Towards Identification of TRISO Pebbles at Arbitrary
Orientation Using X-ray Radiographs

Modern pebble-bed reactor concepts employing TRISO-based pebbles involve the circulation of thousands of such pebbles through the reactor core. There is currently no method for the tagging, identification, and tracking of individual TRISO pebbles as they enter and exit the reactor core. Methods for uniquely identifying TRISO pebbles would significantly assist nuclear material accountancy and safeguards as each individual pebble could be tracked within a TRISO pebble reactor. This paper presents progress in demonstrating the use of computer vision software combined with machine learning approaches (Convolutional Neural Networks) for identifying TRISO pebbles from x-ray radiographs, based on the unique distribution of TRISO particles within each pebble. To identify a pebble an x-ray radiograph is compared with a library of reference radiographs of a set of pebbles. The matching performance is affected by the orientation of the radiographed TRISO pebble compared to the reference radiograph. The orientation can be specified by the off-axis and on-axis rotation angles which are defined by the radiography source and detector setup. Translational shifts of the pebble relative to the axis must also be accounted for. Significant progress has been made in the use of computer vision software for identifying any orientation of a TRISO pebble, when including an embedded reference marker, in a given acquired x-ray radiograph. In parallel, complementary machine learning methods are being developed to determine the orientation and to identify the pebble. This hybrid combination of methods can increase the accuracy of the identification. The potential implementation of this method for safeguarding TRISO pebble reactors is also discussed.