From data-driven constitutive modeling to workability assessment: elucidating multimechanism contributions in dynamic recrystallization of HIPed Ti4522XD alloy

To investigate workability and microstructure evolution of hot isostatic pressed (HIPed) Ti4522XD alloy, this study employs data-driven constitutive modeling coupled with EBSD characterization, focusing on multimechanism contributions. Combining isothermal compression tests (1100–1300 °C, 0.001–1 s⁻¹, 0–0.7 true strain) with machine learning (ML) algorithms, the constitutive framework is established. The Physics-Informed model, incorporating physics-inspired (PI) input features, physics-constrained (PC) loss functions and posterior validation (PV) modules with embedded material parameters, and optimized by adaptive inertia weight together with the trainlm algorithm, achieves superior prediction accuracy (correlation coefficient (R) = 0.9932), outperforming Arrhenius model by 84.59 % and 91.26 % in average absolute relative error (AARE) and root mean square error (RMSE). The optimized processing windows were identified: 1100–1200 °C/0.001–0.0041 s⁻¹ (strain 0.1–0.4) and 1120–1220 °C/0.011–0.014 s⁻¹ (strain 0.4–0.7). The twin-induced dynamic recrystallization (TDRX) is initiated at 60°<111> pseudo twins (PT) and 70°<110> true twins (TT) due to dislocation pile-ups, where the accumulation of stacking faults (SFs) on twin boundaries (TBs) facilitates their transformation into 9 R-type LPSO structures. However, DRX-induced grain refinement raises the critical resolved shear stress CRSS for twin nucleation, suppressing further twin development. Meanwhile, particle-stimulated nucleation (PSN) occurs in particle deformation zones (PDZs) around TiB, driving lattice rotation and recrystallization. In summary, DRX proceeds via multiple mechanisms—DDRX, CDRX, TDRX, and PSN-DRX, with DDRX dominating (>70 %) and strongly enhanced by higher temperatures, larger strains, and lower strain rates, while CDRX and TDRX contribute modestly (10–15 %) and PSN-DRX remains minor (<5 %). These results highlight that DDRX prevails, while the others provide secondary contributions.