Prediction and optimization of stretch flangeability of advanced high strength steels utilizing machine learning approaches

Abstract

Abstract Advanced high strength steels (AHSS) exhibit diverse mechanical properties due to their complex chemical compositions and microstructures. Existing machine learning (ML) studies often focus on specific steel grades, limiting generalizability in predicting and optimizing AHSS properties. Here, an ML framework was presented to predict and optimize the stretch-flangeability of AHSS based on composition-microstructure-property correlations, using datasets from 212 steel conditions. Support vector machine, symbolic regression, and extreme gradient boosting models accurately predicted hole expansion ratio (HER), ultimate tensile strength (UTS), and total elongation (TE). Shapley additive explanations revealed the importance of bainite volume fraction (VB), carbon content (C), and chromium content (Cr) for HER, UTS, and TE, respectively. Multi-objective optimization generated 252 optimized conditions with improved comprehensive mechanical properties. The best optimized chemical compositions (0.12wt.% C-1.10Mn-0.15Si-0.47Cr) along with the carbon equivalent (CE) of 0.44 wt.%, and microstructural features (7.2% ferrite, 44.5% bainite, 40.5% martensite, and 7.8% tempered martensite) yielded HER of 119.8%, UTS of 1013.5 MPa, and TE of 22.7%. This systematic framework enables efficient prediction and optimization of material properties (especially HER), with potential applications across various fields of materials science.