Professional Documents
Culture Documents
E-mail: s.k.spurgeon@kent.ac.uk
Contract/grant sponsor: EPSRC Grant; contract/grant number:
Reference EP/E020763/1
innite dimensional when compared with ordinary
differential equations. Sometimes the delay may affect
the system performance greatly: for example, a small
delay may destabilize a system while a large delay may
stabilize a system; chaotic behaviour may appear if the
delayed state involves a nonlinear function but in other
cases, chaotic systems may be stabilized by a delayed
feedback [1]. This shows that time-delay systems are
very complicated especially when the delay exists in
nonlinear terms. Therefore, the study of time-delay
systems is pertinent and valuable both from theoretical
and applications perspectives.
As early as the 18th century, time-delay systems
have been studied (see the survey paper [2]). Since
Krasovskii extended the Lyapunov theory to time-delay
systems [3] and Razumikhin proposed a method to
avoid the functional in Lyapunov stability analysis [4],
great progress has been made but most of the early
Copyright q 2009 John Wiley & Sons, Ltd.
778 X. G. YAN, S. K. SPURGEON AND C. EDWARDS
work focused on the analysis of unforced time-delay
systems. In more recent years, the advancement of
control theory and control engineering has motivated
the study of control systems. For systems affected by
time delay, contributions have considered cases where
the delay may appear in the system state, input, output
and disturbances experienced by the system [57]. A
variety of control approaches such as sliding mode
control, H
A=
_
A
11
A
12
A
21
A
22
_
,
B=
_
0
B
2
_
,
C=[0
C
2
] (3)
where
A
11
R
(nm)(nm)
,
B
2
R
mm
is nonsingular
and
C
2
R
pp
is orthogonal. Further, it is assumed
that system (
A
11
,
A
12
,
C
1
) with
C
1
dened by
C
1
=[0
( pm)(np)
I
pm
] (4)
is output feedback stabilizable, i.e. there exists a matrix
K R
m( pm)
such that
A
11
A
12
K
C
1
is stable.
It is shown in [21] that a necessary condition for
(
A
11
,
A
12
,
C
1
) to be stabilizable is that the invariant
zeros of (A, B, C) lie in the open left half-plane.
Let
F =F
2
[K I
m
]
C
T
2
(5)
where F
2
R
mm
is any nonsingular matrix. If a
further coordinate change is introduced based on the
nonsingular transformation z =
Tx with
T dened by
T =
_
I
nm
0
K
C
1
I
m
_
then in the new coordinates z, system (1) and (2) has
the following form:
A=
_
A
11
A
12
A
21
A
22
_
,
B=
_
0
B
2
_
,
C=[0 C
2
] (6)
where A
11
=
A
11
A
12
K
C
1
is stable, C
2
R
pp
is
nonsingular and
C satises
F
C=[0 F
2
] (7)
where F
2
R
mm
is nonsingular. In the new coordinate
system z, the system output is described by
y =
Cz (8)
For convenience, the following denition is introduced:
Denition 1
The matrix triple (A, B, C) or linear system (1)(2)
is called normalizable if there exists a nonsingular
transformation z =T x such that in the new coordinate
system z, the system (1) and (2) has the following form:
z
1
= A
11
z
1
+A
12
z
2
(9)
z
2
= A
21
z
1
+A
22
z
2
+B
2
u (10)
y =[0 C
2
]z (11)
where z
1
R
nm
, z
2
R
m
, A
11
is stable, and B
2
R
mm
and C
2
R
pp
are nonsingular. Then (9)(11)
is called the canonical form of system (1) and (2).
From the analysis above the following conclusion is
obtained immediately
Lemma 1
System (1) and (2) is normalizable if (i) rank(CB)=
m, (ii) for the triple (
A
11
,
A
12
,
C
1
) dened by (3)
and (4), there exists a matrix K such that
A
11
A
12
K
C
1
is stable.
Proof
By letting T =
if
a = and lim
r
:(r)=.
Consider a time-delay system
x(t )=
f (t, x(t d(t ))) (12)
with an initial condition
x(t )=[(t ), t [d, 0]
Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2010; 20:777788
DOI: 10.1002/rnc
780 X. G. YAN, S. K. SPURGEON AND C. EDWARDS
where
f : R
+
C
[d,0]
R
n
takes R(bounded sets
of C
[d,0]
) into bounded sets in R
n
, d(t ) is the time-
varying delay and d :=sup
t R
+{d(t )}<.
