Microstructural diagram for steel based on crystallography with machine learning

Abstract A methodology for identifying the microstructures of steel based on crystallographic features obtained using electron back scattering diffraction (EBSD) is presented. The digital data obtained by EBSD, such as variant pair density and kernel averaged misorientation value, provide the basis for recognizing bainite and martensite microstructures. This implies that their own territories can be projected on a crystallographic space as a kind of phase diagram called microstructural diagram. The microstructural diagram can be deduced by incorporating the EBSD data into machine learning. The scheme is applied to low-carbon steels subjected to seven different thermal cycles and is confirmed to successfully classify the appropriate regions for microstructural characterization. The low-angle variant pair density is found to be a key factor for identifying the microstructures of low-carbon steels.