POD-ROM methods for Uncertainty Quantification of CFD

Abstract: Monte-Carlo Methods are established tools to quantify uncertainty by brute force exploring the solution space within an uncertain parameter range. As the approach is numerically expensive it is not suitable if single simulation runs are expensive as it is the case for CFD. We propose to first create surrogate models for CFD to allow exploring the parameter space. The POD-ROM methodology allows creating suitable fast-running models. In the talk it is shown how a non-intrusive POD reduced order model is created from the results of a transient k-epsilon-simulation that can be executed several orders faster than the original CFD model. Additionally a development towards POD-ROM for problems exhibiting repeated geometric subdomains is discussed.

Bio: Habilitation in fluid mechanics (2002); head of flow modeling group at Institute of Nuclear and Energy Technology (since 2005), professor of fluid mechanics at Karlsruhe Institute of Technology (since 2008); head of Framatome Professional School (former: AREVA Nuclear Professional School) (since 2011); director of InnoEnergy Master’s School Double Degree Programme ENTECH (2017-2022).

Research interset: Reduced order models based on proper orthogonal decomposition (POD ROM), industrial CFD, stochastic field method, coarse grid CFD, asymptotic series expansions, spectral methods, validation&verification, stability analysis, evolutionary optimization algorithms, levelsets, thermohydraulics, heavy liquid metal flow, nuclear technology, accelerator driven system (ADS), fuel assemblies, energy storage, micro fluidics, electro kinetics, combustion, premixed flames, spallation targets, pressure waves, supercritical fluids; education: challenge based learning, flipped classroom