Determination of spent nuclear fuel parameters using modelled signatures from non-destructive assay and Random Forest regression

Abstract

Verification of fuel parameters is a central undertaking for nuclear inspectors aiming at verifying the completeness and correctness of operator declarations. Traditionally, such verification is done analysing data from one instrument at a time. Here we present a study based on simulated data from various non-destructive assay measurement techniques applied on modelled PWR nuclear fuel assemblies. The data comprised multiple signatures and were analysed using machine learning algorithms. These signatures included activities from gamma-ray emitting fission product radionuclides, the parametrised early die-away time τ from the prototype Differential Die-away Self-Interrogation (DDSI) instrument, as well as the total Cherenkov light intensity which is directly measurable. The objective of the work is to systematically explore the capability to predict values of the fuel parameters initial enrichment (IE), burnup (BU) and cooling time (CT) independently of operator declarations, using Random Forest regression and modelled pressurised water reactor (PWR) fuel. The results show that passive gamma-ray activities alone can be used to predict IE, BU and CT for CT<20 years, and that by adding a feature proportional to the total gamma-ray activity, the errors in the predictions are significantly reduced. In this work, two measures proportional to total gamma activity have been studied: the sum of all considered gamma-ray intensities, and the total Cherenkov light intensity. From this work it was concluded that for fuels with CTs between 20 and 70 years, CT can be well determined by a multivariate analysis of the activities of 134Cs, 137Cs, 154Eu. For a BU determination, an additional feature corresponding to total gamma activity is required. This is, however, not sufficient to determine IE, which requires inclusion of the neutron signature τ as well.

Topics & Concepts