Model Based Fault Detection and Diagnosis Using Structured Residual Approach in A Multi-Input Multi-Output System

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

SERBIAN JOURNAL OF ELECTRICAL ENGINEERING

Vol. 4, No. 2, November 2007, 133-145

Model Based Fault Detection and Diagnosis


Using Structured Residual Approach in a
Multi-Input Multi-Output System
A. Asokan1, D. Sivakumar2
Abstract: Fault detection and isolation (FDI) is a task to deduce from observed
variable of the system if any component is faulty, to locate the faulty components
and also to estimate the fault magnitude present in the system. This paper
provides a systematic method of fault diagnosis to detect leak in the three-tank
process. The proposed scheme makes use of structured residual approach for
detection, isolation and estimation of faults acting on the process [1]. This
technique includes residual generation and residual evaluation. A literature re-
view showed that the conventional fault diagnosis methods like the ordinary Chi-
square (Ψ2) test method, generalized likelihood ratio test have limitations such as
the “false alarm” problem. From the results it is inferred that the proposed FDI
scheme diagnoses better when compared to other conventional methods.

Keywords: Fault detection and diagnosis, Residual generation, Structured resi-


dual approach

1 Introduction
Faults can occur either in the processing equipment (leak in a tank) or in the
auxiliary equipment like sensors and actuators. These can result in degradation
of closed loop performance and also have an impact on safety, productivity and
plant economy. Therefore, it is necessary to assess the process performance and
diagnose the cause of performance degradation using model based fault
detection and identification. Since the early 1970s fault detection and isolation
have attracted increasing research attention. This lead to the development of
various approaches through use of redundant hardware, Kalman filter and
observer [2], parity equations and directional and structured residual [3, 8, 12].
Among the various FDI schemes, the structured residual approach (SRA)
proposed by J. Gertler, M. Staroswiecki and M. Shen [4] is powerful in isolating
faults. SRA proposed by Gertler [10, 11] is further simplified for a multi input

1
Department of EIE, Annamalai University, Tamilnadu, India; E-mail: asokan_me@yahoo.co.in
2
Department of EIE, Annamalai University, Tamilnadu, India

133
A. Asokan, D. Sivakumar

multi output system and it is considered for this paper for fault detection and
isolation for a three tank system. The SRA involves two steps i) generation of
Primary Residual Vector (PRV) for fault detection and ii) transformation of PRV
into structured residual vector (SRV) for fault isolation.
The implementation procedure of the proposed FDI scheme is illustrated in
Fig. 1. The controller in the system is used to maintain the process variable at its
set point. When there is a fault in the process, its output differs with model
output. This difference is termed as residual. By simply monitoring the residuals
one can say that something is going wrong. But it is not possible to identify the
location of the fault. So the residual has to be processed to enhance isolation. In
this paper the structured residual approach is applied to a MIMO system to
enhance fault isolation.

FAULTS

OUTPUT
CONTROLLER PROCESS

RESIDUAL
GENERATOR

STRUCTURAL
RESIDUAL APPROACH

Fig. 1 – Block diagram representation of proposed FDI scheme.


This paper is organized as follows. In section 2 the system under study
which is the three-tank system is described. In section 3, the identification of un-
measured disturbance variables (faults) using residual approach as reported in
literature is explained. The proposed scheme to identify and estimate the un-
measured disturbance acting on the process is presented in section 4. In section 5
the simulation results are discussed. Finally the conclusions are drawn and scope
of further work is provided in section 6.

2 System Descriptions
The three-tank system considered for study [6] is shown in Fig. 2. The con-
trolled variables are the level of the tank1 ( h1 ) and level of the tank3 ( h3 ). In
flow of tank1 (fin1) and in flow of tank3 (fin3) are chosen as manipulated vari-
ables to control the level of the tank1 and tank3. The unmeasured outflow of that
is leak of tank1, tank2 and tank3 have been considered as fault variables.
134
Model Based Fault Detection and Diagnosis Using Structured Residual Approach…

fin1 fin3

S1 S2 S3

h1 h2 h3

L1 L2 L3

Fig. 2 – Three Tank System.

