Fault Detection and Isolation Scheme Based on Parity

Space Method for Discrete Time-Delay System 119

X6

Fault Detection and Isolation Scheme

Based on Parity Space Method for
Discrete Time-Delay System
Hongyu Wang, Zuohua Tian, Songjiao Shi and Zhenxin Weng
Shanghai Jiaotong University
China

1. Introduction
In recent years, fault detection and isolation (FDI) problem in dynamitic system has been
paid more and more attention. A great number of methods for FDI have been proposed
(Chow & Willsky, 1984; Frank & Ding, 1997; Chen & Patton, 1999; Patton et al., 2000;
Venkatasubramanian et al., 2003). All of the FDI schemes are concerned with a core stage:
the generation of the residual signals. The difference between the measurement of the
system and its estimation is called residual, whose values are zero or near to zero when no
fault occurs while differ distinctly from zero otherwise. Appropriate decisions such as the
occurrence, magnification, type, location, etc. of the faults are called fault isolation, which
are achieved by residual evaluation.
In the field of analytical model-based FDI techniques, the analytical redundancy relations of
the system are used to create residual signal. The approaches can be roughly classified into
observer-based approaches and parameter estimation approaches. Parity space approaches
have been proved to be structurally equivalent to the observer-based though the design
procedures differ (Gertler, 1991). However, the parity space methodology using the
temporal redundancy has its advantages, especially in the discrete system. This method was
firstly generalized by the (Chow & Willsky, 1984).
Time delays are inherent in many real physical processes (i.e. mechanical and chemical
processes, long transmission lines in pneumatic systems, power and water distribution
networks, air pollution systems etc.) Over the past two decades, analysis and synthesis of
dynamic time-delay systems have attracted a great deal of interests (Dugard & Verriest,
1997; Yang & Saif, 1998). However, there are relative fewer research results on FDI of time-
delay systems (KRATZ et al., 1998; Zhong et al., 2004).
This paper proposes a method to deal with the FDI problem for the linear discrete-time
systems with delays. The results in (KRATZ et al., 1998) are extended. Both fault detection
and fault isolation method are proposed. The occurrence of the fault can be detected timely
and the position of the fault can be located exactly. A numerical example is given to
illustrate the design method at the end.

[Link]
120 Fault Detection

2. Mathematical Preliminaries
A time delay operator  is defined according to (KRATZ et al., 1998). f ( k )  f ( k  1) for

any discrete-time function f . It is easy to understand that f ( k )   f ( k )  f ( k  2) ,

2

 f ( k )  f ( k  w) .
w
Consider a linear discrete time-delay system described by

 x (k  1)   Ai x (k  i )  B[u(k )  f a (k )]  E d d (k )
 v

 i 0 (1)
 y (k )  Cx (k )  F d (k )
 d

where x (k )  R n is the state vector, u(k )  R p is the control vector, y ( k )  R m is the output
vector, f a ( k )  [ f a1 (k ),, f ap (k )]T stands for the actuator faults. f ai ( k ), i  1, , p is
corresponding to the ith actuator fault. d (k )  R l
is disturbance
vector, d (k )  d . Ai (i  0, , v) , B , C , E d and Fd are constant matrices with appropriate
dimensions. Integers v  0 denotes the number of time delays in the state.
Using the operator  , the system (1) can be rewritten as

 x (k  1)  A() x (k )  B[u(k )  f a (k )]  E d d (k )

 y (k )  Cx (k )  Fd d (k )
(2)

where,

A()  A0  A1    Av v (3)

3. Parity Space Residual Generation for Fault Detection and Isolation

The task of FDI is to design a residual signal which is zero or near to zero in a fault free case
and non-zero when a fault occurs in the monitored system. Time delay implies that the state
of the system for the next time step is not only determined by the current state but also
concerned with the state of the former intervals. The recursion of equation (2) from time
instant k  L to time instant k yields

yL  H o , L x (k  L)  H u , L uL  H d , L d L  H fa , L f aL (4)
where

[Link]
Fault Detection and Isolation Scheme Based on Parity
Space Method for Discrete Time-Delay System 121

  L  
  
   
  L 
 
   
 
   
m

 y (k  L)    L 1
  y (k ) 
 
 y (k  L  1)       
yL (k )      y (k )   R ( L 1) m ,

  L 1 
       
    
 
 y (k )     y (k ) 
m

  
 1 
  
   
 1 
 
 m 
  Lm

 u( k  L )   u( k )   f a (k  L)   f a (k ) 
 u(k  L  1)   u( k )   f (k  L  1)   
uL (k )       R ( L 1) p , f (k )   a     f a (k )   R ( L 1) p ,
         
