Leveraging supervised learning for large-scale automated characterization of additively manufactured metals

Post-sintering surface treatments, particularly shot peening, can alter defect-induced fatigue behavior in additively manufactured (AM) steels. Automated, high-throughput metrology is essential to understanding the relationship between surface treatments, defects, and their impact on fatigue. In this study, we collected a library of optical microscopy (OM) and scanning electron microscopy (SEM) images and trained Vision Transformers (ViTs) to detect and characterize defects in Direct Metal Laser Sintering (DMLS)-printed 316L stainless steels. To evaluate the dynamic behavior of 3D-printed steels, we used Moore fatigue testing and corroborated our model with experimental data. It was observed that surface defects are key parameters controlling the fatigue response by acting as stress concentrators. The results establish a quantitative link between microstructural attributes and macro-scale properties in AM metals, demonstrating that our Machine Learning (ML)-assisted automated characterization technique can enhance predictive modeling of the mechanical performance of AM metals.