You are on page 1of 6

51st IEEE Conference on Decision and Control

December 10-13, 2012. Maui, Hawaii, USA

Fault Detection for Switched Systems Based on A Deterministic Method


Ali Abdo, Steven X. Ding, Jedsada Saijai, Waseem Damlakhi
Institute for Automatic Control and Complex Systems (AKS)
University of Duisburg-Essen, 47057 Duisburg, Germany.
ali.abdo@uni-due.de, steven.ding@uni-due.de
jedsada.saijai@uni-due.de, waseem.damlakhi@uni-due.de

Abstract— This paper addresses fault detection (FD) issues In the last few years, some methods of FDI have been
for switched systems. The basic idea behind the proposed applied for switched systems, which are reported in [20], [4]
method is to estimate the disturbances and then use it in the and [19]. In these papers, the standard norm-based evaluation
evaluation function. The main contributions of this paper are
summarized as follows: (1) develop a deterministic method for method has been used to evaluate the switched residual
FD. (2) Improve the fault detection performance. (3) Enhance signal, and the threshold has been set constant overall the
the fault detectability for the switched systems by utilizing the switched system. In [1] and [2] we have studied the robust
available information provided by each local sub-model in the FD problem for the switched systems with adaptive threshold
residual generation, evaluation and threshold computation. The setting taking into account the behavior of the switched
system under consideration in this paper is linear discrete-time
switched systems. A switched residual signal will be generated systems. In this paper, we will continue the study on FD
based on the Parity Space approach. The proposed method will the switched systems and propose a new scheme for the
be illustrated by an example. evaluation of residual signal for the following objectives:
• develop a new deterministic method for residual eval-
I. INTRODUCTION uation and threshold setting based on the disturbance
estimation.
Model-based fault detection technique has been
investigated in the last two decades, and it is accepted • show the advantages of the proposed approach in com-
nowadays as a powerful tool to solve fault diagnoses parison with the standard one.
problems in technical processes, [11], [12], [6], [8] and
• achieve a high FD performance for switched systems.
references therein. The theory of the model-based fault
Since, all the available information for each sub-model
diagnosis is well-established for linear time invariant (LTI)
should be taken into account in the FD design.
systems, see e.g. [3], [6], [8] and [12]. In comparison, FD
solutions in the nonlinear systems are limited to some special • use the parity space approach in FD design for switched
kinds of nonlinearities, as in [18] and [16]. Representing systems. By utilize the benefits of this approach in its
the nonlinear system as a group of LTI sub-models is one simplicity in computation and its elimination of the
of the techniques, which has been used to overcome the initial conditions’ effects.
restricted FDI solutions for nonlinear systems, where the This paper is organized as follows. After the introduction
model-based FD approaches can be used, as in the Multiple in Section I, preliminaries and the basic idea are presented
Models, T-S Fuzzy System and Hybrid Switched Systems. in Section II. In Section III, the problem to be addressed is
formulated. The solution and a comparison with the trivial
FD issues for linear discrete-time switched systems solution are given in Section IV. In Section V the proposed
will be investigated in this paper for the following reasons: scheme is illustrated by an example.
• switched systems show a promise and convince ability Notations: The notations used throughout this paper are
in modeling and representing the high complex systems generally standard. X T , X −1 and X − are the transpose,
in sub-models, which are normally linear and simple. inverse and pseudo inverse of the matrix X, respectively. 0
is a zero matrix of appropriate dimension. I represent the
• switched systems exist in physical and practical techni- identity matrix andrL2 -norm for each sub-model is defined
cal systems, as in the control of mechanical system, ∞
process control, automotive industry, power systems, as ||xσ (k) (k)||2 = ∑ xσT (k) (k)xσ (k) (k), where σ (k) is the
k=0
aircrafts, traffic control and many other fields, see e.g. switching signal, which will be defined in the next section.
[5]. Therefore, the reliability and safety of these systems
can be achieved by applying FDI to these systems. II. P RELIMINARIES
• a lot of work has been done in designing a robust Consider the following linear discrete-time switched sys-
control law for switched systems; however, to the best tems:
of the author’s knowledge, FDI problems have not been
intensively investigated for switched systems, and it is xk+1 = Aσ (k) xk + Bσ (k) uk + Ed,σ (k) dk + E f σ (k) fk (1)
still an open issue. yk = Cσ (k) xk + Dσ (k) uk + Fd,σ (k) dk + Ff σ (k) fk (2)

