Machine Learning-Assisted Turbulence Modeling for Flow Separation around Slender Bodies at Non-Zero Angles of Attack

This study develops a machine learning-assisted turbulence modeling approach to represent separated vortical flows around axisymmetric slender bodies at non-zero angles of attack. The original Reynolds-Averaged Navier-Stokes (RANS) model exhibits persistent limitations in capturing flow separation and associated vortices. To address these issues, the field inversion and machine learning (FIML) framework is applied to optimize the baseline turbulence model with experimental pressure data. The inversion results show that reducing turbulent viscosity within the adverse pressure gradient region before flow separation is essential for correcting delayed separation and improving surface pressure representation. A neural network is then trained to generalize the model correction through local flow features. When evaluated on an unseen slender body configuration, the trained model accurately captures the flow separation and the ensuing vortical structures, yielding significant improvements over the original turbulence model. The results demonstrate that FIML provides a physically consistent and generalizable pathway for enhancing RANS modeling of 3D separated flows around slender bodies.