Physics-based Gaussian process for the health monitoring for a rolling bearing

Abstract

An improved Gaussian process regression (GPR) is presented to predict the remaining useful life (RUL) of a gyroscope being integral to its prognostics and health management, despite uncertainties in the mean and variance values. In our approach, the degradation model of the gyroscope innovatively serves as a global model in the GPR methodology to capture the actual trends of the RUL. Moreover, the ball bearing whose rolling contact wear is responsible for the drifting in the gyroscope is considered as an essential component for the gyroscope RUL estimation. Employing the GPR method and the physical degradation model, the prognosis for the ball bearing can successfully predicts the defect before it occurs. Compared to other data-driven algorithms, results obtained for a gyroscope in an inertial navigation system confirm that the proposed method can be applied for drift prognostics with significant efficiency.