Physics-informed symbolic regression for tool wear and remaining useful life predictions in manufacturing

Prognostics and Health Management (PHM) plays a crucial role in enhancing the reliability and safety of engineering systems. Recently, physics-informed machine learning (PIML) methods have gained significant attention for their ability to incorporate domain-specific knowledge into data-driven models. This paper proposes a novel approach that integrates symbolic regression with recursive modeling to develop a robust framework for PHM of dynamic processes. Our framework was applied to a manufacturing process to build a generic model for tool wear prognostics across various machining scenarios. The proposed method integrates domain knowledge of milling processes under different conditions with recursive models using symbolic regression to achieve accurate and robust tool wear predictions. A recursive feature model and a recursive tool wear model were developedto accurately predict future tool wear, taking into consideration the strong correlation between features extracted from sensor signals and tool wear. The Genetic Programming-based Toolbox for Identification of Physical Systems (GPTIPS) was employed for symbolic regression. The results illustrate that the proposed framework can capture the dynamics of tool wear by recursively updating predictions with new data and can derive simple, interpretable mathematical expressions that represent the physical characteristics of the tool wear process. Benchmarking analysis demonstrated the effectiveness of the proposed approach, achieving lower root-mean-square error (RMSE) compared to other tool wear prognostic models in the literature.