The material balance equation for the above three-tank system is given by
d h1 fin1 Az1 L
= − 2 g ( h1 − h2 ) − 1 (1)
dt S1 S1 S1
d h2 Az1 Az L
= 2 g ( h1 − h2 ) − 3 2 g ( h1 − h3 ) − 2 (2)
dt S2 S2 S2
d h3 fin3 Az 3 Az L
= + 2 g ( h2 − h3 ) − 2 2 gh3 − 3 (3)
dt S3 S3 S3 S3
The steady state operating data of the Three-tank system is given in
Table 1.
Table 1
Steady state operating data.

h1 , h2 , h3 [m] 0.7, 0.5, 0.3


fin1, fin3 [ml/s] 100
Product of outflow coefficient and cross section of 2.25*10-5,
the connection pipe (Az1, Az2, Az3) [m2] 3.0571*10-5, 2.307*10-5
Area of tank ( S1 - S3 ) [m2] 0.0154
L1, L2, L3 [ml/s] 0
Acceleration due to gravity, g [m/s ]2
9.81

3 Fault [leak] Detection Using Residual Generator


Residuals are generated from the observable variable of the monitored plant,
that is, from the command values of the controlled inputs and the outputs [5].
Ideally, the residuals should only be affected by the faults. However, the

135
A. Asokan, D. Sivakumar

presence of disturbances, noise and modeling errors also causes the residuals to
become nonzero and thus interferes with the detection of faults. Therefore the
residual generator needs to be designed so that it is maximally unaffected by
these nuisance inputs, which means that it is robust in the face of disturbance,
noise and model errors. Structured residual are so designed that each residual
responds to a different subset of faults and insensitive to the others. When a
particular fault occurs, some of the residuals do respond and others do not. Then
the pattern of the response set, the fault signature or fault code, is characteristic
of the particular fault.
Example for fault code:
R1 R2 R3
L1 ⎡1 0 0⎤
L2 ⎢0 1 0 ⎥⎥

L3 ⎢⎣ 0 0 1 ⎥⎦

The above fault code implies that fault L1 affects only residual R1 like L2
affects R2 and L3 affects R3. In order to perform detection and isolation of set of
faults, structured residuals can be used. The so called signature code describes
the subset of residuals which react to each fault. Since the levels of all three-tank
are assumed to be measurable, there will be three residuals corresponding to
each of the three tanks When there is a fault, all the three residuals get affected.
By simply monitoring the residuals it is possible to predict the change in beha-
vior of the system from normal. But it is not possible to identify the location of
the fault. So the residual has to be transformed to enhance isolation.
Let the plant output is given by
YP ( s ) = G ( s )U ( s ) + GF ( s ) L( s ) , (4)
where: G(s) – transfer function under normal conditions and
GF(s) – fault transfer function.
Let the output of the model be given by
YM ( s ) = G ( s ) U ( s ) , (5)
where
⎡ fin1 ( s ) ⎤ ⎡ L1 ( s) ⎤
⎢ ⎥ and ⎢ ⎥
U (s) = ⎢ 0 ⎥ L( s ) = ⎢ L2 ( s) ⎥ .
⎢⎣ fin3 ( s) ⎥⎦ ⎢⎣ L3 ( s ) ⎥⎦
Residual R(s) is defined as difference between process output and model
output.

136
Model Based Fault Detection and Diagnosis Using Structured Residual Approach…

R ( s ) = YP ( s ) − YM ( s ) . (6)
Substituting the expressions for YP ( s ) and YM ( s ) in equation (6)
R ( s ) = GF ( s ) L ( s ),

⎡ R1 ( s ) ⎤ ⎡ GF 11 GF 12 GF 13 ⎤ ⎡ L1 ( s) ⎤
⎢ ⎥ ⎢ ⎢ ⎥
⎢ R2 ( s ) ⎥ = ⎢GF 21 GF 22 GF 23 ⎥⎥ ⎢ L2 ( s) ⎥ , (7)
⎢⎣ R3 ( s ) ⎥⎦ ⎢⎣GF 31 GF 32 GF 33 ⎥⎦ ⎢⎣ L3 ( s ) ⎥⎦

R1 ( s ) = GF 11 ( s ) L1 ( s ) + GF 12 ( s ) L2 ( s ) + GF 13 ( s ) L3 ( s ) , (8)
R2 ( s ) = GF 21 ( s ) L1 ( s ) + GF 22 ( s ) L2 ( s ) + GF 23 ( s ) L3 ( s ) , (9)
R3 ( s ) = GF 31 ( s ) L1 ( s ) + GF 32 ( s) L2 ( s) + GF 33 ( s) L3 ( s ) . (10)
The above equation is valid only in the absence of plant model mismatch
and in the absence of state and measurement noise. From the above equations, it
is evident that the presence of fault will affect all the three residuals. It is
difficult to identify the location of fault by monitoring the residuals. Therefore in
order to enhance the fault isolation it is required to transform the residuals. The
design of transformation matrix is discussed in the subsequent section.

