Sundarapandian et al. (Eds) : CSE, CICS, DBDM, AIFL, SCOM - 2013 pp. 29–39, 2013. © CS & IT-CSCP 2013 DOI : 10.5121/csit.2013.3304         Yahia Kourd 1 , Dimitri Lefebvre 2 and Noureddine Guersi 3 1 Med Khider Biskra University, Department of Electrical Engineering, Algeria ykourd@ yahoo.fr 2 GREAH – University of Havre 25 rue Philippe Lebon – 76058 Le Havre – France dimitri.lefebvre@univ-lehavre.fr 3 Badji Mokhtar University, Department of Electronic, Annaba, Algeria guersi54@yahoo.fr ABSTRACT This paper presents a new idea for fault detection and isolation (FDI) technique which is applied to industrial system. This technique is based on Neural Networks fault-free and Faulty behaviours Models (NNFMs). NNFMs are used for residual generation, while decision tree architecture is used for residual evaluation. The decision tree is realized with data collected from the NNFM’s outputs and is used to isolate detectable faults depending on computed threshold. Each part of the tree corresponds to specific residual. With the decision tree, it becomes possible to take the appropriate decision regarding the actual process behaviour by evaluating few numbers of residuals. In comparison to usual systematic evaluation of all residuals, the proposed technique requires less computational effort and can be used for on line diagnosis. An application example is presented to illustrate and confirm the effectiveness and the accuracy of the proposed approach. KEYWORDS Neural Network, Fault Detection and Isolation, Faulty Model, & Decision Tree 1. INTRODUCTION In a process, early diagnosis of faults, that might occur, allows performing important prevention actions. Therefore, fault detection is a crucial task in the automatic control of large system as manufacturing systems (Diag, 2009). Moreover, it allows avoiding heavy economic losses due to production stop, replacement of spares parts, etc. The need of performing and reliable developed methods for the systems diagnosis becomes increasingly pressing. These methods should respects the following points: (1) Standards and quality improvement; (2) Diagnostic failure to improve the relationship and (3) The definition of new services as new technologies and economic interests. Most of the fault diagnosis methods found in the literature are based on linear methodology or exact models. Models of industrial processes are often very complex. It is difficult to accurately predict their behavior, especially with corrupted measures, and unreliable sensors. Therefore, a number of researchers have perceived artificial neural networks as an alternative way to represent knowledge about faults (Sorsa et al. 1992, Himmelblau 1992, Patton et al. 1994, Frank 1997, Patton et al. 1999, Calado et al. 2001, Korbicz et al. 2004). This paper presents a FDI method that generates a large number of residuals depending on the set of candidate faults. The residuals are analyzed and evaluated according to their mean values. A decision tree is introduced to manage the residuals evaluation and to decide on line which residual