Epistemic uncertainty in URANS based CFD analysis of buoyancy driven flows—Comparison of URANS and LES

The containment is the last barrier to prevent the release of radioactive materials into the environment in case of a severe reactor accident. Thus, it is important to understand the flow and transport processes of combustible gases and to develop strategies to maintain the integrity of the containment. A holistic OpenFOAM-based CFD software package, containmentFOAM, is developed for containment analysis. During the validation of the code with regard to free convection flows and mixing processes, it became apparent that the simulation results are fairly sensitive to the definition of initial and boundary conditions, which are often not precisely known from the experiment. Therefore, establishing the validity of the models is impossible without a quantification of uncertainties.Generally, the uncertainty is categorized in two types, namely aleatoric and epistemic uncertainties. The first exhibits stochastic variations around the exact value. The latter has two possible effects on the uncertainty quantification: first a different sensitivity of the model to the input variation and second a bias to the exact value. Both effects can affect the significance of an aleatoric uncertainty quantification. The epistemic uncertainty originating from URANS turbulence modelling is investigated in this study by means of a direct comparison with the results obtained by LES, assuming that LES turbulence closures have smaller epistemic uncertainty. For a buoyancy driven natural circulation flow as well as a superimposed mixing process, both epistemic effects of the URANS approach are identified in terms of different sensitivities with respect to a variation of uncertain input parameters and the deviations of the URANS output variation ranges to those of LES. Nevertheless, the analysis of uncertainty bands derived on the basis of the most relevant parameter revealed that these epistemic effects ultimately result in a negligible effect on the aleatoric uncertainty quantification conducted with URANS.