Engineering Applications of Articial Intelligence 28 (2014) 3651

Contents lists available at ScienceDirect

Engineering Applications of Articial Intelligence

journal homepage: www.elsevier.com/locate/engappai

A variant of the particle swarm optimization for the improvement

of fault diagnosis in industrial systems via faults estimation
Ldice Camps Echevarra a, Orestes Llanes Santiago a,n, Juan Alberto Hernndez Fajardo a,
Antnio J. Silva Neto a,b, Doniel Jimnez Snchez a
a
b

Instituto Superior Politcnico Jos Antonio Echeverra, Cujae, La Habana, Cuba

Instituto Politcnico, IPRJ/UERJ, RJ, Brazil

art ic l e i nf o

a b s t r a c t

Article history:
Received 13 January 2012
Received in revised form
6 November 2013
Accepted 6 November 2013
Available online 9 December 2013

This paper proposes an approach for Fault Diagnosis and Isolation (FDI) on industrial systems via faults
estimation. FDI is presented as an optimization problem and it is solved with Particle Swarm
Optimization (PSO) and Ant Colony Optimization (ACO) algorithms. Also, is presented a study of the
inuence of some parameters from PSO and ACO in the desirable characteristics of FDI, i.e. robustness
and sensitivity. As a consequence, the Particle Swarm Optimization with Memory (PSO-M) algorithm, a
new variant of PSO was developed. PSO-M has the objective of reducing the number of iterations/
generations that PSO needs to execute in order to provide a reasonable quality diagnosis. The proposed
approach is tested using simulated data from a DC Motor benchmark. The results and analysis indicate
the suitability of the approach as well as the PSO-M algorithm.
& 2013 Elsevier Ltd. All rights reserved.

Keywords:
Ant colony optimization
Fault diagnosis
Industrial systems
Particle swarm optimization
Robust diagnosis
Sensitive diagnosis

1. Introduction
A fault is an unpermitted deviation of at least one characteristic
property or parameter of a system from the acceptable, usual or
standard operating condition (Simani et al., 2002).
Faults can cause economic losses as well as damage to human
capital and the environment. There is an increasing interest on the
development of new methods for fault detection and isolation, FDI,
also known as Fault Diagnosis, in relation to reliability, safety and
efciency (Isermann, 2005).
The FDI methods are responsible for detecting, isolating and
establishing the causes of the faults affecting the system. They
should also guarantee the fast detection of incipient faults (sensitivity to faults) and the rejection of false alarms that are attributable to disturbances or spurious signals (robustness).
The FDI methods are broken down into three general groups,
the process history based methods (Venkatasubramanian et al.,
2002c), those based on qualitative models (Venkatasubramanian
et al., 2002b), and the quantitative model based methods, also
known as analytical methods (Venkatasubramanian et al., 2002a).
The quantitative model based methods make use of an analytical or computational model of the system. The great variety of
the proposed model based methods is brought down to a few basic
concepts such as: the parity space; observer approach and the
n

Corresponding author. Tel.: 53 7 2663204.

E-mail address: orestes@electrica.cujae.edu.cu (O. Llanes Santiago).

0952-1976/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.engappai.2013.11.007

parameters identication or estimation approach (Isermann, 1984;

Frank, 1990, 1996; Isermann, 2005).
Many papers and books have been devoted to making descriptions
and establishing links among the different approaches for model
based diagnosis (Frank, 1990; Venkatasubramanian et al., 2002a;
Simani et al., 2002; Witczak, 2007; Metenidin et al., 2011). A clear
description of each approach and their limitations are presented in
Witczak (2007). In Witczak (2007) and Metenidin et al. (2011) it is
recognized that observers and parity space approaches do not always
allow the isolation of the actuators faults. For nonlinear models, the
complexity on the observer design increases, while an exact model of
the system is necessary for the parity space approach (Witczak, 2007;
Metenidin et al., 2011).
Parameter estimation approach requires the knowing of the
relationships between such parameters and the physical coefcients
of the system, as well as the inuence of the faults in these coefcients
(Frank, 1990, 1996; Isermann, 2005). This approach does not provide a
good diagnosis for the case of sensor faults. Furthermore, this usually
demands a high computing time, which makes it infeasible for most
situations (Isermann, 2005; Witczak, 2007).
The topics of robustness and sensitivity are of high interest
in FDI. Thus, many robust analytical methods have been developed
(Isermann, 1984, 2005; Frank, 1990; Chen and Patton, 1999; Patton
et al., 2000). However, the unavoidable process disturbances and
the modelling errors make that most FDI methods become
unfeasible in practical applications (Simani et al., 2002; Simani
and Patton, 2008). Therefore, further research on the topic of

L. Camps Echevarra et al. / Engineering Applications of Articial Intelligence 28 (2014) 3651

robust and sensitive FDI methods is indispensable (Isermann,

1984; Simani et al., 2002; Simani and Patton, 2008).
The FDI methods based in observers or parity space demand
large efforts to generate robust and sensitive residuals, related to
the fault detection. The generation of a residual is highly dependent on the model that describes the system.
This paper proposes an approach for FDI in industrial systems via
faults estimation. The proposal allows to diagnose the system based on
a direct fault estimation. Soft computing techniques are recognized as
attractive tools for solving various problems related to modern FDI
(Witczak, 2007; Metenidin et al., 2011). This approach considers the
use of meta heuristics, in order to obtain a robust and sensitive
diagnosis. Some FDI methods that use meta heuristics are reported in
Witczak (2007), Yang et al. (2007), Wang et al. (2008) and CampsEchevarra et al. (2010) and Metenidin et al. (2011).
The meta heuristics Particle Swarm Optimization (PSO) and Ant
Colony Optimization (ACO) have simple structures. Moreover, they
were recently applied to FDI (Liu et al., 2008; Liu, 2009; Samanta
and Nataraj, 2009; Duarte and Quiroga, 2010; Metenidin et al.,
2011). Therefore, they were selected for this approach.
This work is also aimed at studying the inuence of parameters
from PSO and ACO in robustness and sensitivity. This study is the basis
for the development of a new variant of PSO, named Particle Swarm
Optimization with Memory (PSO-M). The new algorithm has the
objective of reducing the number of iterations/generations that PSO
needs to execute, in order to discover reasonable quality solutions. This
means less computational time, which allows faster diagnosis. PSO-M
can be easily extended to other optimization problems.
The proposed approach also permits the direct estimation of
the faults, indistinctly if they take place in the actuator, process or
sensors. This estimation is based on the residual, which is directly
obtained between the measurements of the output of the system
and the output that estimates the model. This avoids restrictions
to the nature of the model.
The main contributions of this paper can be summarized as
follows: the study of a new approach for the development of
robust and sensitive FDI methods based on direct fault estimation
with the meta heuristics PSO and ACO; the study of the inuence
of their parameters in order to increase the robustness and
sensitivity; and the development of a new variant PSO-M, which
uses a pheromone matrix from ACO for storing the history of PSO,
useful for improving the computational cost that is required by
PSO. The viability of the proposal is demonstrated by diagnosing
the simulation data from a DC Motor.
This paper is organized as follows. The second section introduces
the modelling of faults and the model-based FDI methods via
parameters estimation. The proposed approach for FDI is also
described in this section. The third and fourth sections give a brief
description of the algorithm for PSO and the algorithm for ACO,
respectively. The fth section justies and describes the PSO-M
algorithm. Afterward, the next section details the DC motor case study
and its simulations. The other sections present the experimental
methodology, experiments and results, following the same order.
The tenth section presents a comparison between Parity Space,
Diagnostic Observers and our proposal for FDI. Finally, some concluding comments and remarks are presented.

2. Modelling faults and FDI based on direct fault estimation

37

The most used model is the linear time invariant (LTI), which has
two representations: the transfer function or transfer matrix, and
the state space representation. This last representation is also valid
for non-linear models.
Let us express the input/output behavior of SISO (single input
single output) processes by means of ordinary linear differential
equations
yt T tt

where

t a1 an b0 bm 

and

T t  y1 t  yn t  u1 t  um t

3
n

where y t and u t indicate derivatives (y t dy t=dt ).

