Data augmented turbulence modeling for three-dimensional separation flows

Field inversion and machine learning are implemented in this study to describe three-dimensional (3D) separation flow around an axisymmetric hill augment the Spalart–Allmaras (SA) model. The discrete adjoint method is used solve field problem, artificial neural network as A validation process for proposed adjust hyperparameters obtain a physically acceptable solution. result shows that non-equilibrium turbulence effects boundary layer upstream of mean line separating shear dominate structure 3D flow, which agrees with prior physical knowledge. However, effect anisotropy on appears be limited. Two approaches stage overcome problem sample imbalance while reducing computational cost during training. results all satisfactory, proves effectiveness approaches.