Design and time parameters identification for non-Markovian Petri net models: application to reliability analysis

The reliability analysis for dynamic systems is often based on timed stochastic Petri net models. The contribution of this paper concerns the design and identification of non-Markovian timed stochastic Petri net models. In particular, stochastic Petri nets that combine firing periods with normal and exponential probability density functions (pdfs) are considered. A numerical method is proposed for the identification of the parameters that characterize the pdfs of the firing periods. This method is based on a statistical analysis of the collected event sequences that are recorded by supervision systems. The method is applied for the reliability analysis of a single server with variable operation periods.