978-1-4673-2064-1/12/$31.00 ©2012 IEEE 568


where, x ∈ R n is the system state vector, y ∈ R m is the The set of vectors which satisfied (3) is called parity space
measurement output vector and u ∈ R p is the input vector. of the s-th order.
d ∈ R kd represents the disturbance vector and f ∈ R k f is
P(s,σ ) , {v(s,σ ) |v(s,σ ) H(o,s,σ ) = 0} (5)
the vector of the faults to be detected.
Switching signal σ (k) can be defined to be time-dependent, and, ys (k),
 us (k), H(u,s,σ) are givenas follows: 
state-dependent, input/output dependent or parameter y(k − s) u(k − s)
dependent. In this study, it is assumed that, the switching  y(k − s + 1)   u(k − s + 1) 
   
signal is unknown in a priory but its value is real-time ys (k) =  .. , us (k) =  .. ,
 .   . 
available. The switching rule σ (k) takes values in the finite
set I = {1, 2, ..., N}, i.e. σ (k) ∈ I . y(k) u(k)
 
The Index I is related to the sub-models in the Dσ 0 ... 0
switched systems, where: AI , {A1 , A2 , ..., AN },  .. .. 
 Cσ Bσ Dσ . . 
BI , {B1 , B2 , ..., BN }, CI , {C1 ,C2 , ...,CN }, and H(u,s,σ ) =   ∈
 .. .. .. 
DI , {D1 , D2 , ..., DN }. These matrices are known, and it is  . . . 0 
in appropriate dimensions. Cσ Aσs−1 Bσ ... Cσ Bσ Dσ
In the switched systems, each linear sub-model may be R m(s+1)×p(s+1) .
affected by different disturbances and faults. Therefore,
the disturbance and fault matrices are defined as follows: The residual signal for the switched systems given in (1)
© ª and (2) can be formulated as follows:
Ed,I , {Ed1 , Ed2 , ..., EdN }, E f ,I ,© E f 1 , E f 2 , ..., Eªf N ,
Fd,I , {Fd1 , Fd2 , ..., FdN } and Ff ,I , Ff 1 , Ff 2 , ..., Ff N . rs,σ (k) = v(s,σ ) (H(o,s,σ ) x(k −s)+H(d,s,σ ) ds (k)+H( f ,s,σ ) fs (k))
   (6)
In this work, it is assumed that all sub-models are Cσ d(k − s)
stable and there is enough dwell-time between any two  Cσ Aσ   d(k − s + 1) 
   
consecutive switching intervals. Therefore, the global where H(o,s,σ ) =  .. , ds (k) =  .. ,
 .   . 
stability of the switched systems can be ensured by slow
Cσ Aσs d(k)
switching criteria, see [15] and references therein.  
f (k − s)
Generally speaking, a model-based FD system consists of  f (k − s + 1) 
 
residual generation and residual evaluation with threshold fs (k) =  .. ,
 . 
setting, the following subsections will highlight the FD
f (k)
design procedure.  
Fd,σ 0 ... 0
 .. .. 
A. Residual generation - Parity space approach  Cσ Ed,σ Fd,σ . . 
H(d,s,σ ) =   ..
,

 .. .. 
The first step to a successful FD is to generate the residual . . . 0
C A s−1 E ... Cσ Ed,σ Fd,σ 
signal. There are different approaches to generate the residual  σ σ d,σ
signal in model-based FD systems, [8], [6]. Among these Ff ,σ 0 ... 0
approaches, the Parity space approach, which is widely  .. .. 
 Cσ E f ,σ F f ,σ . . 
accepted to generate the residual signal due to its simplicity H( f ,s,σ ) = 
 ..
.

 .. ..
and straightforward computation, [8], [7], [10] and [17]. . . . 0 
The basic idea of designing the parity space based residual Cσ Aσs−1 E f ,σ ... Cσ E f ,σ Ff ,σ
generator is to find a so-called parity vector, vs ∈ R m(s+1) , Based on the parity space relation v(s,σ ) H(o,s,σ ) = 0, Eq. (6)
such that: is rewritten as follows:
vs Ho,s = 0 (3) r(s,σ ) (k) = v(s,σ ) (H(d,s,σ ) ds (k) + H( f ,s,σ ) fs (k)) ∈ R m(s+1)
(7)
where   where, rs,σ (k) ∈ R m(s+1) , H(d,s,σ ) ∈ R m(s+1)×kd (s+1) , ds (k) ∈
C
£ ¤  CA  R kd (s+1) , H( f ,s,σ ) ∈ R m(s+1)×k f (s+1) and fs (k) ∈ R k f (s+1) .
 
