About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Modeling, AI Applications, and Method Development in Nuclear Materials
|
| Presentation Title |
Machine Learning Predicted Thermal Conductivity of Nuclear Fuels |
| Author(s) |
Abigail Hogue, Benjamin Rhoads, Samrat Choudhury |
| On-Site Speaker (Planned) |
Abigail Hogue |
| Abstract Scope |
Thermal conductivity is a critical property governing the performance of nuclear fuels, yet existing experimental and computational thermal conductivity data remain fragmented across the literature and frequently contain incomplete material property information. This work presents a machine learning framework for constructing a unified database of thermal conductivity measurements and associated material characteristics from published studies. Missing material properties are estimated using an uncertainty-aware bootstrap random forest imputation approach, enabling incomplete datasets to be incorporated while quantifying prediction of uncertainty. Feature importance and sensitivity analysis were conducted to identify the key processing conditions and microstructural features governing thermal conductivity in UO2 and metallic fuels. Later, symbolic regression (SR) is employed to derive humanly interpretable analytical expressions relating thermal conductivity to temperature and microstructural features. Finally, SR predicted thermal conductivity is compared with experimental measurements. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Computational Materials Science & Engineering, Machine Learning, Nuclear Materials |