| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Elucidating Microstructural Evolution Under Extreme Environments
|
| Presentation Title |
Machine Learning Potential Development for Advanced Oxide Fuels |
| Author(s) |
Audrey Miles, Bartomeu Monserrat, Sarah C. Finkeldei |
| On-Site Speaker (Planned) |
Audrey Miles |
| Abstract Scope |
Molecular dynamics simulations provide valuable insights into the microstructural evolution of nuclear materials under extreme conditions. However, the accuracy of these simulations depends on the quality of a potential energy surface (PES) describing the forces acting on atoms in the fuel lattice. While these PESs are typically computed using density functional theory (DFT), strong f-electron correlations in actinide-based materials lead to significant increases in computational complexity. Machine learning interatomic potentials (MLIPs) present an alternative to expensive and size-constrained computations, combining the efficiency of empirically-fitted potentials with the accuracy of electronic structure calculations. We use higher-order equivariant message passing to fit an MLIP for UO2. A DFT+U dataset is generated and used to train the MLIP. The model is then used to calculate thermodynamic and kinetic parameters necessary for simulating the microstructural evolution of oxide fuel. This work enables the extension of highly-accurate atomic-scale modeling approaches to grain- and pellet-scale challenges. |
| Proceedings Inclusion? |
Planned: |