A bearing fault diagnosis method based on SE-TCN-SVM

With the intelligent development of mechanical equipment, bearing fault diagnosis is facing more complex and changeable challenges. Traditional neural networks suffer from redundant features, limited feature recognition, and poor classification performance. To solve these problems, this paper proposes a SE-TCN-SVM bearing fault diagnosis model that combines a squeeze-and-excitation (SE) attention mechanism, a temporal convolutional network (TCN), and a support vector machine (SVM). By introducing the squeeze-and-excitation module, the model can adaptively enhance the weights of key channel features and optimize the modeling ability of TCN for the long-term dependence of vibration signals. The SVM classifier is used instead of Softmax to improve the robustness of sample classification by using its structural risk minimization characteristics. Experiments on bearing datasets from Case Western Reserve University and Jiangnan University show that the classification accuracy of SE-TCN-SVM reaches 98.92% and 96.88%, respectively. Compared with other benchmark models, it achieves better classification performance, faster training efficiency, and is suitable for different bearing data. This method enhances feature selection through the SE attention mechanism and improves generalization performance by combining SVM classifier. Our method provides a highly accurate, efficient, and adaptable solution for bearing fault diagnosis under complex working conditions.