Gearbox Fault Diagnosis Method Based on Attention Mechanism WDNN and Gated Recurrent Neural Networks

Abstract In this paper, the traditional diagnostic methods cannot make full use of the temporal nature of the data, and the characteristics of the traditional convolutional neural network CNN are not sensitive to the temporal information feature extraction in the vibration signals, the proposed method based on the Wide and Deep Convolutional Neural Network (WDCNN) and the Gated Recurrent Neural Network(Gated Recurrent Neural Network, GRU) and the WDCNN-SENet-DGRU model combined with the SENet channel attention mechanism. The WDCNN improves the model's ability to extract spatial features from sequence data by widening the kernel size of the first layer of the traditional convolutional neural network; the SENet channel attention mechanism improves the extraction of spatial features from sequence data through the feature recalibration mechanism; and the SENet channel attention mechanism improves the extraction of spatial features from sequence data through the feature recalibration mechanism. The mechanism further improves the feature expression ability of the WDCNN network by increasing the weights of key features through the feature recalibration mechanism; the long-term dependencies in the time series data can be better captured using the GRU neural network and compared with the long and short-term memory network LSTM, the GRU has fewer parameters, which makes it easier to train. The experimental results show that the method can reach 100% accuracy on the training set and 99.59% on the test set, which has better fault diagnosis capability than other models.