Uncertainty Quantification method for CFD applied to the turbulent mixing of two water layers

Highlights: • Validation of an Uncertainty Quantification method for CFD. • The uncertainty resulting from input parameters is evaluated from PDFs with the Latin Hypercube Sampling method. • The numerical uncertainty is obtained by Richardson extrapolation. • Downstream velocity, turbulent kinetic energy and water concentration are the subjects for Uncertainty Quantification. - Abstract: Computer codes contain sources of uncertainty. Therefore, one need to quantify the uncertainty in the physical models, the corresponding inputs, and the applied numerical methods used in a computer code, in order to assess the reliability of the results. Uncertainty Quantification (UQ) is therefore common practice for most fast running codes which easily allow to run thousands of simulations. However, for computationally demanding Computational Fluid Dynamics (CFD) codes, UQ is a challenge! Due to the continuous increase in computer power on the one hand and the development of sophisticated UQ methods on the other hand, UQ for CFD is becoming more and more feasible nowadays. This work aims to evaluate the CFD prediction and associated UQ for the OECD/NEA benchmark based on a GEMIX (GEneric MIxing eXperiment) mixing layer test. Mixing problems are often encountered in nuclear systems and are typical single phase issues for which CFD may bring benefits. The presented CFD-UQ methodology is based on (a) the ASME Verification and Validation (V&V) standard for UQ in CFD applications, (b) the propagation of uncertain input parameters, and (c) Richardson extrapolation to evaluate spatial discretization uncertainty. The CFD-UQ methodology is proven to be efficient thanks to the Latin Hypercube Sampling (LHS) approach, which samples the uncertain input parameters prior to CFD propagation, in contrast to the computationally expensive coupling of CFD simulations and Monte Carlo (MC) Sampling. Namely, 20 computations were sufficient to evaluate the propagation of the uncertain input parameters. Moreover, the calculated uncertainty bands, with the presented CFD-UQ methodology, tied up to the CFD results, are effectively enclosing the experimental data. Overall, a good agreement is obtained with the experimental profiles for velocity, concentration and turbulent kinetic energy. For all three quantities, the presented CFD-UQ methodology was ranked by the benchmark organizers within the top 5 (out of 13 submissions). This shows that the applied CFD-UQ methodology is efficient, effective, and robust.