Lemma 2 (Razumikhin Theorem, Gu et al. [22])
If there exist class K
functions
i
() with i =1, 2,
a class K function
3
() and a continuous function
V
1
(): [d, ]R
n
R
+
satisfying
1
(x)V
1
(t, x)
2
(x), t [d, ], x R
n
such that the time derivative of V
1
along the solution
of system (12) satises
V
1
(t, x)
3
(x) whenever
V
1
(t +0, x(t +0))V
1
(t, x(t )) (13)
for any 0[d, 0], then system(12) is uniformly stable.
If, in addition,
3
(t)>0 for t>0, and there exists a
continuous nondecreasing function
4
() which satises
4
(t)>t for t>0 such that inequality (13) is strength-
ened to
V
1
(t, x)
3
(x) whenever
V
1
(t +0, x(t +0))
4
(V
1
(t, x(t ))) (14)
for any 0[d, 0], then system (12) is uniformly
asymptotically stable.
From Lemma 2. the following conclusion can be
obtained directly:
Lemma 3
If there exist constants >0 and >1 and a function
V
2
(x(t ))=x
T
Px with
P>0 such that the time deriva-
tive of V
2
() along the solution of system (12) satises
V
2
|
(12)
P
1/2
x(t )
2
(15)
whenever
P
1/2
x(t +0)
P
1/2
x(t ) for any 0
[d, 0], then, system (12) is uniformly asymptotically
stable.
Proof
From the denition of V
2
() it follows that
z
min
(
P)x
2
V
2
(t, x(t ))z
max
(
P)x
2
and from (15)
V
2
|
(12)
x(t )
T
Px(t )z
max
(
P)x
2
It is clear that V
2
(x(t +0))
2
V
2
(x(t )) is equivalent to
P
1/2
x(t +0)
P
1/2
x(t ).
Then from Lemma 2 and
P>0, the conclu-
sion follows by letting
1
(t)=z
min
(
P)t
2
,
2
(t)=
z
max
(
P)t
2
,
3
(t)=z
min
(
P)t
2
and
4
(t)=
2
t in
Lemma 2.
The lemmas presented in this section will be used in
the following analysis.
3. SYSTEM DESCRIPTION AND PROBLEM
FORMULATION
Consider a time-varying delay system with time-
delayed disturbance described by
x(t ) = Ax(t )+A
0
x(t d(t ))+B(u(t )
+g(t, x(t ), x(t d(t ))))
+ f (t, x(t ), x(t d(t ))) (16)
y(t ) =Cx(t ) (17)
where x R
n
( is an neighborhood of the origin),
u R
m
and y R
p
are system states, inputs and
outputs, respectively, with mp<n. The matrices
A, A
0
, B and C represent constant matrices of appro-
priate dimensions with B and C of full rank. The
vectors g() and f () represent the matched and
mismatched disturbances affecting the system, respec-
tively. The known function d(t ) is a time-varying delay
which is assumed to be continuous, nonnegative and
bounded in R
+
, that is, d :=sup
t R
+{d(t )}<. The
initial condition for the system is given by
x(t )=[(t ), t [d, 0]
where [() is continuous in [d, 0]. It is assumed that
the nonlinear functions g() and f () are smooth enough
in their domain of denition such that the system has
unique continuous solutions for the initial condition.
Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2010; 20:777788
DOI: 10.1002/rnc
STATIC OUTPUT FEEDBACK SLIDING MODE CONTROL 781
First, it is necessary to impose some basic assump-
tions on system (16)(17):
Assumption 1
The triple (A, B, C) is normalizable, and Im(A
0
)
Im(B).
Remark 2
Assumption 1 is a limitation on the linear part of system
(16)(17). It guarantees that the triple (A, B, C) can be
transformed to a canonical form (9)(11). The assump-
tion Im(A
0
)Im(B) means that the time-delay term
A
0
x(t d(t )) is matched and thus it will not affect the
sliding motion.
