Research on online monitoring of chatter based on continuous wavelet transform and convolutional neural network–vision transformer (CNN-ViT)

Abstract. In order to solve the problem of chatter identification in the machining process of thin-walled parts, this paper proposes an online chatter-monitoring method based on continuous wavelet transform and convolutional neural network–vision transformer (CNN-ViT). Firstly, the continuous wavelet transform is used to convert the one-dimensional time domain force signal into a two-dimensional time–frequency image. Secondly, by using the CNN model's capabilities of adaptive feature learning and hierarchical feature extraction, the local features in the input image are learned and captured through a series of convolutional layers, pooling layers, and other structures, which effectively reduces the complexity of feature extraction. Finally, the self-attention mechanism of the ViT model is used to integrate and model the global feature map of the input feature map, the global contextual information of the captured image is calculated by calculating the global contextual information of the captured image, and then the classification result is outputted according to the integrated global feature. In this paper, by combining the CNN model with the ViT model, the chatter characteristics are comprehensively captured from multiple angles and levels, which improves the robustness of the model, thereby improving the accuracy and efficiency of online monitoring and providing an innovative and effective technical means for online monitoring of chatter.