The respective transfer function becomes, through Laplace
transformation:
n

Gp

ys Bs b0 b1 s bm sm

us As
1 a1 s an s n

The faults affecting the system may eventually change one or

several parameters in the vector t. The FDI based on model
parameters is divided into two steps. The rst is meant for the
estimation of the model parameters vector t. The second for
detecting and isolating the faults based on known relationships
between model parameters, physical coefcients of the system
and faults (Isermann, 1984, 2005).
The main drawback of this approach is that the model parameters should have physical meaning, i.e., they should correspond
with the parameters of the system. In such situations, the detection and isolation of faults are very straightforward. Otherwise, it
is usually difcult to distinguish a fault from a change in the
parameters vector t. Moreover, the process of fault isolation
may become extremely difcult because model parameters do not
uniquely correspond with those of the system. It should also be
pointed out that the detection of faults in sensors and actuators is
possible but rather complicated (Witczak, 2007; Metenidin et al.,
2011).
The two approaches that are commonly used for estimating the
model parameters t are classied with respect to the minimization function they use (Frank, 1990; Isermann, 2005):

 sum of least squares of the equation error;

 sum of least squares of output error.
The FDI based on parameters estimation considering the
minimization of the sum of least squares of the output error
requires numerical optimization methods. These methods give
more precise parameters estimations but the computational effort
is bigger, and on-line applications are, in general, not possible
(Isermann, 2005). Another typical limitation regarding parameters
estimation-based approaches is related to the fact that the input
signal should be persistently excited (Witczak, 2007; Metenidin
et al., 2011).
Instead of estimating the model parameters vector , let us
consider explicitly the faults in a SISO system in a closed loop
described by a LTI model as
ys Gyw sws Gyf u sf u s

FDI based on model parameters, which are partially unknown,

requires online parameters estimation methods. For that purpose
the input vector ut A Rm , the output vector yt A Rp and the basic
model structure must be known (Isermann, 2005).
The models for describing the systems depend on the dynamics
of the process and the objective to be reached with the simulation.

Gyf y sf y s Gyf p sf p s

where ws A R is the reference signal of the control system, fu, fp,

f y A R are faults in the actuator, process and output sensors,
respectively. The transfer function Gyw(s) represents the dynamics
of the system while Gyf u s, Gyf p s and Gyf y are the transfer

38

L. Camps Echevarra et al. / Engineering Applications of Articial Intelligence 28 (2014) 3651

functions that represent the faults fu, fp and fy, respectively (Ding,
2008).
This proposed approach considers the estimation of the faulty
parameters vector f f u f y f p  instead of t. Therefore, it
requires a model that directly represents the effect of the faults
in the actuator, process and sensors of the system. This kind of
model is widely used in other FDI model based methods such as
those based in observers or in parity spaces (Frank, 1990;
Isermann, 2005; Ding, 2008).
The estimation of f allows diagnosing the system in a direct
way: from the minimization of the sum of the square of the output
errors. The optimization is described as follows:
I

min

F^f yt f  y^ t ^f 2
t1

s:a:

f min r ^f r f max

where I is the number of sampling instants and y^t ^ f is the

estimated output at each time instant t. y^t ^ f is obtained from the
solution of the model given by Eq. (5), and yt f is the output
measured by the sensor at the same instant (Ding, 2008).
The proposed approach in this paper permits the direct
estimation of the faults taking place in the actuator, process or
sensors. This allows alleviating one of the limitations of the model
based methods for FDI (Witczak, 2007; Metenidin et al., 2011).
In order to obtain robustness to model uncertainties, disturbances or noise, the algorithms ACO and PSO are applied to the
faults estimation. These two algorithms showed robustness in
other applications. Their parameters can be manipulated in order
to increase this characteristic. Depending on the model used with
the proposed approach and the applications of PSO and ACO there
is no necessity of making additional efforts in the generation of a
robust residual. Thus, this proposal allows avoiding another
limitation of the model based FDI methods (Witczak, 2007;
Metenidin et al., 2011).
The proposed approach requires to meet the following
assumptions:

 There is a known model of the system that represents the

dynamics of the faults.

 The faults cannot be intermittent.

The steps of the proposed methodology are
1 Formulate the optimization problem, see Eq. (6).
!
2 Take the vector ^f 0 as a solution of the optimization
problem and compute the objective function F^f . If
F^f o 0:01 then the system is not under the inuence of
faults. Otherwise, go to step number 3.
3 Apply ACO, PSO or PSO-M to solve the optimization problem:
estimations of f (It is recommended PSO-M).
4 Diagnosis: If any component of ^f is different from zero, then
the fault that corresponds with that component is affecting the
system. The magnitude of the fault coincides with the value of
the estimation.
Due to model uncertainties, noise in the measurements and
other disturbances, the value of the objective function is not equal
to zero, even when the estimation of the fault vector ^ f coincides
with the real fault vector f. Therefore, the fault vector can be
different from zero even when the system is not affected by faults.
This causes uncertainties in the decision.
Step number 2 of the methodology allows avoiding this
disadvantage. A threshold for the objective function is determined.
It is considered that the system is not affected by faults and that
the measurements are affected by noise up to 8%. If this threshold

is exceeded, then it is decided that the system is under the effect of

faults.

3. Particle swarm optimization

Many strategies that mimic different natural behavior have
been proposed for handling difcult optimization problems. The
Swarm Intelligence brings together some optimization algorithms
that are based on the observation of simplied social models. This
is the case of Particle Swarm Optimization (PSO), which was
introduced by Kennedy and Eberhart in 1995 (Kennedy and
Eberhart, 1995; Kennedy, 1997). PSO is based on the social
behavior of ocking birds and schooling shes.
PSO has been applied to different elds requiring parameter
optimization in a high dimensional space. This is a result of its
simplicity, high efciency in searching, easy implementation and
its fast convergence to the global optimum (Kennedy and Eberhart,
1995; Kameyama, 2009). All these advantages, in addition to its
applications in automatic control, system identication, and some
recent results related with the FDI area (Poli, 2007; Samanta and
Nataraj, 2009; Liu, 2009; Duarte and Quiroga, 2010) motivated the
selection of PSO in this study.
3.1. Description of the algorithm PSO
PSO works with a group or population (swarm) of Z agents
(particles), which are interested on nding a good approximation
to the global minimum x0 of the objective function f : D  Rn -R.
Each agent moves throughout the search space D. The position
of the zth particle is identied with a solution for the optimization
problem. On each lth iteration, its value is updated and it is
represented by a vector X zl A Rn .
Each particle accumulates its historical best position X zpbest ,
which represents the achieved individual experience. The best
position that was achieved along the iterative procedure and
among all the Xgbest represents the collective experience.
The generation of the new position needs the current velocity
of the particle V zl A Rn and the previous position X zl 1
X lz X lz 1 V lz
The vector

7
V zl

is updated according to the following expression:

V zl V zl 1 c1 X zpbest  X zl 1 c2 X gbest  X zl 1

where V zl 1 is the previous velocity of the zth particle; denotes a

diagonal matrix with random numbers in the interval [0,1]; and
c1 , c2 are the parameters that characterize the trend during the
velocity updating (Kennedy, 1997; Kameyama, 2009). They are
called cognitive and social parameter, respectively. They represent
how the individual and social experience inuence in the next
agent decision. Some studies have been made in order to determine the best values for c1 and c2. The values c1 c2 2 ,
c1 c2 2:05 or c1 4 c2 with c1 c2 r4:10 are recommended
(Kennedy, 1998; Carlisle and Dozier, 2001; Beielstein et al., 2002).
Some variants of the algorithm have been developed with the
objective of improving some characteristics of PSO, e.g. velocity,
stability and convergence.
Eqs. (7) and (8) represent the canonical implementation of PSO.
Another well known variant is the one with inertial weight, which
considers either constant inertial weight or inertial weight reduction. The idea behind this variant is to add an inertial factor for
balancing the importance of the local and global search (Beielstein
et al., 2002; Kameyama, 2009). This parameter affects the
updating of each particle velocity by the expression
V zl V zl 1 c1 X zpbest  X zl 1

L. Camps Echevarra et al. / Engineering Applications of Articial Intelligence 28 (2014) 3651

39

4.1. Description of the algorithm

Fig. 1. Pseudo-code for PSO algorithm.

c2 X gbest  X zl 1

Nowadays, the most accepted strategy for is to establish

A min ; max  and to reduce its value according the number of
the current iteration l by means of

max 

max  min
Itr max

10

where Itrmax is the maximum number of iterations to be reached. It is

recommended to take max 0:9 and min 0:4 (Liang et al., 2006).
The basic PSO is recognized as a particular case for the
alternative that considers inertial weight if assigning 1 during
all the performance of the algorithm (Beielstein et al., 2002;
Kameyama, 2009).
The parameter called constriction factor (Clerc and Kennedy,
2002; Kameyama, 2009) is introduced in order to facilitate the
control of the particles velocity
V zl V zl 1 c1 X zpbest  X zl 1 c2 X gbest X zl 1 

11

where

j2  

2
p
2  4 j

12

and c1 c2 44.
The literature recommends to set 0:729 with c1 c2 2:05.
This is equivalent to use the inertia weight variant with 0:729
through the entire procedure and establishing c1 c2 1:49
(Eberhart and Shi, 2001).
There are different topologies for PSO. In this work, is used the
Gbest topology. It determines that all the particles are connected to
each other and they are part of a unique neighborhood
(Kameyama, 2009).
A pseudo-code for PSO is given in Fig. 1.