Assumption 2
There exist known continuous nonnegative func-
tions j
i
(): R
+
R
p
R
p
R
+
with i =1, 2 and
=(): R
+
R
p
R
p
R
+
such that for t R
+
, and
x(t ), x(t d)
f (t, x(t ), x(t d(t ))
j
1
(t, y(t ), y(t d(t )))x(t )
+j
2
(t, y(t ), y(t d(t )))x(t d(t )) (18)
g(t, x(t ), x(t d(t ))=(t, y(t ), y(t d(t ))) (19)
Remark 3
Assumption 2 requires that the uncertainties are
bounded by some known continuous functions. It is an
extension of the existing results [13, 15, 23] where it is
required that the uncertainty is bounded by the linear
combination of y(t ) and y(t d) which is called
the linear growth condition.
In this paper, the objective is to design a static output
feedback control with time delay of the form
u =u(t, y(t ), y(t d)) (20)
based-on sliding mode techniques such that the closed-
loop system formed by the control (20) and the system
(16) and (17) is uniformly asymptotically stable in
a domain of the origin even in the presence of the
disturbances. Notably the control (20) only depends on
system output y(t ) and time delay d(t ). Since d(t ) is
assumed to be known in this paper, the term y(t d(t ))
is available and thus the control (20) is called static
output feedback control with time delay.
Remark 4
As in the work in [8, 15, 24, 25], the delay experienced
by the system is assumed to be known, which may
limit its application. However, in some important
industrial systems such as ow through pipes and
web forming processes, the delay existing in the
process is known, and can thus be employed in the
control design and/or the compensator design [25].
Furthermore, the approach proposed in [26] enables
the time delay to be identied in some cases even
when the delay is unknown.
It is well known that the sliding mode control tech-
nique consists of two steps: (i) design of the sliding
surface such that the system possesses the required
performance when it is restricted to the surface and (ii)
design of a variable structure control which drives the
system trajectory to the sliding surface in nite time
and maintains a sliding motion on it. The following
section will consider these two steps.
4. SLIDING MOTION ANALYSIS AND
CONTROL DESIGN
In this section, the main results will be presented.
From Im(A
0
)Im(B) in Assumption 1, there exists
a matrix DR
mn
such that A
0
=BD. Then, from
Section 2, it follows that under Assumption 1 there
exists a coordinate transformation z =T x such that
in the new coordinate system z, system (16)(17) is
described by
z
1
= A
11
z
1
+A
12
z
2
+ f
1
(t, z(t ), z(t d(t ))) (21)
z
2
= A
21
z
1
+A
22
z
2
+B
2
DT
1
z(t d(t ))
+B
2
(u+g(t, T
1
z(t ), T
1
z(t d(t ))))
+ f
2
(t, z(t ), z(t d(t ))) (22)
y =[0 C
2
]z (23)
Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2010; 20:777788
DOI: 10.1002/rnc
782 X. G. YAN, S. K. SPURGEON AND C. EDWARDS
where A
11
R
(nm)(nm)
is stable, B
2
R
mm
and
C
2
R
pp
are nonsingular, and
_
f
1
(t, z(t ), z(t d(t )))
f
2
(t, z(t ), z(t d(t )))
_
:=T[ f (t, x(t ), x(t d(t )))]
x=T
1
z
(24)
where f
1
()R
nm
and f
2
()R
m
.
Remark 5
Since the coordinate transfer matrix T can be obtained
using a systematic approach given in [11], the system in
(21)(23) is well dened and can be directly obtained
from system (16) and (17).
Consider the following sliding surface for system
(16)(17):
S ={x|FCx =0} (25)
where F is dened in (5). Then from (7) and (8), it
follows that
FCx =Fy =F
Cz =[0 F
2
]z =F
2
z
2
Since F
2
is nonsingular, it follows that in the z coor-
dinate system the sliding surface (25) can be described
by the equation
z
2
=0 (26)
Then from the canonical form (21)(23), it follows that
the sliding dynamics associated with the sliding surface
(25) are described by
z
1
= A
11
z
1
+[ f
1
(t, z(t ), z(t d(t )))]
z
2
(t )=0
(27)
where z
1
R
nm
are the sliding mode state variables
and A
11
is stable. It is clear that the mismatched
disturbance affects the sliding motion directly. Obvi-
ously system (27) that describes the sliding motion
involves time delay. The following further assumption
is required:
Assumption 3
There exist known continuous functions [
1
() and [
2
()
such that
[ f
1
(t, z(t ), z(t d(t )))]
z
2
(t )=0
[
1
(t, z
1
(t ), z
1
(t d))z
1
(t )
+[
2
(t, z
1
(t ), z
1
(t d))z
1
(t d(t )) (28)
where the function [
1
(t,r
1
,r
2
) and [
2
(t,r
1
,r
2
) are
both nondecreasing about variables r
2
.
