Large-eddy simulation (LES) has come a long way from its early beginnings in the 1960s. With the advent of the dynamic procedure in 1991, LES has now become an viable tool for the prediction of turbulent flows. One area where accurate predictions are important is heat transfer in turbulent flow. Heat transfer plays an important role in many engineering applications such as heat exchangers and combustors and in chemical processes such as combustion and corrosion. Traditionally, engineers have used empirical correlations, experimental datasets and low-order turbulence model results in the design of these devices. LES offers high fidelity results without the computational expense of direct numerical simulation (DNS).
Comparisons of two subgrid-scale models, the dynamic eddy-viscosity/diffusivity model and the dynamic mixed model, have been made for heat transfer in turbulent channel flow. Differences between the subgrid-scale
models were minimal, however the dynamic mixed model showed slightly better agreement with DNS results for the fluctuating stream-wise velocity. A parametric study of Prandtl number effects on heat transfer in turbulent flow has been performed. Mean temperature profiles and wall heat transfer coefficients showed excellent agreement with empirical correlations over a wide range of Prandtl numbers. The nearwall streaks seen in the velocity field were also found in the temperature field. These streaks were found to vary in width as the Prandtl number was changed, but maintained their streamwise length. At high Prandtl numbers the width of the temperature streaks seemed to limit towards the width of the wall-normal velocity structures. This suggested that at high Prandtl numbers the temperature field directly followed the near-wall ejection and sweep events. Because of the lack of diffusion (and spanwise convection very close to the wall), the temperature field does not diffuse in the
spanwise direction. Budgets of the temperature variance and turbulent heat fluxes were also presented for the entire range of Prandtl numbers.
The second part of this study concerns the prediction of heat transfer in a separated and reattaching flow downstream of a backward-facing step. For the first time, an LES of this flow has been directly compared with experimental data (previous simulations have been at lower Reynolds numbers where experimental data was not available). Both subgrid-scale models gave nearly identical results in this flow. The results of these simulations have been analysed in an attempt to discover the mechanism which causes the peak in the heat transfer coefficient to occur upstream of the mean reattachment point. The peak in the heat transfer coefficient showed an excellent correlation with the peak in the wall shear stress fluctuations. It is proposed that the high level of shear stress fluctuations is caused by eddies, created in the
shear layer, impinging the wall upstream of the reattachment point. They impact upstream of the reattachment point simply because of their size. These eddies were found to cause a "downwash" (i.e. a high towards-the-wall wall-normal velocity fluctuation) of cold fluid which effectively increased the heat transfer coefficient. Budgets of the resolved temperature variance and turbulent heat fluxes are presented to complete the database.