For the continuous case the idea of ACO is to mimic the

behavior of ants (Dorigo and Blum, 2005; Silva-Neto and Becceneri, 2009; Socha and Dorigo, 2008). In this case, the adaptation to
the continuous case that was reported in Silva-Neto and Becceneri
(2009) was applied. This variant was applied to other problems
(Becceneri and Zinober, 2001; Souto et al., 2005). In this variant,
the rst step is to divide the feasible interval of each variable of the
problem into k possible kn values. At each iteration the algorithm
generates a family of Z new ants. This generation uses the
information that was obtained from the previous ants, which is
saved in the pheromone accumulative probability matrix PC (the
matrix has dimensions n  k where n is the number of variables in
the problem). This matrix is updated at each iteration l as
pcij l

jg 1 f ig l

13

kg 1 f ig l

where fij are the elements of the pheromone matrix F A M nk R

and they express the pheromone level of the discrete value jth of
the variable ith. This matrix is updated in each iteration based on
the evaporation factor Cevap and the incremental factor Cinc as
f ij l 1 C evap f ij l  1 ij;best C inc f ij l 1
where
(

ij;best

if ji xbest
i

otherwise

15

the component ith of the best ant X gbest .

being xbest
i
The scheme for generating a new ant X zl at the iteration l needs
n random numbers qrand
, qrand
, , qrand
. For each component xz
n
n
1
2
z
from the ant X l to be generated, the following generation
mechanism is set as
( m
n if qrand
oq0
n
z
xn
16
^
rand
m
if
q
Zq0
n
n
where
m : f nm Z f nm

8 m 1; 2; ; k

17

and
^ : pcnm^ 4 qrand
4 pcnm^ r pcnm
m
n

^
8mZm

18

The control parameter q0 allows controlling the level of

randomness during the ants generation. The pseudo-code for
ACO is given in Fig. 2.

4. Ant colony optimization

Ant Colony Optimization (ACO) was initially proposed for
integer programming problems (Dorigo and Caro, 1992). ACO is
inspired on the behavior of ants seeking a path between their
colony and a source of food. This behavior is due to the deposit and
evaporation of pheromone.
ACO was successfully extended to continuous optimization
problems (Dorigo and Blum, 2005; Silva-Neto and Becceneri,
2009; Socha and Dorigo, 2008). An advantage of this algorithm
is that its parameters can be manipulated in order to achieve more
diversication or intensication during the search. This allows an
efcient hybridization with other algorithms.

14

Fig. 2. Pseudo-code for ACO algorithm.

40

L. Camps Echevarra et al. / Engineering Applications of Articial Intelligence 28 (2014) 3651

5. New variant: Particle swarm optimization with memory

(PSO-M)
PSO has been hybridized successfully with other optimization
methods. The interest of combining PSO with other methods is
mostly based on its recognized characteristic of not improving the
quality of the solutions when the number of iterations is increased
(Angeline, 1998). Instead, PSO can nd good enough solutions
faster than other evolutionary algorithms (Angeline, 1998).
In the particular case of PSO and ACO, some works have focused
on the hybridization between them. For example, in Shelokar et al.
(2007) are proposed an algorithm for which ACO is used to
perform a local search around each particle of PSO, at each
iteration.
Our proposal, called Particle Swarm Optimization with Memory
(PSO-M), has the objective of reducing the number of iterations/
generations that PSO needs to execute, in order to discover reasonable
quality solutions. This means a less computational time, which allows
a faster diagnosis. This is also a desirable characteristic for online
diagnosing systems in industrial processes.
The implementation of the algorithm consists of two stages:

 First stage: A swarm explores the search space, i.e. PSO is



Fig. 3. Pseudo-code for PSO-M algorithm.

applied. A pheromone matrix, as described in Section 4, stores

the information of the search in each iteration.
Second stage: Another swarm makes an intensication of the
promising regions of the search space. For that purpose, its initial
position is generated using the generation scheme of ACO, which is
described in Section 4. For this generation scheme, the algorithm
uses the pheromone matrix achieved in the rst stage.

It can be concluded that PSO-M uses the memory of ACO for

storing the historical behavior of the agents from PSO during the
rst stage. In the second stage, the algorithms use this memory for
addressing the search of another swarm.

 With the new vector GX gbest X gbestd the matrix F is updated

5.1. Description of the algorithm

This part describes the implementation of PSO-M.
In the rst stage, it applies PSO, i.e. a swarm of Z1 agents
explores the search space D following the structure of PSO. The
new position X lz A Rn and the new velocity V lz of each agent z, with
z 1; 2; Z 1 , are updated at the iteration l following Eqs. (7) and
(9), respectively. The values of the PSO parameters are based on
the presented study of their inuence in the robust and sensitive
diagnosis. Following the idea of ACO for the continuous case, see
Section 4, we divide the permissible interval of each variable into k
possible values xnk, and a pheromone matrix F, with dimensions
n  k, is generated and updated at each iteration.
In order to update the pheromone matrix F, we propose the
following strategy:

 On each iteration l, each component of the vector Xgbest is

identied with only one of the k discrete values that are
assigned to the variable that corresponds with that component
nth. This connection generates the vector X gbestd . We dene a
vectorial function G : Rn Rn .
Let be xgbest
and xgbestd
the nth components of the vector Xgbest
n
n
and X gbestd respectively, then it is established
xm
Gn X gbest xgbestd
n
n

19

where

gbest
m : jxgbest
n  xm
n  xm
n
n j minjxn
nj
m

Fig. 4. Block diagram of the DC Motor control system AMIRA DR300.

with m 1; 2k

20

as in ACO, see Eq. (14).

The second stage considers a swarm of Z2 (Z 2 o Z 1 ) agents that
will perform an intensication of the promising regions of the
search space D. For that purpose, the algorithm generates an
initial swarm using the information stored in the pheromone
matrix F from the rst stage. The mechanism for generating the
initial swarm is the same as described in Eqs. (16)(18):
The pseudo-code for the PSO-M method is given in Fig. 3.

6. Benchmark DC motor
This section describes the main characteristics of the DC Motor
control system DR300 (Ding, 2008). This system has been widely
used for studying and testing new methods of FDI due to its
similitude with high speed industrial control systems (Ding, 2008).
The system is formed by a permanent magnet, which is coupled
to a DC generator. The main function of this generator is to
simulate the effect of a fault that results when a load torque is
applied to the axis of the motor. The speed is measured by a
tachometer that feeds the signal to a PI (proportional-integral
controller) speed controller. Fig. 4 shows the block diagram of the
DC Motor control system AMIRA DR300.
The voltage UT(Volts) is proportional to the rotational speed of
the motor's axis W(rad/s). UT is compared with Uref(Volts) in order
to use the error for computing the control signal UC(Volts) for the
PI speed controller. The AMIRA DR300 system also includes an

L. Camps Echevarra et al. / Engineering Applications of Articial Intelligence 28 (2014) 3651

internal control loop for the armature current IA. The controller
computes the motor armature voltage UA(Volts) as a function of
the reference that is obtained by means of the gain K1(Volts/Amp)
and the output IA(Amp).
6.1. Mathematical model
For this study, the internal current loop, which is the process to
be controlled, is a single block. The block diagram of the closed
loop is formed by the process and the PI controller. The parameters
of the laboratory DC motor DR300 are reported in Table 3.1 from
Ding (2008).
This analysis considers that the system can be affected by three
additive faults fu, fp and fy. fu represents a fault in the actuator and
it is modeled as a deviation of the control signal; fp represents a
fault in the process itself due to a load torque, which is applied to
the axis of the motor, and fy represents a fault in the measurement
of the motor speed.
The dynamics of the control system in the open loop is
described in the frequency domain by
U T s Gyu sU C s Gyf p sf p s
Gyu s

8:75
1 1:225s1 0:03s1 0:005s

31:07
Gyf p s 
s1 0:005s

21
22

23

Gc s

U C s
1:6
1:96
Es
s

Es U ref s  U T s

24
25

where E(s) denotes the error signal.