vs = v0,s v1,s . . . vs,s , Ho,s =  ..  ∈ R m(s+1)×n B. Residual evaluation
 . 
CAs Residual signal contains the most useful information
and s > 0 is the order of parity relation. regarding the systems behavior. Therefore, the evaluation
The switched residual signal rs,σ (k) is realized using the of this signal will create a mathematical feature of the
measurements of output and input signals as follows: system behavior, as a result it can be used to detect
the faults. In literatures, there are two existing methods
rs,σ (k) = vs,σ (ys (k) − Hu,s,σ (k)us (k)) (4) for residual evaluation depending on the system under

569
Fig. 1: (A) is the standard model based FDI system and (B) is the proposed design method

consideration. The first one is statistical testing, such as This idea will be investigated in the following two sec-
likelihood ratio and generalized likelihood ratio, which are tions, showing the advantages of this scheme comparing with
used for residual evaluation, threshold setting and decision the standard FD procedure.
making for stochastic systems, see e.g. [8], [3], [13] and [14].
III. P ROBLEM FORMULATION
Norm-based is another method for residual evaluation, Consider the residual signal given in (7):
such as the Peak value and the root mean square (RMS)
value, see e.g. [8] and [9]. This method has been widely rs,σ (k) = v(s,σ ) (H(d,s,σ ) ds (k) + H( f ,s,σ ) fs (k)) (10)
used in practice. Let Mσ = v(s,σ ) H(d,s,σ ) ∈ R 1×kd (s+1) and Gσ =
C. Basic idea v(s,σ ) H( f ,s,σ ) ∈ R 1×k f (s+1) . Then (10) can be transformed to
the form of (8) as follows:
The basic idea of this paper is to utilize the residual signal
which contains rich information about the system features. rs,σ (k) = Mσ ds (k) + Gσ fs (k) (11)
Therefore, a successful FD system should be designed by
The standard FD problem formulation is defined as follows:
making use of the full information in the residual signal.
Consider the following residual signal:
Problem I: Standard FD problem
rs,σ (k) = Mσ ds (k) + Gσ fs (k) (8) Given the switched residual signal (11), then:
1) define an evaluation function Jσ for residual signal.
where the disturbance ds (k) and the fault fs (k) are L2 - 2) set the threshold value.
bounded values. Mσ and Gσ are known matrices in appro- 3) detect the fault by choosing the decision rule.
priate dimensions.
The standard and trivial FD procedure is to evaluate the
residual signal given in (7), as shown in Figure (1.A). In The main contribution in this paper is to reformulate the
this paper, an alternative method for residual evaluation is previous problem to new form:
proposed. The root of this new design scheme is from the
estimation in the stochastic process - generalized likelihood Problem II: The new design scheme
ratio. As shown in Figure (1.B) the evaluation function is Given the switched residual signal (11), then:
based on the disturbance estimation which is derived from 1) estimate the disturbance ds (k) from the residual signal.
the residual signal. The importance of this design scheme is 2) define an evaluation function Jσ based on the distur-
illustrated as follows: bance estimation.
3) set the threshold value.
1) it develops a deterministic method for FD.
4) detect the fault by choosing a suitable decision rule.
2) it will lead to a significant improvement in the FD
performance, as will show in the next sections.
Based on the previous two problems, a discussion will be
After evaluation of the estimated disturbance, the fault can be
hold to prove that, the new design scheme given in Problem
detected based on the decision rule. Therefore, the following
II provides a better fault detection performance than the
two hypotheses have been defined:
½ standard fault detection scheme given in Problem I.
0 , H0 Fault − free
f= (9)
f1 6= 0 , H1 Faulty − case The solutions of these problems will be addressed in

570
the next section. • set Jth,σ = max kds k2 = ρmax (Mσ )kd
On-line computation:
IV. E SSENTIAL RESULTS AND SOLUTIONS OF P ROBLEMS • implement of the residual generator (4) for switched
I − II systems
In this section, the solutions for the previous problems will • residual evaluation krs,σ (k)k2,L =
be given in sequence. s
L

