Artificial Intelligence tools for spent fuel management

After passing through the reactor, fuel elements undergo significant changes in their composition due to the effect of nuclear reactions induced by the neutron flux from the reactor. Ultimately, due to fission and activation processes, the uranium oxide pellets introduced into the reactor end up containing a wide spectrum of isotopes, including virtually all elements from the periodic table in various forms of aggregation. In absolute terms, only a minority of uranium is converted, but this imparts entirely different properties to the fuel element, including the emission of residual heat or neutron and gamma radiation, which impact its management. A key parameter in the management of irradiated fuel is burnup, which expresses the amount of energy obtained per unit of initial uranium. Therefore, it is necessary to know the burnup of each fuel element for its management, typically determined through various means involving theoretical simulations and/or experimental methods to varying degrees.

Within the SICOM equipment family developed with Tecnatom, ENUSA possesses the SICOM-NG-FA equipment, which allows for the measurement of burnup in a fuel element based on measurements of its neutron and gamma emissions. ENUSA has developed new correlations to obtain burnup from the neutron source and the activity of ¹³⁷Cs. These correlations, as well as the calibration methodology, have been incorporated into the Supervisor software, which processes signals from the equipment’s detectors and calculates axial profiles and the average burnup of the inspected fuel.

The new correlations have been adjusted using a dataset generated through calculations with SCALE 6.2, employing a random population of irradiation histories. These simulations were conducted assuming a series of simplifications, the impact of which on the fuel’s gamma and neutron emissions has been evaluated through specific sensitivity calculations. In the case of the neutron source, a strong nonlinear dependence on the presence of consumable poisons such as gadolinium and WABA has been observed.

To obtain a good burnup estimation with the correlation that does not explicitly consider the presence of neutron poisons, a model capable of online prediction of the poisons’ effect on the neutron source is needed. The physics of the problem involves numerous parameters with nonlinear dependencies, necessitating the creation of a complex model using supervised learning tools. Specifically, Gaussian processes have been used, which have shown robust behavior and low sensitivity to the size of the available database.

Regarding gadolinium rods, the effect on the neutron source depends on the number of rods, poison concentration, and fuel burnup, with the overall effect being maximal between 15 000 and 20 000 MWd/MTU and virtually negligible for high burnups. Regarding the presence of WABA, the neutron emission is primarily affected in the low burnup range, with this effect decreasing as the average burnup of the element increases. There is also significant dependence on the initial enrichment of ²³⁵U and the number of WABA, as well as the element’s design, so the calculated bias incorporates a term dependent on the ratio between the number of WABA rods and the uranium mass of the element.

Although not listed as authors of this article, it is necessary to highlight the contributions of Miriam Vázquez Antolín and Carlos Casado Sánchez to the development of this work during their tenure at ENUSA.