Considering the other faults that may affect the system, the
equation that describes it by means of the closed loop transfer
function is
U T s Gyw sU ref s Gyf u sf u s
Gyf p sf p s Gyf y sf y s

Table 1
Faulty situations for the rst and second parts of the numerical experiments.
Cases

fu

fy

fp

Case
Case
Case
Case

0.87
 0.27
0.63
0

 0.12
0.96
0
0.47

0.53
0
0.29
0.86

1
2
3
4

Table 2
Faulty situations for the third part of the numerical experiments.
Cases

fu

fy

fp

Case
Case
Case
Case

 0.08
0.15
0
0

0.09
0
 0.1
0

0.2
0
0
0.12

5
6
7
8

Table 3
Variants of PSO.
Alg

c1

c2

max

min

PSOB
PSOI1

2
2

2
2

30
30

0.4

0.9

30

0.4

0.9

3.5

0.5

30

0.4

0.9

PSOI2

where UT is the controlled variable, UC is the control signal, Gyu(s)

represents the dynamics of the process in the open loop and Gyf p s
is the transfer function of the fault fp (Ding, 2008).
The transfer function of the PI speed controller is

41

PSOI3

max  min
l
Itr max
max  min
max 
l
Itr max
max  min
max 
l
Itr max
max 

Table 4
Variants of ACO.
Alg

qo

ACO1:1
ACO1:2
ACO1:3
ACO2:1
ACO2:2
ACO2:3

63
63
63
127
127
127

0.15
0.55
0.85
0.15
0.55
0.85

30
30
30
30
30
30

26

For the present study, it was considered that the faults are time
invariant and under the following restrictions:

7. Experimental methodology

f u; f y A R :  1 V r f u; f y r 1 V

Three aspects were considered: robustness, sensitivity and

computational cost. This allows analyzing the merits of the
diagnosis based on faults estimation with PSO, ACO and PSO-M.
Also, it was analyzed the inuence of some parameters from PSO
and ACO in the characteristics of the diagnosis.
With this goal in mind, several faulty situations were considered. The cases can represent single fault (only one fault affecting
the system); multiple faults (more than one fault affecting the
system) or incipient fault conditions (faults of low magnitude). The
experiments were divided into three parts:

f p A R : 0 Nm r f p r1 Nm

27

6.2. Simulation of the benchmark DC Motor

It was made simulations of the speed control system for the
closed loop. In all test cases, the system was affected by a noise up
to 2% or 8%. The addition of noise is aimed to simulate more
realistic conditions. The noise affecting the systems is one of the
recognized causes of a wrong diagnosis and leads to the necessity
of robust FDI methods. All the implementations were made in
MATLAB R2008a. The reference speed was 3000 rpm, which
corresponds to 15 V.
The direct estimation of the faults allowed diagnosing the
system. These estimations can be obtained by the solution of the
minimization problem described by Eq. (6).
In the analyzed application, yt f is the measurement of the
speed in different time instants, t, and y^t ^f is the speed
computed from the model at the same time instants.

1 General performance: For the general analysis of the diagnosis

via PSO and ACO. This part includes multiple faults situations.
The output of the system is corrupted up to a 2% level noise.
The faulty situations are shown in Table 1.
2 Robust performance: For the analysis of robustness. The same
situations from the rst part applies but now with a noise level
up to 8%.
3 Sensitive Performance: for the sensitivity analysis. The cases
include simple and incipient faults, see Cases 68 from Table 2;

42

L. Camps Echevarra et al. / Engineering Applications of Articial Intelligence 28 (2014) 3651

Fig. 5. Comparison between the performance of PSOB and PSOI1 when diagnosing the faulty situations in Table 1, up to 2% level noise.

as well as multiple and incipient faults, see Case 5 from Table 2. All
the measurements are corrupted with a noise level up to 8%.
Different values for some parameters in PSO and ACO were
considered. The General, Robust and Sensitive performance of the
variants of PSO and ACO were analyzed. For each faulty situation
was made 30 runs of each algorithm. The abbreviations F ^ f for
the mean value of the objective function and Eval for the
arithmetic average of the minimum number of objective function
evaluations achieved until the nal value of the objective function
was reached were used. The analysis of the computational effort of
the algorithm was based on the number of evaluation of the
objective function.
Based on this study, the best set of parameters for ACO and PSO,
respectively, were selected. For that selection the Sign test, which
is an easy way to compare the overall performance of two
algorithms (Derrac et al., 2011) was used.
A comparison between the best variant of ACO and PSO using
the Wilcoxon signed ranks test was made. This is a simple and safe
nonparametric test for pairwise statistical comparisons (Derrac
et al., 2011). The statistical T was computed and it was compared
with the value of the Wilcoxons distribution for Num degrees of
freedom (critical value of W), where Num is the number of cases
for which the performance of the algorithms is compared (Derrac
et al., 2011). Also, the Wilcoxon signed ranks test for comparing
the best variant of PSO against PSO-M was used.

8. Experiments
8.1. Implementation of PSO
Two variants of PSO were considered: basic (canonical) PSO,
PSOB , and PSO with Inertial Weight, PSOI , see algorithm in Fig. 1.
Table 3 shows the values for the parameters of the algorithm in
each variant. These variants will permit to analyze the inuence of
the parameters , c1 and c2 in the diagnosis. These parameters
have a great inuence over the quality of the solution and the
convergence (Becceneri et al., 2006). The selected values for c1 and
c2 follow the recommendations from Kennedy (1998), Carlisle and
Dozier (2001) and Kameyama (2009).
In all the experiments were selected Z30, following the idea
of taking Z dimD. The values of the coefcients c1, c2 and
permit to establish the balance between the intensication and
diversication of the search. In Table 3 the notation l represents
the number of the current iteration, yielding a reduction on the
inertial weight along the iterative procedure.
8.2. Implementation of ACO
The variants of ACO were based on the different values for the
parameters q0 and k. The parameter q0 permits to establish the level
of randomness in the selection of the discrete value of the variable

(Silva-Neto and Becceneri, 2009), determining the trend of the search.

The values q0 0.15, q0 0.55 and q0 0.85 indicate a greater diversication, a balance between diversication and intensication, and a
greater intensication of the search, respectively. The parameter k
determines the size of the search space. With k 127 the size of the
search is duplicated comparing with k63. The six variants that we
considered in this paper are based on the algorithm in Fig. 2. Table 4
shows the values for the parameters of the algorithm in each variant.
The value of Z30 was used following the same criterion as in PSO.
8.3. Implementation of PSO-M
The variant PSO-M was designed in two stages, see algorithm
in Fig. 3.

 First stage: Variant PSOI1 (c1 2 and c2 2) with Z 1 45,

Itr max1 15. The pheromone matrix has dimensions 3  63.

 Second stage: Variant PSOI1 (c1 2 and c2 2) with Z 2 20.

8.4. Stopping criteria
The implemented criteria for all variants of PSO and ACO are
1. Criterion 1: Maximum number of iterations Itr max 100.
2. Criterion 2: Maximum number of iterations for which the best
value of the objective function remains constant Itr cte 10 and
minimum value for the objective function F^f o 0:01.
For the rst stage of PSO-M only the stopping criterion number
1 was considered. In the second stage of PSO-M all the stopping
criteria with values Itr max2 25, Itr cte 10 and F^f o 0:01 were
considered.

9. Results
9.1. Results of the diagnosis with PSO
9.1.1. General performance
In this part the variants PSOB and PSOI1 to solving the optimization problem given in Eq. (6) were applied.
In Fig. 5 it is shown that both variants PSOB and PSOI1 detect the
fault but PSOI1 is more precise while it presents a greater number
of objective function evaluations.
In order to determine the causes of this behavior, in Fig. 6 the best
value of the objective function versus the iterations for PSOB and PSOI1
is represented, respectively. The gures demonstrate that the greater
number of evaluations of the objective function in the variant PSOI1 is a
consequence of its capability for obtaining better estimations. This is
related with the fact that the algorithm decreases the parameter as
a function of the number of iterations, allowing a more intensication
around the better solutions.

L. Camps Echevarra et al. / Engineering Applications of Articial Intelligence 28 (2014) 3651

FObj vs Iterations

FObj vs Iterations

12

10
8

FObj

FObj

10

6
4

4
2

43

10

15

10

Iterations

20

30

40

Iterations

Fig. 6. Best value of F^ f obtained by PSOB and PSOI at each iteration for the Case 3 in Table 1, up to 2% level noise. (a) PSOB and (b) PSOI.

