Reliability analysis and prediction of the failure process of some traction converters

This paper studies the failure process of the traction converters mounted on some trains produced by AnsaldoBreda SpA and operating in the district of the city of Naples. Each train is equipped with three traction converters and 23 new trains were monitored from their entry in service over a time period of about four years. A preliminary analysis of the observed failure data, based on a non-parametric estimation procedure, showed an overall decreasing trend of the failure intensity during the monitoring period, as it is often observed during the early life of complex systems undergoing minimal repair. As such, the failure process of converters is assumed to be a non-homogeneous Poisson process (NHPP) over the covered mileage. Then, NHPP with different decreasing intensity functions are considered, and the process that better fit the observed data, namely the log-linear process, is chosen to analyse the observed failure process. Maximum likelihood estimate of the process parameters is then obtained, from which several reliability characteristics are estimated under the conservative hypothesis that the failure intensity remains constant beyond the end of the observation period. Finally, the probability distribution of the annual mileage of the train is derived, that allows the reliability characteristics to be estimated in terms of the calendar time, too.