Bearing fault diagnosis Based on 1-D DenseNet-LSTM

Rolling bearing, as a vulnerable part of rotating machinery, acts an essential role in mechanical equipment. Once fails, it will cause huge economic losses, even lead to major safety accidents which threaten human life. Considering that, it's of great significance for the bearing fault diagnosis. Aiming at the issue that the accuracy of the shallow diagnosis model is limited in big data set and generalization ability is weak, an end-to-end fault diagnosis method based on one-dimensional densely connected convolutional neural network and long short-term memory neural network (1-D DenseNet-LSTM) is proposed. First, the bearing vibration signal after preprocessing is input into the fault diagnosis model. Then, Convolution layer can effectively extract the local features of bearing vibration signal, in which, dense connection mechanism will fuse the low-level features and high-level features. After that, LSTM layer pays more attention to temporal information and extracts deep features. Finally, the fault feature information is fully extracted and input to the classification layer and output the fault type. Experiments showed that the fault diagnosis model based on 1-D DenseNet-LSTM can fully extract bearing fault feature information, and classify different fault types effectively. The accuracy of bearing fault diagnosis is more than 99%, which has higher accuracy of diagnosis compared with the depth network model.