Professional Documents
Culture Documents
Review
a r t i c l e i n f o a b s t r a c t
Article history: Fault Diagnosis and Health Monitoring (FD-HM) for modern control systems have been an active area of
Received 4 August 2016 research over the last few years. Model-based FD-HM computational approaches have been extensively
Accepted 13 September 2016
developed to detect and locate faults by considering logical or mathematical description of the mon-
Available online 16 September 2016
itored process. However, because of parametric, measurement and model uncertainties, applicable ap-
Keywords: proaches that endeavor to locate faults with great accuracy are likely to give false alarms. Recently, many
Fault detection research works have been conducted in order to tackle this issue by making a tradeoff between accuracy
Fault diagnosis and robustness during the fault detection phase. Due to the recent advances in sensor technology, com-
Data-driven methods putational capabilities and dedicated software/hardware interfaces, data-driven FD-HM approaches have
Model-based methods demonstrated that highly accurate fault detection is possible when the system monitoring data for nom-
Hybrid methods inal and degraded conditions are available. Therefore, it seems that more than one approach is usually
required for developing a complete robust fault detection and diagnosis tool. In this paper, the features
of different model-based and data-driven approaches are investigated separately as well as the existing
works that attempted to integrate both of them. In this latter context, there have been only few works
published in the literature and hence reviewing and discussing them is strongly motivated by providing
a good reference for those interested in developing hybrid approaches for FD-HM.
© 2016 International Federation of Automatic Control. Published by Elsevier Ltd. All rights reserved.
1. Introduction The occurrence of a fault in the system can generate a failure. The
term fault is defined by Isermann (2006) as an unpermitted de-
The increasing complexity of industrial systems and their re- viation of at least one characteristic property or parameter of the
lated performance requirements have induced the need to develop system from the acceptable condition while failure is a permanent
new approaches for supervising them. Supervision includes moni- interruption of a system’s ability to perform a required function
toring tasks which aim to determine the system’s operating state under specified operating conditions. In general, faults that occur
at each time. This issue can be divided into two distinct but com- in a physical system can be classified into three categories i.e. ac-
plementary steps, namely: tuator faults (parameters changes), sensor faults (bias for example)
and plant faults (leak in a tank for example).
• Detection which aims to identify the presence of an eventual
These faults can lead to critical failures if they remain unde-
fault in the system.
tected. Therefore, there is a need to generate a set of fault indi-
• Diagnosis which aims to determine the root causes of the de-
cators that should be significantly sensitive to these faults. In re-
tected fault. This task encompasses the fault isolation and iden-
cent years, there has been an increasing interest in fault detec-
tification steps which enable to characterize the type of fault,
tion and diagnosis approaches in order to cope with such issue.
its size and its profile.
Among these approaches, one can distinguish between data-driven
A system is considered in normal operating mode when it pro- approaches, model-based approaches and expert knowledge ones
vides a set of desired functions i.e. the system is in fault free case. (Venkatasubramanian et al., 2003a; Venkatasubramanian et al.,
2003b; Venkatasubramanian et al., 2003c).
∗
Corresponding author. Data-driven approaches (Ding et al., 2011; Joe Qin, 2003) con-
E-mail addresses: khaoula.tidriri@univ-angers.fr, tidririkhaoula@gmail.com sider the detection and the diagnosis as classification tasks. This
(K. Tidriri), nizar.chatti@univ-angers.fr (N. Chatti), sylvain.verron@univ-angers.fr classification can be either supervised or unsupervised. Among the
(S. Verron), teodor.tiplica@univ-angers.fr (T. Tiplica).
http://dx.doi.org/10.1016/j.arcontrol.2016.09.008
1367-5788/© 2016 International Federation of Automatic Control. Published by Elsevier Ltd. All rights reserved.
64 K. Tidriri et al. / Annual Reviews in Control 42 (2016) 63–81
most popular data-driven methods, one can cite Neural Networks ture extraction step which maps the high dimensional vectors to
(Zhang, 20 0 0), Bayesian Networks (Nielsen, 20 07; Pearl, 2014), feature space in order to find certain projecting vectors with low-
Control Charts (Joekes & Barbosa, 2013; Montgomery, 2007), dimension. A common way to perform fault diagnosis is to employ
Principal Component Analysis (Yu, 2012), Partial Least Squares statistical models which aim at classifying the data acquired from
(Kruger, Wang, Chen, & Qin, 2001). These methods play an impor- the monitored process into a normal operating condition class and
tant role in modern monitoring systems especially for large-scale a faulty (out of control) class or distinguishing between different
industry applications since they do not require a lot of computa- fault classes. However, it is difficult to anticipate a priori all the
tions and hence they are compatible with real-time constraints of possible ways in which faults can occur. In order to tackle this is-
dynamic complex systems. However, a preprocessing step is nec- sue, different unsupervised approaches have been developed.
essary to extract information before applying the data-driven ap- In the following, a review of supervised and unsupervised ap-
proaches. proaches, as well as discussions on their main advantages and
Model-based approaches (Ding, 2008; Isermann, 2005) are a drawbacks are carried out.
good alternative when an access to a significant amount of data
is not possible but an accurate analytical model is available. This 2.1. Supervised classification
model is generally build on the physics of the process to be mon-
itored. For each time step, residuals, which depict the differences Supervised classification uses historical data to construct a
between the measured process variables and their estimates, are learning model, which is used for the fault detection and diag-
generated. Then, a decision rule is chosen to evaluate those resid- nosis of the new data. Among the most used methods, one can
uals in order to detect a fault. cite Bayesian Networks (Pearl, 2014) and Artificial Neural Networks
The generation of residuals can be accomplished by various (Zhang, 20 0 0).
methods: diagnostic observers (Luenberger, 1966; Yang, Ding, & Li,
2015), parity relations (Gertler, 1997; Zhong, Song, & Ding, 2015), 2.1.1. Bayesian Networks (BN)
Bond Graph (Ould-Bouamama, El Harabi, Abdelkrim, & Gayed, A BN is a directed acyclic probabilistic graphical model in-
2012; Paynter, 1961). A major advantage of these methods con- troduced by Pearl (2014). BN has been successfully applied in
sists on their ability to provide a description of the dynamic be- various application domains including clinical decision support
havior and a physical understanding of the system. However, in (Sesen, Nicholson, Banares-Alcantara, Kadir, & Brady, 2013), diag-
practice, it is very difficult to develop an accurate mathematical nostic diseases (Antal, Fannes, Timmerman, Moreau, & De Moor,
model that takes into account modeling errors and uncertainties, 2003), genotype data analysis (Yan & Cercone, 2010), cancer metas-
because some sources of uncertainty are not quantifiable. In order tasis modeling and prediction (Wang, Makond, & Wang, 2014),
to address such issue, statistics and interval approaches have been fault detection and diagnosis (Atoui, Verron, & Kobi, 2015b; Ver-
developed. The former ones have represented the uncertainty with ron, Li, & Tiplica, 2010a; Verron, Tiplica, & Kobi, 2008; 2010b;
Gaussian stochastic variables while the latter supposed the uncer- Zhao, Wen, & Wang, 2015; Zhao, Xiao, & Wang, 2013). In particu-
tainties to be unknown but to stay within known and acceptable lar, BN provides a powerful tool for knowledge representation and
bounds. reasoning in presence of uncertainties (Correa, Bielza, & Pamies-
Thereby, each approach has its own advantages and drawbacks. Teixeira, 2009; Gaymard & Tiplica, 2014; Lu, Bai, & Zhang, 2009;
The development of hybrid approaches could improve the FDI per- Weidl, Madsen, & Israelson, 2005). Within this scope, BN has been
formances and overcome the limitations of individual methods used to represent Gaussian Mixture Model (GMM) which is a non-
used separately. Besides, the state of the art encourages defin- regular statistical model (Tao, Li, Zhu, & Li, 2012).This strategy en-
ing a common framework that enables the fusion of different ap- abled to deal with non-Gaussianity problem since the regular sta-
proaches (Venkatasubramanian et al., 2003a; Ding, Zhang, Naik, tistical models cannot be applied in such case. Another problem
Ding, & Huang, 2009). concerns modeling of temporal relationships between variables. It
Accordingly, in Sections 2 and 3, an overview of the methods has been addressed by developing the Dynamic Bayesian Network
from data-driven and model-based approaches respectively, is pre- (DBN) which combines static network with temporal information
sented. Section 4 will highlight the benefits and limits of each ap- (Murphy, 2002).
proach through a comparative analysis and emphasize the interest The DBN has been applied to improve the quality of Internet
of hybrid approaches. In Section 5, a review on the existing re- service (Li, Cheng, Qiu, & Wu, 2009), to detect transient faults
searches that aimed to bridge different approaches is carried out. (Jha, Li, & Seshia, 2009), to identify the fault propagation path-
Finally, the last section will discuss some perspectives about FD- ways, and diagnose the root cause variables (Yu & Rashid, 2013).
