Large Eddy Simulation of Turbulent Rayleigh-Bénard Convection: AnAssessment of Subgrid-Scale Models

Abstract

Large eddy simulation of turbulent Rayleigh-Bénard convection was carried out to assess various algebraic eddy viscosity subgrid-scale models: (i) Smagorinsky with Wall-Damping, (ii) Dynamic Smagorinsky, (iii) Wall-Adapting Local Eddy-Viscosity, (iv) Vreman, (v) Mixed-Scale, and (vi) a buoyancymodified Mixed-Scale model that accounts for the buoyancy effects from subgrid-scales. The last model is proposed for the first time in this study. Non-dissipative, kinetic energy conserving, fully implicit method was employed for simulations. To evaluate the models, mean and turbulent (both low- and high-order) flow diagnostics were computed. Some advanced turbulent statistics such as skewness, turbulent heat flux, subgrid-scale kinetic energy and Nusselt number were also calculated and compared with each other and against a reference solution. Since models differ from each other by means of turbulent generation terms, they have their own strengths and weaknesses which are particularly observed in the near-wall treatments. Additionally, unlike the others, the Dynamic Smagorinsky model computes the subgrid-scale viscosity coefficient dynamically which has some effects on results. Overall, the Mixed-Scale and its new, buoyancy-modified variant show different characteristics and mostly the best agreement with Direct Numerical Simulation data. They are also found computationally less expensive. Moreover, buoyancy enhancement in the new model slightlyimproves the predictions of Mixed-Scale model. Although relatively poor performance by the Dynamic Smagorinsky model is observed especially in estimating the integrated Nusselt number, it captures the turbulent heat flux more accurately than the others. A more detailed discussion on the model's performance based on evaluations are also made.