Estimating uranium hexafluoride cylinder contents using autonomous measurement systems

In materials accountancy, radiological measurements of the nuclear material content of storage cylinders can be leveraged as a robust method of verifying declarations. These measurements are usually time-consuming and can result in appreciable radiation dose to personnel. In the near future it will be possible to task robotic systems with making such measurements. Such systems will use computer vision to orient themselves and locate the cylinders, and take measurements using onboard gamma-ray detectors. However, these systems will need real-time and near-real-time methods to assess whether the collected measurements are sufficient to determine the cylinder contents to a high degree of certainty, or whether more measurement locations and/or longer dwell times are necessary. In this work, we developed a simple physical model of a detector making a series of measurements around an arbitrary — but typically cluttered — arrangement of uranium hexafluoride cylinders and evaluated different algorithms for determining whether the containers were filled or empty, focusing on maximum likelihood (ML) and genetic algorithm approaches. We find that such approaches are promising, although there are degeneracies due to different possible uranium hexafluoride fill morphologies that significantly complicate the analysis.