| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Hume-Rothery Symposium on Thermodynamics of Microstructure Stability and Evolution
|
| Presentation Title |
Phase-field models, multiscale models and machine learning |
| Author(s) |
Kaushik Bhattacharya |
| On-Site Speaker (Planned) |
Kaushik Bhattacharya |
| Abstract Scope |
Phase field models have been remarkably successful in describing and understanding microstructure and its evolution in multiphase materials, in fracture, in crystals etc. However, it remains challenging to use phase field models to understand the evolution of history dependent properties at the macroscopic scale, because it is difficult to identify and transfer relevant information from the microstructural to the macroscopic scale. This contribution will describe how machine learning, and specifically neural operators, can be used to overcome this challenge. Neural operators and generalizations of neural networks, and can be used to obtain data driven approximations of maps between function spaces. They enable a discretization-independent method to transfer relevant information from the detailed phase field simulations at the microstructural scale to the macroscopic scale. We describe the general methodology and illustrate these with examples drawn from various phenomena. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Computational Materials Science & Engineering, Mechanical Properties |