Frozen propagation of Reynolds force vector from high-fidelity data into Reynolds-averaged simulations of secondary flows

TL;DR

This paper demonstrates that freezing the propagation of Reynolds force vector (RFV) from high-fidelity data into RANS simulations significantly reduces error propagation, improving the accuracy of secondary flow modeling at various Reynolds numbers.

Contribution

The study introduces a frozen treatment method for RFV propagation in RANS models, showing it outperforms traditional Reynolds stress tensor methods in reducing error propagation across different flow regimes.

Findings

Frozen RFV propagation reduces error by an order of magnitude.

Explicit and implicit RST treatments are less effective at high Reynolds numbers.

Combining correction terms with frozen RFV improves velocity and turbulence predictions.

Abstract

Successful propagation of information from high-fidelity sources (i.e., direct numerical simulations and large-eddy simulations) into Reynolds-averaged Navier-Stokes (RANS) equations plays an important role in the emerging field of data-driven RANS modeling. Small errors carried in high-fidelity data can propagate amplified errors into the mean flow field, and higher Reynolds numbers worsen the error propagation. In this study, we compare a series of propagation methods for two cases of Prandtl's secondary flows of the second kind: square-duct flow at a low Reynolds number and roughness-induced secondary flow at a very high Reynolds number. We show that frozen treatments result in less error propagation than the implicit treatment of Reynolds stress tensor (RST), and for cases with very high Reynolds numbers, explicit and implicit treatments are not recommended. Inspired by the obtained results, we introduce the frozen treatment to the propagation of Reynolds force vector (RFV), which leads to less error propagation. Specifically, for both cases at low and high Reynolds numbers, propagation of RFV results in one order of magnitude lower error compared to RST propagation. In the frozen treatment method, three different eddy-viscosity models are used to evaluate the effect of turbulent diffusion on error propagation. We show that, regardless of the baseline model, the frozen treatment of RFV results in less error propagation. We combined one extra correction term for turbulent kinetic energy with the frozen treatment of RFV, which makes our propagation technique capable of reproducing both velocity and turbulent kinetic energy fields similar to high-fidelity data.