Machine learning neural-network identification for dynamic recrystallization grains during hot deformation of nickel-based superalloy

Abstract

A novel multiscale dynamic standard for electron backscatter diffraction (EBSD) test has been constructed through the neural network to study the recrystallization behavior of Incoloy 925 superalloy during hot deformation. The average grain size distribution and grain orientation spread (GOS) distribution at different stages of dynamic recrystallization were analyzed. The recrystallized grains and deformed grains were described by grain size, length-diameter ratio, GOS, and GOS maximum value. The neural network can obtain the weight ratio of different influencing factors by learning the grain characteristics, which reveals excellent recognition accuracy for the different deformation states. • A novel multiscale dynamic standard was constructed through ANN for DRX recognition. • The “feature engineering” was quantified by grain size, length-diameter ratio, GOS, and GOS max value. • The ANN standard shows higher adaptability in different Ni-based superalloys.