Dynamic Bayesian Network Approach to improve the Preventive Maintenance of Multi-State Systems

This work includes a part of the results of I.W.Soro on performance evaluation of Multi-State Systems (MSS) about the preventive maintenance policy. It was to assess the availability and the rate of production of a multi-state system based on a rate of transitions in the level of β degradation. The formalism of calculation based on Markov chains used and Chapman-Kolmogorov equations induce as many calculations as possible cases of β transition rates to deduce the one that brings the best drift of the availability curves and production rates. Moreover, the representation of multi-state system by a Markov graph quickly becomes dense and difficult to use. In this paper, it will first be presented formalization of the transition process of multi-state system (MSS) by Bayesian Networks (especially compact) and the rules governing promotion from the Markov graph. In a second step, it will be exhibited, the cost function of preventive maintenance and the best method for identifying the β transition rates and thus the best preventive maintenance policy to adopt. The optimization has done by reinforcement learning.