Remark 6
Assumption 3 is a limitation on the mismatched distur-
bance. It implies that when a sliding motion takes
place, the uncertainty f
1
can be bounded by a known
continuous function of variables z
1
(t ) and z
1
(t d(t )).
It should be noted that Assumption 3 is unnecessary
if the disturbance f () in (16) does not include time
delay [20].
Since the matrix A
11
in (27) is stable, it follows
that for any Q>0 (QR
mm
), there exists an unique
matrix P>0 such that
A
T
11
P+PA
11
=Q (29)
For the later analysis, the following lemma is presented:
Lemma 4
If Assumption 3 holds, then there exist known contin-
uous functions
1
() and
2
() such that
P
1/2
[ f
1
(t, z(t ), z(t d(t )))]
z
2
(t )=0
1
(t, z
1
(t ), z
1
(t d))P
1/2
z
1
(t )
+
2
(t, z
1
(t ), z
1
(t d))
P
1/2
z
1
(t d(t )) (30)
where the functions
1
(t,r
1
,r
2
) and
2
(t,r
1
,r
2
) are
both nondecreasing about variables r
2
.
Proof
It follows from the fact
z
1
(t ) z
max
(P
1/2
)P
1/2
z
1
(t ) and
z
1
(t d(t )) z
max
(P
1/2
)P
1/2
z
1
(t d(t )) (31)
that under Assumption 3
P
1/2
[ f
1
(t, z(t ), z(t d(t )))]
z
2
(t )=0
z
max
(P
1/2
)([
1
(t, z
1
(t ), z
1
(t d))z
1
(t )
+[
2
(t, z
1
(t ), z
1
(t d))z
1
(t d(t )))
z
max
(P
1/2
)([
1
(t, z
1
(t ), z
1
(t d(t )))
z
max
(P
1/2
)P
1/2
z
1
(t )
Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2010; 20:777788
DOI: 10.1002/rnc
STATIC OUTPUT FEEDBACK SLIDING MODE CONTROL 783
+[
2
(t, z
1
(t ), z
1
(t d(t )))
z
max
(P
1/2
)P
1/2
z
1
(t d(t ))) (32)
Let
1
(t,r
1
,r
2
)=z
max
(P
1/2
)z
max
(P
1/2
)[
1
(t,r
1
,r
2
)
and
2
(t,r
1
,r
2
)=z
max
(P
1/2
)z
max
(P
1/2
)[
2
(t,r
1
,r
2
)
Then it follows that (28) is true and the functions
1
(t,r
1
,r
2
) and
2
(t,r
1
,r
2
) are both nondecreasing
about variables r
2
since [
1
(t,r
1
,r
2
) and [
2
(t,r
1
,r
2
)
are both nondecreasing about variables r
2
. Hence, the
conclusion follows.
The following theorem that gives a sufcient condi-
tion under which the sliding motion is stable, can be
presented.
Theorem 1
Under Assumption 3, the sliding mode dynamics (27)
are uniformly asymptotically stable if there exists a
domain
0
={z
1
|z
1
R
nm
} of the origin in T()
and a constant >1 such that for any z
1
(t )
0
and
t R
+
:=z
min
(P
1/2
QP
1/2
)
sup
t R
+
,z
1
(t )
0
{(t, z
1
(t ))}>0 (33)
where
(t, z
1
(t ))
:=
1
(t, z
1
(t ), z
max
(P
1/2
)P
1/2
z
1
(t ))
+
2
(t, z
1
(t ), z
max
(P
1/2
)P
1/2
z
1
(t )) (34)
where
1
() and
2
() satisfy (30), and P and
Q satisfy (29).
