Tool wear prediction based on a fusion model of data-driven and physical models in the milling process

Precision tool wear prediction in the milling process is crucial in enhancing product quality and machining efficiency. Both data-driven models and physical models are vital for indirect tool wear prediction. However, data-driven models rely on appropriate model structures and extensive datasets to achieve high prediction accuracy; physical models face challenges when adapting to complex cutting conditions, in which case accurate modeling is difficult. Aiming at employing the advantages of both methods for accurate tool wear prediction, a fusion model is developed by integrating both the data-driven model and the physical model by constructing an indirect prediction layer and a parameter constraint layer. The indirect prediction layer incorporates domain knowledge of tool wear, while the parameter constraint layer utilizes priori knowledge from accumulated data. Validation results show that with the introduction of domain knowledge and prior knowledge as constraints, the range and shape of the fusion model’s prediction result confidence intervals are effectively constrained to more reasonable zones, the area of the confidence intervals is reduced by 73.7%, and the average prediction accuracy of the fusion model is improved by 11.5%.