       
Lp aL Lp
 
 u( k )   u( k )   f a (k )   f a (k ) 
 d (k  L)  d (k )   C 
 d ( k  L  1)  d (k )   
d L (k )       R ( L 1) l , H   CA()   R ( L 1) m n ,
       
     L 
Ll o, L

 d (k )  d (k )  CA () 
 
 CE 
Fd
 Fd 0 

 CA() E d   R ( L 1) m( L 1) l , H fa , L  H u , L
d

 
Hd ,L CEd Fd
     
CA L 1 () B CAL  2 () B   CE Fd 
 d

Define the following parity space:


PL  v L v L H o , L  0  (5)
where v L  R 1 ( L 1) m
is row vector. Vectors belong to parity space are called parity vectors.
Residual signals can be created by the folowing equation:

rL (k )  v L  yL (k )  H u , L uL (k )   R (6)
Substituting equation (6) to equation (4) yields:

rL (k )  v L H d , L d L (k )  v L H fa , L f aL ( k ) (7)

[Link]
122 Fault Detection

It should be noted that the parity vectors v L satisfying equation (5) are not unique, and the
corresponding residual signals rL ( k ) are not unique. The freedom of the v L can be used to
creat specific residual signals, so as to fulfill specific design purpose.
The parity vectors v L can be described as v L  [v1 , v2 ,, v( L 1) m ] . Substituting it to equation (7),
the terms corresponding to disturbance v L H d , L d L (k ) and faults v L H fa , L f aL ( k ) can be
respectively expanded as follows:
v L H d , L d L (k )   1d1 (k )   2 d 2 (k )     l dl (k ) (8)

v L H fa , L f aL (k )  1 f a1 (k )  2 f a 2 (k )     p f ap (k ) (9)
Where  1 , 2 , , p , 1 , 2 ,,  p are polynomials corresponding to v1 , v2 ,, v( L 1) m and  .
The successful detection of a fault is followed by the fault isolation procedure which will
distinguish (isolate) a particular fault from others. While a single residual signal is sufficient
to detect faults, a set of residuals (or a vector of residual) is usually required for fault
isolation. According to (Chen & Patton, 1999), a commonly used scheme in designing the
residual set is to make each residual sensitive to all but one fault, i.e.
 r1  R ( f a 2 ,, f a 3 )

 

(10)
 ri  R ( f a1 ,, f ai 1 , f ai 1 ,, f ap )

 
rp  R ( f a1 ,, f ap 1 )

where R () denotes some functional relation, which works as the residual generator. This is
defined as a generalized structured residual set. The isolation can be performed by the
following logic:
ri  thi 
  f ai  0 (11)
rj  th for j  1, , i  1, i  1, p 
j

where thi means the fault isolation threshold to the corresponding fault.
To achieve the so called generalized structured residual set, let 1 , 2 ,,  p satisfy the
folowing equatinos:
 1  0




i 1  0 ,i  1, 2,, p (12)
  0
 i
i 1  0
Solving the equations (12) respectively can achieve a set of parity
vecters v Lfai i  1, 2, , p ,witch lead to a set of residual signals rLfai (k ), i  1, 2,, p by equation
(6). When the ith actuator fault occurs( f ai (k )  0 ), the corresponding residual signal rLfai ( k ) is
not affected, while the other residual signals rLfa1 (k )，rLfa1 (k )，
，rLfai 1 (k ), rLfai 1 (k ), rLfap (k ) are
affected. The isolation can be fulfilled by equaton (11).

[Link]
Fault Detection and Isolation Scheme Based on Parity
Space Method for Discrete Time-Delay System 123

4. Numerical Example
To illustrate the design process of the proposed mothed and verify its effectiveness, the
following numerical example is demonstrated. Consider a time delay system of the form (l),
  x1 ( k  1)   0.15 0.324 0   x1 ( k )   0 0   x1 (k  1) 
       0.308  x2 ( k  1) 
0.265
  x2 (k  1)   0.102 0.258 0   x2 (k )    0.406 0
  x3 (k  1)   0 0.412   x3 (k )   0 0.252 0.124   x3 ( k  1) 

0
 1 0 0    u1 (k )   f a1 (k )   1
 0 1 0   u (k )    f (k )    1 d (k )
      2   a2    
 0 0 1   u3 (k )   f a 3 (k )   1
 
  y (k )  1 0 0   x (k )  1
  1     
  y2 (k )   0 1 0   x2 (k )   1 d (k )
1

  y (k )  0 0 1   x (k )  1
  3    3   
Where d ( k )  0.01  randN (0,1) , randN (0,1) stands for zero mean, uint Gauss noise. Using the
operator  , the system (1) can be rewritten into the form (2), the system matrix is:

 0.324  0.265 
A()  0.102  0.406 
0.15 0
0.258 0.308 
 0 0.252 0.412  0.124 
In order to fulfill the FDI, a parity space with L  2 is established, the matrix H o , L , H u , L ,etc.
can be computed using the software MAPLE. Solving the equation (5) can get the parity
vecter:

