Dynamic bayesian networks modelling maintenance strategies: Prevention of broken rails

In this paper, an assistant tool for the maintenance of rails in a metro context for normal steel wheeled trains is proposed. The theory of Dynamic Bayesian Networks offers an interesting frame to solve this specific problem. We present a modular modelling of the current rails’ diagnostic process. It simulates the rail degradation, the behavior of the various actors involved in the defect detection, but also the maintenance action decisions. This model provides indicators, such as the non detection rate or the number of false alarms, helpful for the determination of optimal maintenance parameters.