Fault diagnosis of rolling bearing under adaptive attention LSTM-ResNet

The complexity and nonlinearity of rolling bearing signals pose challenges to feature extraction and fault classification. To solve these problems, in this paper, an adaptive attention long short-term memory residual network (LSTM-ResNet) is proposed. The bidirectional LSTM group feature extraction model is designed to obtain the rolling bearing features under complex operating conditions. Then, an adaptive attention LSTM-ResNet is proposed to complete feature learning and adjust the weights of key features in the model. Finally, global average pooling (GAP) method is used to combine softmax model to alleviate overfitting and merge the model to complete fault classification. The rolling bearing fault classification was completed in the data set. The experimental results show that the proposed method has a higher detection accuracy than SVD-ResNet method and wide convolutional model, and can achieve a higher detection accuracy under the condition of less marked samples and noise.with higher accuracy and stronger robustness.