A glass neutron detector with machine learning capabilities

• Published in: Journal of instrumentation Vol. 14; no. 6; p. P06013
• Main Authors: Ademoski, G.L., Simko, S., Teeple, M., Morrow, I., Kralik, P., Wilkinson, C.J., Varney, G., Martinez-Szewczyk, M., Yinong, L., Nimmagadda, J.K., Samant, S., Wu, Y., Pan, L., Jacobsohn, L.G., Wilkinson, Q., Duru, F., Akgun, U.
• Format: Journal Article
• Language: English
• Published: Bristol IOP Publishing 01.06.2019
• Subjects: Artificial intelligence; Artificial neural networks; Borosilicate glass; Computer simulation; Learning theory; Lithium; Machine learning; Model testing; National security; Neutron counters; Neutrons; Nuclear engineering; Sensors
• ISSN: 1748-0221; 1748-0221
• DOI: 10.1088/1748-0221/14/06/P06013

The detection of neutrons above background levels is an indication of nuclear materials, creating significant applications for handheld neutron detectors in homeland security. For such applications, a 10B and 6Li enriched, scintillating glass neutron detector was designed. The model is compact enough to be used as a handheld detector and is equipped with machine learning capabilities to determine the location of the source and discriminate a neutron from a gamma. Lithium Borosilicate glass samples, with up to 70% 10B and 6Li content, and doped with Tb and Eu, were engineered to optimize the performance of the detector. The scintillation properties and neutron/gamma detection capabilities of the glass samples were tested. The model detector's performance was simulated in Geant4 and the simulation data was utilized for machine learning to predict the location of the source with an Artificial Neural Network (ANN). The reported handheld neutron detector design with the implemented artificial intelligence capability can achieve 99.9% accuracy in neutron/gamma discrimination, 8.9% error in radial angle estimates, and 2.9% error in azimuthal angle estimates.