Uncertainty and sensitivity analysis of coupled multiscale simulations in the context of the SESAME EU-Project

Abstract

There are many complex industrial hydraulic systems for which the detailed simulation of their behavior is computationally expensive. Fast three-dimensional (3D) computational approaches need to be developed that can offer the required degree of detail where it is needed, while being computationally efficient. One possibility is to establish multiscale models by developing coupled code systems based on one- or two-dimensional thermal-hydraulic modeling for less relevant system parts and 3D-computational fluid dynamics (CFD) for those locations where high accuracy is important. Despite the increased modeling accuracy achieved by such approaches, their results are always subjected to uncertainty. Safety relevant simulation results should therefore be qualified with uncertainty, so that decission makers can make use of them with the necessary assurance. In addition to uncertainty, sensitivity analysis based on statistical techniques can be applied to the simualtion results to identify weak points in the system design, to quantify the influences of local and global parameters on local/whole system thermal-hydraulics behavior, and to support the design and realization of experiments and measurements involving those systems.In the frame of the EU-Project SESAME, an experimental facility named TALL-3D loop was built to provide experimental data for code validation and assessment in the simulation of liquid metal coolant loops. In the work presented in the paper a multiscale model of TALL-3D was built by using the coupled ATHLET-ANSYS CFX code system to perform flow simulations based on fast 3D-approaches and to conduct uncertainty and sensitivity analysis of the results. Based on information of the variability of input system parameters, a series of code simulations were performed for selected experimental scenarios to investigate the propagation of uncertainty in the coupled multiscale computational chain, and the uncertainty of the results was compared to the experimental data. It was found that the best-estimate results provided by the coupled code system, together with the uncertainty intervals (tolerance limits) were capable of bounding the experimental data. It is demonstrated that the Uncertainty Quantification (UQ) can be applied for the coupled code system and is able to quantify the simulation results with their uncertainties, which provide more comprehensive information of the real system. Sensitivity analysis allowed for the determination of the influences of input variables on the results, which can be used for the dimension reduction and provide a reference for the experimentalists and manufacturers to reduce the system uncertainty.