Performance evaluation of the Adaptive Response Modeling algorithm

A new nuclide identification algorithm, Adaptive Response Modeling (ARM), has been designed to automate robust whole spectrum analysis. ARM rapidly generates simulated spectra based on a limited number of bounded input parameters that are iteratively optimized to match a measured spectrum. Multiple responses based on bounding geometries can be interpolated to fit an unknown geometry. The performance of ARM has been evaluated by generating spectra for all singles and pairs from a library of 25 nuclides at a range of live times for a 2″×2″ NaI(Tl), 1.5″×1.5″ LaBr3(Ce), 1 cm³ CZT, 20% HPGe, and a 1.5″×15″×68″ plastic scintillator detector for unshielded and shielded (5 mm steel) conditions. These spectra were analyzed with ARM and standard peak analysis algorithms for comparison of relative performance. The false negative and false positive rates for each detector were determined at each live time and shielding condition for each analysis algorithm. The results are presented in plots for each detector and shielding condition in terms of probability of detection and false alarm rate. The ARM algorithm yielded the best overall performance, but was shown to be susceptible to failure when the correct response is significantly outside the bounding geometries.