Machine learning-enhanced constitutive modeling and hot deformation behaviour of Ti-stabilized AISI 321 austenitic stainless steel

This study investigates the hot deformation behaviour of Ti-stabilized AISI 321 austenitic stainless steel and develops a hybrid constitutive modelling framework that integrates machine learning with classical Arrhenius analysis. Forty hot (40) compression tests were conducted over temperatures of 800–1200 °C and strain rates from 0.001 to 5 s⁻¹ to characterise the flow response and underlying dynamic restoration mechanisms. The alloy exhibited a transition from work-hardening dominated behaviour at lower temperatures and higher strain rates to dynamic recrystallisation-controlled softening under high-temperature, low strain-rate conditions. A Random Forest regression model was trained using strain, strain rate, and temperature as input features to predict flow stress. The model accurately reproduced full stress–strain curves, achieving high predictive performance and capturing nonlinear deformation behaviour without prescribing analytical functional forms. The ML-predicted stress data were subsequently used to refine Arrhenius-type constitutive fitting, yielding more consistent activation energy and stress exponent values by mitigating experimental noise and localized microstructural variability. The resulting Power dissipation efficiency and flow instability maps identified a stable hot-working domain of 1000–1150 °C and 0.001–0.1 s⁻¹ , corresponding to efficient dynamic recrystallization and enhanced workability, while instability occurred at T < 900 °C and ε̇ > 1 s⁻¹ . The hybrid modelling approach shows improved predictive capability and interpretability compared to either classical or purely data-driven models alone, providing a robust framework for process design and microstructure control in Ti-stabilized austenitic stainless steels.