Application of Deep Learning in Rolling Bearing Fault Diagnostics

Abstract

　This paper presents a diagnostic method for rolling bearings using deep learning instead of conventional rule-based diagnostic methods which requires human and time costs. However, deep learning has two major problems. One is difficulty in acquiring large amount of data under the operation with damaged rolling bearings for training, and the other is difficulty in interpreting the diagnostic results. As a solution to these problems, this paper introduces transfer learning and a method to visualize the input data points contributing to the diagnostic results.