ISBN 9783843955669

978-3-8439-5566-9, Reihe Ingenieurwissenschaften

Philipp Josef Wenig
Uncertainty Quantification of Buoyancy-Induced Mixing Processes using Stochastic Spectral Methods

236 Seiten, Dissertation Universität der Bundeswehr München (2024), Hardcover, B5

Zusammenfassung / Abstract

Despite advancements in computational modeling, accurately representing real-world phenomena remains challenging due to uncertainties inherently present in computational models. Consequently, Uncertainty Quantification (UQ) of numerical simulations is crucial, especially when safety-critical applications are involved, such as in hydrogen facility operations and in nuclear reactor safety. Severe accidents in these fields often encounter buoyancy-induced mixing processes. The complexity of these phenomena, coupled with potential risks involved, highlights the necessity of developing robust and reliable predictive models. However, the Computational Fluid Dynamics (CFD) models are inevitably afflicted by uncertainties in initial and boundary conditions, which, if unaddressed, can lead to adverse consequences. Traditional UQ methods often require numerous simulation runs, posing significant challenges for UQ in the field of CFD, particularly when dealing with computationally intensive simulations. Therefore, the primary objective of this work was to assess and advance efficient forward UQ techniques that are suitable for large-scale applications. During the method development stage, efficient UQ methodologies were initially developed by investigating a generic test case. Subsequently, established techniques were adopted for the UQ of a technical-scale CFD application, which is subject to uncertainties. The applied UQ framework based on Stochastic Spectral Methods such as Polynomial Chaos Expansion and Karhunen-Loeve Expansion in conjunction with a variety of complementary modeling techniques provided promising results. The methods achieved both high computational efficiency and accuracy, underscoring their suitability for UQ of large-scale computational models and motivating their application to broader fields.