As a critical element in rotating machines, remaining useful life (RUL) prediction of rolling bearings plays an essential role in realizing predictive and preventative machine maintenance in modern manufacturing. The physics of defect (e.g. spall) initiation and propagation describes bearing's service life as generally divided into three stages: normal operation, defect initiation, and accelerated performance degradation. The transition among the stages are embedded in the variations of monitored data, e.g., vibration. This paper presents a multi-mode particle filter (MMPF) that is aimed to: 1) automatically detect the transition among the three life stages; and 2) accurately characterize bearing performance degradation by integrating physical models with stochastic modeling method. In MMPF, a set of linear and non-linear modes (also called degradation functions) are first defined according to the physical/empirical knowledge as well as statistical analysis of the measured data (e.g. vibration). These modes are subsequently refined during the particle filtering (PF)-based bearing performance tracking process. Each mode corresponds to an individual performance scenario. A finite-state Markov chain switches among these modes, reflecting the transition between the service life stages. Case studies performed on two run-to-failure experiments indicate that the developed technique is effective in tracking the evolution of bearing performance degradation and predicting the remaining useful life of rolling bearings.
展开▼