HM with hybrid approaches and conclude the paper. In Zhang and Dong (2014), the authors proposed a multi-time-slice
DBN with a mixture of the Gaussian output to handle two princi-
2. Data-driven approaches pal issues: the missing data samples and the non-Gaussian process
data.
Actual processes are increasingly automated, allowing the ac- Beyond its ability to reason with uncertain information, BN can
cess to a sizeable amount of data. Therefore, it is natural to moni- use historical data and expert knowledge to complete the lack
tor the process using methods based on these data. For data-driven of data (Zhao et al., 2013). Moreover, multivariate control charts
approaches, fault diagnosis can be considered as a two stages pro- and other techniques as principal component analysis were mod-
cedure that encompasses (1) fault detection and (2) classification. eled by a BN classifier which enabled to detect and isolate faults
The first stage aims at detecting whether the system behavior within the same framework (Atoui, Verron, & Kobi, 2015c; Verron
matches with the expected one while the second stage concerns et al., 2010a). This strategy had proved its robustness and good
the determination of the class (type) of fault. These two stages can performance. However, the network structure is designed depend-
be performed independently or combined to each other. ing on the prior process knowledge and requires a large amount
Within the data-driven approaches category, one can distin- of training data. Furthermore, the prior probability determination
guish between supervised and unsupervised classification. In su- and conditional probability table (CPT) computation are still chal-
pervised classification, it is necessary to define the classes and la- lenging issues.
bel the training data i.e. provide the category label for each of Thus, the effectiveness of BN depends on the various assump-
them before the training procedure. This latter consists on fea- tions or conditions required for developing an accurate model. In
K. Tidriri et al. / Annual Reviews in Control 42 (2016) 63–81 65
general, BN and statistical classification methods are built on the the Cumulative Sum control chart (CUSUM) [See the pioneering
basis of the Bayesian decision theory. Artificial Neural Networks work of Page (1954), Lucas (1982), and Liu, Xie, and Goh (2006)],
are a good alternative since they can adjust themselves to the data the Exponentially Weighted Moving Average chart (EWMA) [See
without any explicit specification of the distribution in the input the introduction by Roberts (Lucas & Saccucci, 1990; Roberts,
data (Zhang, 20 0 0). 20 0 0) for a performance study], the Weighted CUSUM (WCUSUM)
(Yashchin, 1989) etc.
2.1.2. Artificial Neural Networks (ANN) From the 1920s to the present, there have been many develop-
In recent years, ANNs have been used successfully in pattern ments in control chart methodologies. An important one consists
recognition in order to handle noisy or incomplete data, and for of taking variable sample size (VSS) from the process using a vari-
prediction and classification problems (Zhang, 20 0 0). This method able sampling interval (VSI). VSS and VSI control charts can detect
has been widely used for different applications such as bankruptcy process changes faster than conventional fixed sampling interval
prediction and credit scoring (Tsai & Wu, 2008), medical diagno- control charts [for VSI chart (Amin & Miller, 1993; Chou, Chen, &
sis (Mazurowski et al., 2008), transportation (Garrido, De Oa, & Liu, 2006), for VSI-CUSUM (Reynolds, Amin, & Arnold, 1990)].
De Oa, 2014), image classification (Barat & Ducottet, 2016), senti- Based on the same principle as the univariate control charts,
ment analysis (Vinodhini & Chandrasekaran, 2016). The classifica- multivariate control charts (Lowry & Montgomery, 1995) were de-
tion problem in ANN is reduced to the estimation of the connec- veloped in order to handle a large number of variables, to consider
tion weights, which are learned by using the mismatch between the correlation between these variables and to control the False
the desired and actual outputs. Among the most common net- Alarm Rate (FAR). Among the multivariate control charts, one can
works used for classification, one can cite the single-layer and the cite the Hotelling’s T2 control chart (Lowry & Montgomery, 1995;
multilayer perceptron network (Sorsa, Koivo, & Koivisto, 1991). The Mason, Tracy, & Young, 1995), the MEWMA (Multivariate EWMA)
use of the single-layer perceptron is restricted since it can produce (Bernal-de Lázaro et al., 2016; Lowry, Woodall, Champ, & Rigdon,
only linearly separable decision regions while the multilayer per- 1992; Prabhu & Runger, 1997; Stoumbos & Sullivan, 2002; Sullivan
ceptron network, which consists of three or more layers (an input & Jones, 2002), the MCUSUM (Multivariate CUSUM) (Crosier, 1988).
layer, hidden layers, an output layer), can achieve more general de- Thus, the proposed methods based on multivariate control charts
cision regions (Sorsa et al., 1991). It is worth noting that the ANN’s have become increasingly sensitive to small changes, more robust
performances are characterized by the ANN’s structure, the learn- to the normality assumption and increasingly fast in detection. But
ing algorithm, and the transfer function. Furthermore, an inappro- in return, their development and design have become more com-
priate structure may lead to a weak ANN classifier. plex. However, detecting a small shift in a process very quickly
with tolerable false alarms is still a challenging problem in vari-
2.2. Unsupervised classification ous areas.
Even though the multivariate control charts are much bet-
Unlike supervised methods, unsupervised ones do not make any ter than the univariate control charts for monitoring modern in-
assumption on the knowledge about the individuals belonging to dustrial processes, they are suffering from two inconveniences:
different classes as for control charts, principal component analysis they lose their effectiveness when the number of variables be-
or partial least squares. comes very large and the diagnosis remains a complex task (i.e.
identifying what happened wrong in the process when a fault
2.2.1. Control charts has been detected is not obvious) (Li, Jin, & Shi, 2008; Ma-
The first monitoring method based solely on data was proposed son et al., 1995). In order to cope with dimensionality reduc-
in 1931 by Shewhart and called control chart (Shewhart, 1931). tion, statistical approaches such as Principal Component Anal-
This graphical method, based on statistical hypothesis testing, is ysis (PCA) (Montgomery, 2007) and Partial Least Squares (PLS)
used to monitor a process quality characteristic in order to iden- (Venkatasubramanian et al. (2003a)) are generally used.
tify its special causes of variability (Montgomery, 2007).
One has to distinguish between two phases during the design 2.2.2. PCA and PLS
of control charts: the retrospective analysis in Phase I and the PCA (Montgomery, 2007) is a multivariate statistical analysis
process monitoring in Phase II (Montgomery, 2007). The objec- method that reduces data dimensionality so that essential informa-
tive of the retrospective analysis is to determine the in-control tion is retained and analyzed more easily than in the original data
state of the process and to estimate the parameters when they set. It also captures the maximum variance of the data and the cor-
are unspecified. The most suitable performance index in this step relation between the process parameters. PCA has been widespread
is the False Alarm Probability (FAP) which is the error that oc- used in process monitoring especially in real-time control of con-
curs when a free-fault measurement is considered as a fault tinuous processes (Kourti & MacGregor, 1996; Lu, Wang, & Gao,
(Bernal-de Lázaro, Llanes-Santiago, Prieto-Moreno, Knupp, & Silva- 2003; Zhou, Guo, Celler, & Su, 2014).
Neto, 2016) [See Jones-Farmer, Woodall, Steiner, and Champ (2014), However, PCA is a linear technique while most real processes
Chakraborti, Human, and Graham (2008), and Coelho, Chakraborti, are nonlinear. Therefore, many extended methods have been de-
and Graham (2015) for Phase I control charting review]. In Phase II, veloped to address this issue such as Nonlinear PCA based on auto-
it is assumed that a reference sample is available from the analysis associative neural networks (Kramer, 1992), Kernel PCA which can
performed in Phase I and hence the control chart is used to mon- efficiently compute PCs in a high-dimensional feature space using
itor the process in order to detect changes. The performances dur- nonlinear kernel functions (Lee, Yoo, Choi, Vanrolleghem, & Lee,
ing this step are generally quantified by the Average Time to Sig- 2004; Phillpotts, 2007)...etc.
nal (ATS) and/or the Average Run Length (ARL) (Riaz, Ahmad et al., In order to monitor time-varying systems, adaptive or recursive
2016). This latter represents the number of samples until the de- approaches have been exploited such as the recursive algorithm
tection of an out-of-control signal. for KPCA (Zhang, Li, & Teng, 2012), the moving window kernel
According to Montgomery (Montgomery, 2007), Shewhart con- PCA (Liu, Kruger, Littler, Xie, & Wang, 2009), the adaptive multi-
trol charts are useful for Phase I, since they are effective in de- block and multi-scale PCA which updates the model structure to
tecting large changes. To improve the sensitivity of the control deal with changing process (Lee, Park, & Vanrolleghem, 2005), the
chart to small process shifts, several methods have been pro- weighted adaptive recursive PCA for the slow and normal process
posed such as The Western Electric Run Rules (Electric, 1956) , changes (Portnoy, Melendez, Pinzon, & Sanjuan, 2016)...etc.
