Comparing mechanistic and data-driven models for predicting cutting force and chip morphology of γ-TiAl alloys

γ-TiAl alloys are extensively utilized in the aerospace sector owing to their elevated specific strength and high-temperature capabilities; nonetheless, their inherent brittleness and complex work-hardening behavior often lead to significant discrepancies in traditional mechanistic models. This study presents a comprehensive comparative investigation between a data-driven constitutive framework and the conventional Johnson-Cook (J-C) mechanistic models. A neural network (ANN) model is proposed to adaptively map the highly nonlinear interplay among strain, strain rate, and temperature. To bridge machine learning and physical simulations, the ANN is embedded into a finite element platform via a custom subroutine, where precise analytical derivatives are formulated to guarantee computational stability. Benchmarked against both standard and modified J-C models, the ANN model demonstrates superior fidelity, achieving a global average relative error of merely 0.73% for flow stress and reducing the cutting force prediction error to 4.3%. Sensitivity analysis confirmed the exceptional predictive robustness of the data-driven approach. Furthermore, the simulation reveals that chip formation is governed by a “local plastic instability-transient brittle fracture” mechanism. This study confirms that data-driven methodology provides higher reliability and predictive accuracy compared to mechanistic models for high-fidelity simulations of difficult-to-machine materials.