Assessment of RANS models for flow mixing in large enclosures using high-resolution experiments and LES

Accurate prediction of turbulent mixing in large enclosures is essential for the safety analysis of High-Temperature Gas-cooled Reactors (HTGRs), particularly during extended Loss of Forced Circulation (LOFC) accident scenarios. Despite its industrial relevance, this phenomenon remains challenging to model due to complex three-dimensional flow interactions and multi-scale effects involved, including jet impingement, shear-layer development, and recirculation, which are characteristic flow mechanisms in large plena. In this work, Reynolds-Averaged Navier–Stokes (RANS) turbulence models are systematically evaluated against a high-quality validation database composed of high-resolution experiments and Large Eddy Simulation (LES). The study focuses on an isothermal single-jet injection case in the 1/12-scale Michigan Multi-jet Gas-mixture Dome (MiGaDome) facility, which is representative of upper plenum mixing in HTGRs. Four commonly used turbulence models—the Standard 𝑘 − ɛ Low-Re, Realizable 𝑘 − ɛ , Reynolds Stress Transport, and 𝑘 − 𝜔 SST models—are assessed by comparing first- and second-order turbulence statistics on multiple planes within the enclosure. Quantitative error metrics are employed to evaluate model performance relative to experimental and LES data. Results show that the Standard 𝑘 − ɛ Low-Re and Realizable 𝑘 − ɛ models provide the most accurate predictions of mean velocity fields, while all RANS models significantly underpredict turbulence intensity and second-order statistics. These findings highlight both the strengths and limitations of RANS approaches for confined jet mixing and provide guidance for their application in reactor safety analyses, as well as a foundation for future development of reduced-fidelity models informed by high-resolution data.