Pulse shape discrimination in an organic scintillation phoswich detector using machine learning techniques

Yujin Lee,Jinyoung Kim,Byoung-cheol Koh,Young Soo Yoon,Chang Hyon Ha

2024-03-06

Abstract:We developed machine learning algorithms for distinguishing scintillation signals from a plastic-liquid coupled detector known as a phoswich. The challenge lies in discriminating signals from organic scintillators with similar shapes and short decay times. Using a single-readout phoswich detector, we successfully identified $\gamma$ radiation signals from two scintillating components. Our Boosted Decision Tree algorithm demonstrated a maximum discrimination power of 3.02 $\pm$ 0.85 standard deviation in the 950 keV region, providing an efficient solution for self-shielding and enhancing radiation detection capabilities.

Instrumentation and Detectors,Instrumentation and Methods for Astrophysics,High Energy Physics - Experiment

What problem does this paper attempt to address?

This paper aims to solve the problem of how to distinguish signals from two different scintillator materials in an organic scintillator composite detector (phoswich detector). Specifically, the research team has developed machine - learning algorithms, especially the Boosted Decision Tree (BDT) algorithm, to distinguish the pulse shapes generated by the Plastic Scintillator (PS) and the Liquid Scintillator (LS). Since the decay times of these two scintillators are very close (2.1 ns and 4.0 ns respectively) and the pulse shapes are similar, it is a challenging task to distinguish them.

By using a single - read - out composite detector, the research team has successfully identified gamma - ray signals and separated these signals from the two scintillator components. The experimental results show that the maximum resolution ability of the BDT algorithm in the 950 keV energy region reaches 3.02 ± 0.85 standard deviations, which provides an effective solution for self - shielding and improving radiation detection capabilities.

In addition, the research has also explored the performance of other machine - learning methods (such as Deep Neural Network, DNN) in signal separation, but the results show that the BDT method is more suitable for this specific task. Future work will focus on further optimizing the detector performance to achieve the classification of neutrons and gamma - rays.