This seminar provides an insight into Prognostics and Health Management (PHM) through the use of computational intelligence and industrial networked system. Key topics include a neuro-fuzzy self-built system for prognostics, evidential Markovian classification of real-time neuro-fuzzy predictions and intelligent health prognosis of industrial networked system and processes. This seminar is jointly organised by SIMTech and supported by IEEE Industrial Electronics Chapter, Singapore.
Presentation by Dr Rafael Gouriveau
Part I: A Neuro-fuzzy Self-built System for Prognostics: A Way to Ensure Good Prediction Accuracy by Balancing Complexity and Generalisation
Choosing an efficient technique for prognostics depends on constraints that limit the applicability of the tools: available data-knowledge-experiences, dynamic of systems, implementation requirements, available monitoring devices. Moreover, it can be a non trivial task to provide effective models including the inherent uncertainty of prognostics and there is no systematic way to construct a prognostics tool since the user can make some assumptions: choice of a structure, initialisation of parameters. This last problem is addressed in the paper "How to build a prognostics system with no human intervention, neither a priori knowledge?" The proposition is based on the effective use of a neuro-fuzzy (NF) predictor whose architecture is partially determined by a statistical approach based on the Akaike information criterion (AIC). It illustrates the following fundamentals:
1. Since NFs systems learn from examples and attempt to capture the relationships among data, they are suited for problems where it is easier to gather data than to formalise the behaviour of the system being studied. In order to minimise the user's assumptions, the paper highlights the evolving TS models whose structure and parameters are updated without the intervention of an expert and can be trained in online mode (which is very useful for real applications).
2. Nevertheless, practitioners have to choose the inputs of an NF predictor. An approach based on the use of the AIC is proposed to build a cost function that takes into account simultaneously the accuracy of predictions and the complexity of the model. Various inputs can easily be tested in a computational procedure in order to automatically generate an accurate prediction system that reaches a compromise between complexity and generalisation capability.
The proposition is illustrated with various industrial prediction benchmarks and the performances are discussed and compared with the classical Auto Regressive eXogenous approach (ARX).
Part II: Prognostics in Switching Systems: Evidential Markovian Classification of Real-time Neuro-fuzzy Predictions
Condition-based maintenance is nowadays considered as a key-process in maintenance strategies and prognostics appears to be a very promising activity as it should permit not to engage inopportune maintenance spending. For that purpose, various approaches have been developed and, assuming that it is often possible to instrument industrial plant with sensors, data-driven methods are increasingly applied. Given a set of failure histories with possibly numerous observations (measurements or features), usual data-driven approaches generally consist in two steps: a training step that builds models of failures and an inference step that detects potential failure at each instant. However, the training step generally requires huge data sets since a lot of methods rely on probability theory and/or on artificial neural networks. Therefore, this step is time-consuming and generally made in batch mode which can be very restrictive in practical application when few data are available. A method for prognostics is proposed to face up to this problem of lack of information and missing prior knowledge. The approach is based on the integration of three complementary modules and aims at predicting the failure mode early while the system can switch between several functioning modes. The three modules are:
(a) Observation selection: an information theory-based criterion is first used to isolate the most useful observations and quantify their complementarity and redundancy, and thereby, to select the most relevant observations.
(b) Prediction: an evolving real-time neuro-fuzzy system is then used for a robust on-line prediction of observations at different horizons.
(c) Classification: the predicted observations are classified into one of the possible functioning modes using an evidential Markovian classifier (EMC) based on Dempster-Shafer theory. In the experiments, this association is shown to be very efficient in treating a real-world prognostic problem concerning the prediction of an engine health based on more than twenty observations.
About the Speaker
Dr Rafael Gouriveau received his engineering degree from National Engineering School of Tarbes (ENIT) in 1999. He then obtained his MS (2000) and PhD in Industrial Systems in 2003, both from the Toulouse National Polytechnic Institute (INPT). While doing his PhD, he worked in the field of risk management and dependability analysis. In September 2005, he joined the national high school of mechanics and microtechnology in Besançon, France (Ecole Nationale Supérieure de Mécanique et des Microtechniques, ENSMM) as Associate Professor in the field of production, maintenance, manufacturing, and informatics domains. His current research interests include the development of industrial prognostics systems via neuro-fuzzy methods and the investigation of reliability modelling by using possibility theory.
Presentation by Dr Justin Chee Khiang Pang
Intelligent Health Prognosis of Industrial Networked Systems and Processes
Identification and prediction of lifetime in engineering systems using minimal sensors is crucial to reduce production costs and down-time without halting and intrusion into ongoing industrial processes. This seminar provides autonomous decision-making and formal mathematical software machineries for intelligent diagnosis and prognosis of industrial networked systems using matrix operations. This framework allows tighter integration among various skilled levels and increases the yield of mass-produced devices for rapid design evaluations and refinements. The proposed methodology also improves the reliability of large-scale inter-connected systems and discrete event systems to boost sales and increase revenue in competitive manufacturing industries.
About the Speaker
Dr Justin Chee Khiang Pang holds a doctorate and a masters degree in electrical and computer engineering from National University of Singapore (NUS). He has been working closely with A*STAR's Data Storage Institute (DSI). In 2003, he began his academic career with University of Queensland in Brisbane as a Visiting Fellow in the School of Information Technology and Electrical Engineering (ITEE) to teach probabilistic small signal stability of large-scale interconnected power systems project funded by Electric Power Research Institute (EPRI), Palo Alto, CA, USA. He moved to Japan to work as a researcher at the Central Research Laboratory, Hitachi Ltd in Tokyo in 2006. In 2007, he returned to University of Queensland as a Visiting Academic staff in the School of ITEE, and was invited by IEEE Queensland Section to give a seminar. From 2008 to 2009, he was employed as a Visiting Research Professor in the Automation & Robotics Research Institute, University of Texas at Arlington in Texas, USA. A member of IEEE, he is currently an Assistant Professor in the Department of Electrical and Computer Engineering, National University of Singapore.
He was a Guest Editor for Transactions of the Institute of Measurement and Control and the Journal of Control Theory and Applications in 2008 and 2009 respectively. He was also an International Program Committee Member for the 2007 IEEE Congress on Evolutionary Computation (CEC). He is currently a reviewer of many international referred journals and conferences.
His current research interests include intelligent diagnosis and prognosis of industrial networked systems, small signal stability analysis for large-scale inter-connected power systems, and precision control of actuators for data storage systems. He has authored more than 40 technical publications, and is currently co-authoring a research monograph by C K Pang, F L Lewis, T H Lee, and Z Y Dong, Intelligent Diagnosis and Prognosis of Industrial Networked Systems in readiness for submission to Taylor and Francis Group, CRC Press, December 2010.
2.00-2.45 pm: Presentations by Dr Rafael Gouriveau
3.00pm-3.35pm Presentation by Dr Chee Khiang Pang, Justin
3.50pm-4.15pm Refreshments & Networking
Who Should Attend
Research academic staff, students, senior management, R&D managers and engineers.
Registration for this seminar is free of charge. Seats are confirmed on a first-come, first-served basis.
For technical enquiries, please contact Dr Li Xiang, Email: xli@SIMTech.a-star.edu.sg; Tel: 6793 8264
For registration & general enquiries, please contact Alice Koh, Email: email@example.com; Tel: 6793 8249