Accurate Prediction of RUL under Uncertainty Conditions: Application to the Traction System of a High-speed Train

Abstract Traction systems of high-speed trains suffer with many uncertainties in their degraded process. The uncertainties mainly include the inherent uncertainty associated with the progression of the degradation over time and the inevitable uncertainty caused by noise and disturbance. Handling uncertainties is important to improve the prediction accuracy of remaining useful life (RUL) in the degraded process from fault to failure. This study takes uncertainties into consideration via the relevant vector machine (RVM) approach in order to achieve a good accuracy. Firstly, a target vector is derived by first hitting time (FHT) to characterize the RUL. This is fulfilled off-line based on historical sample data. Then, a stochastic model is established by RVM under Bayesian framework. Further, parameters are updated using the expectation-maximization (EM) algorithm for an optimized RVM model. And then, the on-line RUL prediction is implemented with the updated RVM model. Finally, the proposed method is demonstrated by two real case studies of traction faults occurred in the high-speed train. The prediction results show the effectiveness of the proposed method.