Date of Award
Doctor of Philosophy (PhD)
Environmental Engineering and Earth Science
DeVol, Timothy A.
DeVol , Timothy A.
Fjeld , Robert A.
Powell , Brian A.
Sharp , Julia L.
On-line radiation monitoring is essential to the U.S. Department of Energy (DOE) Environmental Management Science Program for assessing the impact of contaminated media at DOE sites. The goal of on-line radiation monitoring is to quickly detect small or abrupt changes in activity levels in the presence of a significant ambient background. The focus of this research is on developing effective statistical algorithms to meet the goal of on-line monitoring based on time-interval (time-difference between two consecutive radiation pulses) data. Compared to the more commonly used count data which are registered in a fixed count time, time-interval data possess the potential to reduce the sampling time required to obtain statistically sufficient information to detect changes in radiation levels. This dissertation has been formulated into three sections based on three statistical methods: sequential probability ratio test (SPRT), Bayesian statistics, and cumulative sum (CUSUM) control chart. In each section, time-interval analysis based on one of the three statistical methods was investigated and compared to conventional analyses based on count data in terms of average run length (ARL or average time to detect a change in radiation levels) and detection probability with both experimental and simulated data. The experimental data were acquired with a DGF-4C (XIA, Inc) system in list mode. Simulated data were obtained by using Monte Carlo techniques to obtain a random sampling of a Poisson process. Statistical algorithms were developed using the statistical software package R and the programming function built in the data analysis environment IGOR Pro. 4.03.
Overall, the results showed that the statistical analyses based on time-interval data provided similar or higher detection probabilities relative to other statistical analyses based on count data, but were able to make a quicker detection with fewer pulses at relatively higher radiation levels. To increase the detection probability and further reduce the time needed to detect a change in radiation levels for time-interval analyses, modifications or adjustments were proposed for each of the three chosen statistical methods. Parameter adjustment to the preset background level in the SPRT test could reduce the average time to detect a source by 50%. Enhanced reset modification and moving prior modification proposed for the Bayesian analysis of time-intervals resulted in a higher detection probability than the Bayesian analysis without modifications, and were independent of the amount of background data registered before a radioactive source was present. The robust CUSUM control chart coupled with a modified runs rule showed the ability to further reduce the ARL to respond to changes in radiation levels, and keep the false positive rate at a required level, e.g., about 40% shorter than the standard time-interval CUSUM control chart at 10.0cps relative to a background count rate of 2.0cps.
The developed statistical algorithms for time-interval data analyses demonstrate the feasibility and versatility for on-line radiation monitoring. The special properties of time-interval information provide an alternative for low-level radiation monitoring. These findings establish an important base for future on-line monitoring applications when time-interval data are registered.
Luo, Peng, "Time-Interval Analysis for Radiation Monitoring" (2011). All Dissertations. 850.