Multiple model particle filtering for bearing life prognosis

For bearing remaining life prognosis, past research has investigated deterministic material fatigue crack growth models such as Paris law and Newman model. Due to the inherent stochastic nature of defect propagation and varying operating conditions, the accuracy of such models has shown to be limited. This paper addresses this challenge by presenting a stochastic modeling approach, based on interacting multiple models and particle filter. Experiments were conducted on a customized bearing test rig to demonstrate the effectiveness of the developed method. Comparison between the developed method and the traditional particle filter has shown that the developed method improves the accuracy in bearing remaining life prediction.