Using machine learning techniques for neutron signal processing

Nicholas Ferguson


Supervised by Matthias Treder; Moderated by Hiroyuki Kido

This project will be carried out in partnership with the ISIS Neutron and Muon Source at the STFC Rutherford Appleton Laboratory in Oxfordshire.

Neutron beams can be used to probe the structure of materials on an atomic scale. One method of generating these neutron beams is to exploit the collision of a proton beam with a target. This produces neutrons which are guided into narrow beams and directed towards experimental areas. However, alongside neutrons, many gamma-ray photons are also produced, which present as noise in the output of neutron detectors. Currently, discrimination between neutrons and gamma rays is performed by analogue electronic signal processing, which works well. This project will explore the feasibility of using a machine learning-based classifier for performing neutron-gamma signal discrimination, with the aim of improving on the accuracy of the current process. I will train a variety of models using data from the detector test facility at ISIS and compare the relative merits of each. The final model will be compared against the electronic signal processing procedure. Ideally, a standalone program will be created, which will serve as part of the data processing routine as a whole.

Final Report (18/09/2020) [Zip Archive]

Publication Form