Fig. 7. Comparison between the performance of PSOI1, PSOI2 and PSOI3 when diagnosing the faulty situations in Table 1, up to 8% level noise.

Fig. 8. Comparison between the performance of PSOI1, PSOI2 and PSOI3 when diagnosing the faulty situations in Table 2, up to 8% level noise

1.5

1.5

fp

fp

0.5
0

0
0.5

1.5
1
1.5

0
fy 0.5

0
1 1

1.5
1

0.5
fu

fy

0.2

0.5
1

fu

Fig. 9. Comparison between the behavior of the search space by the variant PSOI1 and PSOI3. (a) PSOI1 and (b) PSOI3.

The results of this part point to that PSOI1 permit a better

diagnosis than PSOB. Thus, we only keep the variants of PSOI1 in
the subsequent studies.

9.1.2. Robust performance

In this part the variants PSOI1, PSOI2 and PSOI3 were applied.
In Fig. 7 it is shown that there are not evident differences
between the estimations of the faults that provide PSOI1, PSOI2 and
PSOI3. This indicates that the differences between the values of the

parameters c1 and c2 do not affect signicantly the robustness.

On the other hand, there are differences between the number of
objective function evaluations: PSOI3 shows better results in
this point.

9.1.3. Sensitive performance

In this part the objective is the sensitivity analysis of the
proposal for FDI, as well as the inuence of some parameters from

44

L. Camps Echevarra et al. / Engineering Applications of Articial Intelligence 28 (2014) 3651

PSO in the quality of the diagnosis, see description of the experiments in Section 7.
In Fig. 8 it is shown that the worse estimations were exhibited
by PSOI3. The performance of PSOI1 and PSOI2 was quite similar
respecting the quality of the estimations. This indicates that a
higher diversication of the search space is important for a
sensitive diagnosis.
In order to analyze the effect of diversication in sensitivity, the
search of PSOI1 and PSOI3 was compared, see Fig. 9.
In Fig. 9 it is shown that PSOI1 performs a greater diversication
than PSOI3. Based on the better estimations of PSOI1 over PSOI3, and
based on Fig. 9, it is possible to conclude that the sensitivity is
improved with greater diversication of the search space. On the
other hand, sensitivity does not necessarily imply a higher
computational cost.
Taking into account all these results also it was concluded that
the better variants for obtaining a sensitive diagnosis are PSOI1 and
PSOI2. Based on the Sign test, the variant PSOI1 was selected as the
best variant with a level of signicant 0:05, see the results of
the test in Table 5.

ACO1:1 shows the best performance for Case 1. For Case 2, ACO2:1 is
more efcient than ACO1:1 ; for Cases 3 and 4 both variants are
similar.
In Fig. 11 a comparison between ACO1:1 and ACO2:1 is shown. This
time the gures show the evolution of the best value of the objective
function obtained for each iteration. Taking into account these results,
the conclusion is that the greater search space of ACO2:1 with a low
value for the parameter q0 0.15 produces greater variations in the
value of the objective function than ACO1:1 .

9.2.2. Robust performance

Fig. 12 shows the performance of six variants of ACO from
Table 4. The results uctuated within a wide range for some of
these variants. Therefore, the variants ACO1:1 and ACO2:1 were
selected. The values of the parameters from ACO1:1 and ACO2:1 lead
to a robust variant for the case of study while the others do not.
The variants ACO1:1 and ACO2:1 give the most accurate estimation.
Thus, it was concluded that the higher diversication obtained
with the parameter q0 0.15 has a good inuence on the robustness of the diagnosis.

9.2. Results of the diagnosis with ACO

9.2.1. General performance
In this part the variants ACO1:1 and ACO2:1 were applied.
In Fig. 10 the performance of ACO1:1 and ACO2:1 is shown. The
results indicate that both variants detect the faults. The variant
Table 5
Sign test results: PSOI1 versus PSOI2.
Comparison

Criterion

PSOI1 wins

PSOI1 lost

Num

critical value

PSOI1vs PSOI2

F^ f

0.05

9.2.3. Sensitivity performance

In Fig. 13 the performance of the six variants of ACO from
Table 4 is shown. The variants ACO1:1 and ACO2:1 give the most
accurate estimations, which coincide with best performing algorithms in the robustness analysis. Thus, the conclusion is that the
higher diversication obtained with the parameter q0 0.15 also
has a good inuence on the sensitivity of the diagnosis.
Considering all the results, ACO1:1 and ACO2:1 were selected as
the better variants. Based on the Sign test, ACO1:1 was selected as
the best variant with a level of signicant 0:05, see Table 6 for
the results of the test.

Fig. 10. Comparison between the performance of ACO1:1 and ACO2:1 when diagnosing the faulty situations in Table 1, up to 2% level noise.

FObj vs Iterations

FObj vs Iterations

2
1.8

1.6
1.4
FObj

FObj

1.2
1
0.8

0.6
0

10
Iterations

15

20

0.4

10

15

20

25

Iterations

Fig. 11. Comparison between the best value of F^ f obtained by ACO1:1 and ACO2:1 , until each iteration, when diagnosing the Case 3 in Table 1, up to 2% level noise. (a) ACO1:1
and (b) ACO2:1 .

L. Camps Echevarra et al. / Engineering Applications of Articial Intelligence 28 (2014) 3651

45

Fig. 12. Comparison between the performance obtained by six variants of ACO from Table 4 for the faulty situations from Table 1, up to 8% level noise.

9.3. Comparison between PSO and ACO

After the previous study, PSOI1 and ACO1:1 were chosen as the
best variant of PSO and ACO, respectively. Table 8 shows the
comparison between them. The faulty situations are the same
from Robust Performance and Sensitive Performance analysis.
The results in Table 8 point out that the PSOI1 gave absolutely
better faults estimations than ACO1:1 in six of the eight faulty
situations. On the other two situations, PSOI1 was better than
ACO1:1 in the estimations of two of the three faults. In Fig. 14 it is
shown that the number of objective function evaluations needed
by ACO1:1 was less than those needed by PSOI1, while the value of
the objective function was always less in the diagnosis with PSOI1.
Table 7 shows the results of the application of the Wilcoxon's
test. For the rst line R represents the sum of the ranks for which
PSOI1 outperformed ACO1:1 , taking as criterion the value of the
objective function. For the second line R represents the sum of
the ranks for which ACO1:1 outperformed PSOI1, taking as criterion
the number of evaluations of the objective function.
From the results in Table 7, it was concluded that PSOI1 shows a
signicant improvement over the estimations obtained by ACO1:1 ,
with a level of signicant 0:01. On the other hand, ACO1:1
permits to obtain acceptable estimations with a computational
cost signicantly better than PSOI1, with 0:01. This issue is
important for online diagnosis.
Taking into account the above results the new algorithm PSO-M
is proposed. The objective with PSO-M is keeping the quality of the
estimations provided by PSOI1, which means robust and sensitive
diagnosis, while reducing its number of objective function evaluations for improving its performance in online diagnosis.

9.4. Validation of PSO-M

The comparison between the results obtained by PSOI1, ACO1:1
and PSO-M is shown in Table 8. This table also shows that PSO-M
obtained more accurate faults estimations in six of the eight cases.
The algorithm PSO-M also reduced the number of function
evaluations needed by PSOI1 in all cases.
In order to evaluate the performance of PSO-M, the Wilcoxon'
test for the comparison between PSOI1 and PSO-M was applied.
Table 9 shows the test results. R represents the sum of the ranks
for which PSO-M outperformed PSOI1 in the rst and second line.
The value of the objective function and the number of evaluations
of the objective function, respectively, were chosen as criteria for
the comparison.
The results on the rst line from Table 9 indicate that the null
hypothesis cannot be rejected, which means that in the estimations of faults with PSO-M and with PSOI1 there are no differences.
The results from the second line also indicate that there is an over
performance of PSO-M, with a level of signicant 0:01, taking
the number of evaluations of the objective function, which implies
less computational cost.

Table 6
Sign test results: ACO1:1 versus ACO2:1 .
Comparison

Criterion ACO1:1 wins ACO2:1 lost Num Critical

value

ACO1:1 vs ACO2:1 F^
f

10

12

0.05

The comparison between PSOI1, ACO1:1 and PSO-M is also

shown in Fig. 15. The algorithm PSO-M obtained similar estimations to PSOI1 as the Wilcoxon signed ranks test showed, and it
also permitted to reduce its computational time. The PSO-M
improves the disadvantages of the diagnosis via PSO and ACO.

