Abstract
An accurate prediction of the austenite-to-ferrite (γ → α) phase transformation using the multi-phase-field (MPF) model requires both the incorporation of thermodynamic calculations based on the CALPHAD method and the identification of model parameters. In this study, a Markov chain Monte Carlo (MCMC)-based parameter-estimation framework was developed to evaluate model parameters and their uncertainties using experimental data. To reduce the computational cost of the MPF simulation and enable parameter estimation within a reasonable computational time frame, neural-network surrogate models were constructed to perform thermodynamic calculations. By introducing surrogate models, the MPF simulation was accelerated by more than 220 times compared with conventional MPF simulations coupled with Thermo-Calc software. Subsequently, the developed framework was validated through numerical experiments in which we checked whether the framework can estimate ground-truth parameters using synthetic data obtained from an MPF simulation. The results demonstrated that the framework can accurately estimate the parameters and their uncertainties over a wide range of experimental uncertainties. Furthermore, the developed framework was applied to estimate the interfacial mobility and its uncertainty from the evolution of the α-phase volume fraction during the isothermal γ → α transformation in an Fe–0.19C–1.00Mn (wt.%) alloy. The estimated interfacial mobility parameter was of the same order of magnitude as values reported in previous studies, supporting its physical plausibility. Overall, this study contributes to the accurate prediction of microstructural evolution during the γ → α phase transformation by providing a framework for estimating the model parameters and quantifying their uncertainties based on experimental data.