Multivariate analysis quantifying amount of fissile isotopes for pebble-bed high temperature reactors

Pebble-bed high-temperature gas-cooled reactors operate under a continuous refueling regime, rendering the trajectory of individual fuel spheres within the reactor core indeterminate. This inherent lack of tracking capability results in an obscured burnup history for the fuel spheres. Consequently, the conventional approach of deducing the nuclide inventory in the nuclear fuel of pebble-bed reactors through burnup calculations, predicated on initial fuel composition, is infeasible. Presently, while the online Burnup Measurement System integrated with High Temperature Reactor-Pebble bed Modules allows for the acquisition of gamma-ray spectra, enabling the measurement of certain nuclide activities, the direct non-destructive measurement of fissile isotopes remains elusive. This limitation poses a significant challenge to the field of nuclear material accounting, necessitating innovative methodologies for accurate inventory assessment.

This study presents a comprehensive approach to enhance the predictive accuracy of fissile isotopes and Zr-95 in pebble-bed reactors, encompassing three research facets. Initially, the Discrete Element Method (DEM) model is employed to simulate the core of a pebble-bed reactor. The model is calibrated against three-dimensional pebble flow experimental data by selecting an appropriate contact model and tuning model parameters to ensure consistency in flow dispersion, radial velocity ratio, and other key metrics. Subsequently, irradiation histories for the fuel spheres are constructed based on their trajectories within the reactor core. This data is then integrated with a nuclide inventory calculation software to simulate burnup information for the fuel spheres. Finally, the study employs multivariate analysis techniques, including Principal Component Analysis (PCA), ridge regression, and the random forest model, to predict fissile isotopes and Zr-95 from activity of online-measurable radionuclides. The predictive accuracy of our approach is appraised by correlating the outcomes yielded by a multivariate linear model, trained with simulated datasets, against those derived from experimental data.

This paper establishes a computational method for fissile isotopes in pebble-bed high-temperature reactors, enhancing the accuracy of nuclear material accounting, with a prediction error of only 4.9% for Zr-95, which is the most accurate experimental data that can be obtained. It demonstrates that online nuclide content contains valuable information worth further exploration and validates the applicability of multivariate analysis methods in nuclear material calculations.