An efficient method for input uncertainty propagation in CFD and the application to buoyancy-driven flows

Abstract

Severe accident scenarios address the release of large amounts of hydrogen and steam to the containment. The formation of a flammable gas cloud could lead to a combustion and even failure of containment structures. In order to support the hydrogen mitigation method development, a detailed understanding of the gas transport and mixing process is crucial. Efforts in terms of numerical simulations such as Computational Fluid Dynamics (CFD) models have been made, which allow to investigate the complex 3D gas mixing process. One of the uncertainty sources that challenge the reliability of CFD validation results is the input uncertainty. It was assessed efficiently using the deterministic sampling method, which requires e.g., in the present case only eight binary samples for seven uncertain input parameters. However, the lean number of samples makes the direct derivation of a probability density function as well as a 95% confidence interval impossible. The assumption of a normal distribution does not always yield convincing and physically consistent output uncertainty bands, in particular for measurements inherent to oscillations. In this context, a new method has been proposed, which enables the generation of reasonable pseudo-samples without additional CFD simulations and the derivation of 95% confidence interval through the statistical analysis on these pseudo-samples. It was assessed against the Monte Carlo sampling method with a simple test case and confirmed an improved prediction. This method has been applied to the large scale application-oriented validation case THAI-TH32 in this work, in order to assess the impact of input uncertainties on the CFD results.