Methodology for Sample Size Reduction in Uncertainty Analysis with High-Fidelity Fuel Performance Tools

This paper presents an efficient computational approach for modeling the propagation of uncertainties in input variables to output variables in fuel rod thermal-mechanical simulations. Our primary goal was to develop a methodology to identify a reduced sample size capable of providing information on uncertainties and sensitivities while remaining cost effective for computation-intensive high-fidelity three-dimensional simulations or full-core calculations.

Our method uses the best-estimate code TRANSURANUS (TU), which is equipped with a built-in Monte Carlo engine. This framework allows for the introduction of uncertainties into the selected input parameters through minor modifications in the input file used for the reference case. We applied this methodology to analyze a representative fuel rod proposed for use in the conceptual molten-salt fluoride-cooled high-temperature reactor (FHR), adapted to the geometry of the advanced gas-cooled reactor (AGR).

The computational efficiency of our approach lies in the reduced number of input/output operations. Consequently, we can execute numerous TU runs, enabling a comprehensive comparison of the results generated with a smaller number of statistical runs. To support statistical postprocessing, we developed the TUPython tool. With this tool, we can quantitatively assess both temporal and spatial variations as well as the sensitivity of fuel behavior model responses. The study showed that the sample size of 153, defined by the fourth-order Wilks’ method, can be used to economically model uncertainty propagation and perform sensitivity analyses in this specific case.