estimation and control issues concerning MJSs have been studied such
as [12,29,30]. On the other hand, since stochastic modelings can play
very important role in many areas, the study of stochastic systems and
stochastic Markovian jump systems (SMJSs) has been of great interest
such as [2,1 3,22,25,3 1,32,39]. It should be pointed out that time delays
and parameter uncertainties are frequently sources of instability and
poor performance of a real system as well as a Markovian jump
system. Therefore, it is necessary and reasonable to take into account
time delays and parameter uncertainties when investigating MJSs. To
the best of the authors' knowledge, however, until now , the problem
of fault detection for SMJSs with parameter uncertainties and mode-
dependent time-varying delays has not been fully studied, which still
remains open and challenging.
This paper considers the robust H
1
fault detection problem for a
class of SMJSs with mode-dependent time-varying delays and norm-
bounded paramet er uncertainties. W e deal with the fault detection by
designing a mode-dependent detection filter that generates a residual
signal to estimate the fault signal and based on this, provide a residual
evaluation function to compare with a predefined threshold, when the
residual evaluation function has a value larger than the threshold, an
alarm of fault is generated. A simulation example and an industrial
nonisothermal continuous stirred tank reactor (CSTR) system are used
to illustrate the effectiveness of the proposed method.
The main contributions of this paper are summarized as follows:
(1) a new kind of stochastic Markovian jump system model with
mode-dependent time-varying delays and norm-bounded parameter
uncertainties, which has not been considered in the existing r efer-
ences, is proposed; (2) a stochastic Ly apunov–Krasovskii function is
constructed to reflect the mode-dependent time-varying delays,
some novel mode-dependent and delay-dependent sufficien t condi-
tions in t erms of linear matrix inequality (LMI) is proposed to
guarantee the exist ence of the desired fault detection filter ; (3) expli-
cit expression of the desired filter parameters is char act erized by
matrix decomposition, congruence transformation, and conve x opti-
mization technique; (4) weighting fault signal approach is employed
to improve the performance of the fault detection system.
Notation: The notation used throughout this paper is fairly stan-
dard. R
n
and R
nm
, denote, respectivel y, the n-dimensional Euclidean
space and the set of all n m real matrices; R
þ
refers to the set of all
nonnegative real numbers; C
2;1
ðR
n
R
þ
; RÞ represents the family of
all real-valued functions Vx; tðÞdefined on R
n
R
þ
which are con-
tinuously twice differentiable in xA R
n
and once differentiable in
t A R
þ
; S
δ
denotes the family of xA R
n
and satisfies x
jj
r δ.The
notation M
T
represents the transpose of the matrix M and for sym-
metric matrices X and Y, while the notation X 4 Y (respectivel y , X Z Y)
means that the matrix X Y is positive definite (respectively, positive
semi-definite). I and O refer to the identity matrix and a zero matrix
with appropriate dimensions, respectiv ely. Let ð
Ω; F ; fF
t
g
t Z 0
; PÞ be a
complete probability space with a filtr ation fF
t
g
t Z 0
satisfying the
usual conditions (i.e., the filtration contains all Pnull sets and is right
continuous); Efg refers to the expectation operator with respect to
the giv en probability measure P; L
2
½0; 1Þ is the space of square-
integrable vect or functions over ½0; 1Þ; ‖ ‖
2
stands for the usual
L
2
½0; 1Þ norm, ‖ ‖
E
2
denotes the norm in L
2
ððΩ; F ; PÞ; ½0; 1ÞÞ,
while jj is the Euclidean norm in R
n
. The notation trðÞ stands for
the trace of a matrix; diagð⋯Þ denotes a block diagonal matrix. In
block symmetric matrices or long matrix expressi ons, we use ð
n
Þ to
represent a term that is induced by symmetry . Matrices, if their
dimensions are not explicitly stated, are assumed to have compatible
dimensions for algebraic operations.
