NEAMS IRP Challenge Problem 3: Mixing in Large Enclosures and Thermal Stratification

Abstract

Mixing in large enclosures and thermal stratification play an important role in several advanced reactor designs, including liquid metal cooled reactors and high temperature gas reactors. Lessons learned from a recent international benchmark [1] using system-level codes for the simulations of Sodium-Cooled Fast Reactors (SFRs) have pointed out that, in order to improve the predictive capabilities of system codes, model developments are needed to correctly capture mixing and thermal stratification in the reactor hot pool upper plenum, the propagation of stratification fronts, and the overall effect of thermal stratification and mixing on natural circulation and heat transfer between primary and intermediate loops. CFD codes have also been shown to under-perform when simulating buoyancy-driven flows, yielding inaccurate predictions of the extension and propagation of stratified fronts. This is because the simulation and transport of scalar quantities (e.g., temperature, density, species concentration, etc.) in current CFD RANS-based turbulence models rely on the Simple Gradient-Diffusion Hypothesis (SGDH) to take into account turbulent fluxes whereas a simplified model to take into account turbulent kinetic energy production and dissipation induced by density differences. The NEAMS IRP Challenge Problem 3 (CP3) seeks to develop a multi-fidelity, multi-scale set of models for the mixing in large enclosures with or without thermal stratification, from CFD URANS to system-level code models including reduced order models (ROMs) in order to provide accurate and computationally affordable predictions for mixing and stratification in large enclosures. This Challenge Problem generalizes specific needs related to the TerraPower, Westinghouse and General Atomics designs by developing a set of benchmarks in order to advance, demonstrate and quantify the accuracy of mixing and stratification in large plena. Results of high-resolution experiments and LES/DNS are used in combination with machine learning techniques to feed the development of lower fidelity models, following a hierarchical approach. In the paper an overview of the ongoing experimental and modeling activities is presented.