General-purpose Data-driven Wall Model: Predictions

A general-purpose wall model for large-eddy simulation is introduced. The model is grounded in the building-block flow principle, where essential physics from simple canonical flows are leveraged to train a generalizable model applicable to arbitrary geometries. This approach addresses key limitations of traditional equilibrium wall models (EQWM) and improves upon shortcomings in previous building-block-based strategies. The model is designed to capture a wide range of flow phenomena, including turbulence under wall curvature, zero/adverse/favorable mean pressure gradients (PGs), flow separation, and laminar flow. The approach is formulated as a regression problem to predict the wall shear stress using a feedforward artificial neural network. The inputs to the model are local-in-space and dimensionless, with their optimal selection guided by an information-theoretic method. Training data includes, among other cases, a newly generated direct numerical simulation dataset of turbulent boundary layers under varied PG conditions. The model is validated through both a priori and a posteriori testing. The a priori evaluation spans 34 diverse experimental and high-fidelity numerical datasets, including cases involving turbulent boundary layers, airfoils, Gaussian bumps, and aircraft geometries, to name a few. The proposed wall model demonstrated superior performance to the EQWM across most test scenarios, yielding an average relative error of 7% over all cases.