66 K. Tidriri et al. / Annual Reviews in Control 42 (2016) 63–81
Many researches have been conducted to address fault isolation server (UIO) or the Unknown Input Filter (UIF) (Hwang, Kim, Kim,
issues using PCA, and they aim to determine the type, location and & Seah, 2010; Liu, Gao, & Han, 2015; Sotomayor & Odloak, 2005).
detection time of a fault. Contribution charts and multi-block ap- On the other hand, to deal with the robust fault isolation, a set
proaches have been proposed to improve fault interpretation using of robust structured or directional residuals have been designed
only process data or using some process knowledge for block par- (Patton & Chen, 1997). These enhanced residuals can be obtained
titioning when it is available (MacGregor, Jaeckle, Kiparissides, & from consistency relations, or from a full PCA model (Gertler &
Koutoudi, 1994). Sparse PCA and shrinking PCA (Xie, Lin, & Zeng, Cao, 2005).
2013) can also enhance fault isolation by resolving an optimization Thus, the observer based techniques have become very robust
problem. to model uncertainties and disturbances but in return, their sensi-
Similarily to PCA, PLS is another powerful statistical tool widely tivity to faults with slow time constants has decreased. To over-
used for fault detection and diagnosis purposes. This method con- come this limitation, Bastin and Gevers (1988) proposed to use
sists on projecting the information contained in high dimension adaptive observers. This solution is applicable not only to linear
data spaces onto low dimension spaces, defined by a number of la- but also to a class of nonlinear systems (Caccavale, Pierri, Iamarino,
tent variables. These new latent variables summarize all the impor- & Tufano, 2009). Basically, the linear observers theory for residual
tant information contained in the original data sets. Equivalently to generation was extended to certain classes of nonlinear systems
PCA, Multi-block and Multiway PLS (Kourti, Nomikos, & MacGregor, (Zarei & Shokri, 2014) by using fuzzy techniques (Li, Ding, Yang,
1995) were developed to facilitate the diagnostic procedure. & Zhang, 2016), H− and H∞ formulation (Chadli, Abdo, & Ding,
It is clear that data-driven approaches are suitable for complex 2013) etc. Nevertheless, many physical systems cannot be modelled
and large-scale systems and can reduce time and cost since they in this way.
do not require the development of models (Alzghoul, Backe, Lf- Furthermore, filtering approaches have been investigated to
strand, Bystrm, & Liljedahl, 2014). However, they mainly depend address the problem of robust fault detection and diagnosis in
on the quantity and the quality of data (Chiang, Braatz, & Rus- the presence of uncertainties. He, Liu, and Hua (2015) proposed
sell, 2001). This brings out a crucial point of distinction between a strategy based on an adaptive Extended KF (EKF) which en-
approaches based on historical process data and model-based ap- ables to reduce the measurement noises. However, this approach
proaches. presents a limitation since it deals only with sensors fault, includ-
ing the bias fault in current and voltage sensors. Recently, another
3. Model-based approaches strategy was proposed by Xiong, Jauberthie, Trave-Massuyes, and
Le Gall (2013) which combines the interval analysis and the sta-
Model-based approaches compare available measurements with tistical behavior of the noise based on the improved Interval KF
a prior information, represented by a mathematical model. The in- and hence it leads to an extension of conventional Kalman filtering.
consistencies between the real and predicted behavior are consid- The output prediction of the KF is considered as the reference of
ered as fault indicators. They are called residuals and they can be the healthy system and the standard deviation is used to build the
generated by using observer based techniques (Narasimhan, Vach- fault detection thresholds. This approach, used for detecting addi-
hani, & Rengaswamy, 2008), parity space (Odendaal & Jones, 2014), tive sensor faults, considers bounded parameter uncertainties and
parameter estimation (Mulumba, Afshari, Yan, Shen, & Norford, centered Gaussian noise. A comparison between a statistical deci-
2015), graph theoretic approaches (Bouamama, Biswas, Loureiro, & sion and an interval-based approach can be found in Gelso, Biswas,
Merzouki, 2014) and so on. Those residuals are then evaluated in Castillo, and Armengol (2008).
order to detect a fault and to reveal abnormal situations on the
monitored process. 3.2. Parity space
parity relation-based residual generation (Hwang et al., 2010). In uncertainties. However, this approach cannot address the measure-
order to obtain a good performance index with a low computa- ments uncertainties. To overcome the limitations of the BG and
tional cost, Ye, Zhang, Ding, and Wang (20 0 0) proposed a wavelet- to generate qualitative and quantitative diagnosis on the basis of
based parity vector of low order but this approach presents a a single representation, Chatti, Ould-Bouamama, Gehin, and Mer-
problem of misdetection. The authors introduced then an infi- zouki (2014) proposed a new formalism called Signed Bond Graph
nite impulse response filter to improve the performance index (SBG).
without increasing the computational effort. However the intro- A Signed Bond Graph G(X,A) is a directed labeled and signed
duction of the filter leads to slow response when faults occur graph built from the BG. This formalism exploits its qualitative
(Ye, Wang, Ding, & Su, 2002). To overcome these issues, Ye, Wang, and quantitative structural properties enabling the generation of
and Ding (2004) used a stationary wavelet transform (SWT) which multiple behavior predictions (i.e. possible conflicts). In order to
presents a low misdetection rate, a satisfactory response speed to enhance the existing fault isolation algorithms based on residuals
faults and a good performance index. generation, it incorporates qualitative features to the BG that take
Since observer and parity space based schemes require the de- into account temporal orders of measurement deviations within a
velopment of the complete model and given that a model cannot specific sliding time window, to address multiple faults. A refined
be easily obtained due to nonlinearity, complexity, and high di- description of the SBG can be found in Chatti et al. (2014).
mensionality of a process, an alternative data-driven approach has Although research on model-based approaches for fault detec-
been proposed recently, using only the primary form of the resid- tion and diagnostic has been actively carried out with noteworthy
ual generator (Ding, Zhang, Huang, & Ding, 2005). results (Isermann, 2006), it is still a challenging task, especially in
researches related to robustness in presence of different sources of
3.3. Parameter estimation uncertainties (Keliris, Polycarpou, & Parisini, 2015), nonlinear sys-
tems and hybrid system issues. During the last years, different fil-
Since the basic model structure is known, the process supervi- tering and parameters estimation approaches have been developed
sion can be fulfilled by monitoring the evolution of its structural and they were based on statistical hypothesis tests (Li, Chan, &
parameters (Isermann, 1984; Venkatasubramanian et al., 2003b). Zhang, 2005). However, such statistical tests require to define the
Indeed, the non-measurable quantities such as process parameters statistical distributions of the residuals in faulty situation. This is
and state variables can be estimated based on the measurable in- why various works focused on the so-called set-membership ap-
put and output variables. This estimation can be performed by an proach such as the robust fault detection strategy proposed by
algorithm based on measured data such as the KF (Kalman, 1960), Tornil-Sin, Ocampo-Martinez, Puig, and Escobet (2012) and formu-
the EKF (Pan et al., 2016), the particle filter (Walker, Rayman, & lated as a set-membership state estimation problem, for both time-
White, 2015), the Bayesian method and other derived methods variant and time-invariant system models.
(Ljung, 1987). Furthermore, special attention has been given to fault detection
Parameter estimation for fault detection and diagnosis relies on approaches for Linear Parameters Varying (LPV) systems since they
a comparison between nominal parameters as computed in fault- represent an attractive alternative to consider nonlinear dynamics
free operation and parameters estimated during online operation (López-Estrada, Ponsart, Theilliol, Zhang, & Astorga-Zaragoza, 2015;
over a given time horizon. Therefore, a fault is detected once the Varga & Ossmann, 2014). Many efforts have also been made to deal
values estimated online are inconsistent with the nominal ones with fault detection and diagnosis in decentralized and distributed
(Isermann, 1993). This approach has been extensively studied in systems (Boem, Ferrari, Parisini, & Polycarpou, 2013; Keliris et al.,
system identification (Ljung, 1987), then applied to fault detection 2015). Finally, it is worth mentioning that various works focused
and diagnosis as reviewed in Isermann (2006). Parameter estima- recently on the design of consistency tests for dynamical systems
tion can be very simple when the model parameters are directly with additive and multiplicative parameter uncertainties by deal-
related to the physical coefficients (such as resistance, loads etc.) ing with intervals analysis (Puig, Oca, & Blesa, 2013).
