Sylvain Verron

Bridging data-driven and model-based approaches for process fault diagnosis and health monitoring: A review of researches and future challenges

Fault Diagnosis and Health Monitoring (FD-HM) for modern control systems have been an active area of research over the last few years. Model-based FD-HM computational approaches have been extensively developed to detect and locate faults by considering logical or mathematical description of the monitored process. However, because of parametric, measurement and model uncertainties, applicable approaches that endeavor to locate faults with great accuracy are likely to give false alarms. Recently, many research works have been conducted in order to tackle this issue by making a tradeoff between accuracy and robustness during the fault detection phase. Due to the recent advances in sensor technology, computational capabilities and dedicated software/hardware interfaces, data-driven FD-HM approaches have demonstrated that highly accurate fault detection is possible when the system monitoring data for nominal 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.

Fault diagnosis of industrial systems by conditional Gaussian network including a distance rejection criterion

The purpose of this article is to present a method for industrial process diagnosis with Bayesian network, and more particularly with conditional Gaussian network (CGN). The interest of the proposed method is to combine a discriminant analysis and a distance rejection in a CGN in order to detect new types of fault. The performances of this method are evaluated on the data of a benchmark example: the Tennessee Eastman Process. Three kinds of fault are taken into account on this complex process. The challenging objective is to obtain the minimal recognition error rate for these three faults and to obtain sufficient results in rejection of new types of fault.

Distance rejection in a bayesian network for fault diagnosis of industrial systems

The purpose of this article is to present a method for industrial process diagnosis with Bayesian network. The interest of the proposed method is to combine a discriminant analysis and a distance rejection in a bayesian network in order to detect new types of fault. The performances of this method are evaluated on the data of a benchmark example: the Tennessee Eastman Process. Three kinds of fault are taken into account on this complex process. The challenging objective is to obtain the minimal recognition error rate for these three faults and to obtain sufficient results in rejection of new types of fault.

Fault Diagnosis with Bayesian Networks: Application to the Tennessee Eastman Process

The purpose of this article is to present and evaluate the performance of a new procedure for industrial process diagnosis. This method is based on the use of a Bayesian network as a classifier. But, as the classification performances are not very efficient in the space described by all variables of the process, an identification of important variables is made. This feature selection is made by computing the mutual information between each process variable and the class variable. The performances of this method are evaluated on the data of a benchmark problem: the Tennessee Eastman process. Three kinds of faults are taken into account on this complex process. The objective is to obtain the minimal recognition error rate for these 3 faults. Results are given and compared with results of other authors on the same data.

A Bayesian network dealing with measurements and residuals for system monitoring

The purpose of this paper is to present an original method for system monitoring with Bayesian networks. Our proposal is to associate a data-driven method to another model-based under a common tool. The two methods are first modeled under a Bayesian network (conditional Gaussian network), and then combined to evaluate the system state. In the proposed framework the residuals and measures coexist under a probabilistic framework. This approach is tested on a simulation of a water heater process under some various circumstances and shows better results than the two methods used alone.

Fault detection with Conditional Gaussian Network

The main interest of this paper is to illustrate a new representation of the Principal Component Analysis (PCA) for fault detection under a Conditional Gaussian Network (CGN), a special case of Bayesian networks. PCA and its associated quadratic statistics such as T2 and SPE are integrated under a sole CGN. The proposed framework projects a new observation into an orthogonal space and gives probabilities on the state of the system. It could do so even when some data in the sample test are missing. This paper also gives the probabilities thresholds to use in order to match quadratic statistics decisions. The proposed network is validated and compared to the standard PCA scheme for fault detection on the Tennessee Eastman Process and the Hot Forming Process.

Fault detection of univariate non-Gaussian data with Bayesian network

The purpose of this article is to present a new method for fault detection with Bayesian network. The interest of this method is to propose a new structure of Bayesian network allowing to detect a fault in the case of a non-Gaussian signal. For that, a structure based on Gaussian mixture model is proposed. This particular structure allows to take into account the non-normality of the data. The effectiveness of the method is illustrated on a simple process corrupted by different faults.

Procedure based on mutual information and bayesian networks for the fault diagnosis of industrial systems

The aim of this paper is to present a new method for process diagnosis using a Bayesian network. The mutual information between each variable of the system and the class variable is computed to identify the important variables. To illustrate the performances of this method, we use the Tennessee Eastman Process. For this complex process (51 variables), we take into account three kinds of faults with the minimal recognition error rate objective.

Model-based approach for fault diagnosis using set-membership formulation

This paper describes a robust model-based fault diagnosis approach that enables to enhance the sensitivity analysis of the residuals. A residual is a fault indicator generated from an analytical redundancy relation which is derived from the structural and causal properties of the signed bond graph model. The proposed approach is implemented in two stages. The first stage consists in computing the residuals using available input and measurements while the second level leads to moving horizon residuals enclosures according to an interval consistency technique. These enclosures are determined by solving a constraint satisfaction problem which requires to know the derivatives of measured outputs as well as their boundaries. A numerical differentiator is then proposed to estimate these derivatives while providing their intervals. Finally, an inclusion test is performed in order to detect a fault upon occurrence. The proposed approach is well suited to deal with different kinds of faults and its performances are demonstrated through experimental data of an omni-directional robot.

Monitoring of Complex Processes with Bayesian Networks

1. IntroductionIndustrial processes are more and more complex and include a lot of sensors giving measure-ments of some attributes of the system. A study of these measurements can allow to decideon the correct working conditions of the process. If the process is not in normal working con-ditions, it signies that a fault has occurred in the process. If no fault has occurred, thus theprocess is in the fault-free case. An important research eld is on the Fault Detection and Di-agnosis (FDD) (Isermann (2006)). The goal of a FDD scheme is to detect, the earliest possible,when a fault occurs in the process. Once the fault has been detected, the other important stepis the diagnosis. The diagnosis can be seen as the decision of which fault has appeared in theprocess, what are the characteristics of this fault, what are the root causes of the fault.One can distinguish three principal categories of methods for the FDD (Chiang et al. (2001)):the knowledge-based approach, the model-based approach and the data-driven approach.The knowledge-based category represents methods based on qualitative models (FMECA -Failures Modes Effects and Critically Analysis; Fault Trees; Decision Trees; Risk Analysis)(Stamatis (2003); Dhillon (2005)). For the model-based methods, an analytical model of theprocess is constructed based on the physical relations governing the process (Patton et al.(2000)). The model gives the normal (fault free) value of each sensor or variable of the systemfor each sample instant, then residuals are generated (residuals are the differences betweenmeasurements and the corresponding reference values estimated with the model of the fault-free system). If the system is fault free, residuals are almost nil, and so their evaluations allowto detect and diagnose a fault. Theoretically, the best methods are the analytical ones, but themajor drawback of this family of techniques is the fact that a detailed model of the process isrequired in order to monitor it efciently. Obtaining an effective detailed model can be verydifcult, time consuming and expensive, particularly for large-scale systems with many vari-ables. Thelastcategoryofmethodsaretheprocesshistory(ordata-driven)methods(Venkata-subramanian et al. (2003)). These techniques are based on rigorous statistical developmentsof process data. In literature, we can nd many different data-driven techniques for FDD. Forthefaultdetectionofindustrialprocessesmanymethodshavebeensubmitted: univariatesta-tisticalprocesscontrol(Shewhartcharts)(Montgomery(1997)),multivariatestatisticalprocesscontrol ( T