Bearing Fault Diagnosis Method Based on Parallel Feature Extraction
Aiming at the improvement of the robustness and generalization performance of existing bearing fault diag-nosis methods in industrial noise environment,a model of parallel features extraction by convolutional autoencoder-long short-term memory (CAE-LSTM) was proposed.This method used the local spatial features frozen by the encoder to fuse the temporal correlation features extracted by long short-term memory-1-dimensional convolutional neural network (LSTM-1DCNN) to complete the model training.In the process of feature fusion,an effective channel attention (ECA) mechanism was introduced to complete the weight distribution of features,and the full extraction of global feature information was real-ized.Finally,the fault diagnosis results of the bearing were output through the Softmax function.The experimental verifica-tion was carried out on the self-test and public bearing data sets.The comparative experimental results show that:the parallel feature extraction model proposed in this study has excellent bearing fault diagnosis performance,as well as good noise ro-bustness and generalization ability.
万方数据
维普
