Remaining Useful Life Prediction Of Machinery Subjected To Two-Phase Degradation Process
2018 IEEE INTERNATIONAL CONFERENCE ON PROGNOSTICS AND HEALTH MANAGEMENT (ICPHM)(2018)
摘要
Remaining useful life (RUL) prediction of machinery is a major task in condition-based maintenance, which is able to provide crucial guidance for preventive maintenance. To guarantee the accuracy for the RUL prediction of machinery subjected to two-phase degradation process, the interactive multiple model (IMM) filtering technique has been used because of its capability in estimating the state and the phase dynamically. However, there are two limitations in the IMM based methods. 1) A crucial parameter of the IMM, i.e., the transition probabilities matrix (TPM) of the IMM, is set manually in existing IMM based methods, which often leads to inaccurate state estimation results. 2) The phase estimation is derived as one-step filtering results without considering the overall evolution of the degradation trend, which is unable to describe the phase transition, thus causing inaccurate phase estimation results. To tackle these two limitations, an improved RUL prediction method is proposed in this paper for machinery subjected to two-phase degradation process. In the proposed method, a two-phase degradation model is constructed to describe the degradation process. A nonlinear IMM technique, i.e., the interactive multiple model particle filter (IMMPF) is utilized for the state and the phase estimation, where the TPM is estimated using the numerical-integration TPM estimation (NI-TPME) algorithm instead of being pre-specified manually. The transition point (TP) distribution is adopted to reflect the overall evolution of the degradation trend, and is further used to modify the phase estimation from the IMMPF. Finally, the RUL is predicted by Monte Carlo simulation. The effectiveness of the proposed method is demonstrated by a numerical simulation study.
更多查看译文
关键词
remaining useful life prediction, two-phase degradation process, interactive multiple model particle filter
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络