10. Comparison between this proposal and other FDI methods

A comparison with other FDI methods is included in this
section. The main model based approaches for FDI are Parity
Space and Diagnostic Observers. Recent techniques for model
based FDI are mostly a variation of them and they maintain the
principal limitations of the original methods taken into account in
the comparison (Odgaard and Matajib, 2008; Li and Dahhou, 2008;
Narasimhana et al., 2008; Fliess et al., 2004). Therefore, Parity
Space and Diagnostic Observers will be used to compare with the
proposal of this paper.
Both approaches generate residuals, which are used to form
appropriate decision functions.
The idea is to generate a structured set of residual. Then, at
least one measured quantity has no impact on a specic residue. In
case of a faulty measurement, the decoupled residue remains
small, while the others increase their value. This feature helps to
locate the fault (Frank, 1990; Hoeing and Isermann, 1996).
Both approaches are formulated based on the state representation but, in the linear case it can be connected with the transfer
function models (Ding, 2008).

10.1. Parity space approach

The introduction of the Parity Space approach for FDI was made
in the early 8s. Different forms of Parity Space Approach have been
developed since then. In this paper is considered the original one,
which is based on the assumption of a state space model of a
linear discrete time system (Chow and Willsky, 1984) and have the
form:
xk 1 Axk Buk Ef f k
yk Cxk F f f k

28

where x A Rn is the vector of state variables; u A Rm and y A Rp the

measurable input and output signals, respectively. The matrices A,
B, C, Ef and Ff are known and with appropriate dimensions.

46

L. Camps Echevarra et al. / Engineering Applications of Articial Intelligence 28 (2014) 3651

Fig. 13. Comparison between the performance obtained by six variants of ACO from Table 4 when diagnosing the faulty situations in Table 2, up to 8% level noise.

Fig. 14. Comparison between the performance obtained by ACO1:1 and PSOI1 for the faulty situations from Tables 1 and 2, up to 8% level noise.

Table 7
Wilcoxon signed ranks test results: PSOI1 versus ACO1:1 .

Table 8
Results of the comparison between PSOI1, ACO1:1 and PSO-M, up to 8% level noise.

Comparison

Criterion

R

T minfR ; R  g

Critical value
of W

PSOI1 vs ACO1:1

F^ f

36

0.01

ACO1:1 vs PSOI1

Eval

36

0.01

Case

It is further assumed that (C,A) is observable and rankC p.

A continuous time parity space can be generated similar to the
design of the discrete one (Hoeing, 1993).
Using the notations:
ys k yk  s yk  s 1ykt ;

29

us k uk  s uk  s 1ukt ;

30

f ;s k f k  s f k  s 1f k ;
t

2
6
6
H u;s 6
4
and

6
6
H f ;s 6
6
4

CB

7
7
7
05

CAs  1 B

CB

Ff

CEf

Ff

CEf

33

The matrix H o;s is called the extended observability matrix of

the system. It is dened as (Ding, 2008)

CAs  1 Ef

s t

H o;s C CACA 

34

510
1491
1037

(0.96)
0.7556
0.9944
0.9587

(0)
 0.0857
0.01734
 0.0015

9.5828
9.1491
9.0020

591
2244
1083

ACO1:1
PSOI1
PSO  M

(0.63)
0.6159
0.6428
0.6412

(0)
0.0889
0.0275
0.0859

(0.29)
0.3270
0.3040
0.3293

2.4367
2.1905
2.1881

549
1734
991

ACO1:1
PSOI1
PSO  M

(0)
 0.0063
 0.0010
0.0087

(0.47)
0.4889
0.4576
0.4614

(0.86)
0.8667
0.8545
0.8570

1.0940
0.3261
0.3548

483
1761
947

ACO1:1
PSOI1
PSO  M

(  0.08)
 0.1397
 0.0775
 0.0720

(0.09)
0.0444
0.0951
0.1126

(0.2)
0.1746
0.2023
0.2113

3.3207
2.8132
2.8339

660
1716
1051

ACO1:1
PSOI1
PSO  M

(0.15)
0.1714
0.1633
0.1483

(0)
0.0349
0.0175
0.0247

(0)
0.0222
0.0088
0.0109

4.9194
4.3135
4.3073

576
1941
963

ACO1:1
PSOI1
PSO  M

(0)
0.0032
 0.0103
 0.0039

(  0.1)
-0.2127
-0.0884
-0.0547

(0)
 0.0508
0.0043
0.0194

4.3432
3.7786
3.6969

606
1473
1077

ACO1:1
PSOI1
PSO  M

(0)
 0.0190
 0.0028
0.0118

(0)
0.0063
0.0135
0.0315

(0.12)
0.1175
0.1261
0.1358

3.6654
3.1399
3.1353

633
1830
1109

7
7
7
07
5
Ff

1.4702
0.7018
0.5966

(  0.3)
 0.2349
 0.2496
 0.2995

(0.53)
0.5429
0.5402
0.5526

ACO1:1
PSOI1
PSO  M

Eval

(  0.2)
 0.0794
 0.0985
 0.1496

32

F^ f

fp

(0.87)
0.8508
0.8778
0.8635

31

fy

fu

ACO1:1
PSOI1
PSO  M

Variant

L. Camps Echevarra et al. / Engineering Applications of Articial Intelligence 28 (2014) 3651

the equations:

The parity relation is obtained by

ys k H o;s xs  k H u;s us k H f ;s f ;s k

35

If s Z n exists, it means that at least one vector vs A Rps 1

(parity vector) with vs a 0 such that vs H o;s 0. The parity relation
based residual generator, in the absence of faults, is constructed by
r s k vs H o;s xk  s vs ys k  H u;s us k 0

36

In the presence of faults the residual becomes

r s k vs H o;s xk  s vs H f ;s f ;s k;

vs : vs H o;s 0

37

r s k vs H f ;s f ;s k a 0

38

10.2. Diagnostic observers

x^_ A  HCx^ Bu Hy
^
y^ C x:

40

where H denotes the feedback gain matrix that has to be chosen

properly to achieve a desired performance of the observer.
^ and the
The relations for the state estimation error, x  x,
^ become
output estimation error, e y  y,

_ A HC Ef f HF f f
e C F f f

41

where e is used as the residual, r, for the purpose of detection and

isolation of faults.
10.3. Connections between parity space and diagnostic observers

The basic idea of the observers approach is to reconstruct the

outputs of the system from the measurements or subsets of the
measurements with the aid of observers, using the estimation
error as the residual for the detection and isolation of the faults
(Frank, 1990; Ding, 2008).
In the case of a linear process with the state equations
_ Axt But Ef f t
xt
yt Cxt F f f t

39

^ and output x^ of a full-order observer are governed by

the state x,
Table 9
Wilcoxon signed ranks test results: PSO-M versus PSOI1.
R

R

T minfR ; R  g

Critical
value of W

Comparison

Criterion

PSO-M vs PSOI1

F^ f

27

0.1

Eval

36

0.01

PSO-M vs PSOI1

47

It was demonstrated that exists a one-to-one mapping between

the design parameters of diagnostic observers and parity relation
based residual generators. It was also shown that for a given parity
relation based residual generator there exists a set of corresponding observer-based residual generators. It was also shown how to
calculate the respective parity vector when an observer-based
residual generator is provided, and vice versa (Frank, 1996; Ding,
2008).
The parity space based system design is characterized by its
simple mathematical handling. It only deals with matrix- and
vector-valued operations. In the case of observers, the design is
more complex. Due to the connection between the parity space
approach and observer-based approach, a strategy called parity
space design-observer-based implementation was developed
(Ding, 2008). This strategy makes use of the computational
advantage of parity space approach for the system design (selection of a parity vector or matrix) and then realizes the solution in
the observer form to ensure a numerically stable and less consuming online computation.

Fig. 15. Comparison between the performance obtained by obtained by ACO1:1 , PSOI1 and PSO-M for the faulty situations in Tables 1 and 2, up to 8% level noise.

0.5

Diagnostic observer based residual

Parity relation based residual

x 10

1
0

Residual

0.5
1

1.5

2.5

5
0

10

20

30

40

50

60

70

80

90

100

10

20

30

40

50

60

70

80

90

100

Time [s]

Fig. 16. Residual obtained with parity space and diagnostic observers; no faults affecting the system and no noise affecting the output. (a) Parity space and (b) diagnostic
observers.