Proof
For system (27), consider as a Lyapunov function
candidate V(z
1
(t ))=(z
1
(t ))
T
Pz
1
(t ). It follows from
(28) and (29) that the time derivative of V along the
trajectories of system (27) is given as
V(z
1
(t ))|
(27)
=(z
1
(t ))
T
(A
T
11
P+PA
11
)z
1
(t )
+2(z
1
(t ))
T
P[ f
1
(t, z(t ), z(t d(t )))]
z
2
(t )=0
=(z
1
(t ))
T
P
1/2
(P
1/2
QP
1/2
)P
1/2
z
1
(t )
+2(z
1
(t ))
T
P
1/2
P
1/2
[ f
1
(t, z(t ), z(t d(t )))]
z
2
(t )=0
z
min
(P
1/2
QP
1/2
)P
1/2
z
1
(t )
2
+P
1/2
z
1
(
1
(t, z
1
(t ), z
1
(t d))P
1/2
z
1
(t )
+
2
(t, z
1
(t ), z
1
(t d))P
1/2
z
1
(t d(t )))
(35)
where Lemma 4 has been used to obtain the above.
Since
1
(t,r
1
,r
2
) and
2
(t,r
1
,r
2
) are both nonde-
creasing about variables r
2
, it follows from (31) that
1
(t, z
1
(t ), z
1
(t d))
1
(t, z
1
(t ), z
max
(P
1/2
)P
1/2
z
1
(t d(t ))) (36)
2
(t, z
1
(t ), z
1
(t d))
2
(t, z
1
(t ), z
max
(P
1/2
)P
1/2
z
1
(t d(t ))) (37)
When P
1/2
z
1
(t +0)P
1/2
z
1
(t ) for any 0
[d, 0] and some >1, by substituting (36) and (37)
to (35), it follows that
V(z
1
(t ))|
(27)
z
min
(P
1/2
QP
1/2
)P
1/2
z
1
(t )
2
+
1
(t, z
1
(t ), z
max
(P
1/2
)
P
1/2
z
1
(t d(t )))P
1/2
z
1
(t )
2
+
2
(t, z
1
(t ), z
max
(P
1/2
)P
1/2
z
1
(t d(t )))
P
1/2
z
1
(t )P
1/2
z
1
(t d(t ))
z
min
(P
1/2
QP
1/2
)P
1/2
z
1
(t )
2
Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2010; 20:777788
DOI: 10.1002/rnc
784 X. G. YAN, S. K. SPURGEON AND C. EDWARDS
+
1
(t, z
1
(t ), z
max
(P
1/2
)P
1/2
z
1
(t ))
P
1/2
z
1
(t )
2
+
2
(t, z
1
(t ), z
max
(P
1/2
)
P
1/2
z
1
(t ))P
1/2
z
1
(t )
2
=(z
min
(P
1/2
QP
1/2
)(t, z
1
(t )))
P
1/2
z
1
(t )
2
P
1/2
z
1
(t )
2
(38)
Hence, the conclusion follows directly from Lemma 3.
Remark 7
From Theorem 1, it follows that the stability of the
sliding motion is completely robust to the matched
uncertainty g() but is affected by the mismatched
uncertainty f (). Since the sliding mode is a reduced-
order system, it is clear that only f
1
() affects the
sliding mode and thus in the proposed conguration
the limitation on the mismatched uncertainty is weaker
than in other work [8, 9, 23] where a similar limitation
is imposed on f () instead of f
1
().
Remark 8
From the proof of Theorem 1, it follows that in order to
establish the stability of the sliding motion, it is neces-
sary to estimate P
1/2
[ f
1
(t, z(t ), z(t d(t )))]
z
2
(t )=0
which involves uncertainty and a time-varying delay.
It should be pointed out that when the structure of f
1
is available, (30) may give a less conservative bound
than (28).
Theorem 1 above has shown that, under appropriate
conditions, the sliding motion on sliding surface (25)
is stable. The objective now is to design a controller
to drive the system to the sliding surface in nite time.
Comparing the linear part of system (16) and (17) with
the linear part of system (21)(23), it follows that
CT
1
=[0 C
2
], T B=
_
0
B
2
_
where C
2
R
pp
and B
2
R
mm
are nonsingular.
From (6) and (7), it follows that
FCB=FCT
1
TB=F [0 C
2
]
. ,, .