(Dai & Gao, 2013). However, this technique requires a process in-
put excitation and it is especially suitable for the detection of mul- 4. Comparison between data-driven and model-based
tiplicative faults. approaches
3.4. Bond Graph (BG) Despite numerous studies published in the literature, no single
approach is adapted to all the requirements for a diagnostic sys-
Among the graphical approaches, the BG has proved its ade- tem.
quacy to represent energy exchanges in mixed systems and has On the one hand, data-driven approaches rely on a huge
been used to generate residuals in a systematic and generic way. amount of data and focus on the analysis of the systems output
BG is a unified graphical description that presents a domain- signals. These approaches can handle high dimensional and corre-
independent and energy-based methodology for modeling the dy- lated process variables (Venkatasubramanian et al., 2003a). Thus,
namic behavior of physical systems from different domains (elec- they are suitable for complex and large-scale systems and can re-
trical, mechanical, hydraulic, thermodynamic ...etc.). duce time and cost since they do not require the development of
The BG methodology encompasses various kinds of informa- models (Alzghoul et al., 2014). Although data-driven approaches
tion due to its causal and structural features that enable to de- are simpler than model-based ones, their application require a pre-
duce directly a set of fault indicators. Recently, the BG model has processing step to extract the useful information from data with a
been extended by incorporating the parameters uncertainties in high computational cost. Therefore, the performances rely on train-
order to generate robust and adaptive thresholds for the resid- ing data and may degrade once the system is working in an un-
uals evaluation stage. The procedure for obtaining residuals is known condition (i.e. outside of the training data) or is affected by
based on covering causal paths (Samantaray & Bouamama, 2008) unknown faults. Fig. 1 shows a comparison study of various data-
and is implemented in dedicated software (Chatti, Gehin, Ould- driven methods in terms of desirable criteria for a diagnostic sys-
Bouamama, & Merzouki, 2013; Xiaotian & Anlin, 2014). Notice that tem.
Djeziri, Merzouki, Bouamama, and Dauphin-Tanguy (2007) pro- On the other hand, model-based approaches need small amount
posed BG modeling approach in the LFT (Linear Fractional Transfor- of online data but require an explicit mathematical model, avail-
mation) configuration that enables to take into account parameter able from first principles or identified through system estimation
68 K. Tidriri et al. / Annual Reviews in Control 42 (2016) 63–81
techniques. As the system complexity increases, the estimation be- In recent years, there has been a growing interest in inte-
comes more difficult. In addition, a model validation step is neces- grating different methods to each other. The main motivation for
sary since most models rely on a set of assumptions and approxi- developing hybrid frameworks is that no single method is able
mations (Sargent, 2013). Indeed, these approaches have focused on to satisfy all the requirements of an accurate diagnostic system.
linear system description and some very specific class of nonlin- The purpose of the brief comparison presented above is to high-
earities. According to Chiang et al. (2001), the model-based ap- light and reveal the need for a hybrid approach. Furthermore, ac-
proaches are not suitable for large-scale and complex systems and cording to several studies (Ding et al., 2011; Ding et al., 2009;
their performances depend on how good and precise the model Venkatasubramanian et al., 2003a), there is a need for integrat-
is. Because it integrates the physical understanding of the system ing data-driven and model-based approaches to benefit from the
and incorporates the interactions with the environment into the advantages while avoiding the drawbacks of both approaches. This
fault detection and diagnosis process, these approaches can per- integration has many interests:
form better than the data-driven approaches when an accurate
model is available (Chiang et al., 2001). • It overcomes the weakness of a diagnostic method with the
The main features of existing model-based approaches are sum- strength of another method to achieve a better performance.
marized in Fig. 2. • It enables to use a variety of information sources such as ser-
However, most of the fault diagnosis applications in process in- vice history, operation and maintenance records, historical and
dustries are based on data-driven approaches since they are able to on-line data, mathematical models, causal relationships...etc.
detect faulty situations faster, are easier to implement and require when they are available. Indeed, the more knowledge and in-
less a priori knowledge. formation are merged, the more efficient and reliable the diag-
Furthermore, integrating various strategies of fault detection nosis would be.
and diagnostic within the same framework seems to be interesting. • It enables to benefit from all the progress and achievements
For example, parameter estimation method has been integrated to made by each community.
observers and parity space (Dai & Gao, 2013). Indeed, the state es-
timation based schemes are efficient for fault detection because
they are inherently fast (Patton & Chen, 1997) when compared to 5. Hybrid approaches
parameter estimation method.
Statistical techniques like PCA and PLS have been combined Recently, there has been a growing interest in data-driven
with supervised methods like ANN (Ghate & Dudul, 2011) or with and model-based diagnosis to understand and integrate both ap-
the time-frequency analysis (Lu et al., 2003) to extract statistical proaches in order to provide better diagnostic systems and im-
features, multiple classifier systems have been mixed to improve prove health monitoring. In the following, a review of hybrid ap-
the performance (Ghosh, Ng, & Srinivasan, 2011)...etc. proaches is carried out.
K. Tidriri et al. / Annual Reviews in Control 42 (2016) 63–81 69
One of the earliest examples of a hybrid approach is the hybrid SDG QTA PDF
framework for large scale process fault diagnosis called DKit, pro-
Root cause analysis ++ − −−
posed in 1997 (Mylaraswamy & Venkatasubramanian, 1997). The Large scale, complex, nonlinear process − + ++
motivation for designing such an alternative approach relies on the Robustness to process noise, modeling uncertainties + + ++
fact that a perfect method does not exist. The essential compo- Quickness of response −− ++ ++
nents of DKit are illustrated in Fig. 3. It consists of an input-output Ease of deployment − + ++
Adaptability to new steady states + − −−
interface which receives process measurements, a collection of di-
agnostic methods, a scheduler (responsible for resolving the con-
flicts in results) and an operator interface that presents diagnostic
results to the operator. Assuming that there are n diagnostic methods and that each
The design of Dkit depends on three parameters: method (j) will compute the probability Pij of the hypothesis Hi
given the sensor evidence e, the conflict resolution can be done by
• The collection of diagnostic methods using the weighted voting scheme to estimate the probability of
• The scheme for triggering diagnostic methods the overall hypothesis Hi :
• The algorithm for conflict resolution. 1 n−1
P a j Pi j or
In this paper, the framework combined causal model-based di- P (Hi ) = n−1 i,SDG j=1
−1
Pi,SDG if nj=1 Pi j = 0
agnosis (Signed Digraph) with Probability Density Function-based
statistical classifier (PDF) and qualitative trend analysis (QTA). With 0 ≤ aj ≤ 1.
Thus, a comparison between SDG, PDF and QTA is presented to ex- It implies that SDG is used as a validation for the hypothesis
plain the need for a new hybrid diagnostic system. The Table 1 proposed by QTA and PDF. Indeed, according to the authors, pro-
summarizes this comparison study. cess history based methods are often short-sighted, hence a fault
70 K. Tidriri et al. / Annual Reviews in Control 42 (2016) 63–81
Fig. 4. The proposed framework combining SDG and QTA for fault diagnosis.
Fig. 5. The proposed framework combining PCA and SDG for fault detection and diagnosis.
the combination of SDG and QTA in the second approach is that ology is depicted in Figs. 6 and 7. The performance of this study
both methods are sharing the diagnosis task. Indeed, SDG is used is evaluated on the TEP for static linear systems. Extension to dy-
as a filter to reduce the possible root-causes and QTA is used there- namic and nonlinear systems is not discussed in details in this pa-
after to improve the diagnostic resolution while in the PCA-SDG per.
approach, PCA is only used for detection and SDG for diagnosis. Ding et al. (2009) were also interested in identifying a pri-
mary form of residual generator from data, using parity space
5.2. Identifying a primary form of residual generator from data or observers. The main objective consisted in proving that ob-
server and parity space residuals can be obtained directly from
Some researches focused on combining the ease of deploy- test data, without using the process model. This approach is based
ment of statistical methods with the powerful isolation capa- on the work of Wang and Qin (2002) which combines the ideas
bilities of analytical redundancy. Indeed, Gertler, Li, Huang, and of PCA/DPCA and observer based schemes. The main contributions
McAvoy (1999) showed that there is equivalence between PCA and of the study are the algorithms for the identification of the noise
parity relations and proved this close duality strictly. The idea be- model and for the construction of parity space and observer based
hind this work is to transfer to PCA, which is a powerful tool in residuals. Ding et al. assumed that the processes under consider-
detecting faults in complex systems, some concepts of analytical ation can be modelled as linear time invariant (LTI) systems and
redundancy like the fault isolation capability. proved the applicability of their idea from the TEP.
Thus, the paper presented how structured residuals with good The two previous approaches (Ding et al., 2009; Gertler et al.,
isolation properties can be obtained by PCA. The overall method- 1999) benefit from isolation abilities of model-based approaches
72 K. Tidriri et al. / Annual Reviews in Control 42 (2016) 63–81
Fig. 10. Framework for fault detection and diagnosis using a three-layer BN.
Fig. 11. Framework using measurements and residuals for fault detection and diagnosis.
domains, the conditional probability for a fault to propagate from 5.5. Integrating the reliability data
a domain to the other has a high probability.
