The accuracy of Reynolds-averaged Navier-Stokes results for turbulent scalar transport are affected by epistemic uncertainty in the Reynolds stress model in two ways: by altering the mean velocity field that advects the scalar, and by altering the inputs required for scalar flux models. We investigate these effects by propagating uncertainty in the Reynolds stress model to the prediction of scalar quantities in simulations of a pin-fin heat exchanger. The Reynolds stress model uncertainty is quantified by perturbing the anisotropy of the predicted stress tensor towards the three limiting realizable states. This uncertainty is then propagated to the scalar turbulence transport via the conservation law for the mean scalar and the turbulent scalar flux model. We consider three different scalar flux models that explicitly take the Reynolds stresses as input, and evaluate the performance of the models by verifying if high-fidelity large eddy simulation results are encompassed by the model predictions. The results indicate that the predicted uncertainty depends on both the general level of momentum transport and the most-relevant stress component, which is affected by the anisotropy perturbations. The predictions provide a similar bounding behavior for the overall temperature field as for the momentum field, but fail in bounding the local heat transfer rates in several locations. The bounding behaviors are further analyzed in terms of the predicted uncertainty in the flux magnitude, direction, and divergence to identify opportunities for further improvement of the proposed methods.
Read full abstract