| dc.description.abstract | Nuclear signatures, whether for forensics or safeguards applications, utilize two broad classes: comparative and predictive. While comparative methods analyze a sample against a database of previously measured samples, predictive signatures utilize the underlying physics of the system to draw conclusions about the origin of the sample. Both sets of signatures would ideally be used for any thorough analysis; however, for uranium ore concentrate (UOC) the use of predictive signatures has been scarce. This work evaluates the potential use of 236U variation in uranium ores as a predictive signature for UOC.
Improvements in accelerator mass spectrometry (AMS) have allowed variations to be seen in measurements of 236U for a variety of ore samples. Work was done to evaluate the current capabilities/limitations of AMS systems in regards to 236U measurements. The abundance sensitivity was shown to be the primary limitation for AMS measurements, as some evaluated systems have a cutoff above the natural range of 236U. Improvements in sensitivity can lower precision, and further work is needed to determine potential bias between AMS systems. The physics of 236U production was evaluated next, and it was determined that the primary production pathway was neutron capture of 235U. A model was created to simulate the variations in neutron capture whether through changes in neutron yield and flux. Benchmarking of the model was performed against a set of measured samples, with an average deviation of approximately 40%.
Data analysis was performed using Bayesian methods, due to the incorporation of uncertainty in the parameters and use of additional data through prior distributions. Class selection was performed for uranium mineralization and deposition. In both cases, true positive results were only found for a fraction of the samples. However, the analysis indicated a low false positive rate (important in nuclear forensics). Parameter estimation was also evaluated and showed promise in individually analyzing an element of interest, although computation time and model error will be limiting factors. While each of these capabilities shows promise, work needs to be performed to validate the techniques by utilizing a larger and better characterized data set. | en |
