A data-driven distributed force model for wall-modeled large-eddy simulations of rough-wall turbulence

In this work, a data-driven distributed force model for wall-modeled large-eddy simulations of rough-wall turbulence is proposed. In the proposed model, a body force with its distribution given by the area fraction of solid is distributed in the near-wall roughness layer to model the impact of the rough wall on the outer flow. The amplitude of the body force is computed based on the local streamwise velocity and the equivalent sandgrain roughness height, with the latter calculated using a deep neural network (DNN) model. Ideally, the proposed model can be applied to any type of rough wall if the geometrical parameters required by the DNN model are available. The proposed model is tested by simulating the turbulent channel flow for three cube-roughened surfaces and three ellipsoid-roughened surfaces at different grid resolutions and different Reynolds numbers. An overall good agreement with the direct numerical simulation results is achieved.