1D CNN-Based Transfer Learning Model for Bearing Fault Diagnosis Under Variable Working Conditions

Abstract

Classical machine learning approaches have made remarkable contributions to the field of data-driven techniques for bearing fault diagnosis. However, these algorithms mainly depend on distinct features, making the application of such techniques tedious in real-time scenarios. Under variable working conditions (i.e., various fault severities), the acquired signals contain variations in the signal amplitude values. Therefore, the extraction of reliable features from the signals under such conditions is important because it could discriminate the health conditions of the bearings. In this paper, a transfer learning approach based on a 1D convolutional neural network (CNN) and frequency domain analysis of the vibration signals is presented to solve the problem. Transfer learning enables the developed model to utilize information obtained under a given working condition to diagnose faults under other working conditions. The proposed approach has a classification accuracy of 99.67% when tested with the data acquired from the bearings with various fault severities. We also observe that a frequency spectrum enhances the performance of the transfer learning-based fault diagnosis model.