A Data-driven non-equilibrium wall model for LES of transitional flows
In session: MON 1.1 - Machine Learning I
Please login to view abstract download link
A data-driven non-equilibrium wall model for Large Eddy Simulation (LES) is developed from high-fidelity data from the Direct Numerical Simulation (DNS) of the flow through a diffuser. The model is trained with non-dimensional and reference frame invariant input features to predict the wall-shear stress. The wall shear stress obtained from the model is used as an approximate boundary condition in wall-modeled LES(WMLES). The trained model is tested a posteriori on the NASA hump experiment. It has been shown to perform remarkably better than a state-of-the-art wall model, especially at the turbulent to the laminar transition region.