Bearing fault diagnosis method using the joint feature extraction of Transformer and ResNet

Abstract The failure of rotating machinery can be prevented and eliminated by a regular diagnosis of bearings. In deep learning (DL) models of bearing fault diagnosis driven by big data, problems, such as data acquisition difficulties, data distribution imbalance, and high noise, often exist in the samples. This study proposes a novel bearing fault diagnosis method using the joint feature extraction of Transformer and residual neural network (ResNet) coupled with transfer learning (TL) strategy to overcome the aforementioned issues. First, the data are transmitted to the Transformer encoder and ResNet architecture, respectively, where the input obtained by the encoder must separate features and word embedding via a one-dimensional convolutional layer. Next, the feature sequences mined using encoder and ResNet are connected and classified. Moreover, the TL strategy with model fine-tuning is exploited to reduce the training difficulty of the proposed method in new tasks. Experiments on two bearing fault datasets demonstrate that the proposed method can effectively combine the characteristics of both architectures. Moreover, the prediction accuracy outperforms traditional DL networks in high-noise environments.