Quantifying turbulence model uncertainty in Reynolds-averaged Navier–Stokes simulations of a pin-fin array. Part 1: Flow field

Abstract

The assumptions that form the basis for Reynolds stress closure models have been formulated by considering canonical flows. As a result, the accuracy of Reynolds-averaged Navier–Stokes simulations can deteriorate significantly when modeling complex flows, and engineering applications would benefit from methods that can quantify the corresponding uncertainty in the predictions. This paper analyzes the performance of a previously proposed turbulence model uncertainty quantification (UQ) framework for simulations of flow through a pin-fin array. The method is a physics-based, data-free, interval approach that perturbs the Reynolds stress tensor shape towards the three limiting realizable states of anisotropy. The performance of the method is evaluated by determining whether large-eddy simulation results for the quantities of interest are encompassed by the intervals predicted by the UQ method. The results demonstrate that perturbing the stress shapes towards the one-component or two-component limit generally enhances the momentum transport between the bulk flow and wake regions, whereas perturbations towards the three-component limit suppress this transport. For quantities of interest that depend on the mean velocity and pressure field this results in predictions for uncertainty intervals that encompass the reference solution. For the turbulence kinetic energy the method fails to predict an adequate upper bound in regions with larger-scale turbulent structures, and it predicts overly conservative bounds in the pin stagnation regions. Based on these results several suggestions for improving the framework are made.