Identification of recrystallized grains in EBSD maps: Automated adaptive thresholding using a Gaussian mixture approach

Modeling dynamic recrystallization requires a clear knowledge of how grain populations evolve during hot deformation. Electron backscatter diffraction (EBSD) is a common technique for discriminating between populations of new recrystallized grains and strain-hardened grains. One conventional method for identifying grain populations using EBSD data is to apply a predetermined threshold to grain orientation spread (GOS) maps. For a given alloy, the threshold must be manually adjusted, as it depends on the thermo-mechanical conditions investigated, making this conventional method user-dependent. This paper introduces a novel approach that aims to provide a more robust and reproducible method for identifying grain populations from EBSD data. This method uses principal component analysis (PCA) combined with Gaussian mixture modeling (GMM) to achieve automatic adaptive thresholding of grain populations. It incorporates an integrated uncertainty control for errors caused by threshold selection. This approach was used to monitor microstructural changes during the dynamic recrystallization of a 304 L austenitic stainless steel. Adaptive thresholding is expected to provide an alternative to conventional methods for recrystallization tracking. • Under variable thermo-mechanical conditions, fixed GOS thresholds cause grain misidentification. • GOS and GMKAM identified as key microstructural descriptors for monitoring recrystallization in 304 L. • PCA-GMM approach enables adaptive, data-driven grain identification from EBSD data. • Uncertainty band improves the reliability of recrystallized grain identification.