Uncertainty Quantification method for CFD applied to the turbulent mixing of two water layers - II: Deterministic Sampling for input uncertainty

Computer simulations are frequently used for design and safety analyses of nuclear installations. In such computer simulations, uncertainties in the outcome are present, for instance, due to uncertainties in the input parameters of the analyses. It is important to assess and quantify these uncertainties, especially when the simulations form the basis for nuclear reactor safety or licensing. Quantification of the uncertainties in Computational Fluid Dynamics (CFD) analyses is currently seldom done, since the existing and robust way of Uncertainty Quantification (UQ) involves a large number of computations, and CFD computations are computationally very demanding. The long run-times and large computational efforts do simply not allow for thousands of CFD simulations. Therefore, smart UQ approaches for CFD are being tested, developed and validated at NRG, which allow a full quantification of the uncertainties, while the required number of computations is still acceptable. The UQ methodology used by NRG is based on the ASME Verification and Validation (V&V) standard for UQ in CFD applications, combined with a propagation technique to evaluate the input uncertainty, and Richardson extrapolation to evaluate the spatial discretization uncertainty. This CFD-UQ methodology proved to be efficient thanks to the Latin Hypercube Sampling (LHS) technique, which samples the uncertain input parameters prior to CFD propagation, in contrast to the computationally expensive and fully random Monte Carlo sampling (MCS) technique. Nevertheless, MCS and LHS are both Random Sampling (RS) techniques, which will always be affected to some degree by a random sampling error. Therefore, Deterministic Sampling (DS) techniques are tested in this work to evaluate the input uncertainty, instead of LHS. The DS techniques proved to be more efficient, producing similar results with less computations.