| Large-eddy simulation (LES) requires very high resolution in attached turbulent boundary layers due to the need to capture the small, dynamically important near-wall eddies. Therefore, the computational expense of LES scales almost as strongly with Reynolds number as direct numerical simulation in these flows. Wall models enable LES to be performed on grids that do not resolve the wall layer by providing approximate boundary conditions to the LES at solid boundaries. This allows a much weaker scaling of the LES grid size with the Reynolds number.; Unfortunately, wall models based on purely physical reasoning often lead to an inaccurate LES, particularly on coarse grids and at high Reynolds numbers, because they do not account for the numerical and SGS modeling errors that become large in these types of simulations. While control-based wall models can account for these errors, they are limited by their large computational expense and lack of predictability.; The goal of this work is to address these two issues. To reduce the expense, the adjoint equations, used to determine the gradients needed for the optimization, are reformulated so they can be evaluated efficiently. Further, the optimization algorithm has been modified to only use near-wall information, eliminating work wasted in regions of the flow which are insensitive to the control. The approach reduces the computational cost of the method by an order of magnitude without a loss of accuracy.; To make the method predictive, a near-wall Reynolds-averaged Navier-Stokes (RANS) model has been coupled to the system. The LES provides the velocity boundary conditions for RANS away from the wall, while the RANS feeds back into the LES through the definition of the cost function. An additional coupling enables the RANS to provide the mean wall stress for the LES. The control then provides the fluctuating wall stress which minimizes the cost function. Using this method in plane channel flow, an accurate prediction of the mean velocity profile has been obtained over a range of Reynolds numbers and on different grids. The results are comparable to non-predictive control-based methods and are much more accurate than traditional wall models. |
