Use of causal probabilistic networks for the improvement of the maintenance of railway infrastructure

This paper deals with the problem of classification of sequential events that occur one after the other and when the different prior transition probabilities can be approached with the help of a labeled database. Dynamic Bayesian networks (DBN) and input output HMM (IOHMM) are employed to formalize such complex dynamic process. For a compact representation of our system, DBN modeling IO-HMM are used. They allow simulating these processes, taking into account information about time detection. DBN modeling were tested, until the third order, on an experimental problem that concerns the classification of faults on transport system infrastructure.