Revisiting the bias factor methodologies for the validation of fast test reactors

Abstract In this paper, we investigate different bias factor methodologies, including the standard bias factor method, the generalized bias factor method, and the product of exponentiated experimental values bias factor method. One new method is proposed—the representativity weighted bias factor method, which is based on weighted statistical dispersion. Finally, an old method, the extended bias factor method, is recast and reformulated in order to provide accurate results and new quantities to assess the usefulness of the integral experiments. These methods have been implemented and tested on a practical application targeting the Keff of a typical FTR (Fast Test Reactor) using up to four integral experiments. Results are compared against those of an adjustment taken as a reference using the same set of integral experiments. Based on results, we make several recommendations, including to adopt the extended bias factor method because it reproduces the results of the adjustment. In selecting experiments, the most important requirement is to have a large representativity factor, followed by an attached low experimental uncertainty. If two experiments are highly correlated, only one needs to be included. If an experiment has a very high representativity factor (e.g. 0.95), there is no need to include more experiments in determining the bias factor and the associated uncertainty reduction, and it is crucial to use experiments that capture the physics of the integral parameter under consideration.