Machine learning–assisted grain boundary segmentation reveals nanoscale heat treatment behavior of additively manufactured 316L stainless steel

Additive manufacturing produces inhomogeneous and anisotropic microstructures that differ from those in conventionally processed materials. Heat treatment is a crucial process to regulate the microstructural properties. Although estimating the resultant microstructure is a challenging issue, visualizing grain boundaries (GBs) would provide fundamental understanding of the mechanism during heat treatment. In this study, we developed a visualization method using electron diffraction mapping in transmission electron microscopy combined with machine learning. By extracting the area where the feature of electron diffraction patterns varies in the mapping data, GB region was successfully visualized. The developed image processing was applied to ex-situ heating observation of an additively manufactured 316L stainless steel (316L SS), and the curvature variation of GB before and after heat treatment was discussed. It was shown that GB bulging was likely observed around existing sub-GBs, causing unevenness in GB migration.