Degradation modeling and RUL prediction using Wiener process subject to multiple change points and unit heterogeneity

• Published in: Reliability engineering & system safety Vol. 176; pp. 113 - 124
• Main Authors: Wen, Yuxin, Wu, Jianguo, Das, Devashish, Tseng, Tzu-Liang(Bill)
• Format: Journal Article
• Language: English
• Published: Barking Elsevier Ltd 01.08.2018; Elsevier BV
• Subjects: Bayesian analysis; Computer simulation; Condition monitoring; Degradation; Degradation modeling; Estimating techniques; Heterogeneity; Internet; Life prediction; Mathematical models; Medical prognosis; Model updating; Modelling; Monte Carlo simulation; Multiple change-point model; Reliability engineering; Remaining useful life prediction; Signal processing; Wiener process; Multiple change-point model; Remaining useful life prediction; Wiener process; Degradation modeling
• ISSN: 0951-8320; 1879-0836
• DOI: 10.1016/j.ress.2018.04.005

• A multiple change-point Wiener process based prognostic framework is proposed.• To consider unit-to-unit heterogeneity, a fully Bayesian approach is developed.• An empirical two-stage process is proposed for model estimation.• Online individual model updating is achieved through an exact recursive algorithm.• The effectiveness is demonstrated through simulation and real case studies. Degradation modeling is critical for health condition monitoring and remaining useful life prediction (RUL). The prognostic accuracy highly depends on the capability of modeling the evolution of degradation signals. In many practical applications, however, the degradation signals show multiple phases, where the conventional degradation models are often inadequate. To better characterize the degradation signals of multiple-phase characteristics, we propose a multiple change-point Wiener process as a degradation model. To take into account the between-unit heterogeneity, a fully Bayesian approach is developed where all model parameters are assumed random. At the offline stage, an empirical two-stage process is proposed for model estimation, and a cross-validation approach is adopted for model selection. At the online stage, an exact recursive model updating algorithm is developed for online individual model estimation, and an effective Monte Carlo simulation approach is proposed for RUL prediction. The effectiveness of the proposed method is demonstrated through thorough simulation studies and real case study.