Probabilistic radwaste characterization: Findings of a multi-method multi-mockup exercise using interpolation-based surrogate efficiencies

Abstract

In this paper, we present the lessons learned from a multi-method, multi-mockup probabilistic radiological characterization exercise conducted within the EU project CHANCE “Characterization of Conditioned Nuclear Waste for its Safe Disposal in Europe”. A Bayesian approach that accounts for uncertainty in the measurement efficiencies is used to interpret combinations of (i) open geometry gamma spectrometry (OGGS), (ii) passive neutron coincidence counting (PNCC) and (iii) large-volume calorimetry (LVC) measurements, performed on five different combinations of mockup drum and source position. The used approach treats uncertainty in the measurement efficiencies by doing multilinear interpolation between reference efficiencies representing potential “end-member” matrices with respect to both composition and density, for the considered drum. The end-member proportions are then jointly inferred with the other unknowns. Our results indicate that the success of this method critically depends on the accuracy of the set of reference end-member efficiencies obtained from physics-based Monte Carlo particle transport modeling. We find that Bayesian inversion of OGGS data alone can recover reasonably accurate nuclide masses, with relatively high precision and low bias. Furthermore, when little prior information about the drum is available then jointly inverting OGGS and PNCC data is found to allow for a much higher precision compared to inverting OGGS data alone. In addition, calorimetry appears to be a promising method and brings some added value to OGGS as well in certain cases. In some setups and for some nuclides, errors in the reference efficiencies cause the marginal posterior distribution to be fairly biased. We thus design a simple approach to increase posterior uncertainty (that is, decrease precision) such that the true mass values are included in the bulk of the posterior distribution, in a controlled way. The approach consists of multiplying the interpolated efficiencies with a series of multipliers, that are jointly inferred under a relatively strong prior.