Physics-informed meta learning for machining tool wear prediction

Tool wear prediction plays an important role in ensuring the reliability of machining operation due to their wideranging application in smart manufacturing. Massive effort has been devoted into exploring the methods of tool wear prediction. However, it remains a challenge to improve the accuracy of tool wear prediction under varying tool wear rates. To address this issue, this paper presents a new physics-informed meta-learning framework for tool wear prediction under varying wear rates. First, a physics-informed data-driven modeling strategy is proposed by employing the empirical equations' parameters to improve the interpretability of the modeling and optimization of the data-driven models. The piecewise fitting is adopted to ensure the accuracy of the parameters. Second, the physics-informed model input is investigated to help the data-driven models explore the hidden information about the tool wear under varying tool wear rates. Third, the physics-informed loss term is presented to constrain the optimization of the meta-learning model. An experimental study on a milling machine is performed to validate the effectiveness of the presented method.

Automated summary

This paper introduces a new physics-informed meta-learning framework for tool wear prediction under varying wear rates, improving prediction accuracy by enhancing modeling strategy and constraining optimization process with a loss term informed by physics.