Bearing Fault Diagnosis Based on Deep Separable Convolutional Neural Network
In real industrial environment,it is necessary to make fast and accurate diagnosis of equipment faults.Requirements of low latency and high accuracy make traditional Convolutional Neural Network(CNN)severely restricted in fault diagnosis process.To solve this problem,a bearing fault diagnosis model based on depth one-dimensional separable convolutional neural network is proposed.First,construct a backbone convolutional neural network that can directly process one-dimensional vibration signals.Then,a one-dimen-sional Separable Convolutional Neural Network(SCNN)is constructed by performing separable processing on the convolutional layer in the backbone CNN,which realizes separation of channels and regions in the convolution process,and reduces parameters required in the convolution calculation process,reducing the calculation delay thereby.Finally,in order to ensure the accuracy of diagnosis on the basis of reducing calculation delay,a residual layer is added to the constructed SCNN,and the accuracy of the convolution process is guaran-teed through residual connections.In order to compare the effectiveness of the constructed model,traditional VGG16 and ResNet50 net-works were reconstructed one-dimensionally for verification,and the classified CWRU fault bearing data were analyzed.Results show that the model can improve fault diagnosis efficiency while ensuring recognition accuracy.
万方数据
维普
