial Methods in
Conûex Programming: Theory and Applications, SIAM, Philadelphia, Pennsyl-
vania, 1993.
21. GAHINET, P., and APKARIAN, P., A Linear Matrix Inequality Approach to HS
Control, International Journal of Robust and Nonlinear Control, Vol. 4,
pp. 421–448, 1994.
372 JOTA: VOL. 113, NO. 2, MAY 2002

22. XIE, S., XIE, L., and DE SOUZA, C. E., Robust Dissipatiûe Control for Linear
Systems with Dissipatiûe Uncertainty, International Journal of Control, Vol. 70,
pp. 169–191, 1998.
23. CHILALI, M., and GAHINET, P., HS Design with Pole Placement Constraints:
An LMI Approach, IEEE Transactions on Automatic Control, Vol. 41, pp. 358–
367, 1996.
24. GAHINET, P., NEMIROVSKY, A., LAUB, A. J., and CHILALI, M., LMI Control
Toolbox for Use with Matlab, The Math Works, Natick, Massachussets, 1995.
25. XIE, L., and SOH, Y. C., Robust Control of Linear Systems with Generalized
Positiûe Real Uncertainty, Automatica, Vol. 33, pp. 963–967, 1997.
26. KHARGONEKAR, P. P., PETERSEN, I. R., and ZHOU, K., Robust Stabilization of
Uncertain Linear Systems: Quadratic Stabilization and HS Control Theory,
IEEE Transactions on Automatic Control, Vol. 35, pp. 356–361, 1990.
27. XIE, L., and SOH, Y. C., Guaranteed Cost Control of Uncertain Discrete-Time
Systems, Control Theory and Advanced Technology, Vol. 10, pp. 1235–1251,
1995.
28. XIE, L., Output Feedback HS Control of Systems with Parameter Uncertainty,
International Journal of Control, Vol. 63, pp. 741–750, 1996.
29. PACKARD, A., ZHOU, K., PANDEY, P., and BECKER, G., A Collection of Robust
Control Problems Leading To LMIs, Proceedings of the 30th IEEE Conferences
on Decision and Control, Brighton, United Kingdom, pp. 1245–1250, 1991.

You might also like