Robust deep learning for emulating turbulent viscosities

From the simplest models to complex deep neural networks, modeling turbulence with machine learning techniques still offers multiple challenges. In this context, present contribution proposes a robust strategy using patch-based training learn turbulent viscosity from flow velocities, and demonstrates its efficient use on Spallart-Allmaras model. Training datasets are generated for past two-dimensional (2D) obstacles at high Reynolds numbers used train an auto-encoder type convolutional network local patch inputs. Compared standard technique, not only yields increased accuracy but also reduces computational cost required training.