C
_
0
B
2
_
=[0 F
2
]
_
0
B
2
_
= F
2
B
2
is nonsingular since both F
2
R
mm
and B
2
R
mm
are nonsingular. Partition the matrices AT
1
, A
0
T
1
and T as
AT
1
:= [
1
2
], A
0
T
1
:=[
1
2
]
T :=
_
T
1
T
2
_
(39)
where
1
R
nm
and
1
R
nm
are the rst m
columns of AT
1
and A
0
T
1
, respectively; T
1
R
mn
and T
2
R
(nm)n
are the rst m and the last nm
rows of T. Then, from the analysis above,
T x =
_
T
1
x
T
2
x
_
=
_
z
1
z
2
_
=
_
T
1
x
C
1
2
y
_
(40)
Now, consider system (16)(17) in
1
2
where
1
:= {x(t )|T
1
xj
1
}
2
:= {x(t d(t ))|T
1
x(t d(t ))j
2
}
(41)
and T
1
is dened in (39). Then the following output
feedback sliding mode controller with time delay is
proposed for the system
u =k(t, y(t ), y(t d(t )))(FCB)
1
sgn(Fy(t )) (42)
where sgn is the usual signum vector function and the
scalar function k() is dened by
k(t, y(t ), y(t d(t )))
=
1
j
1
+
2
C
1
2
y
+FCB=(t, y(t ), y(t d(t )))
+FCT
1
(j
1
(t, y(t ), y(t d(t )))
(j
1
+C
1
2
y(t ))+j
2
(t, y(t ), y(t d(t )))
(j
2
+C
1
2
y(t d(t )))+p (43)
for some p>0 where matrices
1
and
2
are dened
by (39), the positive constants j
1
and j
2
are given in
(41), and the functions =(), j
1
() and j
2
() are given
in Assumption 2.
Remark 9
From the analysis above, it is clear to see that the
sliding mode controller (42) with k() dened by (43) is
Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2010; 20:777788
DOI: 10.1002/rnc
STATIC OUTPUT FEEDBACK SLIDING MODE CONTROL 785
well dened since the matrix FCB is nonsingular and
the functions =(), j
1
() and j
2
() are assumed to be
known. Obviously, the proposed control only depends
on the time t , the known time delay d(t ) and system
output y(t ).
Theorem 2
Consider system (16)(17) in
1
2
. Under Assump-
tions 1 and 2, the controller (42) with the gain k()
dened by (43) drives the system (16)(17) to the
sliding surface (25) in nite time and maintains a sliding
motion on it thereafter.
Proof
Let o(x):=FCx. Then the sliding surface (25) can be
described by equation o(x)=0. From (16) and (42), it
follows that
o
T
(x) o(x)
=o
T
(x)FC(Ax(t )+A
0
x(t d(t ))+B(u(t )
+g(t, x(t ), x(t d(t ))))+ f (t, x(t ), x(t d(t ))))
o(x)(FC(Ax(t )+A
0
x(t d))+FCB
g(t, x(t ), x(t d(t )))
+FC f (t, x(t ), x(t d(t ))))
k(t, y(t ), y(t d(t )))o(x) (44)
where the fact that o
T
(x)sgn(o(x))=o(x) is used
above. From (40) it follows that in
1
2
dened by
(41),
T x(t ) j
1
+C
1
2
y(t )
T x(t d(t )) j
2
+C
1
2
y(t d(t ))
(45)
From (39) and (40),
FC(Ax(t )+A
0
x(t d(t )))
=FC(AT
1
T x(t )+A
0
T
1
T x(t d(t )))
=FC
_
[
1
2
]
_
T
1
x
C
1
2
y
_
+[
1
2
]
_
T
1
x(t d)
C
1
2
y(t d)
__
=FC
1
T
1
x +FC
2
C
1
2
y +FC
1
T
1
x(t d)
+FC
2
C
1
2
y(t d)
Therefore, from (45),
FC(Ax(t )+A
0
x(t d(t )))
FC
1
j
1
+FC
2
C
1
2
y(t )+FC
1
j
2
+FC
2
C
1
2
y(t d) (46)
From (18) and (45)
f (t, x(t ), x(t d(t )))
j
1
(t, y(t ), y(t d(t )))T
1
T x(t )
+j
2
(t, y(t ), y(t d(t )))T
1
T x(t d(t ))
T
1
(j
1
(t, y(t ), y(t d(t )))
(j
1
+C
1
2
y(t ))+j
2
(t, y(t ), y(t d(t )))
(j
2
+C
1
2
y(t d(t )))) (47)
Substituting (19), (46), (47) and (43) to (44) yields
o
T
(x) o(x)po(x)
This shows that the reachability condition [10, 11] is
satised and thus the conclusion follows.
