ML and discriminant analysis for predicting properties of heat-treated LPBF-DSS

Additive manufacturing using laser powder bed fusion (LPBF) offers a powerful method for fabricating complex duplex stainless steel (DSS) components, such as S2507, which require high strength and corrosion resistance. However, post-processing heat treatments significantly impact their mechanical behavior by altering microstructural features like grain size, boundary misorientation, and dislocation density. This study evaluates the tensile performance of LPBF-fabricated S2507 DSS under three conditions: as-built, stress-relieved, and aged. Mechanical testing, scanning electron microscopy/electron backscatter diffraction analyses, and hardness measurements were performed to determine ultimate tensile strength, yield strength (YS), and % elongation. Statistical methods, including analysis of variance, multivariate analysis of variance, and canonical discriminant analysis, confirmed significant differences among heat treatments, while stepwise discriminant analysis identified YS, elongation, and grain size as primary discriminants. Furthermore, machine learning models linear regression, tuned support vector machine (SVM), extreme gradient boosting (XGBoost), and K-nearest neighbors were developed using six key variables. The tuned SVM and XGBoost models outperformed others, achieving R2 values of 0.941 (YS) and 0.910 (% elongation). These results validate the integration of multivariate statistical analysis and machine learning as a robust approach for predicting mechanical behavior and optimizing post-processing strategies for LPBF-fabricated DSS components.