Hybrid Prediction Model of the Remaining Useful Life of Bearing Based on Long Short-Term Memory Network

Rolling bearings remaining useful life (RUL) prediction is a key research issue in the field of prediction and health management (PHM). Thus, accurate prediction is an important guarantee for maintaining the stability of its applied systems. However, the current research has two main shortcomings. The first one is when the bearing contains fault, the difference in faults causes instability of the prediction model. The second point is that the subhealth state of the bearing life can be easily ignored using the health index (HI). A new framework is proposed for RUL prediction in large bearing samples. First, aiming at bearing containing fault, using signal decomposition and feature extraction, and then combined with strategies to achieve the final forecasts. Second, when the bearing is under normal degradation, Mahalanobis distance is used to convert to modified health index (MHI), which captures the beginning of the subhealth state