LAUSR

Abstract Prediction of highly turbulent flows using the computational fluid dynamics (CFD) tools is not an easy task. Besides the uncertainty in the choice of turbulence model parameters, the physical properties of the fluid and experimental boundary conditions also can be largely affected by uncertainties. The objective of the study is uncertainty quantification of CFD simulation to obtain figures of merits, downstream velocity, and turbulence kinetic energy. The water-mixing experiment in the GEneric MIxing Experiment (GEMIX) facility performed at Paul Scherrer Institute is used as a benchmark case. The NEPTUNE_CFD code that solves Reynolds-averaged Navier Stokes equations with k-eps turbulence model has been used to perform a series of simulations. For uncertainty quantification with the Monte Carlo method the Optimal Statistical Estimator (OSE) was used for response surface (RS) generation from the set of CFD calculations. The results of the uncertainty analysis show that OSE is a very suitable method for RS generation, which is then used in uncertainty analysis using the Monte Carlo method to determine the 5% lower limit and 95% upper limit with 95% confidence level. In this way, the impact of some sources of uncertainty is evaluated. Also, OSE can reproduce the CFD simulation with high accuracy at the CFD calculation points, even in the case when only 5 out of 40 calculation points are used for RS generation. The results further suggest that it is very important to perform accurate reference calculation and select appropriate ranges of variation for uncertain input parameters.