A Comparison of Deep Transfer Learning Methods on Bearing Fault Detection

Abstract

In rotating machinery, bearings are widely used as universal components. Bearings are placed in critical positions, therefore, in predictive maintenance, it is crucial to diagnose bearing faults accurately and in a timely manner. In this paper, three diverse pre-trained networks on bearing fault diagnosis are discussed. A generic intelligent bearing fault diagnosis system based on AlexNet, GoogLeNet and ResNet-50 with transfer learning is proposed to distinguish and classify different bearing faults. Three bearing faults at all various loads and speeds selected from the Case Western Reserve University (CWRU) bearing dataset were converted to time-frequency images, in order to improve the performance of the proposed networks. Results showed that when compared to previous methods, the proposed method achieved outstanding execution, with overall classification training accuracy of 100%, validation accuracy of 99.27%.