Masters Theses

Orcid ID


Date of Award


Degree Type


Degree Name

Master of Science


Nuclear Engineering

Major Professor

Jamie B. Coble

Committee Members

Steven E. Skutnik, Richard T. Wood


For nuclear nonproliferation safeguards, the ability to characterize used nuclear fuel (UNF) is a vital process. Fuel characterization allows for independent verification by inspectors of operator declarations of the special nuclear material flow and nuclear related activities within a facility, and an estimation of fissile material remaining in a fuel assembly. Current methods to verify this information rely heavily on non-destructive assay techniques, such as gamma spectroscopy and neutron detection measurements. While these measurements are effective tools for estimating a specific characteristic of the fuel, such as burnup or cooling time, they often require an accurate estimation of a select few isotopes in the fuel. This requirement means that the characterization is based on a very small amount of information that is contained in radiation emissions. To help overcome this limitation, this work investigates the use of empirical modeling to predict the burnup, initial enrichment, and cooling time of a Westinghouse 17x17 UNF assembly. This technique utilizes the entire spectrum of gamma emissions and gross neutron counts to predict each output to explore the full suite of information contained in these signatures. Three primary parametric modeling techniques are investigated for their performance in modeling this system: Ordinary Least Squares Regression (OLS), Principal Component Regression (PCR), and Partial Least Squares Regression (PLS). The models created are evaluated based on their root mean square percent error and condition number. The uncertainty of the best performing model is then quantified to understand the prediction interval of the predicted characterization and how this compares to the uncertainty of current measurement and characterization techniques. The PLS models are able to provide the best predictions while being stable. The PCR models have a consistent trade-off between accurate prediction results and stability. The OLS model provides fairly accurate results but is highly unstable due to correlations in the input data. The best model is the PLS model based on cross validation because it is stable, yields the lowest RMSPE values for burnup and enrichment predictions, and yields the second lowest percent of cooling time predictions that are more than 1 year away from the actual value. When used with the validation data set, this model yields RMSPE values of 0.42%, 1.39% and 4.61% for the burnup, enrichment, and cooling time, respectively. The total uncertainty of the predictions of this model are calculated to be 0.220 GWd/MTU, 0.051% U-235, and 0.694 years, for the burnup, enrichment, and cooling time, respectively.

Files over 3MB may be slow to open. For best results, right-click and select "save as..."