A quantitative framework for through-thickness toughness assessment in laser-welded press hardening steel using DIC and ANN-based machine learning

A novel quantitative framework for comprehensive toughness assessment across heterogeneous regions in laser-welded press hardening steel was established. Three-point bending tests integrated with through-thickness digital image correlation (DIC) were conducted to capture full-field strain distributions in both base material and welded zones. An artificial neural network (ANN) was subsequently developed and trained to predict localized strain extremes and failure behavior at experimentally inaccessible regions. Results revealed approximately 20% reduction in maximum bending angle at the weld compared to base material, significantly dependent on loading direction relative to the weld. Contrary to conventional understanding, specimens achieving larger bending angles exhibited lower maximum tensile strains at fracture (0.132 vs. 0.143), attributed to decreased strain gradients in the outermost layer and more uniform microcrack distribution. The DIC-ANN framework quantitatively demonstrated that pressure head positioning relative to the weld pool significantly impacts strain distribution patterns and ultimate bending performance in VDA 238-100 testing.