Application of Neural Network in Micromechanical Deformation Behaviors of Inconel 740H Alloy

The rise of machine learning (ML) has taken materials development into a radically distinct realm. In this study, a framework based on ML and rate-dependent crystal plasticity finite elements were established to predicted flow stress and texture evolution of Inconel 740H alloy under uniaxial compression. First, the initial characterization data were used to construct a representative volume unit (RVE) model with an approximate structure to the real material microstructure, and then, the crystal plasticity finite element method (CPFEM) and its verification were carried out. Second, the dataset obtained from CPFEM and experimental data were used as training and test sets for the genetic algorithm optimized neural network (GA-BP) model. The results indicated that the proposed framework can well describe the macroscopic and microscopic response of Inconel 740H during uniaxial compression, which was in line with the experimental findings. Moreover, the GA-BP model had higher prediction accuracy and better prediction performance than the CPFE model; the root-mean-square error (RMSE) and mean square correlation coefficient (R-2) of stress and texture were 0.46156, 0.99282 and 2.7567, 0.9655, respectively. It is clear that the GA-BP is more efficient than the physical mechanism based CPFE model.