Dynamic unified RANS-LES simulations of high Reynolds number separated flows

The development of hybrid RANS-LES methods is seen to be a very promising approach to enable efficient simulations of high Reynolds number turbulent flows involving flow separation. To contribute to further advances, we present a new, theoretically well based, dynamic hybrid RANS-LES method, referred to as DLUM. It is applied to a high Reynolds number flow involving both attached and separated flow regimes: a periodic hill flow is simulated at a Reynolds number of 37 000. Its performance is compared to pure LES, pure RANS, other hybrid RANS-LES (given by DLUM modifications), and experimental observations. It is shown that the use of this computational method offers huge cost reductions (which scale with Re/200, Re refers to the Reynolds number) of very high Reynolds number flow simulations compared to LES, it is much more accurate than RANS, and more accurate than LES, which is not fully resolved. In particular, this conclusion does also apply to the comparison of DLUM and pure LES simulations on rather coarse grids, which are often simply required to deal with simulations of very high Reynolds number flows: the DLUM provides mean velocity fields which are hardly affected by the grid, whereas LES velocity fields reveal significant shortcomings. We identified the reason for the superior performance of our new dynamic hybrid RANS-LES method compared to LES: it is the model's ability to respond to a changing resolution with adequate turbulent viscosity changes by ensuring simultaneously a physically correct turbulence length scale specification under the presence of interacting RANS and LES modes. Published by AIP Publishing.