Monitoring and predictive maintenance: Modeling and analyse of fault latency Zineb Simeu-Abazi 1 , Zouhir Bouredji * Laboratoire d’Automatique de Grenoble (CNRS-INPG-UJF), L.A.G-E.N.S.I.E.G-B.P. 45, 38402 Saint Martin d’He `res Cedex, France Accepted 9 February 2006 Abstract This paper presents an effective way of modeling complex systems through identified functioning modes. In the proposed approach, the integration of monitoring in the manufacturing system is facilitated by the development of a generic model. The aim is to propose a monitoring system able of absorbing internal degradation of any variables and ensuring the continuity of the service. The outline of the optimization of the fault latency method is based on two steps is proposed. The first step is the evaluation of fault latency and the second one is the performance evaluation of monitoring process. Timed automata are the modeling tool used for these two steps. The proposed method can be applied to various kinds of processes and gives good results. Indeed, the simulation results, including a serial manufacturing line, substantiate the feasibility of the proposed method and provide a promising potential to spin-off applications in industrial manufacturing system. # 2006 Published by Elsevier B.V. Keywords: Discrete–event systems; Detection; Fault latency; Predictive maintenance; Timed automata 1. Introduction In manufacturing systems, wear-out and eventual failures are unavoidable. However, to reduce the rate of their occurrences and to improve the lifetime of equipments, maintenance can be performed with an adequate monitoring. In fact, monitoring in production systems may alert the maintenance team when a given degradation increases above a specified threshold [1,2,3,20]. This allows a solution to be found before the occurrence of the failure. Modern industry deals with efficient monitoring to improve reliability of equipment and reduce high maintenance cost [5,16]. Now, the mission allowed to the monitoring system is not only the detection task but also the identification of fault. To detect and identify any faults occurring in the dynamical system, it is necessary to find the kind location, and a time of fault occurrence [1,5]. In discrete–event systems area, the most common monitor- ing and diagnostic approach are based on dynamic model, which represents only the good functioning. The inputs and outputs of the system under supervision are used to detect the fault [4,9,15]. Model-based diagnostic algorithms use an explicit model of dynamical system under investigation. This model incorporates the knowledge about the faultless and the faulty system behaviour in systematic way for the analysis of the fault symptoms [7,8]. For large manufacturing systems, monitoring integration requires specific developments due to the complexity of the models involved [3–5]. Furthermore, these developments consume large computing time resources. Recently, process monitoring and diagnostic methods have been developed for discrete–event systems by using timed Petri nets, stochastic automata, timed automata, template languages or Semi-Markov processes [10,12,14,18]. The main idea of these methods is to simulate nominal or faulty system behaviour with the discrete model. These methods need the structural and functional models of the system. The faulty behaviour is modelled by using a predefined list of eventual faults which can affect the system. The proposed monitoring function for predictive main- tenance is a part of global system supervision. Thanks to the www.elsevier.com/locate/compind Computers in Industry 57 (2006) 504–515 * Corresponding author at: Polytech’Grenoble-Universite ´ Joseph Fourier, B.P. 53, 38041 Grenoble Cedex 9, France. Fax: +33 438 02 91 01. E-mail addresses: Zineb.Simeu-Abazi@inpg.fr (Z. Simeu-Abazi), zouhir.bouredji@wanmes.com (Z. Bouredji). 1 Fax: +33 475 82 53 88. 0166-3615/$ – see front matter # 2006 Published by Elsevier B.V. doi:10.1016/j.compind.2006.02.017