Abstract This study presents a novel approach for estimating cutting forces in CNC machining by integrating feed drive dynamics and motor current signals with machine learning (ML) algorithms. The methodology leverages the mechanical behavior of feed drive components, modeling the interactions among servomotors, leadscrews, and other elements to analytically estimate cutting forces, incorporating a Stribeck friction model to accurately represent frictional losses. By utilizing data directly from CNC controllers, the proposed method offers a practical and scalable solution for in-process force monitoring, without the need for external sensors or costly instrumentation. To enhance prediction accuracy, a hybrid Physics-Informed Machine Learning (PIML) framework based on physics-based feature augmentation was developed combining analytically estimated forces based on feed drive model with servo signals and machining parameters. This framework not only improves the accuracy of the analytical model but also reduces the need for extensive experimentation used in direct data driven approaches. The results highlight the effectiveness of this approach, with R² values exceeding 98% and significantly reduced root mean square error (RMSE) values. Notably, the proposed PIML framework achieved an improvement of approximately 28–32% in prediction accuracy compared to the standalone analytical model, and 15–22% over conventional ML models that rely solely on data-driven inference. This integration not only captures nonlinearities and dynamic behavior but also retains the interpretability of the analytical model. The findings underline the added value of incorporating domain knowledge into ML algorithms, enabling more reliable force estimation directly from CNC controller data. Furthermore, the model demonstrated strong generalization capabilities when tested on unseen data, reinforcing its applicability in real-world manufacturing environments.