Theorems 1 and 2 together show that the closed-
loop system formed by applying control (42) with k()
dened by (43) to system (16) and (17) is uniformly
asymptotically stable.
Remark 10
In this paper, coordinate transformations are employed
to derive the regular form and the sliding mode
dynamics, enabling the stability of the sliding motion
to be analysed. It should be noted that only static
output feedback control is considered in this paper.
In the control design, a state transformation (40) is
introduced to separate the known parts C
1
2
y(t ) and
C
1
2
y(t d(t )) from T x(t ) and T x(t d(t )), respec-
tively, so that they can be used in the control design
to reduce conservatism and avoid unnecessary control
action. This ensures that the conclusion holds, possibly
Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2010; 20:777788
DOI: 10.1002/rnc
786 X. G. YAN, S. K. SPURGEON AND C. EDWARDS
in an unbounded domain, since the constraint (41) only
corresponds to a subset of the state variables.
5. NUMERICAL SIMULATION
Consider the time-varying delay system with delayed
disturbance described by
x =
10 1 0
1 0 0
0 1 5
. ,, .
A
x +
0 0 0
1 0 1
0 0 0
. ,, .
A
0
x(t d(t ))
+
0
1
0
. ,, .
B
(u(t )+g(t, x(t ), x(t d(t ))))
+
2[
1
(t, x(t ), x(t d(t )))x
1
(t )+[
2
(t, x(t ), x(t d(t )))x
1
(t d(t ))
0
[
1
(t, x(t ), x(t d(t )))x
3
(t )+[
2
(t, x(t ), x(t d(t )))x
3
(t d(t ))
. ,, .
f (t,x(t ),x(t d(t )))
(48)
y =
_
0 0 1
0 1 0
_
. ,, .
C
x (49)
where x =col(x
1
, x
2
, x
3
), u and y =col(y
1
, y
2
) are,
respectively, the state variables, the inputs and the
outputs of the system. The unknown functions [
1
() and
[
2
() are time-delayed disturbances that are assumed
to satisfy
|[
1
()| (y
2
(t ))
2
| sin y
1
(t d(t ))|
|[
2
()| |y
1
(t d(t ))| sin
2
y
1
(t )+(y
2
(t ))
2
The matched delayed disturbance g() has unknown
structure but satises
g()y
4
2
(t )sin
2
y
1
(t d(t ))
. ,, .
=()
The domain considered here is
={(x
1
, x
2
, x
3
)|x
2
R,
1
2
x
2
1
+x
2
3
<12}
Obviously,
f (t, x(t ), x(t d(t )))
2(y
2
(t ))
2
| sin y
1
(t d(t ))|
. ,, .
j
1
()
x(t )
+(|y
1
(t d(t ))| sin
2
y
1
(t )+(y
2
(t ))
2
)
. ,, .
j
2
()
x(t d(t ))
and Assumption 2 holds.
Clearly, CB=[0 1]
T
is full rank. According to the
algorithm given in [21], the coordinate transformation
x =
T x with
T =
1 0 0
0 0 1
0 1 0
A
11
A
12
A
21
A
22
_
=
10 0 1
0 5 1
1 0 0
_
0
B
2
_
=
0
0
1
, [0
C
2
] =[0 I
2
]
(50)
Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2010; 20:777788
DOI: 10.1002/rnc
STATIC OUTPUT FEEDBACK SLIDING MODE CONTROL 787
It is clear that the triple (A, B, C) is output feedback
normalizable with the choice K =0 due to the stability
of
A
11
. Further Im(A
0
)Im(B) since A
0
=BD with
D=[1 0 1]. Therefore, Assumption 1 is satised.
Since (50) already has the canonical form (9)(11),
it follows that T =
T, A
11
=
A
11
, A
12
=
A
12
, A
21
=
A
21
, A
22
=
A
22
, B
2
=
B
2
, C
2
=
C
2
=I
2
. Let Q=10I
2
.
