RANS simulation of natural convection in an enclosure with curved surface heating from below

Prediction accuracy of natural convection is strongly dependent on the choice of turbulence model. In this paper, different turbulence models as well as turbulent heat flux models are employed to predict the natural convection in an enclosure with curved surface heating from below and investigate their performance. Based on the simulation results, it is found that k-ε model which is well known to overestimate the production of kinetic energy provides a higher prediction of Nusselt number while k-ε-ζ-f model provides a relatively smaller estimation. A systematic CFD simulation is carried out for Rayleigh numbers in the range 10⁹ to 10¹¹, with which the influence from aspect ratio is also investigated. The effect of Rayleigh number on heat transfer is found to be significant and the dependency of Nusselt number (Nu) is obtained as a function of Rayleigh number (Ra). The influence from the aspect ratio of the simulation domain is found to be dependent on the choice of turbulence model. For k-ε model, Nu increases as aspect ratio increases, while an opposite dependency is observed when k-ε-ζ-f model is used. For this discrepancy, arguments are provided based on the streamlines and the profiles of Nu over the heating surface. (authors)