The optimized local sparse parallel multi-channel deep convolutional neural network-LSTM and the application in bearing fault diagnosis under noise and variable working condition

Since the actual operation of the bearing inevitably exists in both noise and variable working conditions, most of the traditional networks can only deal with them alone, and the fault identification result will be significantly reduced under such complex conditions. Therefore, Parallel Multichannel Deep Convolution Neural Network (PMDCNN) and Long Short-Term Memory (LSTM) are proposed as PMDCNN-LSTM model to enable better performance. And a local sparse structure is used to greatly reduce the number of model parameters, the Exponential Linear Unit (ELU) activation function is used to further improve accuracy and stability. The results show that the proposed method is strongly resistant to noise and variable working conditions, as verified by the Case Western Reserve University (CWRU) bearing dataset and bearing fault simulation experimental platform dataset. The comparison with other model shows that the proposed model has good applicability, strong stability and high accuracy of bearing fault identification.