Sensitivity and uncertainty analysis for the UAM-SFR sub-exercises with linear regression from random sampling

Abstract

The paper presents solutions for the neutron transport sub-exercises regarding pin cells, fuel assemblies, and supercells of the OECD/NEA UAM-SFR Benchmark. For this, the GRS XSUSA methodology is applied with a recently introduced approach performing linear regression analysis on the output samples, denoted as XSUSA(LR). The main purpose for this is to calculate sensitivity profiles and the main contributions to the total output uncertainties via the first-order uncertainty propagation formula, i.e. by multiplying the corresponding total/partial nuclear data covariance matrices with the total/partial sensitivity vectors. Confidence intervals were estimated by applying Jackknife resampling. Implicit effects are considered. All output uncertainties are compared to the corresponding values calculated with deterministic linear perturbation theory using TSUNAMI from SCALE 6.2.3. Various optimizations were applied with respect to nuclear data and problem geometry. The output uncertainties are significant – around 1.5% for multiplication factors and 5% for Doppler and sodium void reactivity effects.