Deep Diffusion for Training Chatter Detection Systems for Machine Tools

In machine tool operations, chatter can cause machining errors and reduce a tool’s lifespan and the production quality, and artificial intelligence (AI) is commonly used for predicting vibrations and chatter. In this study, 1-D chatter data were preprocessed through time delay mapping (TDM) to generate 2-D images. Diffusion models were then trained on the images and used to generate additional images to augment a small training dataset. The dataset was further used to train a convolutional neural network (CNN) model for detecting chatter. The data augmentation method can reduce data imbalance and improve a model’s feature recognition capability, enhancing its accuracy on small datasets. An optimization algorithm was used in this study to obtain the CNN hyperparameters that maximized the model’s robustness and accuracy. Overall, the TDM, model-based augmentation, and optimization methods were discovered to effectively enhance the chatter detection accuracy of the final model. The results of image generation can be used to construct a large database, providing more training data for models and reducing time and labor costs. In addition, because the images retain critical features, the model’s complexity can be substantially reduced.