A. Solution of Problem I ∑ rs,T σ (k + i)rs,σ (k + i)


i=0
For the residual signal given in (11), without any loss of • switched decision logics krs,σ (k)k2,L > Jth,σ ⇒ H1 ,
generality, the disturbance under consideration is assumed to
be: otherwise fault-free ⇒ H0 .
• L2 -bounded disturbance, i.e. ds (k) ∈ R kd (s+1) :
L2,[k,k+L] -bounded, where L is the evaluation time It can be said that, the previous solution is the standard
window. one, and it will be used to demonstrate the advantages of the
• a normalized value of disturbance is considered, i.e. proposed scheme presented in Problem II.
L
∑ ds,2 j (k + i) 6 1, j = 1, ..., kd . B. Solution of Problem II
i=0
The intuitive evaluation function Jσ for the switched residual The solution of this problem presents a new design scheme
signal given in (10) can be defined as follows: for FD systems.
à !1/2 In fault free case, Eq. (11) will turn to:
L
Jσ = ∑ rs,T σ (k + i)rs,σ (k + i) (12) rs,σ (k)| f =0 = Mσ ds (k) (14)
i=0

Normally, the threshold is set as the maximum influence of It is important to notice that; Mσ is known matrix, and
unknown inputs on the residual signal in fault-free case. In rs,σ (k) is available instantly. Based on these notations, the
this paper, the following form of the threshold will be set: disturbance ds (k) can be estimated as follows:

Jth,σ = sup krs,σ (k)k2,L dˆs (k) = Mσ− rs,σ (k) (15)
ds, j , f =0
à !1/2 where Mσ−
is the pseudo inverse of the matrix Mσ , which
L
provides a least squares estimation of the disturbance.
= sup
ds, j , f =0
∑ rs,T σ (k + i)rs,σ (k + i) To this end, the evaluation function Jσ is defined based on
i=0
à !1/2 the disturbance estimation given in (15) as follows:
L
à !1/2
= sup ∑ ds,T j (k + i)MσT Mσ ds, j (k + i) L
ds, j , f =0 i=0
à !1/2
Jσ = ∑ dˆs,T j (k + i)dˆs, j (k + i)
L i=0
à !1/2
≤ ρmax (Mσ ) ∑ ds,T j (k + i)ds, j (k + i) L
i=0 = ∑ rs,T σ (k + i)(Mσ− )T Mσ− rs,σ (k + i) (16)
≤ ρmax (Mσ )kd , j = 1, ..., kd i=0

Then the threshold will be set as follows: The threshold will be set as the maximum influence of
unknown inputs on the evaluation function given in (16),
Jth,σ = ρmax (Mσ )kd (13) i.e.
where ρmax is the maximum singular value for Mσ . Jth,σ = sup kdˆs, j (k)k2,L
ds, j , f =0
For switched systems, the following detection logic à !1/2
L
can be used: = sup ∑ rs,T σ (k + i)(Mσ− )T Mσ− rs,σ (k + i)
ds, j , f =0 i=0
Jσ > Jth,σ =⇒ H1 : false − alarm à !1/2
L
Jσ ≤ Jth,σ =⇒ H0 : fault − free = sup ∑ ds,T j (k + i)MσT (Mσ− )T Mσ− Mσ ds, j (k + i)
ds, j , f =0 i=0
The solution of Problem I is summarized in the following à !1/2
L
algorithm:
≤ ∑ ds,T j (k + i)ds, j (k + i)
i=0
Algorithm I: Solution of Problem I ≤ kd , j = 1, ..., kd
Off-line computation: Then the threshold will be set as follows:
• compute H(0,s,σ ) , vs,σ , H(u,s,σ ) and H(d,s,σ )
Jth,σ = kd (17)