This idea was simulated on a continuous-stirred tank reactor In addition to providing an interesting framework for fault de-
and on a biochemical wastewater treatment plant to prove its effi- tection and diagnosis, BN were also used to improve the decision
ciency. Indeed, slow drifts, intermittent and multiple faults were making in ARR. In Weber, Theilliol, Aubrun, and Evsukoff (2007),
detected and associated to their domains. Hence, the reliability the authors integrated reliability data in a DBN with nodes rep-
of isolation was improved by using different methods. However, resenting the residuals and others associated to the failure of
since PCA is used first to reduce the data dimension, the remaining the components. However, this method only considers components
methods are only applied to the resulting scores. Thus, a fault that with exponential probability distribution functions to simplify the
occurs in the residual space may not be detected, which represents computational tasks.
a major limit. Within this scope, Zaidi, Tagina, and Bouamama (2010) pro-
posed to combine the BG method for residual generation with
K. Tidriri et al. / Annual Reviews in Control 42 (2016) 63–81 75
Fig. 12. Framework for model identification, residual generation, fault detection and diagnosis.
BN for data integration. BG is a powerful graphical tool for the rior probabilities of the components faults. The discrete part of the
generation of ARR which are explicitly associated to components BN consists of the residuals and the failures of the components.
faults. This information about the reliability of each component is This structure depends on the failures signatures: if a residual ri
used to improve the decision-making step, making it possible to is sensitive to the failure of the component Cj , then an arc joins
handle unknown and identical signatures of failures. The residu- Fj to ri . The continuous part represents the reliability associated to
als and the components reliability are integrated into a hybrid BN every component and gives the prior information on the failure of
(with continuous and discrete nodes) which calculates the poste- the component.
76 K. Tidriri et al. / Annual Reviews in Control 42 (2016) 63–81
Table 3
Hybrid approaches considered in this review.
bridging different theories and concepts. This issue requires to in- an effective way to combine the results in order to keep the
vestigate the possibilities for modifying some features of the exist- correct decisions while removing incorrect ones.
ing approaches in order to make them compatible with others.
To our knowledge, a work that compares and discusses the per-
6.2. Main assumptions formance of hybrid approaches was not published yet in the liter-
ature. Hence, the development of a database to measure and com-
Integrating various methods seems to be a straightforward exer- pare the performance of hybrid approaches is worth investigating.
cise but it is never the case in practice, since each one of them re- This represents another research opportunity for the FD-HM com-
lies on its own assumptions. Indeed, a method that performs well munity.
in one condition may not provide a satisfactory performance in an-
other condition. Thus, taking into account several assumptions may
6.4. Generic framework for implementation
lead to reduce the scope of the hybrid approaches to be consid-
ered.
It seems important to point out the fact that once the hybrid
approach is designed, one needs to implement it properly. Indeed,
6.3. Desirable characteristics (criteria) a hybrid approach represents a combination of several methods
that are implemented separately and in different ways. Thus, en-
The utility of hybrid approaches is to arrive at a more com- countering interface issues when transposing it into computer code
prehensive solution in order to increase the overall confidence in is inevitable. Therefore, it is useful to propose a generic framework
the diagnosis and the performance. But, how to measure such im- in order to simplify the implementation of hybrid approaches.
provement quantitatively? This work has pointed out several research opportunities that
Obviously, it is useful to have a set of criteria to compare vari- must be addressed for the development of effective hybrid ap-
ous hybrid methods between each other. Diagnostic systems based proaches. It is worth mentioning that detection and diagnosis
on hybrid approaches should meet the desirable characteristics of could interact with other modules of fault management such as
classical diagnostic systems, such as quick detection and diagnosis prognosis module which is devoted to estimate the remaining
with a tolerable false alarm rate, robustness to noise and uncer- useful life of the critical component/subsystem/system. Prognosis
tainties, novelty and multiple fault identifiability, isolability, stor- helps to avoid the occurrence of abnormal events and to reduce
age and computational requirements...etc. A detailed discussion in production loss. Combined with fault detection and diagnosis, it
these criteria can be found in Venkatasubramanian et al. (2003a). could provide global and comprehensive solutions to fault manage-
In addition to these criteria, the design of an effective hybrid fault ment. Thus, achieving unified frameworks for detection, diagnosis
detection and diagnosis method can include other desirable fea- and prognosis, based on hybrid approaches will represent an active
tures such as: field of research for the next years.
References Djeziri, M. A., Merzouki, R., Bouamama, B. O., & Dauphin-Tanguy, G. (2007). Robust
fault diagnosis by using bond graph approach. Mechatronics, IEEE/ASME Transac-
Alzghoul, A., Backe, B., Lfstrand, M., Bystrm, A., & Liljedahl, B. (2014). Comparing a tions on, 12(6), 599–611.
knowledge-based and a data-driven method in querying data streams for sys- Electric, W. (1956). Statistical quality control handbook. Easton: The Mack Printing
tem fault detection: A hydraulic drive system application. Computers in Industry, Company.
65(8), 1126–1135. Garrido, C., De Oa, R., & De Oa, J. (2014). Neural networks for analyzing service qual-
Amin, R. W., & Miller, R. W. (1993). A robustness study of X charts with variable ity in public transportation. Expert Systems with Applications, 41(15), 6830–6838.
sampling intervals. Journal of Quality Technology, 25, 36. Gaymard, S., & Tiplica, T. (2014). Conditional respect towards the pedestrian: Dif-
Antal, P., Fannes, G., Timmerman, D., Moreau, Y., & De Moor, B. (2003). Bayesian ference between men and women and risk modeling by the Bayesian approach.
applications of belief networks and multilayer perceptrons for ovarian tumor Quality & Quantity, 48(1), 91–110.
classification with rejection. Artificial Intelligence in Medicine, 29(1), 39–60. Gelso, E. R., Biswas, G., Castillo, S. M., & Armengol, J. (2008). A comparison
Atoui, M. A., Verron, S., & Kobi, A. (2015a). A Bayesian network dealing with mea- of two methods for fault detection: A statistical decision, and an inter-
surements and residuals for system monitoring. Transactions of the Institute of val-based approach. In 19th international workshop on principles of diagnosis dx
Measurement and Control. 0142331215581446. (pp. 261–268). Citeseer.
Atoui, M. A., Verron, S., & Kobi, A. (2015b). Fault detection and diagnosis in Gertler, J. (1991). Analytical redundancy methods in fault detection and isolation. In
a Bayesian Network classifier incorporating probabilistic boundary 1. IFAC- Preprints of IFAC/IMACS symposium on fault detection, supervision and safety for
PapersOnLine, 48(21), 670–675. technical processes SAFEPROCESS91 (pp. 9–21).
Atoui, M. A., Verron, S., & Kobi, A. (2015c). Fault detection with Conditional Gaussian Gertler, J. (1997). Fault detection and isolation using parity relations. Control engi-
Network. Engineering Applications of Artificial Intelligence, 45, 473–481. neering practice, 5(5), 653–661.
Barat, C., & Ducottet, C. (2016). String representations and distances in deep con- Gertler, J., & Cao, J. (2005). Design of optimal structured residuals from partial prin-
volutional neural networks for image classification. Pattern Recognition, 54, cipal component models for fault diagnosis in linear systems. Journal of Process
104–115. Control, 15(5), 585–603.
Bastin, G., & Gevers, M. R. (1988). Stable adaptive observers for nonlinear time-vary- Gertler, J., Li, W., Huang, Y., & McAvoy, T. (1999). Isolation enhanced principal com-
ing systems. Automatic Control, IEEE Transactions on, 33(7), 650–658. ponent analysis. AIChE Journal, 45(2), 323–334.
Benkouider, A. M., Buvat, J. C., Cosmao, J. M., & Saboni, A. (2009). Fault detection in Ghate, V. N., & Dudul, S. V. (2011). Cascade neural-network-based fault classifier for
semi-batch reactor using the EKF and statistical method. Journal of Loss Preven- three-phase induction motor. Industrial Electronics, IEEE Transactions on, 58(5),
tion in the Process Industries, 22(2), 153–161. 1555–1563.
Benkouider, A. M., Kessas, R., Yahiaoui, A., Buvat, J. C., & Guella, S. (2012). A hybrid Ghosh, K., Ng, Y. S., & Srinivasan, R. (2011). Evaluation of decision fusion strategies
approach to faults detection and diagnosis in batch and semi-batch reactors by for effective collaboration among heterogeneous fault diagnostic methods. Com-
using EKF and neural network classifier. Journal of Loss Prevention in the Process puters & chemical engineering, 35(2), 342–355.
