A two-stage Gaussian process regression model for remaining useful prediction of bearings

Abstract

Bearing is one of the most important supporting components in mechanical equipment and its health status has a significant impact on the overall performance of equipment. The remaining useful life (RUL) prediction of bearings is critical in adopting a condition-based maintenance strategy to ensure reliable equipment operation. To accurately predict the RUL of bearings, this paper proposes a two-stage Gaussian process regression (GPR) model, which combines the flexibility of the Gaussian process and the physical mechanism of the Wiener process. Compared with the conventional GPR model, the proposed model can reasonably adapt to the statistical characteristics of bearings degradation and provide more stable predictions. In addition, the paper proposes a new degradation detection approach based on the Euclidean distance to distinguish the two stages of the bearing service life cycle, which considers the global characteristics of bearing degradation and can accurately detect the beginning point of bearing degradation. The experimental results show that the proposed two-stage GPR model can help to improve the precision and accuracy of degradation path tracking and RUL prediction.