| About this Abstract |
| Meeting |
2021 TMS Annual Meeting & Exhibition
|
| Symposium
|
Frontiers in Solidification Science VIII
|
| Presentation Title |
A Method of Estimation of Solid-liquid Interface Anisotropy Based on Machine Learning Combined with Phase-field Simulations |
| Author(s) |
Geunwoo Kim, Tomohiro Takaki, Yasushi Shibuta, Munekazu Ohno |
| On-Site Speaker (Planned) |
Geunwoo Kim |
| Abstract Scope |
Anisotropy parameters of solid-liquid interface energy are important parameters that determine the preferred growth direction of dendrites in alloy solidification. It is important for description of microstructural processes to measure or estimate anisotropy parameters including its concentration-dependences with accuracy. However, the experimental measurement or estimation is not straightforward. In this study, we propose a method for estimating anisotropy parameters in fcc alloy, ε1 and ε2, by combining phase-field simulations with machine learning. The phase-field simulations for solidification microstructures in fcc model alloy with different sets of ε1 and ε2 were carried out and various microstructure thus simulated were characterized by the distribution of local curvedness (C) and shape indicator (S) of solid-liquid interface. Datasets of (ε1,ε2) and corresponding (C,S) distribution were trained by deep neural networks. This method allows determination of anisotropy parameters from the given (C,S) distribution accurately. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Solidification, Computational Materials Science & Engineering, Machine Learning |