![]() | Only 14 pages are availabe for public view |
Abstract Nuclear data are fundamental parameters in nuclear systems calculations that have a large amount of associated uncertainties. This is believed to be the main reason for discrepancies between real life experimental measurements and code-computations for different nuclear problems responses. Hence, an adjustment strategy for nuclear data has to be carried out to reduce the associated uncertainties requiring extensive sensitivity analysis. In this work, a probabilistic adjustment procedure for the model response based on the Generalized Liner Least Squares method (GLLS) is derived and investigated. The method does not require the use of sensitivity coefficients, but instead captures the derivatives effect through sampling of nuclear data according to their prior uncertainties. Criticality safety benchmark experiments were used in the investigation to calculate the application bias and its associated uncertainty. The procedure was derived, verified using sensitivity derivatives, and investigated. In addition, the ordering of experiments used in the analysis is investigated to reach the optimal bias and associated uncertainty rapidly with the least number of experiments. Based on the new approach, the probabilistic and deterministic GLLS methods were shown to provide highly similar adjusted bias and uncertainty for the required response. The probabilistic approach results were compared to results provided from TSURFER module in SCALE code which implements adjoint sensitivities and were proven to approach equivalence in the studied case at about 700 samples used in the analysis. Newly proposed experiment arrangements showed promising results, where the final bias is approached rapidly using just a few experiments. |