In the context of the revival of the nuclear industry, the objective of this thesis is to contribute to the R&D (Research and Development) on fast neutron reactors, which offer the possibility of efficiently using mixed oxide (MOx) fuels. This type of fuel indeed allows for a better utilization of nuclear resources and a reduction in high-level nuclear waste. Numerical simulations are an extremely beneficial resource for modeling the thermomechanical and physico-chemical behavior of reactor fuel. Scientific computational tools used to simulate this behavior are based, among other factors, on material property behavior laws derived from experimental measurements that are challenging to obtain at high temperatures, sometimes resulting in a lack of data in important application areas.
The objective of this thesis project is to propose more precise behavior laws using machine learning and a multi-fidelity model. This mathematical model will be developed by combining data from atomic-scale calculations, which can be more easily obtained at high temperatures, and experimental data. This will be a major scientific advancement, as this model will integrate data from different sources for the first time. Thermodynamic properties, especially thermal conductivity and specific heat, will be at the heart of this study. The multi-fidelity model will also guide future experiments to improve these behavior laws by identifying areas where they are less accurate.
The thesis will be conducted within the Department of Fuel Studies (IRESNE-CEA Cadarache Institute), and the candidate will join a team of experts in multiscale material modeling. The work will benefit from several collaborations with experts in applied mathematics. The candidate will use various generic techniques applicable to numerous materials science domains. The research will lead to participation in national and international conferences and the preparation of publications.