Tool wear state recognition study based on an MTF and a vision transformer with a Kolmogorov-Arnold network

Abstract

In manufacturing, tool wear directly affects the quality, production cost and efficiency of machine tools. Accurately monitoring tool wear states is critical. However, 1D tool wear signal features are nonobvious. The existing models entail feature extraction processes that have difficulty considering long-sequence tool wear information in parallel and low robustness. Herein, an adaptive vision transformer (ViT) incorporating a Kolmogorov-Arnold network (KAN) based on a Markov transition field (MTF), MTF-AViTK, is proposed. The network can perform parallel processing and addressing long-term dependencies, better identifying tool wear states from cutting signals. A signal is converted into a 2D image via 2D time series encoding. The ViT performs token patch position encoding on 2D images and self-attention-based tool wear feature extraction in its encoder blocks. An adaptive multilayer perceptron (AdaptMLP) module is introduced to fine-tune the feature extraction process. The extracted features are mapped by the KAN to identify the tool wear state. PHM2010 provides two datasets for validating the performance and robustness of MTF-AViTK model. The highest model recognition accuracies are 95.76 % and 95.71 %. The results show that this model has better robustness and generalizability than the competing approaches. Finally, ablation experiments verify that each module has a positive effect on model performance. This method can accurately identify tool wear states based on machining data, enabling more accurate tool change decisions.