PNNL

Abstract

Passive gamma-ray spectrometers composed of attenuation filters and integrating detector materials provide important advantages in terms of zero-power operation and ruggedness for long-term monitoring scenarios (e.g. national security or environmental remediation). However, the many design parameters, including attenuation filter material and thickness and number of pixels (filter/integrating material combinations), present a challenging optimization problem in designing spectrometers for different applications. In many of these applications, the goal is simply one of anomaly detection—deciding that there is a gamma-ray source not normally found in the nuisance source populations of that particular measurement environment. A passive spectrometer design study approach using an anomaly detection metric is presented here, and is founded on “injecting” target sources of interest (e.g. 57Co, 133Ba, 137Cs) into a nuisance source population that represents the widely varying backgrounds typical of long-term monitoring scenarios. The design evaluation metric is quantified by the probability of detection given a required probability of false alarm. A genetic algorithm employs this metric to probe the large design space and identify superior spectrometer designs. Mapping the false alarm and detection probabilities against each other for each design produces receiver-operator characteristic curves that can be used to compare many instrument designs over a wide range of operating constraints.