48

L. Camps Echevarra et al. / Engineering Applications of Articial Intelligence 28 (2014) 3651

approaches provide false alarms. This fact is related with the lack
of robustness. In order to increase robustness, some thresholds
could be established. In Fig. 17 (c) is shown the results of the
estimation of fu in this case, which is next to zero but not equal. In
this case, it is necessary some thresholds which are represented
with red lines and indicates when the estimation values are within
the 0.1% of the maximum values allowed to this fault.
The residual and the result of the fault estimation approach
when the system is affected by fu 0.9 at t 50 s is shown in
Fig. 18. A zoom of the residual obtained by Diagnostic Observer is
shown in Fig. 19. The residual exceeded the thresholds values at
this time, in both approaches. Thus, it indicates that system is
under a fault fu. Both approaches detected the fault. On the other
hand, our proposal also detected the fault and it also allowed to
obtain fu 0.8940 as an estimation of its magnitude, see Fig. 18 (c).

10.4. Results of the comparison between parity space, diagnostic

observers and our proposal when diagnosing faults in the DC motor
In the DC motor is necessary to diagnose three faults with just one
output (one sensor). Parity Space and Diagnostic Observers do not
permit to diagnose additive faults when their number is larger than
the number of the sensors (Ding, 2008). Therefore, the system cannot
be diagnosed with a Parity Space or Diagnostic Observers, only the
detection of the faults is possible. That is the rst advantage of our
proposal: it can diagnose the three faults using only one sensor.
Considering that only one fault can affect the system, it allows
to make a comparison with the Parity Space and Diagnostic
Observers methods. It was chosen the fault fu. In order to
determine the parity vector and to design the observer, the model
of the DC Motor was transformed into its state space representation. This permits to compute the elements of both approaches in
an easier way. In the case of the proposal of this paper, it can be
applied independently on the kind of model describing the system
and no further elements need to be designed or computed.
The parity vector vs 0:1057 0:4499 0:7162  0:5056
0:1336t and matrix H  194:1163 1:8394 0:021  1:3084t
were computed.

10.4.2. Sensitivity analysis

The residual when the system is affected by an incipient fault
fu 0.08 at t 50 s is shown in Fig. 20. It is also shown in Fig. 20 (a)
that the parity space based residual did not exceed the thresholds
values needed for a robust diagnosis. Thus, for Parity Space the
thresholds that avoid false alarms do not allow detecting incipient
faults. On the other hand, in Fig. 20 (b) it is shown that diagnostic
observers detected the fault. It is shown a zoom of the residual
obtained by the Diagnostic Observer in Fig. 21. Our proposal detected
the fault and it also allowed to obtain fu 0.08 as an estimation of its
value, with the same thresholds established in order to obtain
robustness.
The main disadvantage of the proposal of this paper when it is
compared with the parity space and diagnostic observers
approach is the computational time. Once the parity vector is
computed, the parity space approach only needs matrix multiplication, while the PSO-M algorithm needs to make many
simulations of the model of the system. For cases in Figs. 18 and 20,

10.4.1. Robustness analysis

In Fig. 16 are showed the residual obtained by Parity Space and
Diagnostic Observers when no faults are affecting the system and
the output of the system is not affected by noise. In this case, the
residual is equal zero.
In Fig. 17, is shown the residual when the system is not affected
by faults but the output is corrupted with a noise level up to 8%.
In Fig. 17 (b) a zoom of the residual obtained by Diagnostic
Observer is shown. Both approaches, Parity Space and Diagnostic
Observers generate residuals different from zero due to the noise
affecting the output of the system. As a consequence, both

Parity relation based residual

x 10

Fault alarm

Fault alarm

2
Residual

Diagnostic observer based residual

x 10

1
0

1
1

2
3

4
0

10

20

30

40

50

60

70

80

90

100

50

55

60

65

70

x 10

80

85

90

95

100

Fault Estimation

75
Time [s]

Time [s]

Fault Alarm Area

Value of fu

4
2
0
2
4

Fault Alarm Area

6
8

20

40

60

80

100

Time[s]
Fig. 17. Residual obtained with parity space and diagnostic observers; no faults affecting the system and up to 8% level noise. (a) Parity space, (b) diagnostic observers
and (c) fault estimation.

L. Camps Echevarra et al. / Engineering Applications of Articial Intelligence 28 (2014) 3651

49

Parity relation based residual

x 10

Diagnostic observer based residual

1

Fault alarm

4
0
1
Residual

Residual

2
3

4
4

Fault alarm

5
0

10

20

30

40

50

60

70

80

90

100

20

40

60

80

100

Time [s]

Time [s]

Fault Estimation

Fault Alarm
fu=0.894

Value of fu

0.5

0.5

20

40

60

80

100

Time[s]
Fig. 18. Residual obtained with parity space and diagnostic observers, actuator fault affecting the system fu 0.9 and up to 8% level noise. (a) Parity space, (b) diagnostic
observers, and (c) fault estimation.
6

Diagnostic observer based residual

x 10

Fault alarm

Residual

2
0
2
4
6

10

20

30

40

50

60

70

80

90

100

Time [s]

Fig. 19. Zoom of the residual presented in Fig. 18 (b).

the PSO-M algorithm took 906 model simulations (174 s) and took
1081 model simulations (305 s), respectively. On the other hand,
diagnostic observers took around 16 s for computing the residual.
Considering that most industrial processes have high time constants,
the processing time required by this proposal does not make
impracticable its use.

11. Conclusions
This study indicates that the application of meta heuristics, in
particular PSO and ACO, characterizes a promising methodology
for fault diagnosis problems based on direct faults estimation.

The advantages of the proposed approach are: it does not need to

divide the FDI into many steps, such as model parameters estimation,
determination of the relationship between model parameters and
physical parameters as well as the relationship between the last one
and the faults, which requires more than one technique; it permits the
direct estimation of the faults, indistinctly if they took place in the
actuator, process or sensors, which allows alleviating one of the
described limitations of the model based methods for FDI.
After comparing the proposal of this paper with Parity Space and
Diagnostic Observers, other advantages of it were identied: it permits
to diagnose the system even when the number of faults is longer than
the number of sensors; it allowed to obtain robust and sensitive
diagnosis in a simple way; it is a general methodology that does not
depend on the kind of model used to describe the system, no further
elements are needed to be designed or computed, only the simulations of the model of the system are required.
The study of the inuence of the parameters from PSO and ACO
permitted the analysis of the inuence of diversication and
intensication in the quality of the diagnosis. The manipulation
of these parameters in order to increase robustness and sensitivity
avoid the efforts demanded by the robust residual generation in
other model based FDI methods. The study revealed that an
adequate balance between these two tendencies, diversication
and intensication is essential for the development of a robust and
sensitive diagnosis based on PSO and ACO. The variants that gave
the best results for each algorithm were those that performed
higher diversication during the search.
A new algorithm, PSO-M, was proposed. It has an easy structure
and simple implementation. The application of the Wilcoxon
signed ranks test indicated that PSO-M over performed PSOI1
taking as criterion the computational cost, while keeping the

50

L. Camps Echevarra et al. / Engineering Applications of Articial Intelligence 28 (2014) 3651

Parity relation based residual

x 10

Diagnostic observer based residual

1
4

Fault alarm

0
1
Residual

Residual

4
0

10

20

40

Time [s]

60

80

100

Time [s]

Fig. 20. Residual obtained with parity space and diagnostic observers, actuator fault affecting the system fu 0.08 and up to 8% level noise. (a) Parity space, (b) diagnostic
observers, and (c) fault estimation.

Residual

x 106

Diagnostic observer based residual

Fault alarm

5
0

10

20

30

40

50 60
Time [s]

70

80

90

100

Fig. 21. Zoom of the residual presented in Fig. 20 (b).

quality of the estimations. The variant PSO-M was described

without loss of generality, which permits its application to other
continuous optimization problem.