2. Problem formulation and preliminaries
Fix a probability space ð
Ω; F ; PÞ, we consider a class of SMJSs
with parameter uncertainties and time-varying delays described
by the following model ð
ΣÞ:
dxðtÞ¼½Aðt; r
t
ÞxðtÞþA
1
ðt; r
t
Þxðt τ
1
ðt; r
t
ÞÞ
þB
0
ðt; r
t
ÞuðtÞþBðt; r
t
ÞωðtÞ
þB
1
ðt; r
t
Þf ðtÞ dt þ Eðt; r
t
Þxðt τ
2
ðt; r
t
ÞÞ dϖðtÞ; ð1Þ
dyðtÞ¼½Cðt; r
t
ÞxðtÞþC
1
ðt; r
t
Þxðt τ
1
ðt; r
t
ÞÞ
þD
0
ðt; r
t
ÞuðtÞþDðt; r
t
ÞωðtÞ
þD
1
ðt; r
t
Þf ðtÞdt þ Fðt; r
t
Þxðt τ
2
ðt; r
t
ÞÞ dϖðtÞ; ð2Þ
xðtÞ¼
ϕðtÞ; 8t A ½τ; 0; ð3Þ
where xðtÞ A R
n
is the state vector; yðtÞ A R
p
is the measured output;
uðtÞA R
m
is the kno wn input , ωðtÞ A R
q
is the unknown disturbance
input; f ðtÞ A R
l
is the fault to be detected, uðtÞ; ωðtÞ and f(t)belongto
L
2
½0; 1Þ; ϕðtÞ is the initial condition. ϖðtÞ is a zero-mean real scalar
Brownian motion (Wiener process) on ð
Ω, F , PÞ relative to an
increasing family ðF
t
Þ
t A ½0;1Þ
of salgebras F
t
F satisfying
Efd
ϖðtÞg ¼ 0; EfdϖðtÞ
2
g¼dt: ð4Þ
fr
t
g is a continuous-time Markovian process with right contin-
uous trajectories and taking values in a finite set S ¼f1; 2; …; Ng
with transition probability matrix
Π ¼fπ
ij
g given by
Pfr
t þ h
¼ jjr
t
¼ ig¼
π
ij
hþoðhÞ; iaj
1þ
π
ii
hþoðhÞ; i ¼ j
(
ð5Þ
where h 4 0, lim
h-0
oðhÞ=h ¼ 0 and π
ij
Z 0, for ia j, is the transition
rate from mode i at time t to mode j at time t þh, satisfying
π
ii
¼ ∑
N
j ¼ 1;j a i
π
ij
; ð6Þ
and let
η ¼ maxfjπ
ii
j; iA Sg: ð7Þ
τ
1
ðt; r
t
Þ, τ
2
ðt; r
t
Þ are the mode-dependent time delays, for
8r
t
¼ i A S, satisfying
0o
τ
1
ðt; r
t
Þr τ o 1; 0o τ
2
ðt; r
t
Þr τo 1;
_
τ
1
ðt; r
t
Þr μ
1i
o 1;
_
τ
2
ðt; r
t
Þr μ
2i
o 1; ð8Þ
where
τ
,
μ
1i
and
μ
2i
are constant scalars. In (3), ϕðtÞ is a vector-
valued initial continuous function defined on the interval ½
τ; 0,
initial mode r
0
A S.
Remark 1. The motivation we consider system ð
ΣÞ containing
time delays stems from the fact that time delays arise quite
naturally and exist in many practical applications, such as biolo-
gical systems, hydraulic processes, chemical systems, temperature
processes, electrical networks, and are frequently a primary source
of instability and poor performance of a system. It is necessary and
reasonable to take into account time delays when investigating
various kinds of practical systems as well as Markovian jump
systems [6,26,29–32].
For the sake of simplicity, in the sequel, for each r
t
¼ iA S,a
matrix Oðt; r
t
Þ will be represented by O
i
ðtÞ; for example, Aðt; r
t
Þ is
represented by A
i
ðtÞ, and A
1
ðt; r
t
Þ is represented by A
1i
ðtÞ, etc. And
the parameters in system ð
ΣÞ are described as follows:
Aðt; r
t
Þ¼A
i
ðtÞ¼A
i
þΔA
i
ðtÞ; A
1
ðt; r
t
Þ¼A
1i
ðtÞ¼A
1i
þΔA
1i
ðtÞ;
B
0
ðt; r
t
Þ¼B
0i
ðtÞ¼B
0i
þΔB
0i
ðtÞ; B
1
ðt; r
t
Þ¼B
1i
ðtÞ¼B
1i
þΔB
1i
ðtÞ;
Eðt; r
t
Þ¼E
i
ðtÞ¼E
i
þΔE
i
ðtÞ; Cðt; r
t
Þ¼C
i
ðtÞ¼C
i
þΔC
i
ðtÞ;
C
1
ðt; r
t
Þ¼C
1i
ðtÞ¼C
1i
þΔC
1i
ðtÞ; D
0
ðt; r
t
Þ¼D
0i
ðtÞ¼D
0i
þΔD
0i
ðtÞ;
D
1
ðt; r
t
Þ¼D
1i
ðtÞ¼D
1i
þΔD
1i
ðtÞ; Fðt; r
t
Þ¼F
i
ðtÞ¼F
i
þΔF
i
ðtÞ;
Bðt; r
t
Þ¼B
i
ðtÞ¼B
i
; Dðt; r
t
Þ¼D
i
ðtÞ¼D
i
; ð9Þ
G. Zhuang et al. / ISA Transactions 53 (2014) 1024–1034 1025