Deep Learning Algorithms for Bearing Fault Diagnostics - A Review

Abstract

This paper presents a comprehensive review on applying various deep learning algorithms to bearing fault diagnostics. Over the last ten years, the emergence and revolution of deep learning (DL) methods have sparked great interests in both industry and academia. Some of the most noticeable advantages of DL based models over conventional physics based models or heuristic based methods are the automatic fault feature extraction and the improved classifier performance. In addition, a thorough and intuitive comparison study is presented summarizing the specific DL algorithm structure and its corresponding classifier accuracy for a number of papers utilizing the same Case Western Reserve University (CWRU) bearing data set. Finally, to facilitate the transition on applying various DL algorithms to bearing fault diagnostics, detailed recommendations and suggestions are provided for specific application conditions such as the setup environment, the data size, and the number of sensors and sensor types. Future research directions to further enhance the performance of DL algorithms on healthy monitoring are also presented.