Industries, 25(4), 694–702. He, H., Liu, Z., & Hua, Y. (2015). Adaptive extended kalman filter based fault detec-
Boem, F., Ferrari, R. M., Parisini, T., & Polycarpou, M. M. (2013). Distributed fault tion and isolation for a lithium-ion battery pack. Energy Procedia, 75, 1950–1955.
diagnosis for continuous-time nonlinear systems: The inputoutput case. Annual Hwang, I., Kim, S., Kim, Y., & Seah, C. E. (2010). A survey of fault detection, isolation,
Reviews in Control, 37(1), 163–169. and reconfiguration methods. Control Systems Technology, IEEE Transactions on,
Bouamama, B. O., Biswas, G., Loureiro, R., & Merzouki, R. (2014). Graphical meth- 18(3), 636–653.
ods for diagnosis of dynamic systems: Review. Annual Reviews in Control, 38(2), Isermann, R. (1984). Process fault detection based on modeling and estimation
199–219. methodsa survey. Automatica, 20(4), 387–404.
Caccavale, F., Pierri, F., Iamarino, M., & Tufano, V. (2009). An integrated approach to Isermann, R. (1993). Fault diagnosis of machines via parameter estimation and
fault diagnosis for a class of chemical batch processes. Journal of process control, knowledge processingtutorial paper. Automatica, 29(4), 815–835.
19(5), 827–841. Isermann, R. (2005). Model-based fault-detection and diagnosisstatus and applica-
Chadli, M., Abdo, A., & Ding, S. X. (2013). H-/Hâê¤ fault detection filter design for tions. Annual Reviews in control, 29(1), 71–85.
discrete-time TakagiSugeno fuzzy system. Automatica, 49(7), 1996–2005. Isermann, R. (2006). Fault-diagnosis systems: An introduction from fault detection to
Chakraborti, S., Human, S. W., & Graham, M. A. (2008). Phase I statistical process fault tolerance. Springer Science & Business Media.
control charts: an overview and some results. Quality Engineering, 21(1), 52–62. Jha, S., Li, W., & Seshia, S. A. (2009). Localizing transient faults using dynamic
Chan, C. W., Hua, S., & Hong-Yue, Z. (2006). Application of fully decoupled parity bayesian networks. In High level design validation and test workshop, 2009. HLDVT
equation in fault detection and identification of DC motors. Industrial Electronics, 2009. IEEE international (pp. 82–87). IEEE.
IEEE Transactions on, 53(4), 1277–1284. Joe Qin, S. (2003). Statistical process monitoring: Basics and beyond. Journal of
Chatti, N., Gehin, A., Ould-Bouamama, B., & Merzouki, R. (2013). Functional and be- chemometrics, 17(8–9), 480–502.
havior models for the supervision of an intelligent and autonomous system. Au- Joekes, S., & Barbosa, E. P. (2013). An improved attribute control chart for moni-
tomation Science and Engineering, IEEE Transactions on, 10(2), 431–445. toring non-conforming proportion in high quality processes. Control Engineering
Chatti, N., Ould-Bouamama, B., Gehin, A.-L., & Merzouki, R. (2014). Signed bond Practice, 21(4), 407–412.
graph for multiple faults diagnosis. Engineering Applications of Artificial Intelli- Jones-Farmer, L. A., Woodall, W. H., Steiner, S. H., & Champ, C. W. (2014). An
gence, 36, 134–147. overview of Phase I analysis for process improvement and monitoring. Journal
Chiang, L. H., Braatz, R. D., & Russell, E. L. (2001). Fault detection and diagnosis in of Quality Technology, 46(3), 265.
industrial systems. Springer Science & Business Media. Kalman, R. E. (1960). A new approach to linear filtering and prediction problems.
Chou, C.-Y., Chen, C.-H., & Liu, H.-R. (2006). Economic design of EWMA charts with Journal of basic Engineering, 82(1), 35–45.
variable sampling intervals. Quality and Quantity, 40(6), 879–896. Keliris, C., Polycarpou, M. M., & Parisini, T. (2015). A robust nonlinear observer-based
Chow, E. Y., & Willsky, A. S. (1984). Analytical redundancy and the design of ro- approach for distributed fault detection of inputoutput interconnected systems.
bust failure detection systems. Automatic Control, IEEE Transactions on, 29(7), Automatica, 53, 408–415.
603–614. Kourti, T., & MacGregor, J. F. (1996). Multivariate SPC methods for process and prod-
Coelho, M. L. I., Chakraborti, S., & Graham, M. A. (2015). A comparison of Phase I uct monitoring. Journal of quality technology, 28(4), 409–428.
control charts. South African Journal of Industrial Engineering, 26(2), 178–190. Kourti, T., Nomikos, P., & MacGregor, J. F. (1995). Analysis, monitoring and fault diag-
Correa, M., Bielza, C., & Pamies-Teixeira, J. (2009). Comparison of Bayesian networks nosis of batch processes using multiblock and multiway PLS. Journal of process
and artificial neural networks for quality detection in a machining process. Ex- control, 5(4), 277–284.
pert systems with applications, 36(3), 7270–7279. Kramer, M. A. (1992). Autoassociative neural networks. Computers & chemical engi-
Crosier, R. B. (1988). Multivariate generalizations of cumulative sum quality-control neering, 16(4), 313–328.
schemes. Technometrics, 30(3), 291–303. Kruger, U., Wang, X., Chen, Q., & Qin, S. J. (2001). An alternative PLS algorithm for
Dai, X., & Gao, Z. (2013). From model, signal to knowledge: A data-driven perspec- the monitoring of industrial process. In American control conference, 2001. Pro-
tive of fault detection and diagnosis. Industrial Informatics, IEEE Transactions on, ceedings of the 2001: vol. 6 (pp. 4455–4459). IEEE.
9(4), 2226–2238. Bernal-de Lázaro, J., Llanes-Santiago, O., Prieto-Moreno, A., Knupp, D., & Sil-
Ding, S. (2008). Model-based fault diagnosis techniques: Design schemes, algorithms, va-Neto, A. (2016). Enhanced dynamic approach to improve the detection of
and tools. Springer Science & Business Media. small-magnitude faults. Chemical Engineering Science, 146, 166–179.
Ding, S. X., Zhang, P., Huang, B., & Ding, E. L. (2005). Subspace method aided data– Lee, D. S., Park, J. M., & Vanrolleghem, P. A. (2005). Adaptive multiscale principal
driven design of observer based fault detection systems. In Proceedings of the component analysis for on-line monitoring of a sequencing batch reactor. Jour-
16th IFAC world congress: vol. 16 (pp. 167–172). nal of Biotechnology, 116(2), 195–210.
Ding, S. X., Zhang, P., Jeinsch, T., Ding, E. L., Engel, P., & Gui, W. (2011). A survey Lee, J.-M., Yoo, C., Choi, S. W., Vanrolleghem, P. A., & Lee, I.-B. (2004). Nonlinear
of the application of basic data-driven and model-based methods in process process monitoring using kernel principal component analysis. Chemical Engi-
monitoring and fault diagnosis. In Proceedings of the 18th IFAC world congress neering Science, 59(1), 223–234.
(pp. 12380–12388). Li, J., Chan, C., & Zhang, H.-Y. (2005). Asymptotic local approach in fault detec-
Ding, S. X., Zhang, P., Naik, A., Ding, E. L., & Huang, B. (2009). Subspace method tion based on predictive filters. Journal of guidance, control, and dynamics, 28(6),
aided data-driven design of fault detection and isolation systems. Journal of pro- 1112–1122.
cess control, 19(9), 1496–1510. Li, J., Jin, J., & Shi, J. (2008). Causation-based t2 decomposition for multivariate pro-
Ding, X., Guo, L., & Jeinsch, T. (1999). A characterization of parity space and its ap- cess monitoring and diagnosis. Journal of Quality Technology, 40(1), 46.
plication to robust fault detection. IEEE Transactions on Automatic Control, 44(2), Li, L., Ding, S. X., Yang, Y., & Zhang, Y. (2016). Robust fuzzy observer-based fault de-
337–343. tection for nonlinear systems with disturbances. Neurocomputing, 174, 767–772.
K. Tidriri et al. / Annual Reviews in Control 42 (2016) 63–81 79
Li, Z., Cheng, L., Qiu, X.-s., & Wu, L. (2009). Fault diagnosis for high-level applications Riaz, M., Ahmad, S., Khaliq, Q., Riaz, M., & Ahmad, S. (2016). On designing a new
based on dynamic Bayesian network. In Management enabling the future Internet tukey-ewma control chart for process monitoring. The International Journal of
for changing business and new computing services (pp. 61–70). Springer. Advanced Manufacturing Technology, 82(1–4), 1–23.
Liu, J. Y., Xie, M., & Goh, T. N. (2006). CUSUM chart with transformed exponential Roberts, S. W. (20 0 0). Control chart tests based on geometric moving averages. Tech-
data. Communications in StatisticsTheory and Methods, 35(10), 1829–1843. nometrics, 42(1), 97–101.