 
T

 
t1
 1.767t1  2.160t1 - 0.03768t2 - 0.7173 t2 - 0.6968t2  0.2520t3  
  -0.01221t - 0.2357t - 0.1901 3t - 0.4170 2t - 0.1074 2t - 0.009721t  
2

  5 
  -0.1798t5  0.1133t6  0.2068 2t6  0.1807512 3t6  
4 4 4 4 5

   
  -0.308t2 - 0.412t3 - 0.124t3 - 0.09979t4 - 0.08162 2t4 - 0.2064t5 -  
   
  0.03819 t5  0.06224 t6 - 0.1023t6 - 0.1697t6  
vL   
2 2

 -6.667t1 - 0.6800t2  2.7067t2 - 0.3703t4  0.6968t4  0.7173 t4  1.104t5  

(13)
  
2

 -0.2774t5 - 0.6821 t6  0.1714t6 

2

 
 t2 
 
 
t3
 t 
 
4

 
t5
 t6 
Where t1 , t2 , t3 , t5 , t5 , t6  R and t1 , t2 , t3 , t5 , t5 , t6 are not all zeros. vL is a row 9 dimensions vector
with 6 fredom.

[Link]
124 Fault Detection

In order to achieve fault detection and isolation, a generalized residual set consists in three
residual signals should be created as show in Fig.1.

Fig. 1 Generalized residual set

In Fig.1, f ai stands for the ith actuator fault. Residual singal r1 is sensitive to f a 2 and f a 3 while
insensitive to f a1 . The situation of r2 and r3 are the same with r1 .
Substituting vL to equation (7), and expanding the terms corresponding to the actuator
fault v L H fa , L f aL (k ) into equation (9) yields, v L H fa , L f aL (k )  1 f a1 (k )  2 f a 2 (k )  3 f a 3 ( k ) .
Where 1 , 2 , 3 are polynomials corresponding to t1 , t2 , t3 , t5 , t5 , t6 and  . Using equation (12),
let

1  0

2  0
  0
 3
t1 , t2 , t3 , t5 , t5 , t6 can be achieved by solving the above equation, and substituting them into (13)
can get the parity vector corresponding to the first actuator fault f a1 :
 
T

 
0
 0.2520  0.07762 2

 0.4120  0.2509  0.03819 
 
2

 0.1020  0.4060 
vL fa1   0.2580 
 
 1 
 
 
0
 
 
1
 0 
Substituting vL fa1 to equation (6) can get generalized residual signal rLfa1 (k ) corresponding
to f a1 , denoted by r1 .
r1  (0.1020  0.40600 2 ) y1 (k )  (1  0.2580  0.2520 3  0.07762 4 ) y2 (k ) 
(  0.4120 2  0.2509 3  0.03819 4 ) y3 (k )  ( )u2 (k )  ( 2  0.3080 3 )u3 (k )
By the same process, let

[Link]
Fault Detection and Isolation Scheme Based on Parity
Space Method for Discrete Time-Delay System 125

1  0 1  0
 
2  0, 2  0
  0   0
 3  3
The residual signals correspond to f a 2 and f a 3 can be designed, denoted by r2 and r3 .
Now, the disign of the generalized residual set for actuator fault detection and isolation is
accomplished. The result of the simulation is show in Fig. 2 to Fig. 5.

input
1.5

1
u

u1
0.5
u2
u3

0
0 50 100 150 200 250 300 350 400 450 500
t/s
Fig. 2 System input signals

fault
1.5

f1
f2
1 f3
f

0.5

0
0 50 100 150 200 250 300 350 400 450 500
t/s
Fig. 3 Fault signals

[Link]
126 Fault Detection

output
3.5

2.5

1.5
y

0.5 y1
y2
0 y3

-0.5
0 50 100 150 200 250 300 350 400 450 500
t/s
Fig. 4 System output signals

residual

3
||r1||
||r2||
2 ||r3||
r

0 50 100 150 200 250 300 350 400 450 500

t/s
Fig. 5 Residual signals

There are 5 stages in the simulation process:

1. from time 0 second to time 100 second, the system works properly. r1 , r2 and r3 are
near to zreo.
2. from time 100 second to time 200 second, the actuator 1 suffers from fault, while
actuator 2 and actuator 3 work properly. r2 and r3 differ from zero while r1 keeps
zero nearby.
3. from time 200 second to time 300 second, the actuator 2 suffers from fault, while
actuator 1 and actuator 3 work properly. r1 and r3 differ from zero while r2 keeps
zero nearby.
4. from time 300 second to time 400 second, the actuator 3 suffers from fault, while
actuator 1 and actuator 2 work properly. r1 and r2 differ from zero while r3 keeps
zero nearby.
5. from time 400 second to time 500 second, the actuator 1 and actuator 2 suffer from fault
at the same time, while actuator 3 works properly. r1 , r2 and r3 differ from zero
simultaneously.

