WRESTLING SYSTEM FAILURES BY ADVANCED COMPUTATIONAL METHODS
Failures are phenomena which occur in the uncertain and 'obscure' domains outside the systems design envelopes. They are unavoidable in all technological systems and their investigation entails the exploration of the functional and physical limits of the systems in an effort to understand how, why and when they may not function properly. In this respect, the required approach is complementary to the traditional engineering viewpoint which focuses on how and when a machine functions in an optimal way.
Whatever particular failure one is considering, proper control and management of it become essential. This entails considering several failure-oriented and failure-driven areas such as Reliability, Availability, Maintainability, Safety (RAMS), Risk, Quality control (QC), Fault Detection and Identification (FDI), Security and others. The related analyses present strong connotations of uncertainty and multi-disciplinarity which significantly add to their complexity. For this, appropriate modeling and analysis tools are needed.
The present lecture addresses the above issues and illustrates the modern computational techniques available for their analysis and evaluation. In particular, the following topics will be touched upon:
- Neural Networks techniques, Fuzzy Logic Systems and Monte Carlo methods for real time online monitoring, diagnostics and prognostics of complex components, plants and processes on the basis of data time series
- Monte Carlo simulation, Genetic Algorithms and Fuzzy Logic for the probabilisic and possibilistic modeling, computation and optimization of the availability and reliability of complex systems
- Monte Carlo simulation, Cellular Automata and Statistical Physics for the evaluation of the reliability, security and vulnerability of complex network systems
- Monte Carlo methods and Genetic Algorithms for the evaluation of toxic and/or radioactive contaminant dispersion in artificial or natural, homogeneous or fractured media.
The presentation is intended to give an overview of the engineering issues and of the methods, without entering into the specific technical details. The material presented draws from the research experience of the Laboratorio di Analisi di Segnale ed Analisi di Rischio (LASAR, lasar.cesnef.polimi.it ) of the Dipartmento of Nuclear Engineering of the Politecnico di Milano.