Research on Fault Diagnosis of Rolling Bearings Based on CNTCSA-KAN

Aiming at the problems of difficult feature extraction and low classification accuracy of bearing faults, this paper's proposes a bearing fault diagnosis method based on CNTCSA-KAN network. The method deeply explores the temporal features of bearing fault signals by fusing the local feature extraction capability of convolutional neural network (CNN) with the advantage of temporal convolutional network (TCN) in capturing long time series dependencies. Then, a self-attention mechanism is introduced to enable the model to autonomously pay attention to and adjust the weights among the elements, thus enhancing the characterization ability of key features. Finally, the feature vectors are inputted into the KAN network to improve the model's ability to fit nonlinear functions, while the data features are downsized and visualized using T-SNE to further verify the reliability of the results. The results show that the average accuracy of the model reaches 99.010/0, which is 6.9%, 3.81^%, and 4.87% higher than that of the traditional CNN, CNN-TCN, and CNN-BiLSTM models, respectively, and achieves a more accurate fault identification. This result fully verifies the stability and effectiveness of the method on the rolling bearing fault diagnosis task, and at the same time reflects the advantages of the model such as small network parameters, high accuracy, and fast convergence, which provides important engineering value for the practical application of rolling bearing fault diagnosis.