A Deep Transfer Learning Method for Rolling Bearing Fault Diagnosis Based on ResNeXt-50

Abstract

Rolling bearings are key components in the bogie of locomotive vehicles, their condition monitoring and fault diagnosis are widely concerned. Deep learning, as one of the most widely used methods in rolling bearing fault diagnosis, has achieved proper recognition and promotion. However, compared with other fields (such as computer vision, etc.), the depth of neural network model used for the task of rolling bearing fault diagnosis is relatively shallow due to the number of labeled samples, which limits its final prediction accuracy. In this paper, a data-driven fault diagnosis method based on Resnext-50 is proposed and applied to multiclassification tasks and condition transfer diagnosis tasks of bearing faults. First, a signal-image conversion method based on wavelet transform is utilized to transform the original vibration signals of bearings into RGB format. Second, all the layers of the pre-trained ResNeXt-50 model are further trained directly by the generated target data set. Finally, the trained ResNeXt-50 model was applied to multiclassification tasks and condition transfer diagnosis tasks of bearing faults. The method was verified on the bearing data set from Case Western Reserve University, and the results show that the accuracy rate of the proposed method is close to 100%, which has good application prospects.