Optimization of processing parameters to improve ductile-to-brittle transition of low-carbon structural steel

Literature-derived dataset comprising nine compositional variants of Nb-microalloyed ferrite–pearlite HSLA steels (0.08–0.14 wt% C) was used to model Charpy impact behavior. A data-driven framework integrating evolutionary machine learning (EvoNN and EvoDN2) with multi-objective optimization (NSGA-II and cRVEA) predicted and optimized the complete DBTT curve. The framework captured nonlinear relationships between input parameters, including composition and thermomechanical processing, and output parameters, including 27 J and 54 J impact transition temperatures and upper shelf energy (USE). Optimized outputs were used to reconstruct DBTT curves using a hyperbolic tangent formulation. The EvoDN2–cRVEA framework predicted significant performance gains, with USE increasing by 6% and 27 J ITT decreasing by 45%. Optimization indicated that higher slab soaking, lower reduction ratios, and reduced P, Mn, and Nb contents improve toughness. Notably, grain size and sulfur content exhibited non-intuitive trends, likely due to complex nonlinear interactions among the input parameters.