Applying Machine Learning to Neutron-Gamma Ray Discrimination from Scintillator Readout Using Wavelength Shifting Fibers
Date
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Advances in machine learning have found wide applications including radiation detection. In this work, machine learning is applied to neutron-gamma ray discrimination of an organic liquid scintillator (OLS) readout using wavelength shifting (WLS) fibers. The objective of using WLS fiber is to enable the transfer of the light signal from the scintillation medium, with almost any active volume geometry, to a low-profile photomultiplier. This is a common practice in high-energy physics research and has proven to be very effective for such applications. The drawback of this approach is the light pulses carried to the photomultiplier through the WLS fibers do not perfectly replicate the original OLS light pulses’ intensities or timing. This drawback causes traditional pulse shape discrimination algorithms applied to the degraded light pulses to fail to discriminate between neutron and gamma ray events. However, differences in the degraded light pulses for neutrons and gamma rays still exist and various machine learning algorithms can be applied to identify these differences. An experimental system was constructed to simultaneously capture part of the scintillation medium signal and the corresponding signal through the WLS fibers. Using the known neutron-gamma ray discrimination characteristics directly measured in the scintillation medium to provide the ground truth, supervised machine learning algorithms were applied to the corresponding light pulses carried to the photomultiplier through the WLS fibers. The results indicate that this approach will en-able enhanced recovery of neutron-gamma ray discrimination information. This research effort will focus on two aspects of the OLS-WLS system: 1) developing an experimental system to create machine learning training data and 2) applying and evaluating various machine learning algorithms.