Estimating RANS model uncertainty using machine learning

In this work we present a machine-learning strategy developed to estimate the uncertainty introduced by a turbulence model for the prediction of a turbulent separated flows. The approach is based on the introduction of eigenvalue perturbations of the Reynolds stress anisotropy; the amount of perturbation is predicted by a random forest algorithm trained on high-fidelity simulations of the flow over a wavy wall. The proposed method is applied to the flow in an asymmetric diffuser and demonstrates how the approach correctly identifies the regions in which modeling errors occur and accurately quantifies the amount of errors when compared to experimental observations.