Inverse analysis of the relationship between three-dimensional microstructures and tensile properties of dual-phase steels
Takayuki Shiraiwa, Fabien Briffod, Manabu Enoki, Kazuhiko Yamazaki
Abstract
This study presents a forward model to calculate tensile properties (tensile strength and total elongation) from the three-dimensional microstructures of dual-phase steels, and an inverse analysis method to explore the optimal microstructure from the required tensile properties. The forward model was constructed by a crystal plasticity model with a damage law that takes into account the effect of stress triaxiality. It was calibrated and validated by microstructural analysis and tensile testing of dual-phase steel specimens of various martensite volume fractions. In the inverse analysis method, a large number of synthetic dual-phase microstructure models with different morphologies were randomly generated, and 24 microstructural descriptors were extracted by two-point spatial correlation functions and persistent homology. Feature importance analysis using random forest regression was used to further select the five important descriptors for predicting tensile properties. Using the selected descriptors as explanatory variables and the forward model as the objective function, microstructures that maximize the product of tensile strength and total elongation (TS×EL) were explored. The optimal microstructures found by Bayesian optimization and other exploration methods are consistent with conventional materials engineering findings, demonstrating that the proposed inverse analysis method is effective in solving the structure-properties linkages in the inverse direction.