Simulation of radiological background data for benchmarking statistical algorithms to enhance current radiological detection capabilities
Date
2017
Authors
LaBrake, Michael A., author
Brandl, Alexander, advisor
Johnson, Thomas, advisor
Biedron, Sandra, committee member
Journal Title
Journal ISSN
Volume Title
Abstract
Reliable methods for detection of nuclear material are essential, particularly when the material is a weak or shielded source. Since weak or shielded sources often require longer measurement times to distinguish between the source distribution and the background distribution, a statistical approach is being developed that will utilize prior information obtained by measurement systems, such as portal monitors, which collect data on a continuous basis. The hypothesis is that patterns can be identified in sequences of repeated count rate measurements and used in conjunction with classical statistics to identify and locate a source. By measuring background distributions and establishing standard data for specific locations, it is possible to use the probability of observing individual count measurement results in successive measurement intervals at or above the critical limit, y*, and use this information to help pinpoint a weak or shielded source. Because radioactive decay and detection of radioactive material are stochastic events, pseudo-random numbers are used in conjunction with a mathematical method that enables simulation of various background distributions. The related y* values are calculated for each distribution and repeated measurements at or above y* are counted and compared to those expected for the given distribution. Four distributions were investigated: the triangular, sinusoidal, normal, and Poisson distributions. For each distribution, large random number samples were generated to confirm the expected probabilities for various sequences of values at or above the decision threshold y*. All investigated sequences found the 95% confidence interval for the expected number of sequences greater than y* to include the observed number of sequences greater than y*.