Bearing fault diagnosis based on ID CNN attention gated recurrent network and transfer learning

Here, to solve the problem of bearing fault diagnosis model' s accuracy and generalization ability dropping due to obtaining bearing fault data being difficult in practical applications, a bearing fault diagnosis method based on one-dimensional convolutional neutral network (1DCNN), attention and gated recurrent unit(lDCNN-Attention-GRU) as well as transfer learning was proposed. Firstly, a fault diagnosis network based on 1DCNN, GRU and Attention mechanism was constructed to solve the problem of feature extraction of traditional fault diagnosis methods depending too much on human experience. Then, transfer learning was introduced to train the above network with sufficient source domain data, freeze the trained network' s infrastructure, finely tune the network top-level structure with a small amount of target domain data, and obtain the target network model. Finally, Softmax function was used for fault classification. The test results showed that the fault diagnosis accuracy of the proposed method under different training sample proportions is higher than those of 1 DCNN-GRU, GRU and support vector machine (SVM): the proposed method can obtain higher fault diagnosis accuracy under variable working conditions and small sample data; when 3% target domain data is used for fine tuning, the fault diagnosis accuracy is higher than 98%. © 2023 Chinese Vibration Engineering Society. All rights reserved.