571
Based on the hypothesis in (9) the following detection The evaluation function for the new design scheme
logic is defined: leads to:
Fault f (k) is detected when:
Jσ > Jth,σ =⇒ H1 : false − alarm
Jσ ≤ Jth,σ =⇒ H0 : fault − free Jσ > Jth,σ = kd
L
The solution of Problem II is summarized in the following
algorithm: ∑ dˆs,T j (k + i)dˆs, j (k + i) > kd
i=0
Algorithm II: Solution of Problem II L
Off-line computation: ∑ rs,T σ (k + i)(Mσ− )T (Mσ− )rs,σ (k + i) > kd
− i=0
• compute H(0,s,σ ) , vs,σ , H(u,s,σ ) , H(d,s,σ ) , Mσ , Mσ
L
• set Jth,σ = max kds k2 = kd ∑ (Fσ (d, f ))T (Mσ− )T (Mσ− ) (Fσ (d, f )) > kd (19)
i=0
On-line computation:
Since,
• implement of the residual generator (4) for switched
systems (Mσ− )T (Mσ− ) ≥ ρmax
−1
(Mσ )
• disturbance estimation dˆs (k) = Mσ
− r (k)
s,σ this leads to:
• disturbance evaluation L
s
L Jσ = ∑ (Fσ (d, f ))T (Mσ− )T (Mσ− ) (Fσ (d, f ))
kds, j,k k2,L = ∑ ds,T j (k + i)ds, j (k + i) ∀ j = 1, ..., kd . i=0
L
i=0
• switched decision logics kds, j,k k2,L > Jth,σ ⇒ H1 , ≥ ∑ (Fσ (d, f ))T ρmax
−1
(Mσ ) (Fσ (d, f ))
i=0
otherwise fault-free ⇒ H0 .
It becomes clear that, the evaluation function of the new
design scheme is grater than the evaluation function for the
The improvement on fault detectability will be investigated standard detection scheme. Which means that, if a fault has
in the following subsection. been detected based on relation (18) - the standard detection
C. Discussion and remarks scheme -, then it is also detectable by relation (19) - the
new design scheme -.
There are different ways to evaluate the performance
of the FD system, [6] and [8]. In this paper, the Fault In addition to the previous explanation, the new design
Detection Rate (FDR) and the False Alarm Rate (FAR) will scheme can detect some faults, which are otherwise
be used to evaluate the FD performance. undetectable by the standard detection scheme. Therefore,
False Alarm Rate (FAR): the proposed scheme in this paper provides higher fault
False alarms caused by disturbances, but since the detectability.
threshold is set based on the maximum disturbance, then the
FAR for the standard solution and the new scheme is “zero”. Remark 1: The parity space approach has been chosen to
generate the residual signal due to:
Fault Detection Rate (FDR):
In order to study the FDR for the standard solution and the • the design can be implemented straightforward manner,
new design scheme, the evaluation function for the solutions and its computation is simple.
of Problem I and II will be investigated. • based on this approach the residual signal is independent
of the sub-models initial states.
The standard detection scheme leads to: • this approach helps to eliminate the transient effect
Fault f (k) is detected when: which appears between two switched sub-models.
• this approach provides the simplest way for estimation
Jσ > Jth,σ = ρmax (Mσ )kd of ds (k).
L
∑ rs,T σ (k + i)rs,σ (k + i) > ρmax (Mσ )kd
Remark 2: The norm-based residual evaluation is widely
i=0
L used in practice due to:
∑ (Fσ (d, f ))T (Fσ (d, f )) > ρmax (Mσ )kd • it required less on-line computation.
i=0
L
• this method allows a systematic computation of
∑ (Fσ (d, f ))T ρmax
−1
(Mσ ) (Fσ (d, f )) > kd (18) threshold using a well established robust control theory.
i=0

where Fσ (d, f ) = Mσ ds, j (k + i) + Gσ fs (k). Remark 3: It is important to mention that, the new scheme
can be applied to LTI system as well as switched systems.

572
8 8
Fault Standard evaluation
6 6 Threshold based algorithm 2
4 Evaluation based disturbance estimation
4 Threshold based algorthim 3
2
2
0
0
0 1000 2000 3000 4000 5000 6000
SamplingTime 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000
Sampling time
Fig. 2: Sensor faults.
Fig. 3: Standard and the proposed evaluation scheme for FD

