Seminar on Prognostics in Intelligent Maintenance: Fault Diagnostics and Prognostics Using Artificial Intelligence Tools

Date: 17 Feb 2009 - 17 Feb 2009

Venue: SIMTech Training Room 2, Tower Block, Level 3


This seminar aims to provide participants with a better understanding of the latest prognostic research by leveraging on two novel approaches, namely (i) neuro-fuzzy based approach; and (2) Bayesian Networks and dynamic modelling approach. Industry participants will benefit from the demonstration of the application of predictive and proactive maintenance systems. 

The high costs in maintaining complex equipments make necessary to enhance maintenance support systems and traditional concepts like preventive and corrective strategies are progressively completed by new ones like predictive and proactive maintenance. For corrective maintenance tasks, fault diagnostic proves to be a suitable support for the operators as it allows detecting and isolating the faults which may affect a given system. Fault diagnostic techniques can be grouped into three main categories, namely quantitative, qualitative and historical approaches. Among the qualitative techniques, Case Based Reasoning can be very useful in many industrial applications. 

However, one must admit that the fault diagnostic activity, and consequently, the corrective maintenance strategy, is not the ideal solution as the faults are undergone and not anticipated. This should be completed or completely replaced by a proactive maintenance strategy for which fault prognostic is a key feature. Indeed, prognostic reveals to be a very promising activity as it should allow avoiding inopportune maintenance spending. According to ISO 13381-1:2004 standard, prognostic corresponds to an estimation of the remaining useful life of equipment before fault occurrence and the risk of existing or future appearance of one or more faulty modes. The estimation of the remaining useful life depends on the current state and on the future usage of the system. Thus, prognostic can use data issued from diagnostic procedures in order to estimate the system’s future performances. 

Prognostic methods can be classified into three main categories: model-based approach, data-driven approach and experience-based approach whose applicability depends on the characteristics of the system under study. Real systems are complex, so deriving analytical models reflecting exactly their dynamics is not an easy task. Moreover, the uncertainty which characterises the available information and data renders prognostic process more difficult. Consequently, the three aforementioned approaches are often used together in order to perform a prognostic which can be as reliable as possible. 

A central challenge is to improve the accuracy of a prognostic system to approximate and predict the degradation of equipment. Neurofuzzy techniques offer an appealing problem solving framework for fault prognostic. This is due to their capability for simultaneously handling numerical and linguistic information. Adaptive neuro-fuzzy inference system that provides accurate prediction with non linear data, knowledge representation, flexibility, computation time makes it a suitable tool for forecasting and fault prognostic. 

The use of dynamic Bayesian Networks allows, on one hand, to model the causal relationships between events and variables of the system and, on the other hand, to take into account the stochastic aspect of these variables. The dynamic modelling approach (state space representation, linear on nonlinear analytical models, etc.), derived by using deterministic laws, is used to represent the dynamic behavior of the system, its sub-systems and components. 
The industrial oriented research work in these techniques have been carried out in various European and France National projects on intelligent maintenance systems like FP5 European Integrated Project of ITEA programme (Information Technology for European Advancement) PROTEUS, NEMOSYS (Naval E-Maintenance Oriented SYStem) with DCNS, AMIMAC-FAME (Reliability Improvement of Embeded Machine) with ALSTOM and CEGELEC. 

In this session, the research achievements will be presented and some applications of the aforementioned techniques will be demonstrated. 

About the Speakers
Associate Professor Rafael Gouriveau graduated from National Engineering School of Tarbes (ENIT) with an engineering degree in 1999. He received his PhD degree in Industrial Systems from the Toulouse National Polytechnic Institute in 2003. During his PhD pursuit, he worked in the field of risk management and dependability analysis. He began his academic career as Associate Professor of AS2M Department of Femto-st Institute in Besancon, France since September 2005. His main teaching activities are concerned with production, maintenance, manufacturing, and informatics domains. Currently, his research interests include the development of industrial prognostic systems using neuro-fuzzy methods and the investigation of reliability modelling by using possibility theory. 

Associate Professor Kamal Medjaher received his PhD in control and industrial computing from University of Lille 1 (LAGIS laboratory) in 2005. During his PhD study, he worked in the field of fault diagnosis and supervision. In September 2006, he joined the AS2M Department of Femto-st Institute in Besancon, France, also known as the national high school of mechanicsand microtechniques of Besançon ( as Associate Professor. His main teaching activities are concerned with control, fault diagnosis and fault prognosis domains. Nowadays, his research interests concern the development of fault prognostic methods and procedures, for complex systems, based on artificial intelligence tools (Bayesian networks) and physical modelling. 

 1.45pm     Registration

 2.00pm     Presentation by Associate Professor Rafael Gouriveau

 2.45pm     Q&A  

 3.00pm     Refreshment & Networking

 3.30pm     Presentation by Associate Professor Kamal Madjaher

 4.30pm     End
Who Should Attend
R&D managers, researchers, scientists, engineers, academic staff and engineering students.

Pre-registration for the lecture is free of charge. Seats are available on a first-come, first-served basis.
For technical enquiries, please contact: Ms Zhou Junhong, Email:; or Dr Ian Chan, Email:

For general enquiries, please contact: Alice Koh, Email: