TY  - JOUR
AU  - Hu, Yang
AU  - Wang, Kai
AU  - Spatschek, Robert
TI  - A data-driven strategy for phase field nucleation modeling
JO  - npj Materials degradation
VL  - 8
IS  - 1
SN  - 2397-2106
CY  - [London]
PB  - Macmillan Publishers Limited, part of Springer Nature
M1  - FZJ-2024-06088
SP  - 109
PY  - 2024
AB  - We propose a data-driven strategy for parameter selection in phase field nucleation models using machine learning and apply it to oxide nucleation in Fe-Cr alloys. A grand potential-based phase field model, incorporating Langevin noise, is employed to simulate oxide nucleation and benchmarked against the Johnson-Mehl-Avrami-Kolmogorov model. Three independent parameters in the phase field simulations (Langevin noise strength, numerical grid discretization and critical nucleation radius) are identified as essential for accurately modeling the nucleation behavior. These parameters serve as input features for machine learning classification and regression models. The classification model categorizes nucleation behavior into three nucleation density regimes, preventing invalid nucleation attempts in simulations, while the regression model estimates the appropriate Langevin noise strength, significantly reducing the need for time-consuming trial-and-error simulations. This data-driven approach improves the efficiency of parameter selection in phase field models and provides a generalizable method for simulating nucleation-driven microstructural evolution processes in various materials.
LB  - PUB:(DE-HGF)16
UR  - <Go to ISI:>//WOS:001345098600001
DO  - DOI:10.1038/s41529-024-00529-8
UR  - https://juser.fz-juelich.de/record/1032241
ER  -