[Link]
Fault Detection and Isolation Scheme Based on Parity
Space Method for Discrete Time-Delay System 127

It can be concluded that when there is one actuator goes into fault, the above generalize
residual set based on pariy space can detect the fault and isolate which actuator corrupted
by the fault. However, when there are more than one actuators break into faults, the method
can only detect the fault, while it have no idea that which actuators corrupted by the fault.

5. Conclusion
A fault detection and isolation scheme for discrete time-delay system has been proposed in
this chapter. The scheme can not only detect the faults but also isolate (locate) the faults. To
fulfill the FDI, a generalized residual set in form of parity space is designed by the recursion
of the system equations. Each residual is sensitive to all but one actuator faults. The actuator
with fault can be isolated from the normal ones exactly. A time delay operator is used to
deal with the problem brought by the time-delay system. The effectiveness of the proposed
method has been verified by a numerical example.
However, further studies are required which include the follow aspects:
1. To determine an optimal recursion step L . Such that the residuals can obtain a certain
freedom to complete fault isolation, while the computation is minimized.
2. To extend the fault isolation result. The sensor faults and the actuator faults should be
discerned.
3. To enhance the reliability and robust performance of the FDI system.

6. References
Chen, J.; Patton, R. J. (1999). Robust Model-based Fault Diagnosis for Dynamic Systems, Kluwer
Academic Publishers
Chow, E. Y.; Willsky, A. S. (1984). Analytic redundancy and the design of robust fault
detection systems. IEEE Transaction on Automatic Control, Vol. 29, Jul. 1984, 603-614
Dugard, L.; Verriest, E. I. (1997). Stability and Control of Time-delay Systems. Lecture Notes in
Control and Information Sciences, Springer
Frank, P. M.; Ding, X. (1997). Survey of robust residual generation and evaluation methods
in observer-based fault detection systems. Journal of Process Control, Vol. 7, Jun. 1997
pp. 403-424
Gertler , J. Li, B. (1991). Analytical redundancy methods in fault detection and isolation,
Proceedings of the IFACIIMACS Symposium SAFEPROCESS 9, pp. 9-21, Baden-Baden
1991
KRATZ, F.; NUNINGER, W. & PLOIX, S. (1998). Fault detection for time-delay systems: a
parity space approach, Proceedings of the American Control Conference, pp. 2009-2011,
Philadelphia, Pennsylvania, 1998
Patton, R. J.; Frank, P. M. and Clark, R. N. (2000). Issues of Fault Diagnosis for Dynamic
Systems, Springer
Venkatasubramanian, V.; Rengaswamy, R.; Yin, K. and Kavuri, Surya N. (2003). S A review
of process fault detection and diagnosis: Part I: quantitative model-based methods.
Computers and Chemical Engineering, Vol. 27, 2003 pp. 293-311
Yang, H.; Saif, M. (1998). Observer design and fault diagnosis for state retarded dynamical
systems. Automatica, Vol. 34, Feb. 1998 pp. 217-227

[Link]
128 Fault Detection

Zhong, M.; Ye, H. & Wang, G. (2004). Multi-freedom design of fault detection filter for linear
time-delay systems, Proceedings of2004 8th Inernational Conference on Control,
Automation, Robotics and Vision, pp. 1630-1634, Kunming, China, 2004

[Link]
 Fault Detection
Edited by Wei Zhang

ISBN 978-953-307-037-7
Hard cover, 504 pages
Publisher InTech
Published online 01, March, 2010
Published in print edition March, 2010

In this book, a number of innovative fault diagnosis algorithms in recently years are introduced. These
methods can detect failures of various types of system effectively, and with a relatively high significance.

How to reference
In order to correctly reference this scholarly work, feel free to copy and paste the following:

Hongyu Wang, Zuohua Tian, Songjiao Shi and Zhenxin Weng (2010). Fault Detection and Isolation Scheme
Based on Parity Space Method for Discrete Time-Delay System, Fault Detection, Wei Zhang (Ed.), ISBN: 978-
953-307-037-7, InTech, Available from: [Link]
isolation-scheme-based-on-parity-space-method-for-discrete-time-delay-system

InTech Europe InTech China

University Campus STeP Ri Unit 405, Office Block, Hotel Equatorial Shanghai
Slavka Krautzeka 83/A No.65, Yan An Road (West), Shanghai, 200040, China
51000 Rijeka, Croatia
Phone: +385 (51) 770 447 Phone: +86-21-62489820
Fax: +385 (51) 686 166 Fax: +86-21-62489821
[Link]