Nuclear data are widely considered as one of the largest sources of uncertainty in reactor analyses, including rector criticality and safety. As a consequence, already several years ago the nuclear data evaluators have taken up the initiative to provide nuclear data uncertainty estimates that describe our data knowledge in terms of best fit to experimental measurements. Recently, nuclear models uncertainties were also included to provide for missing experimental information. All these quantities were included as covariance matrices into public evaluations along with their corresponding nuclear data. Also, the standard format ENDF-6 used to store nuclear data into computer readable files was adjusted to accept the new covariance information.
As a second step, algorithms to assess linear sensitivity measures to nuclear data were implemented for a few neutron transport codes and mostly for the effective neutron multiplication factor keff. The resulting sensitivity profiles could be used to propagate nuclear data covariances making use of perturbation theory or generalized perturbation theory (GPT).
Several shortcomings were identified for this technology:
- sensitivities can only be calculated with few specific codes;
- perturbation theory and GPT are often cumbersome to implement for models other than criticality;
- calculations become too constraining for a large number of selected responses;
- sometimes a linear surrogate does not correctly represent the physical model.
Consequently, several research institutes started investigating a different approach to propagate nuclear data covariances based on nuclear data sampling. In this sense, some codes were developed to produce random nuclear data files with perturbed data. Following this trend, SCK CEN developed their own nuclear data sampling tool, SANDY.
SANDY is a python package that can read, write and perform a set of operations on nuclear data files written in the standard ENDF-6 format. Its primary objective is to produce perturbed nuclear data files containing sampled parameters that represent the information stored in the evaluated nuclear data covariances. Such files can be ultimately used to propagate uncertainties through any given compatible system using a brute force technique.
The scope of this Master’s thesis is to use SANDY to identify the existing shortcomings in the covariance information provided by the national and international nuclear data evaluation programs such as JEFF, ENDF/B and JENDL. SANDY will be adopted to propagate nuclear data covariances in realistic problems such as criticality, shielding and burnup models. Conclusions will be drawn on the quality of the nuclear data covariances and corrections will be proposed.