Abstract
During severe accidents in nuclear reactors, the core and internal structures can melt down and relocate into the reactor pressure vessel (RPV) lower head (LH) forming there a stratified molten corium pool. The pool generally consists of superheated oxidic and metallic liquid layers imposing thermo-mechanical loads on the RPV wall. The in-vessel retention (IVR) strategy employs external cooling with water to maintain RPV integrity. Investigating the thermo-fluid behaviour of corium and predicting heat flux distribution on the vessel wall are crucial. The molten pool exhibits natural convection, which can typically consist of two stratified layers. There exists internally heated (IH) natural convection in the oxidic layer and Rayleigh-Bénard (RB) convection in the surface metallic layer.
This study starts by illustrating the mathematical models that involve the numerical study of natural convection flow in molten corium. A verification work of the model has been done using a previous direct numerical simulation (DNS) study, and the results show good agreement. In addition, a scaling theory of the natural convection flow is demonstrated to facilitate the pre-estimation based on the Rayleigh number (Ra) and Prandtl number (Pr). After that, the numerical approaches involved in the numerical simulation of the corium are illustrated, especially focusing on the DNS method. A DNS mesh strategy is proposed in the form of a pipeline from the pre-estimation to the post-check. A scalability study of Nek5000 is performed on four different HPC clusters based on a DNS case of the IH molten convection in a hemispherical geometry with Ra=1.6×1011. The results show a super-liner speedup property of Nek5000 on each cluster within a certain range.
Then, three numerical studies focusing on turbulent natural convection flow within both the oxidic and metallic layers of corium are demonstrated and discussed. Through these simulations, the thermos-fluid behaviour of the system is examined in detail, including flow configuration, temperature distribution, heat flux profiles on cooling boundaries, and turbulent quantities.
1. A DNS investigation is performed on the IH molten pool convection within a hemispherical domain, employing a Rayleigh number of 1.6×1011 and a Prandtl number of 0.5. The results show a turbulent flow characterized by three distinct regions, consistent with the observation from the BALI experiments. Detailed information regarding turbulence, including turbulent kinetic energy (TKE), turbulent heat flux (THF), and temperature variance, is presented. Furthermore, the study offers comprehensive 3D heat flux distributions along the boundaries, showing heat flux fluctuations along the top boundary due to nearby turbulent eddies and a nonlinear increase in heat flux along the curved boundary from bottom to top.
2. A numerical study investigates the effect of Prandtl number on the natural convection of an IH molten pool in a 3D semi-circular test section. Prandtl numbers of 3.11, 1.0, and 0.5 are considered, with a Ra= 6.54×1011. Smaller Prandtl numbers result in more vigorous turbulent motion and a thicker layer of intense turbulent mixing in the upper region. The descending flow extends further down the bottom, creating a stronger circulation at the bottom with smaller Pr. Additionally, smaller Pr leads to more thermal stripping structures and less stable stratification layers. Comparing heat fluxes on the top and curved walls reveals higher fluctuation frequency with smaller Pr for heat fluxes to the top boundary. However, the maximum heat fluxes to the side walls are lower with smaller Pr.
3. A numerical study investigates the turbulent natural convection in a 3D fluid layer based on the BALI-Metal 8U experiment. Different methods, including DNS and three Reynolds-averaged Navier-Stokes (RANS) models, are employed. The results are compared with experimental data, and the performance of the RANS models is evaluated using DNS as a reference. DNS reproduces a two-distinct region flow structure observed in experiments, while the k-ω SST model exhibits similar flow patterns and TKE profiles. However, all simulations overpredict temperature compared to experimental data, with DNS providing the closest results. The DNS results also achieve better agreement with experimental data in terms of heat flux distribution and energy balance, specifically capturing the transient maximum heat flux on the lateral cooling wall. This transient behaviour plays a crucial role in accurately estimating the ‘focusing effect’.