4 Design of Transformation Matrix


To transform raw residual R ( s ) into structured form Rt ( s ) , multiply R ( s )
with weighting matrix W ( s )
Rt ( s ) = W ( s ) R ( s ) . (11)
Weighting matrix is chosen as
W ( s) = Z ( s )GF −1 ( s) . (12)
Substituting W ( s ) in expression (11)
Rt ( s ) = [ Z ( s)GF−1 ( s)][ R( s)],
Rt ( s ) = [ Z ( s)GF−1 ( s)][GF ( s) L( s)], (13)
Rt ( s ) = Z ( s ) I L( s ),
where I is the identity matrix,
Rt ( s ) = Z ( s ) L( s ) . (14)
If Z ( s ) is the diagonal matrix then
R1t ( s ) = Z11 ( s ) L1 ( s), (15)

137
A. Asokan, D. Sivakumar

R2t ( s ) = Z 22 ( s ) L2 ( s ), (16)
R3t ( s ) = Z 33 ( s ) L3 ( s ). (17)
It is inferred that the first element of Rt ( s ) that is R1t ( s ) affected only if
there is a leak in the first tank. The second element of Rt ( s ) that is R2t ( s )
affected only if there is a leak in the second tank. Like that the third element of
Rt ( s ) that is R3 ( s ) affected only if there is a leak in the third tank.
The transformation matrix W ( s ) is
⎡W11 W12 W13 ⎤
W ( s ) = ⎢⎢W21 W22 W23 ⎥⎥ , (18)
⎢⎣W31 W32 W33 ⎥⎦
where:
−0.0011
W11 = 1, W12 = , W13 = 0,
0.0154s + 0.00011
−0.0011 −0.0015
W21 = , W22 = 1, W23 = ,
0.0154 s + 0.00011 0.0154 s + 0.00011
−0.0015
W31 = 0, W32 = , W33 = 1.
0.0154s + 0.00011

The user specified residual specification matrix Z ( s ) is given by

⎡ Z11 0 0 ⎤
Z ( s ) = ⎢⎢ 0 Z 22 0 ⎥⎥ , (19)
⎢⎣ 0 0 Z 33 ⎥⎦
where:
1 1 1
Z11 = , Z 22 = , Z 33 = .
0.0154 s + 0.000112 0.0154 s + 0.00026 0.0154 s + 0.000193

5 Simulation Results
The proposed FDI scheme has been implemented on a three-tank system
and its performance is observed. The controlled variables are the level of tank1
( h1 ) and tank3 ( h3 ). Inflow of tank1 (fin1) and tank3 (fin3) are chosen as manipu-
lated inputs. Outflow of tank1 ( L1 ), tank2 ( L2 ) and tank3 ( L3 ) are considered as
leak variables.

