Remaining life prediction of rolling bearings based on parallel feature extraction and deep learning

Rolling bearings are critical components in rotating machinery and play a vital role in industrial production. However, due to complex working conditions and high operational loads, the risk of bearing failure increases significantly. Accurately predicting the remaining useful life (RUL) of rolling bearings is essential for enhancing the reliability and safety of industrial equipment. To address the challenges in accurately determination of the First Prediction Time (FPT) under weak early-stage faults and the insufficient utilization of temporal information in existing methods, this study proposes the following improvements: First, the data preprocessing procedure is optimized by introducing a frequency-energy ratio-based method for FPT determination, enabling precise identification of incipient weak faults. Second, a synthetic time-frequency graph is innovatively constructed to comprehensively capture the evolutionary trends of time-frequency features. Finally, a parallel deep learning architecture is designed to separately extract temporal and feature information from the bearing degradation process, followed by deep fusion for accurate RUL prediction. Experimental validation conducted on the XJTU-SY bearing dataset demonstrates that the proposed parallel model significantly enhances the accuracy of life prediction.