익형 주변 박리 유동 예측 향상을 위한 데이터 기반 난류 모델 개선

Abstract

A data-driven approach is investigated to improve the Spalart-Allmaras (SA) turbulence model. The field inversion and machine learning framework is chosen for the model improvement. Flow around the S809 airfoil is chosen for the test case. In this approach, a spatial-varying correction term is obtained using the discrete adjoint method in the field inversion process. Then, an artificial neural network is constructed to generalize the correction term with relevant flow features. This study shows that the corrected SA model reduces the turbulence production near the separation point, which leads to the improved prediction of the stalled airfoil at high angles of attack. Detailed investigation on modified flow fields and airfoil pressure distribution is conducted in this study to explain how the machine-learned model improves the turbulence model for separated flow.