Turbulence model correction method based on turbulence modeling machine learning

Abstract

The invention discloses a turbulence model correction method based on turbulence modeling machine learning, and belongs to the field of aero-engine internal flow calculation fluid simulation software. Due to complex flow characteristics such as local acceleration, complex vortex structure, separation, shock wave and boundary layer interference and high-altitude low-Reynolds-number separation in the rotating part, a turbulence model correction method for the internal flow characteristics of the rotating part is developed based on the basic theory of an RANS turbulence model. Firstly, based on a machine learning algorithm, existing domestic and overseas high-precision simulation and test databases are combed to obtain main flow characteristics of the complex flow field of the turbine, a turbulence modeling machine learning method is adopted to correct an RANS turbulence model, influences of pressure gradient, rotation, curvature and separation are emphatically considered, the precision of turbine simulation problems such as blade tip leakage flow is improved, the precision and applicability of an existing Reynolds average turbulence model are improved, and verification shows that the precision is improved by more than 10%.

Description

technical field [0001] The invention belongs to the field of aero-engine internal flow calculation fluid simulation software, and in particular relates to a turbulence model correction method based on turbulence modeling machine learning of a Reynolds stress turbulence model (RANS). Background technique [0002] High-speed rotation will produce a strong wall shear layer, and the turbulent shear stress is very sensitive to streamline bending and model rotation. In the concave corner area and the suction side of the rotating pipe, the turbulent flow is strengthened, while on the convex surface and the pressure of the rotating pipe side, the turbulent flow is weakened. The traditional turbulence model based on the Boussinesq linear vortex assumption cannot accurately capture this characteristic. This is because when the model rotates, although the Coriolis force can explicitly appear in the Reynolds stress tensor in the form of a rotation generating term, in the equation of th...

AI Technical Summary

Problems solved by technology

Traditional turbulence models based on the Boussinesq linear vortex assumption cannot accurately capture this characteristic

This is because when the model rotates, although the Coriolis force can explicitly appear in the Reynolds stress tensor in the form of a rotation generating term, in the equation of the eddy viscous turbulence model, the diagonal terms of the stress tensor cancel each other out, and the rotation effect Not obvious

Images