Adaptive Kernel Auxiliary Particle Filter Method for Degradation State Estimation

Abstract The system degradation processes are usually uncertain due to multi-source variability. Accurate estimation of system degradation states is important for system safety and system health management. In general, degradation processes are indirectly monitored but can be inferred through particle filter (PF) methods, which combine real-time monitoring and degradation models. The degeneration and impoverishment problems of PF methods can reduce the accuracy of the estimation results, especially when the uncertainties in models are large. In this paper, to alleviate the problems, an adaptive kernel auxiliary particle filter method is proposed, which incorporates the kernel density estimation-based resampling strategy to increase the diversity among the resampled particles. An adaptive kernel bandwidth selection method is further developed to improve the estimation results by adaptively assigning appropriate kernel bandwidths to each particle considering its own weight. The effectiveness of the proposed method is verified through a numerical example and a case study on degradation state estimation of fatigue cracks in rotorcraft structures.