It follows that the Lyapunov equation (29) has a
unique solution P =diag{0.5, 1} and thus P
1/2
=diag
{
2 0
0 1
_
[
2
()P
1/2
z
1
(t d(t )) (52)
When sliding motion takes place, y
2
(t )=0, and thus
we have
[
1
()=0, |[
2
()|y
1
(t d(t ))| sin
2
y
1
(t )
Then,
P
1/2
[ f
1
(t, z(t ), z(t d(t )))]
z
2
(t )=0
z
1
(t d(t ))(sin y
1
(t ))
2
P
1/2
z
1
(t d(t )) (53)
By comparing (53) with (30), it follows that
1
()=0
and
2
()=z
1
(t d(t ))(sin y
1
(t ))
2
. Therefore,
(t, z
1
(t ))=
2
2(sin y
1
(t ))
2
P
1/2
z(t )
Let =1.01. By direct computation, it follows that
>0.0026>0 in T(), and thus the conditions of
Theorem 1 hold in the domain T(). From (42) and
(43), the control is given as follows:
u =(10.0499j
1
+
_
(y
2
(t ))
2
+(y
2
(t )5y
1
(t ))
2
+y
4
2
(t )sin
2
y
1
(t d(t ))+
2(y
2
(t ))
2
| sin y
1
(t d(t ))|(j
1
+y)
+(|y
1
(t d(t ))| sin
2
y
1
(t )
+(y
2
(t ))
2
)(j
2
+y(t d(t )))+p)sgn(y
2
(t )) (54)
0 1 2 3 4 5
0
0.5
1
1.5
2
2.5
3
time [sec]
s
t
a
t
e
s
x
1
x
2
x
3
Figure 1. The time responses of the state variables of system
(48)(49) under control (54).
0 1 2 3 4 5
0
20
40
60
time [sec]
c
o
n
t
r
o
l
0 1 2 3 4 5
0
0.5
1
1.5
2
time [sec]
s
l
i
d
i
n
g
f
u
n
c
t
i
o
n
Figure 2. The time responses of control signal (54) (upper)
and sliding function (bottom).
For implementation purposes, choose j
1
=j
2
=2 and
p=1. The time-varying delay d(t ) is chosen as d(t )=
2+sint . A simulation with the initial condition [(t )=
(cos(t ), 1, 2sin(t )) is shown in Figures 1 and 2 and
conrms that the proposed approach is effective.
6. CONCLUSIONS
A robust static output feedback sliding mode controller
has been developed to stabilize a class of time-varying
Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2010; 20:777788
DOI: 10.1002/rnc
788 X. G. YAN, S. K. SPURGEON AND C. EDWARDS
delay systems in this paper. An approach to deal with
nonlinear matched and mismatched disturbances is
shown when time-varying delay is involved in the
nonlinear bounds on the disturbances. Compared with
existing results, the nonlinear bounds are fully used
in the control design. The conservatism is reduced
by using the system structure and the property that
the sliding mode dynamics are of reduced-order. The
advantages of the approach when compared with the
existing literature have been highlighted. Simulations
have shown the effectiveness of the proposed control
scheme.
ACKNOWLEDGEMENTS
The authors gratefully acknowledge the support of EPSRC
for this work via grant reference EP/E020763/1 entitled
Robust Output Feedback Sliding Mode Control for Time-
delay Systems.
REFERENCES
1. Niculescu SI. Delay Effects on Stability: A Robust Control
Approach. Springer: London, 2001.
2. Gu K, Niculescu SI. Survey on recent results in the stability
and control of time-delay systems. ASME Journal Dynamics
Systems, Measurement and Control 2003; 125:158165.
3. Krasovskii NN. Stability of Motion (Translation form
Russian). Stanford University Press: Stanford, CA, U.S.A.,
1963.
4. Razumikhin BS. Application of Liapunovs method to
problems in the stability with a delay (in Russian). Automatika
i Telemekhanika 1960; 21:740749.
5. Levaggi L, Punta E. Analysis of a second-order sliding-mode
algorithm in presence of input delays. IEEE Transactions on
Automatic Control 2006; 51(8):13251332.
6. Fridman E, Shaked U, Suplin V. Input/output delay approach
to robust sampled-data H