Detection of performance deterioration of high-speed train wheels based on Bayesian dynamic model

This paper presents a novel technique, in the context of Bayesian dynamic linear model (BDLM) and Bayesian forecasting, for detecting the performance deterioration of high-speed train wheels using online monitoring data of strain acquired from in-service train bogies. The BDLM is a tool for time series analysis and Bayesian forecasting enables to calculate one-step ahead forecast distribution. The change detection is carried out by checking the current observation against the current model (forecast distribution generated by the BDLM for current instant) as well as against an alternative model (whose mean value is shifted by a prescribed offset). The detection rule is that if the alternative model better fits the actual observation, a potential change is alarmed. To further determine whether the current observation is an outlier or the beginning of a change, a specific logic is developed by introducing the Bayes factors and cumulative Bayes factors. The proposed method is demonstrated by using the monitoring data acquired from an in-service high-speed train under different wheel quality conditions.