Bearing Fault Diagnosis Based on Multi-Channel GAF-MTF and Res2Net

To address the problems that convolutional neural network training data cannot fully extract data features and cannot fully utilize the correlation information among data, this paper proposes a composite processing method based on Gramain Angular Fields (GAF), Markov Transition Field (MTF) and combined with a deep residual network (ResNet) to complete fault identification. After enhancing the original timing signal data with sliding windows, the GAF and MTF techniques are used to form a more informative multi-channel image training set. Based on this, ResNet is used to extract features from the processed data and construct a network framework suitable for bearing fault classification. The results show that the composite processing enables the network to learn more complete fault features and has good fault recognition accuracy.