Novel tool wear prediction method based on multimodal information fusion and deep subdomain adaptation

Reliable tool wear prediction is of great importance for the improvement of machining quality and efficiency. With the advent of the big data era, data-driven tool wear prediction methods have proven to be highly effective. However, these methods have also revealed issues such as shallow feature extraction and limited generalization of models across different machining processes. The objective of this research is to propose a tool wear prediction method based on multimodal information fusion and deep subdomain adaptation to solve the existing problems. First, the original one-dimensional time-series tool monitoring signals are encoded into images to generate a two-dimensional image dataset. Secondly, a two-channel prediction model combining Residual Network and Gated Recurrent Unit is constructed to extract features from the two-dimensional image signals and the one-dimensional time-series signals respectively, and the extracted spatial and temporal features are fused. Thirdly, the dataset is divided into subdomains based on wear values, and the generalization ability of the model is improved by reducing the feature differences between source and target domains through the subdomain adaptive method, thus achieving the prediction of the tool wear values under different situations. Finally, through the validation on two milling wear datasets and comparison with the prediction results of other models, the experimental results prove the accuracy and good generalization of the method, which can provide a reference to improve the machining quality and efficiency, and is suitable for practical industrial application scenarios.