Liu, X., Gao, X., & Han, J. (2015). Robust unknown input observer based fault detec- Rusinov, L. A., Vorobiev, N. V., & Kurkina, V. V. (2013). Fault diagnosis in chemical
tion for high-order multi-agent systems with disturbances. ISA transactions. processes and equipment with feedbacks. Chemometrics and Intelligent Labora-
Liu, X., Kruger, U., Littler, T., Xie, L., & Wang, S. (2009). Moving window kernel tory Systems, 126, 123–128.
PCA for adaptive monitoring of nonlinear processes. Chemometrics and Intelli- Samantaray, A. K., & Bouamama, B. O. (2008). Model-based process supervision: A
gent Laboratory Systems, 96(2), 132–143. bond graph approach. Springer Science & Business Media.
Ljung, L. (1987). System identification: Theory for the user. Englewood Cliffs. Sargent, R. G. (2013). Verification and validation of simulation models. Journal of
Lowry, C. A., & Montgomery, D. C. (1995). A review of multivariate control charts. simulation, 7(1), 12–24.
IIE transactions, 27(6), 800–810. Schubert, U., Kruger, U., Arellano-Garcia, H., de S Feital, T., & Wozny, G. (2011). Uni-
Lowry, C. A., Woodall, W. H., Champ, C. W., & Rigdon, S. E. (1992). A multivariate ex- fied model-based fault diagnosis for three industrial application studies. Control
ponentially weighted moving average control chart. Technometrics, 34(1), 46–53. Engineering Practice, 19(5), 479–490.
Lu, J., Bai, C., & Zhang, G. (2009). Cost-benefit factor analysis in e-services using Sesen, M. B., Nicholson, A. E., Banares-Alcantara, R., Kadir, T., & Brady, M. (2013).
Bayesian networks. Expert Systems with Applications, 36(3), 4617–4625. Bayesian networks for clinical decision support in lung cancer care. PloS one,
Lu, N., Wang, F., & Gao, F. (2003). Combination method of principal component and 8(12), e82349.
wavelet analysis for multivariate process monitoring and fault diagnosis. Indus- Shewhart, W. A. (1931). Economic control of quality of manufactured product. ASQ
trial & engineering chemistry research, 42(18), 4198–4207. Quality Press.
Lucas, J. M. (1982). Combined Shewhart-CUSUM quality control schemes. Journal of Siswantoro, J., Prabuwono, A. S., Abdullah, A., & Idrus, B. (2016). A linear model
Quality Technology, 14(2), 51–59. based on Kalman filter for improving neural network classification performance.
Lucas, J. M., & Saccucci, M. S. (1990). Exponentially weighted moving average control Expert Systems with Applications, 49, 112–122.
schemes: Properties and enhancements. Technometrics, 32(1), 1–12. Sorsa, T., Koivo, H. N., & Koivisto, H. (1991). Neural networks in process fault diag-
Luenberger, D. G. (1966). Observers for multivariable systems. Automatic Control, nosis. Systems, Man and Cybernetics, IEEE Transactions on, 21(4), 815–825.
IEEE Transactions on, 11(2), 190–197. Sotomayor, O. A., & Odloak, D. (2005). Observer-based fault diagnosis in chemical
López-Estrada, F. R., Ponsart, J.-C., Theilliol, D., Zhang, Y., & Astorga- plants. Chemical Engineering Journal, 112(1), 93–108.
Zaragoza, C.-M. (2015). LPV model-based tracking control and robust sen- Stoumbos, Z. G., & Sullivan, J. H. (2002). Robustness to non-normality of the multi-
sor fault diagnosis for a quadrotor UAV. Journal of Intelligent & Robotic Systems, variate EWMA control chart. Journal of Quality Technology, 34(3), 260.
1–15. Sullivan, J. H., & Jones, L. A. (2002). A self-starting control chart for multivariate
MacGregor, J. F., Jaeckle, C., Kiparissides, C., & Koutoudi, M. (1994). Process monitor- individual observations. Technometrics, 44(1), 24–33.
ing and diagnosis by multiblock PLS methods. AIChE Journal, 40(5), 826–838. Tao, J., Li, Q., Zhu, C., & Li, J. (2012). A hierarchical naive Bayesian network clas-
Mason, R. L., Tracy, N. D., & Young, J. C. (1995). Decomposition of T2 for multivariate sifier embedded GMM for textural image. International Journal of Applied Earth
control chart interpretation. Journal of quality technology, 27(2), 99–1108. Observation and Geoinformation, 14(1), 139–148.
Maurya, M. R., Rengaswamy, R., & Venkatasubramanian, V. (2007). A signed directed Tarantino, R., Szigeti, F., & Colina-Morles, E. (20 0 0). Generalized Luenberger ob-
graph and qualitative trend analysis-based framework for incipient fault diagno- server-based fault-detection filter design: An industrial application. Control En-
sis. Chemical Engineering Research and Design, 85(10), 1407–1422. gineering Practice, 8(6), 665–671.
Mazurowski, M. A., Habas, P. A., Zurada, J. M., Lo, J. Y., Baker, J. A., & Tornil-Sin, S., Ocampo-Martinez, C., Puig, V., & Escobet, T. (2012). Robust fault de-
Tourassi, G. D. (2008). Training neural network classifiers for medical decision tection of non-linear systems using set-membership state estimation based on
making: The effects of imbalanced datasets on classification performance. Neu- constraint satisfaction. Engineering Applications of Artificial Intelligence, 25(1),
ral networks, 21(2), 427–436. 1–10.
Montgomery, D. C. (2007). Introduction to statistical quality control. John Wiley & Tsai, C.-F., & Wu, J.-W. (2008). Using neural network ensembles for bankruptcy pre-
Sons. diction and credit scoring. Expert Systems with Applications, 34(4), 2639–2649.
Mulumba, T., Afshari, A., Yan, K., Shen, W., & Norford, L. K. (2015). Robust Varga, A., & Ossmann, D. (2014). LPV model-based robust diagnosis of flight actuator
model-based fault diagnosis for air handling units. Energy and Buildings, 86, faults. Control Engineering Practice, 31, 135–147.
698–707. Vedam, H., Dash, S., & Venkatasubramanian, V. (1999). An intelligent operator deci-
Murphy, K. P. (2002). Dynamic bayesian networks: Representation, inference and learn- sion support system for abnormal situation management. Computers & Chemical
ing. University of California, Berkeley Ph.D. thesis.. Engineering, 23, S577–S580.
Mylaraswamy, D., & Venkatasubramanian, V. (1997). A hybrid framework for large Venkatasubramanian, V., Rengaswamy, R., Yin, K., & Kavuri, S. N. (2003a). A review
scale process fault diagnosis. Computers & Chemical Engineering, 21, S935–S940. of process fault detection and diagnosis: Part I: Quantitative model-based meth-
Narasimhan, S., Vachhani, P., & Rengaswamy, R. (2008). New nonlinear residual ods. Computers & Chemical Engineering, 27(3), 293–311.
feedback observer for fault diagnosis in nonlinear systems. Automatica, 44(9), Venkatasubramanian, V., Rengaswamy, R., & Kavuri, S. N. (2003b). A review of pro-
2222–2229. cess fault detection and diagnosis: Part II: Qualitative models and search strate-
Nielsen, T. D. (2007). Bayesian Networks and decision graphs. Springer. gies. Computers & Chemical Engineering, 27(3), 313–326.
Odendaal, H. M., & Jones, T. (2014). Actuator fault detection and isolation: An opti- Venkatasubramanian, V., Rengaswamy, R., Kavuri, S. N., & Yin, K. (2003c). A review
mised parity space approach. Control Engineering Practice, 26, 222–232. of process fault detection and diagnosis: Part III: Process history based methods.
Ould-Bouamama, B., El Harabi, R., Abdelkrim, M. N., & Gayed, M. B. (2012). Bond Computers & Chemical Engineering, 27(3), 327–346.
graphs for the diagnosis of chemical processes. Computers & chemical engineer- Verron, S., Li, J., & Tiplica, T. (2010a). Fault detection and isolation of faults in a
ing, 36, 301–324. multivariate process with Bayesian network. Journal of Process Control, 20(8),
Page, E. S. (1954). Continuous inspection schemes. Biometrika, 41(1/2), 100–115. 902–911.
Pan, S., Xiao, D., Xing, S., Law, S. S., Du, P., & Li, Y. (2016). A general extended Kalman Verron, S., Tiplica, T., & Kobi, A. (2008). Fault detection and identification with a
filter for simultaneous estimation of system and unknown inputs. Engineering new feature selection based on mutual information. Journal of Process Control,
Structures, 109, 85–98. 18(5), 479–490.