138
Model Based Fault Detection and Diagnosis Using Structured Residual Approach…

The synthesis method [8] is used for the design of PI controller. The PI
controllers are designed so that the closed loop process behaves like a first order
system with unity gain and time constant same as the open loop time constant.
The resulting parameter for controlling the height of tank1 using the inflow of
tank1 is given by K c =2.54 ⋅ 10-4 [ml/ s/ m] and Ti = 222[s] and that of tank3 using
the inflow of tank3 is given by K c =7.69 ⋅10-4 [ml/ s/ m] and Ti = 200[s] .
The process is simulated using the non-linear first principles model,
whereas the FDI is based on the time invariant linearized model (Transfer
function model). The closed loop behavior of the process when a leak of
magnitude 50[ml/s] introduced at time t = 3000[s] in tank1 is shown in Fig. 3.
The behavior of the residuals is shown in Fig. 4. From the Fig. 4 one can infer
the presence of leak in tank1 affect all the three residuals. By simply monitoring
either the Process output or the residual it is not possible to identify the location
of the fault. The Structured residual output is shown in Figs. 5, 6 and 7. From
these figures one can conclude that there is leak only in the first tank.
Closed loop response of the System when leak occurs in all the three-tanks
is shown in the Fig. 8 that is, a leak of magnitude 50[ml/s] given in tank1 at
t = 2750[s] , leak of magnitude 50[ml/s] given in tank2 at t = 4000[s] and leak
of magnitude 100[ml/s] given in tank3 at t = 6000[s] . It is observed that the le-
vels of the tank are maintained even though the fault occurs in the process. So
simply monitoring the process output it is not possible to detect the fault. The
behavior of the residuals is shown in Fig. 9. With the residual one cannot find
the location of fault.
The Structured residual outputs are shown in Figs. 10, 11 and 12. From
these figures one can conclude that there is leak in the all the three tanks. The
Structured residual approach is tested for modeling errors that is 10% deviation
in time constant is considered. The closed loop behavior of the process when a
leak of magnitude 50[ml/s] introduced at time t = 6000[s] in tank3 is shown in
Fig. 13. The behavior of the residuals is shown in Fig. 14. The Structured residu-
al outputs are shown in Figs. 15, 16 and 17. From these figures one can conclude
that there is leak in the third tank only.
Structured residual approach is also tested for another set of controller
parameters. The PI controller settings are obtained so that the closed loop
process behaves like a first order system with unity gain and time constant 10%
less than the open loop time constant. The closed loop behavior of the process
when a leak of magnitude 100[ml/s] introduced at time t = 6000[s] in tank3
under the new settings is shown in Fig. 18. The behavior of the residuals is
shown in Fig. 19. The Structured residual outputs are shown in Figs. 20, 21 and
22. From these figures one can conclude that there is leak in the third tank only.
139
A. Asokan, D. Sivakumar

Fig. 3 – Closed loop response of the system Fig. 4 – Evolution of residuals when
when a leak of magnitude 50 [ml/s] step change in leak of tank1 of magnitude
occurs in tank1 at t = 3000[s]. 50 [ml/s] introduced at t = 3000[s].

Fig. 5 – Structured residual1 of the system Fig. 6 – Structured residual2 of the system

Leak of magnitude 50 [ml/s]


is given in tank1 at t = 3000[s].

Fig. 7 – Structured residual3 of the system.

140
Model Based Fault Detection and Diagnosis Using Structured Residual Approach…

Fig. 8 – Closed loop response of the system. Fig. 9 – Evolution of residuals.

Fig. 10 – Structured residual1 of the system. Fig. 11 – Structured residual2 of the system.

Leak of magnitude 50 [ml/s] is given in


tank1 at t = 2750[s].
Leak of magnitude 50 [ml/s] is given in
tank2 at t = 4000[s].
Leak of magnitude 100 [ml/s] is given in
tank3 at t = 6000[s].

Fig. 12 – Structured residual3 of the system.

141
A. Asokan, D. Sivakumar

Fig. 13 – Closed loop response of the system Fig. 14 – Evolution of residuals when
when a leak of magnitude 50 [ml/s] step change in leak of tank3 of magnitude
occurs in tank3 at t = 6000[s]. 50 [ml/s] introduced at t = 6000[s].

Fig. 15 – Structured residual1 of the system Fig. 16 – Structured residual2 of the system

Leak of magnitude 50 [ml/s]


is given in tank3 at t = 6000[s].

Fig.17 – Structured residual3 of the system.


142
Model Based Fault Detection and Diagnosis Using Structured Residual Approach…

Fig. 18 – Closed loop response of the system Fig. 19 – Evolution of residuals when
when a leak of magnitude 100 [ml/s] step change in leak of tank3 of magnitude
occurs in tank3 at t = 6000[s]. 100 [ml/s] introduced at t = 6000[s].

Fig. 20 – Structured residual1 of the system. Fig. 21 – Structured residual2 of the system.

Leak of magnitude 100 [ml/s]


is given in tank3 at t = 6000[s].

Fig.22 – Structured residual3 of the system.

