This thesis is part of the drive to revive the nuclear industry, and in particular the safety studies associated with Severe Accidents (SA) in fourth-generation sodium-cooled fast breeder reactors (SFR). During such a hypothetical event, a jet of Corium (a mixture of molten fuel and core structural elements) interacts with the reactor coolant. This phenomenon, known as FCI (Fuel Coolant Interaction), generates an energy transfer to the coolant that can lead to steam explosions. The aim of this thesis is to improve FCI simulation tools for SFRs.
The development of the SCONE CFI code is based on TRUST, CEA's open source numerical platform. TRUST is a mature tool from an HPC and numerical robustness point of view, but suffers from certain numerical limitations for the simulation of GAs: transport terms to order 1, schemes ill-suited to the simulation of highly compressible flows and therefore steam explosion, numerical diffusion.
This thesis proposes to overcome these limitations by following a three-part research approach. The first part consists in extending the existing SCONE schemes to make them compatible with the simulation of strong shocks. A derivation via a principle of least action will be used for thermodynamic consistency. In a second part, the schemes developed will be extended to multiphase using the two classic models in the literature (pressure equilibrium and Baer-Nunziato). Finally, the last part will consist in the reduction of numerical diffusion of phase indicators, thanks to the rise in order (with multidimensional limitation) of advection terms and an interface reconstruction (of the VOF-PLIC type). As these algorithms are classically numerically expensive, Machine Learning techniques will be used to accelerate the identification of interface configurations.
The proposed work opens up professional prospects in the field of simulation and numerical methods, in particular towards research centers and R&D departments in industry.
Un stage de master 2 est proposé par l’équipe en complément de la thèse.