Computational analysis of buoyancy driven flows across scales

Abstract

The present thesis focuses on the analysis of the performances of Reynolds-averaged turbulence models, and in particular of second moment closure models based on the concept of elliptic blending, to reliably simulate buoyancy driven flows. These models were identified as the most promising tools for industrial applications as they are more consistent with the near wall physics than earlier models whilst remaining user-friendly. These models were implemented as part of the research in the open-source CFD toolbox OpenFOAM. A particular emphasis was put on the different type of turbulent heat fluxes closure, which are essential in the prediction of buoyancy driven flows.

Three flow configurations were analysed in details in the present work, namely the flow in a differentially heated square cavity at Rayleigh numbers up to 1e11, buoyancy induced single-phase counter-current flow in a pipe and the formation of cold-traps in nuclear reactor loop seals.

In order to provide insight into the statistical closure relations, high fidelity (LES or DNS) simulations were also performed. These deterministic simulations allowed the generation of a large amount of flow statistics that further the knowledge of buoyancy driven flows. Term by term comparison of closure relations helped identifying areas where the modelling could be improved. Calculations were validated against experimental data where available.