A Deep Learning Approach for Failure Prognostics of Rolling Element Bearings

Abstract

Remaining useful life (RUL) prediction of rotating machine components is essential to enabling predictive maintenance of industrial and agricultural machinery. This paper presents a novel deep learning approach for failure prognostics of rolling element bearings. The proposed approach has three unique features: (1) it employs a new data augmentation technique to improve the accuracy and robustness of RUL prediction in cases where a deep learning model only has access to a small amount of training data; (2) it incorporates a robust feature learning strategy that integrates a physics-based feature extraction process with a data-driven process; and (3) it implements a new similarity-based approach for effectively capturing the true degradation trend of each individual bearing unit. A practical case study involving run-to-failure experiments of rolling element bearings on the PRONOSTIA platform is provided to assess the performance of the proposed approach. Results from the case study show the proposed deep learning approach produced higher accuracy in RUL prediction than an existing machine learning approach.