Uncertainty Quantification of a Industrial Scale CFD Application Using Stochastic Spectral Methods

Abstract

Accurate prediction of flow phenomena is crucial for a wide range of industrial and scientific applications. However, due to the complexity of many fluid systems, even state-of-theart Computational Fluid Dynamics (CFD) simulations involve various sources of uncertainty, which need to be quantified. Therefore, this work focuses on the Uncertainty Quantification (UQ) of a CFD simulation of a technical scale experiment, which investigates buoyancy-driven mixing processes between two miscible fluids. The CFD model is subject to uncertainties in initial and boundary conditions, as well as in thermo-physical properties. Since computational runs of the CFD model are computationally expensive, the uncertainty analysis requires the use of efficient methods. For this reason, stochastic spectral methods, such as Polynomial Chaos Expansion (PCE) and Karhunen-Loève Expansion (KLE), are applied along with novel approaches like Stochastic Model Composition (SMC) for the uncertainty quantification of responses. Fluctuations due to turbulence are identified as considerable source of uncertainty and are modeled with tailored stochastic models. Results include a comprehensive probabilistic representation of response quantities with probability density functions (PDFs), statistical moments, variance-based decomposition and supplementary error estimates of the underlying stochastic models. This enables well-founded result interpretation and informed decision making. The UQ results show that the uncertain input parameters considerably affect the duration of the mixing process. Through the investigations, the applied UQ methods were tested on an engineering application and demonstrate great potential as a viable technique for uncertainty quantification in technical-scale computations.