V. E XAMPLE
space approach is used in FD design for switched systems,
Three sub-models are defined as follows: utilizing the benefits of this approach in its simplicity in
Sub-model 1: computation and its elimination of the initial conditions’
" # " # " #
0.53 −0.03 0.3 0.015 effects and the high transient effects between the switched
A1 = , B1 = , Ed1 =
−0.92 0.47 2.3 −0.012 sub-models.
" # " # " #
−286.2 −14.1 207.3 1 R EFERENCES
C1 = , D1 = , Fd1 =
0 1 0 0 [1] A. Abdo, S. X. Ding, W. Damlakhi and J. Saijai. ”Robust Fault
Detection Filter Design for Uncertain Switched Systems with Adaptive
Sub-model 2: Threshold Setting”. 50th IEEE CDC and ECC 2011, USA, 2011.
[2] A. Abdo, S. X. Ding, J. Saijai and W. Damlakhi. Integration
" # " # " # of Residual Evaluation and Threshold Computation into Switched
0.82 −0.01 0.08 −0.02 Fault Detection System. 8th IFAC Symposium on Fault Detection,
A2 = , B2 = , Ed2 =
0.19 0.77 1.16 −0.002 Supervision and Safety of Technical Processes, Mexico City, Mexico,
" # " # " # 2012.
−139.5 1.08 72.5 1 [3] M. Basseville and I. V. Nikiforov. ”Detection of Aprupt Changes
C2 = , D2 = , Fd2 = - Theory and Application”. Prentice Hall, Information and System
0 1 0 0 Sciences Series, 1993.
[4] D. E. C. Belkhiat, N. Messai, N. Manamanni. ”Design of a robust
Sub-model 3: fault detection based observer for linear switched systems with external
" # " # " # disturbances”. Nonlinear Analysis: Hybrid Systems, 2010.
0.89 −0.017 0.04 −0.02 [5] M. S. Branicky. ”Introduction to hybrid systems”. Handbook
A3 = , B3 = , Ed3 = of Networked and Embedded Control Systems, pp. 91-116, Boston:
0.28 0.87 0.85 −0.003
Birkhauser, 2005.
" # " # " # [6] J. Chen and R. Patton. ”Robust Model-based Fault diagnosis for
−103.0 1.01 50.1 1 Dynamic Systems”. Kluwer Academic Publisher, 1999.
C3 = , D3 = , Fd3 = [7] X. Ding, L. Guo and T. Jeinsch. ”A characterization of parity space
0 1 0 0
and its application to robust fault detection”. IEEE Trans. Autom.
Two faults of different gains have been applied as in Contr., Vol. 44, Feb., S. 337 343, 1999.
[8] S. X. Ding ”Model-Based Faults Diagnosis Techniques”. Springer
Fig. (2). The difference in fault detectability between the Verlag, 2008.
standard evaluation technique and the proposed one is clearly [9] A. Emami-Naeini, M. Akhter and S. Rock. ” Effect on model
remarked. Fig. (3) shows that, the large faults can be detected uncertainty on failure detection: The threshold selector ”. IEEE
Transactions on Automatic Control , vol. 33, pp. 11061115, 1988.
by both evaluation methods. However, the proposed evalua- [10] P. M. Frank. ”Fault diagnosis in dynamic systems using analytical
tion method based on disturbance estimation can detect the and knowledge-based redundancy A survey”. Automatica, vol. 26,
small fault which the standard method cannot detect it. This pp. 459474, 1990.
[11] P. M. Frank, S. X. Ding. ”Survey of robust residual generation and
result can be explained that, the proposed method provides evaluation methods in observer-based fault detection systems”. Journal
greater evaluation than the standard one, which leads to of Process Control, vol. 7, pp. 403-424, 1997.
improve the fault detectability. Furthermore, the threshold [12] J. J Gertler. ”Fault Detection and Diagnosis in Engineering systems”.
Marcel Dekker, 1998.
value for the proposed method is less than the standard [13] F. Gustafsson. ”Adaptive Filtering and Change Detection”. John Wiley
method. This property increases the sensitivity of the FD and Sons, LTD, 2000.
system in case of small faults. [14] E. Lehmann. ”Testing Statistical Hypothesis”. Wadsworth, Inc., 1991.
[15] D. Liberzon. ”Switching in Systems and Control”. Birkhauser, 2003.
It is important to note that, the generated residual signal [16] K. Nguang, P. Shi and S. X. Ding. ”Fault detection filter for uncertain
via parity space approach has reduced the transient effects fuzzy systems: an LMI approach”. IFAC world congress 2005.
between the switched sub-models. [17] R. J. Patton and J. Chen. ”Parity space approach to model-
based fault diagnosisA tutorial survey and some new results”. in
VI. C ONCLUSIONS Proc.IFAC/IMACS Symp. SAFEPROCESS, BadenBaden, 1991.
[18] J. Patton, J. Chen, J. Lopez-Toribio. ”Fuzzy observer for nonlinear
In this paper, a deterministic method for fault detection has dynamic system fault diagnosis”. Proc. for IEEE Conference 1998.
been developed. The proposed scheme has been compared [19] G. Sun, M. Wang, X. Yao, L. Wu. ”Fault detection of switched linear
systems with its application to turntable systems”. Journal of Systems
with the standard FD scheme to illustrate the proficiency of Engineering and Electronics Vol. 22, No. 1, 2011.
the new design scheme. The proposed design scheme can [20] D. Wang, W. Wang and P. Shi. ”Robust Fault Detection for Switched
achieve higher fault detectability than the standard one, and Linear Systems With State Delays”. IEEE Tran. on systems, man, and
Cypernetics-Part B, Vol. 39, No. 3, 2009.
by that, it improves the model-based FD approach. Parity

573

You might also like