Cutting chatter is a common phenomenon during metal cutting operations, which can severely impair the surface integrity of machined components and reduce overall production efficiency. Therefore, the precise detection of chatter during the cutting process is of critical importance. In recent years, the widespread integration of artificial intelligence (AI) techniques into chatter detection has led to satisfac-tory results. Although deep learning methods exhibit excellent capabilities in feature learning and classi-fication, their generalization and accuracy are highly dependent on data labeling quality and training procedures. To address these issues, this study introduces a chatter detection approach for thin-walled parts milling based on the hybrid deep convolutional neural network (HDCNN) and names it as Chatter-CNN. The Chatter-CNN model combines two Inception-Chatter modules and two Squeeze-and-Excitation ResNet blocks (SR-blocks). The Inception-Chatter module is capable of automatically extract-ing multi-scale features from cutting force signals. The SR-block mitigates the risk of gradient vanishing, accelerates network convergence, and adaptively assigns weights to different feature channels. The proposed method is benchmarked against two convolutional neural networks (CNNs) that have been extensively applied in the field of chatter detection over the past decade. A series of milling experiments on wedge-shaped thin-walled workpieces are conducted under various cutting conditions. The experi-mental results indicate that the proposed Chatter-CNN model achieved classification accuracies of 100% on the validation set and 97.5% on the test set, outperforming existing methods.
Science Journal 