Parametrizing phase field models for microstructure evolution: an integrated approach

Phase field models enable the simulation of microstructure evolution and phase transformations at length scales ranging from tens of nanometers to millimeters. They represent the evolving microstructure implicitly as diffuse interfaces in one or more continuous phase fields. Since phase field methods include the geometry of the microstructure, they can be higher fidelity than analytical approaches to modeling microstructure evolution. For the complex alloy systems of practical interest, however, phase field models are limited by computational performance and the need to specify many kinetic and thermodynamic model parameters. Performant open-source phase field codes have been developed, but the challenge of appropriately parametrizing phase field models remains complex and time-consuming, even for experienced users. To address this challenge, we are developing AMMBER, the AI-enabled Microstructure Model BuildER, which will provide an automated process for extracting phase field parameters from many different sources of kinetic and thermodynamic data. In the current stage of this project, we are focused on generating thermodynamic parameters for phase field simulations of microstructure evolution in metallic alloys with three or more components and two or more phases. Our approach considers atomistic and experimental thermochemical data as well as coarse-grained experimental data such as volume fractions. We integrate data from multiple sources using Bayesian optimization. We additionally compare tradeoffs between computational performance and fidelity in thermodynamics formulations for phase field models.