Data-Augmented Design of Turbulence Modeling

Abstract

As for the high Reynolds flow, aerodynamic design is mostly based on Reynolds-Averaged Navier–Stokes equations (RANS). As introducing the ensemble average hypothesis, the accuracy of RANS equation is widely doubted in predicting the transition and flow separation, for example, the laminar separation bubble or stalls under high angle of attack. This article takes an airfoil as an example and conducts a research on the data-augmented turbulence modeling design. Based on the high-fidelity prior data from experiment, a spatially varying term which will act as a multiplier of the viscous production term in Spalart–Allmaras model equation can be constructed using primal N-S flow and adjoint flow. In order to handle the issue of the extreme high dimension of this optimization problem (which is close to the number of grids), an adjoint method is used to solve the derivatives efficiently. The posterior result states that using a data-augmented turbulence modeling could predict the flow characteristics more accuracy, which can let the prediction of aerodynamic parameters like lift and drag more precise.