On sensitivity of RANS simulations to uncertain turbulent inflow conditions

Abstract

The present study deals with gPC-based analysis of the sensitivity of URANS simulations to uncertain inflow conditions. The massively separated flow around a square cylinder, which is a classical test case, is selected. Three popular turbulence models are selected, i.e. standard low-Reynolds k–ε model, k–ω model and a realizable k–ε model. The turbulent viscosity ratio, which is observed to vary a lot according to different authors in this configuration, is taken as an uncertain parameter to illustrate the potential applications of the present methodology. It is found that the realizable k–ε model behaves quite stably, and predicts nearly consistent results in all simulations. In contrast, the other two models are very sensitive to random turbulent inflow condition in both cases, especially the standard k–ω model, resulting in a significant weakness of usual validation procedures for turbulence models.