Predictive Maintenance Optimization of High-Speed Train Bearing Under Bayesian Framework

A Bayesian-driven predictive maintenance planning model is proposed for high-speed train bearings. The nonlinear stochastic Wiener process with uncertain drift coefficient is employed to extract the non-monotonic health trends from real-time vibration signals. The degenerate parameters are estimated through the integration of offline estimation via Maximum likelihood estimation (MLE) and online updating under Bayesian framework. Then, the online distribution of asset lifetime is predicted by dynamic spatio-temporal scale transformation, which further supports dynamic sequential replacement decisions. The operational cost in terms of replacement decision variables is optimized, whose value is dynamically updated under Bayesian framework. The availability of the model is demonstrated by a practical case study on health management of high-speed train bearings.