Acknowledgments
The authors acknowledge the support provided by FAPERJ, Fundao de Amparo Pesquisa do Estado do Rio de Janeiro, CNPq, Conselho
Nacional de Desenvolvimento Cientco e Tecnolgico, and CAPES,
Coordenao de Aperfeioamento de Pessoal de Nvel Superior.
References
Angeline, P.J., 1998. Evolutionary Programming VII: Proceedings of the Seventh
Annual Conference on Evolutionary Programming (EP98), Lecture Notes in

Computer Science. Springer-Verlag, vol. 1447. Chapter Evolutionary Optimization versus Particle Swarm Optimization: Philosophy and Performance Differences, pp. 601611.
Becceneri, J., Stephany, S., Campos-Velho, H.F., Silva-Neto, A.J., 2006. Solution of the
inverse problem of radiative properties estimation with PSO technique. In:
Inverse Problems in Engineering Symposium (IPES), Iowa State University, USA.
Becceneri, J.C., Zinober, A., 2001. Extraction of energy in a nuclear reactor. In: XXXIII
Simpsio Brasileiro de Pesquisa Operacional, Campos do Jordo, SP, Brazil.
Beielstein, T., Parsopoulos, K.E., Vrahatis, M.N., 2002. Tuning PSO Parameters
Through Sensitivity Analysis. Technical Report. Reihe Computational Intelligence CI 124/02, Collaborative Research Center (SFB 531), Department of
Computer Science and University of Dortmund.
Camps-Echevarra, L., Llanes-Santiago, O., Silva-Neto, A.J., 2010. An approach for fault
diagnosis based on bio-inspired strategies. In: Nature Inspired Cooperative
Strategies for Optimization (NICSO 2010) Studies in Computational Intelligence,
pp. 5363.
Carlisle, A., Dozier, G., 2001. An off-the-shelf PSO. In: Proceedings of the Particle
Swarm Optimization Workshop, pp. 16.
Chen, J., Patton, R.J., 1999. Robust Model-Based Fault Diagnosis for Dynamic
Systems. Kluwer Academic Publishers, Dordrecht.

L. Camps Echevarra et al. / Engineering Applications of Articial Intelligence 28 (2014) 3651

Chow, E.Y., Willsky, A., 1984. Analytical redundancy and the design of robust failure
detection systems. IEEE Trans. Autom. Control 29, 603614.
Clerc, M., Kennedy, J., 2002. The particle swarm-explosion, stability and convergence in a multidimensional complex space. IEEE Trans. Evol. Comput. 6, 5873.
Derrac, J., Garca, S., Molina, D., Herrera, F., 2011. A practical tutorial on the use of
nonparametric statistical test as a methodology for comparing evolutionary
and swarm intelligence algorithms. Swarm Evol. Comput. 1, 318.
Ding, S.X., 2008. Model-Based Fault Diagnosis Techniques. Springer.
Dorigo, M., Blum, C., 2005. Ant colony optimization theory: a survey. Theor.
Comput. Sci. 344, 243278.
Dorigo, M., Caro, G.D., The Ant Colony Optimization Meta-Hueristic (Ph.D. thesis),
Universite Libre de Bruxelles, 1992.
Duarte, C., Quiroga, J., 2010. Algoritmo PSO para identicacin de parmetros en un
motor DC. Rev. Fac. Ing. Univ. Antioquia 55, 116124.
Eberhart, R.C., Shi, Y.H., 2001. Comparing inertia weights and constriction factors in
particle swarm optimization. In: Proceedings of the IEEE Congress on Evoltionary Computing, pp. 84 88.
Fliess, M., Join, C., Mounier, H., 2004. An introduction to nonlinear fault diagnosis
with an application to a congested internet router. Advances in communication
and control networks. In: Lecture Notes Control Information Science, Springer.
Frank, P.M., 1990. Fault diagnosis in dynamic systems using analytical and knowledgebased redundancy a survey and some new results. Automatica 26, 459474.
Frank, P.M., 1996. Analytical and qualitative model-based fault diagnosis- a survey
and some new results. Eur. J. Control 2, 628.
Hoeing, T., 1993. Detection of parameter variations by continuous-time parity
equations. In: 12th IFACWorld-Congress, pp. 511516.
Hoeing, T., Isermann, R., 1996. Fault detection based on adaptive parity equations
and single-parameter tracking. Control Eng. Pract. 4, 13611369.
Isermann, R., 1984. Process fault detection based on modelling and estimation
methods a survey. Automatica 20, 387404.
Isermann, R., 2005. Model based fault detection and diagnosis. Status and applications. Annu. Rev. Control 29, 7185.
Kameyama, K., 2009. Particle swarm optimization a survey. IEICE Trans. Inf. Syst.
E92-D, 13541361.
Kennedy, J., 1997. The particle swarm: social adaptation of knowledge. In: IEEE
International Conference on Evolutionary Computation. IEEE, pp. 303308.
Kennedy, J., 1998. The Behavior of Particles. Evolutionary Programming VII.
Springer, 581590.
Kennedy, J., Eberhart, R., 1995. Particle swarm optimization. In: IEEE International
Conference on Neural Networks. IEEE, Perth, Australia, pp. 19421948.
Li, Z., Dahhou, B., 2008. A new fault isolation and identication method for
nonlinear dynamic systems: application to a fermentation process. Appl. Math.
Model. 32, 28062830.
Liang, J.J., Qin, A.K., Suganthan, P.N., Baskar, S., 2006. Comprehensive learning
particle swarm optimizer for global optimization of multimodal functions. IEEE
Trans. Evol. Comput. 10, 281295.
Liu, L., Liu, W., Cartes, D.A., 2008. Particle swarm optimization-based parameter
identication applied to permanent magnet synchronous motors. Eng. Appl.
Artif. Intell. 21, 10921100.

51

Liu, Q.L.W., 2009. The study of fault diagnosis based on particle swarm optimization
algorithm. Comput. Inf. Sci. 2, 8791.
Metenidin, M.F., Witczak, M., Korbicz, J., 2011. A novel genetic programming
approach to nonlinear system modelling: application to the damadics benchmark problem. Eng. Appl. Artif. Intell. 24, 958967.
Narasimhana, S., Rengaswamya, P.V.R., 2008. New nonlinear residual feedback
observer for fault diagnosis in nonlinear systems. Automatica 44, 22222229.
Odgaard, P.F., Matajib, B., 2008. Observer-based fault detection and moisture
estimating in coal mills. Control Eng. Pract. 16, 909921.
Patton, R.J., Frank, P.M., Clark, R.N., 2000. Issues of Fault Diagnosis for Dynamic
Systems. Springer, London.
Poli, R., 2007. An Analysis of Publications on Particle Swarm Optimisation Applications. Department of Computer Science, University of Essex.
Samanta, B., Nataraj, C., 2009. Use of particle swarm optimization for machinery
fault detection. Eng. Appl. Artif. Intell. 22, 308316.
Shelokar, P., Siarry, P., Jayaraman, V., Kulkarni, B., 2007. Particle swarm and ant
colony algorithms hybridized for improved continuous optimization. Appl.
Math. Comput. 188, 129142.
Silva-Neto, A.J., Becceneri, J.C., 2009. Bioinspired computational intelligence techniques- application in inverse radiactive transfer problems. Notes in Applied
Mathematics. SBMAC, So Carlos.
Simani, S., Fantuzzi, C., Patton, R.J., 2002. Model-Based Fault Diagnosis in Dynamics
Systems using Identications Techniques. Springer.
Simani, S., Patton, R.J., 2008. Fault diagnosis of an industrial gas turbine prototype
using a system identication approach. Control Eng. Pract. 16, 769786.
Socha, K., Dorigo, M., 2008. Ant colony optimization for continuous domains. Eur.
J. Oper. Res. 185, 11551173.
Souto, R.P., Stephany, S., Becceneri, J.C., Campos Velho, H.F., Silva Neto, A.J., 2005. On
the use of the ant colony system for radiative properties estimation. In: 5th
International Conference on Inverse Problems in Engineering Theory and
Practice (V ICIPE), Leeds University Press, Leeds, Inglaterra, pp. 110.
Venkatasubramanian, V., Rengaswamy, R., Yin, K., Kavuri, S.N., 2002a. A review of
process fault detection and diagnosis. Part 1: quantitative model-based
methods. Comput. Chem. Eng. 27, 293311.
Venkatasubramanian, V., Rengaswamy, R., Yin, K., Kavuri, S.N., 2002b. A review of
process fault detection and diagnosis. part 2: qualitative models and search
strategies. Comput. Chem. Eng. 227, 313326.
Venkatasubramanian, V., Rengaswamy, R., Yin, K., Kavuri, S.N., 2002c. A review of
process fault detection and diagnosis. Part 3: process history based methods.
Comput. Chem. Eng. 27, 327346.
Wang, L., Niu, Q., Fei, M., 2008. A novel quantum ant colony optimization algorithm
and its application to fault diagnosis. Trans. Inst. Meas. Control 30, 313329.
Witczak, M., 2007. Modelling and Estimation Strategies for Fault Diagnosis of NonLinear Systems From Analytical to Soft Computing Approaches. Springer.
Yang, E., Xiang, H., Zhang, D.G.Z., 2007. A comparative study of genetic algorithm
parameters for the inverse problem-based fault diagnosis of liquid rocket
propulsion systems. Int. J. Autom. Comput. 4, 255261.