Patton, R. J., & Chen, J. (1997). Observer-based fault detection and isolation: robust- Verron, S., Tiplica, T., & Kobi, A. (2010b). Fault diagnosis of industrial systems by
ness and applications. Control Engineering Practice, 5(5), 671–682. conditional Gaussian network including a distance rejection criterion. Engineer-
Paynter, H. M. (1961). Analysis and design of engineering systems. MIT press. ing applications of artificial intelligence, 23(7), 1229–1235.
Pearl, J. (2014). Probabilistic reasoning in intelligent systems: Networks of plausible in- Vinodhini, G., & Chandrasekaran, R. M. (2016). A comparative performance evalua-
ference. Morgan Kaufmann. tion of neural network based approach for sentiment classification of online re-
Phillpotts, D. (2007). Nonlinear fault detection and diagnosis using kernel PCA applied views. Journal of King Saud University-Computer and Information Sciences, 28(1),
to a pilot distillation column. university of Pretoria Master’s thesis. 2–12.
Pierri, F., Paviglianiti, G., Caccavale, F., & Mattei, M. (2008). Observer-based sensor Walker, E., Rayman, S., & White, R. E. (2015). Comparison of a particle filter and
fault detection and isolation for chemical batch reactors. Engineering Applica- other state estimation methods for prognostics of lithium-ion batteries. Journal
tions of Artificial Intelligence, 21(8), 1204–1216. of Power Sources, 287, 1–12.
Portnoy, I., Melendez, K., Pinzon, H., & Sanjuan, M. (2016). An improved weighted Wang, J., & Qin, S. J. (2002). A new subspace identification approach based on prin-
recursive PCA algorithm for adaptive fault detection. Control Engineering Practice, cipal component analysis. Journal of Process Control, 12(8), 841–855.
50, 69–83. Wang, K.-J., Makond, B., & Wang, K.-M. (2014). Modeling and predicting the occur-
Potter, I. E., & Sunman, M. C. (1977). Threshold-less redundancy management with rence of brain metastasis from lung cancer by Bayesian network: A case study
arrays of skewed instruments. of Taiwan. Computers in biology and medicine, 47, 147–160.
Prabhu, S. S., & Runger, G. C. (1997). Designing a multivariate EWMA control chart. Weber, P., Theilliol, D., Aubrun, C., & Evsukoff, A. (2007). Increasing effectiveness of
Journal of Quality Technology, 29(1), 8. model-based fault diagnosis: A dynamic bayesian network design for decision
Puig, V., Oca, S. M., & Blesa, J. (2013). Adaptive threshold generation in robust making. In In proceedings of the conference on fault detection, supervision and
fault detection using interval models: Time-domain and frequency-domain ap- safety of technical processes (pp. 90–95).
proaches. International Journal of Adaptive Control and Signal Processing, 27(10), Weidl, G., Madsen, A. L., & Israelson, S. (2005). Applications of object-oriented
873–901. Bayesian networks for condition monitoring, root cause analysis and decision
Reynolds, M. R., Amin, R. W., & Arnold, J. C. (1990). CUSUM charts with variable support on operation of complex continuous processes. Computers & chemical
sampling intervals. Technometrics, 32(4), 371–384. engineering, 29(9), 1996–2009.
80 K. Tidriri et al. / Annual Reviews in Control 42 (2016) 63–81
Xiaotian, L., & Anlin, W. (2014). Definitions of causality in bond graph model for Zaidi, A., Tagina, M., & Bouamama, B. O. (2010). Reliability data for improvement
efficient simulation mechanism. Mechanism and Machine Theory, 80, 112–124. of decision-making in Analytical Redundancy Relations Bond Graph based di-
Xie, L., Lin, X., & Zeng, J. (2013). Shrinking principal component analysis for en- agnosis. In Advanced intelligent mechatronics (AIM), 2010 IEEE/ASME international
hanced process monitoring and fault isolation. Industrial & Engineering Chem- conference on (pp. 790–795). IEEE.
istry Research, 52(49), 17475–17486. Zarei, J., & Shokri, E. (2014). Robust sensor fault detection based on nonlinear un-
Xiong, J., Jauberthie, C., Trave-Massuyes, L., & Le Gall, F. (2013). Fault detection using known input observer. Measurement, 48, 355–367.
interval kalman filtering enhanced by constraint propagation. In Decision and Zhang, G. P. (20 0 0). Neural networks for classification: A survey. Systems, Man,
control (cdc), 2013 IEEE 52nd annual conference on (pp. 490–495). IEEE. and Cybernetics, Part C: Applications and Reviews, IEEE Transactions on, 30(4),
Yan, L. J., & Cercone, N. (2010). Bayesian network modeling for evolutionary genetic 451–462.
structures. Computers & Mathematics with Applications, 59(8), 2541–2551. Zhang, X., & Hoo, K. A. (2011). Effective fault detection and isolation using bond
Yang, Y., Ding, S. X., & Li, L. (2015). On observer-based fault detection for nonlinear graph-based domain decomposition. Computers & chemical engineering, 35(1),
systems. Systems & Control Letters, 82, 18–25. doi:10.1016/j.sysconle.2015.05.004. 132–148.
Yashchin, E. (1989). Weighted cumulative sum technique. Technometrics, 31(3), Zhang, Y., Li, S., & Teng, Y. (2012). Dynamic processes monitoring using recursive
321–338. kernel principal component analysis. Chemical Engineering Science, 72, 78–86.
Ye, H., Wang, G., & Ding, S. X. (2004). A new parity space approach for fault detec- Zhang, Z., & Dong, F. (2014). Fault detection and diagnosis for missing data systems
tion based on stationary wavelet transform. Automatic Control, IEEE Transactions with a three time-slice dynamic Bayesian network approach. Chemometrics and
on, 49(2), 281–287. Intelligent Laboratory Systems, 138, 30–40.
Ye, H., Wang, G. Z., Ding, S. X., & Su, H. Y. (2002). An IIR filter based parity space Zhao, Y., Wen, J., & Wang, S. (2015). Diagnostic Bayesian networks for diagnosing
approach for fault detection. In Proceedings of the 15th Triennial world congress air handling units faultsPart II: Faults in coils and sensors. Applied Thermal En-
of the international federation of automatic control. gineering, 90, 145–157.
Ye, H., Zhang, P., Ding, S. X., & Wang, G. Z. (20 0 0). A time-frequency domain Zhao, Y., Xiao, F., & Wang, S. (2013). An intelligent chiller fault detection and di-
fault detection approach based on parity relation and wavelet transform. In agnosis methodology using Bayesian belief network. Energy and Buildings, 57,
Decision and control, 2000. Proceedings of the 39th IEEE conference on: vol. 4 278–288.
(pp. 4156–4161). IEEE. Zhong, M., Song, Y., & Ding, S. X. (2015). Parity space-based fault detection for linear
Yu, J. (2012). Local and global principal component analysis for process monitoring. discrete time-varying systems with unknown input. Automatica, 59, 120–126.
Journal of Process Control, 22(7), 1358–1373. Zhou, J., Guo, A., Celler, B., & Su, S. (2014). Fault detection and identification span-
Yu, J., & Rashid, M. M. (2013). A novel dynamic Bayesian network-based networked ning multiple processes by integrating PCA with neural network. Applied Soft
process monitoring approach for fault detection, propagation identification, and Computing, 14, 4–11.
root cause diagnosis. AIChE Journal, 59(7), 2348–2365.
K. Tidriri et al. / Annual Reviews in Control 42 (2016) 63–81 81
K. Tidriri was born in 1992. She received her Engineering degree in Electronic and Automatic Control from Institut Polytechnique de Grenoble
(Grenoble INP) in the year of 2015. Since october 2015, she is performing a PhD in the monitoring of dynamic systems in the Laboratoire Angevin
de Recherche en Ingenierie des Systemes (LARIS, EA 7315). Her research activities are mainly devoted to data-driven, model-based and hybrid
approaches for fault detection and diagnosis.
N. Chatti was born in 1985. He obtained a M.Sc. from Ecole Centrale de Lille and a PhD in Automatic Control From Ecole Polytechnique de Lille in
2010, 2013 respectively. Since september 2014, he is associate professor in the public engineering school ISTIA (University of Angers) and researcher
in LARIS laboratory EA 7315. His research interests include Safety System Design, Robust Fault Diagnosis and interval analysis.
S. Verron received the M.S. degree in Industrial Engineering from Institut National Polytechnique de Toulouse (INPT) in 2004. He received his
doctoral degree in Industrial Engineering from University of Angers in 2007. He is currently associate professor at LARIS Laboratory of ISTIA of the
University of Angers (France). His research interests are in the monitoring of complex processes (fault detection and root cause diagnosis).
T. Tiplica, Ph.D., is an associate professor at the Institute of Engineering Sciences and Technologies of LUNAM University, France. He is a member
of the Laboratoire Angevin de Recherche en Ingnierie de Systemes (LARIS, EA 7315). His research interests include statistical process control, fault
detection and isolation, and Bayesian networks.