Machine Learning Applications for Radioisotope Neutron and Gamma Pulse Shape Categorization*

When determining neutron and gamma signatures from radioisotope measurements, discrimination between the two events at low pulse height (PH) spectrums becomes ambiguous. This research overviews the applications of using AI in the form of unsupervised machine learning (UML) to create an algorithm to distinguish between neutron and gamma signatures. The UML model is trained on over 200 radioisotope measurements taken at Duke Free Electron Laser Laboratory (FEL) within the soccer ball arrangement. In addition to creating the UML algorithm, the mechanical process of encasing a fissionable 252-Californium (252Cf) sample was completed using CAD software, light reflective simulations, and workshop techniques for coating and assembly. The data was then compared to previous iterations of the sample encasing structure to determine reflective material efficiency and structure. After creating the UML algorithm and processing the collected data from the 252CF encasement , the source is further analyzed using graphical analysis.