A New Methodology to Deal with the Multi-phase Degradation in Rolling Element Bearing Prognostics

Abstract

Predicting the residual-life (RL) of rolling element bearings (REBs) is an essential procedure of rotating machinery condition monitoring, in industries. Since the bearing fault propagation has a complicated stochastic nature, the multi-phase phenomenon is probable to happen in the degradation process. By occurrence of the multi-phase phenomenon, the degradation process experiences different phases, with distinct failure rate in each phase. Therefore, the mentioned phenomenon leads to large error in the prediction results of single-phase degradation models. The main concern of the present paper is to consider the effect of multi-phase phenomenon on the prediction confidence level of REBs. In order to deal with the unforeseeable effect of multi-phase on the prediction confidence level, several threshold levels are considered at each time step, in contrary to previous works that only consider one distant failure threshold in the prediction. Accordingly, a new introduced belief score function is utilized, to assign the belief value to the estimated RLs corresponding to each threshold, at each time step. This function is constructed based on the probability distribution of multi-phase occurrence, up to the corresponding threshold level. Finally, the belief values are utilized to modify the confidence levels of single-phase predictions. The proposed methodology has been applied to a set of bearing accelerated run to failure test. The results show that the presented method can comprehend the occurrence of multi-phase in the degradation signal and also modify the confidence level, properly. Thus, it can be implemented as an effective tool in taking optimum maintenance decisions.