143
A. Asokan, D. Sivakumar

6 Conclusions
The performance of the proposed scheme has been evaluated on a three-
tank process for leak in the tanks. The proposed FDI scheme can provide fault
information even when there is simultaneous change in more than one leak. It
should be noted that the proposed method is independent of the controller
design. From the structured residuals, the magnitude of leak and time of
occurrence of leak are also found. And one can conclude that the estimated
magnitude and time of occurrence of the leak variable (fault) are close to the true
value. So from the proposed method one can identify the fault as soon as it
occurs in the process. The proposed method is found to be robust to plant model
mismatch.

7 References
[1] J. Gertler: Fault Detection of Dynamical Systems, Marcel Dekker, Inc. USA, 1998.
[2] J. Gertler: A Survey of Model Based Failure Detection and Isolation in Complex Plants,
IEEE Control Systems Magazine, Vol. 8, No. 6, pp. 3-11, 1988.
[3] J. Gertler and M. Staroswiecki: Structured Fault Diagnosis in Mildly Nonlinear Systems:
Parity Space and Input-Output Formulation, FAC 15th World Congress, Barcelona, Spain,
July 2002.
[4] J. Gertler, M. Staroswiecki, M. Shen: Direct Design of Structured Residuals for Fault
Diagnosis in Linear Systems, American Control Conference, Anchorage, Alaska, May 2002.
[5] J. Gertler: Residual Generation in Model Based Fault Diagnosis, Control-Theory and
Advanced Technology, Vol. 9, pp. 259-285, March 1993.
[6] J. Wu, G. Biswas, S. Abdelwahed, E. Manders: A Hybrid Control System Design and
Implementation for a Three Tank Testbed, Proceedings of the 2005 IEEE Conference on
Control Applications, pp. 645–650, August 2005.
[7] E. Marlin: Process Control: Designing Processes and Control Systems for Dynamic
Performance, McGraw-Hill International Editions 1995.
[8] G. Stephanopoulos: Chemical Process Control, An Introduction to Theory and Practice,
Prentice-Hall India Publications 2000.
[9] J. Gertler: A Survey of Model Based Failure Detection and Isolation in Complex Plants,
IEEE Control Systems Magazine, Vol. 8, No. 6, pp. 3-11, 1988.
[10] J. Gertler: Residual Generation in Model Based Fault Diagnosis, Control-Theory and
Advanced Technology, Vol. 9, pp. 259-285, March 1993.
[11] J. Gertler: A Numerical-Structural Approach to Failure Detection and Isolation in Complex
Plants, Proceedings of the IEEE Conference on Decision and Control, Athens, Greece,
pp. 1576-1580, 1986.
[12] R. Isermann: Process Fault Detection Based on Modeling and Estimation Methods – a
Survey, Automatica 20, pp. 387-404, 1984.
[13] B. Käoppen-Seliger, E. Alcorta-Garca, P.M. Frank: Fault Detection – Different Strategies for
Modeling Applied to the Three Tank Benchmark – A Case Study, European Control
Conference, Karlsruhe, Germany, 1999.

144
Model Based Fault Detection and Diagnosis Using Structured Residual Approach…

[14] E. Alcorta-Garca, P.M. Frank: Observer-Based Fault Detection via Control System
Decomposition, SAMS, 24(51), pp. 177-184, 1996.
[15] P.M. Frank, X. Ding: Survey of Robust Residual Generation and Evaluation Methods in
Observer-Based Fault Detection Systems, Journal of Process Control, 7(6), pp. 403-424,
1997.
[16] Q. Zhang, M. Basseville, A. Benveniste: Fault Detection and Isolation in Nonlinear Dynamic
Systems: A Combined Input-Output and Local Approach, Automatica 34, pp. 1359-1373,
1988.
[17] D. Theilliol, H. Noura, J.C. Ponsart: Fault Diagnosis and Accommodation of a Three-Tank
System Based on Analytical Redundancy, ISA Transactions 41, 2002.
[18] A. Akhenak, M. Chadli, D. Maquin, J. Ragot: State Estimation via Multiple Observer the
Three Tank System, Proc. 5th IFAC SAFEPROCESS, Washington, 2003.
[19] J. Chen, R.J. Patton: Robust Model-Based Fault Diagnosis for Dynamic Systems, Kluwer
Academic Publishers, ISBN 0-7923-8411-3, 1999.

145

You might also like