Data-augmented turbulence modeling for separated compressible flow around axisymmetric bodies

Abstract

Flow around an axisymmetric body at high speeds involves flow separation coupled with compressibility effects, which imposes challenges for conventional Reynolds-averaged Navier-Stokes (RANS) turbulence models. The field inversion and machine learning (FIML) approach is employed to improve a RANS model. This study introduces novel flow features to incorporate compressible and rotational effects in a data-augmented approach. The proposed flow features are based on physics-and knowledge-driven model corrections, considering practices recommended for modeling turbulent flow. The current trained model with proposed flow features covers a wide range of the Mach number, including subsonic, transonic, and supersonic conditions. The eddy viscosity in the separated region is adequately reduced in the trained RANS model, which improves the predictive capability of separated compressible flow around axisymmetric bodies. The trained RANS model with two-dimensional flow data is also validated with a three-dimensional axisymmetric body at non-zero angles of attack. The improvement in modeling separated compressible flow is discussed further with conventional RANS corrections.