A SENet-TSCNN model developed for fault diagnosis considering squeeze-excitation networks and two-stream feature fusion

Abstract

The increase in the number of channels for extracting bearing fault features can to some extent enhance diagnostic performance. Therefore, this article proposes a SENet (Squeeze and Excitation Network) - TSCNN (Two Flow Convolutional Neural Network) model with high accuracy and generalization characteristics for fault diagnosis of rolling bearings. Firstly, use convolutional pooling layers to construct a basic diagnostic model framework. Secondly, due to the unsatisfactory performance of feature extraction solely on one-dimensional frequency domain signals or two-dimensional time-frequency signals, there may be misjudgments. Therefore, a dual stream convolutional model is integrated to process both one-dimensional and two-dimensional data. Fast Fourier Transform (FFT) is used to process one-dimensional frequency domain data, and Continuous Wavelet Transform (CWT) is used to process two-dimensional time-frequency maps. Once again, integrating the SENet module into the dual stream diagnostic model, the addition of attention mechanism can enable the model to better understand key features of input data. Finally, the data obtained from the processing of two channels is fused and classified in the Softmax layer. This article uses the rolling bearing fault standard data from Case Western Reserve University (CWRU) and the American Society for Mechanical Fault Prevention Technology (MFPT), and verifies through multiple controlled experiments that the model established in this article has high accuracy and good generalization characteristics.