A Bayesian Prognosis Framework for Rolling Bearings Based on Total Harmonic Distortion Health Indicator and Nonlinear Wiener Process

Abstract

Rolling bearings are one of the key components of rotating machinery. Accurate prediction of the remaining useful life (RUL) of bearings can reduce maintenance costs, as well as increase the reliability of the mechanical system and prevent catastrophic accidents. Accurate and reliable estimation of the RUL of bearings requires a suitable health indicator (HI). Therefore, a novel health indicator called the Total Harmonic Distortion (THD) is constructed using advanced frequency domain analysis to accurately reflect the health status of the bearing. The results confirm that the THD health indicator has excellent monotonicity, robustness, and trendability. In this work, the Wiener process with different drifts is used to predict the RUL. Also, the parameters of the Wiener model are estimated online using the Bayesian theory method. Then, the model is validated using an accelerated experimental test bench. The results showed that the use of the Wiener process with the power law model (PLM) has higher accuracy and stability than other Wiener models.