A methodology for alpha particles identification in liquid scintillation using a cost-efficient Artificial Neural Network

• Published in: Nuclear instruments & methods in physics research. Section A, Accelerators, spectrometers, detectors and associated equipment Vol. 1064; p. 169369
• Main Authors: Carlini, Alessandro, Bobin, Christophe, Paindavoine, Michel, Thevenin, Mathieu
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.07.2024; Elsevier
• Subjects: Alpha; Beta; Discrimination; Instrumentation and Detectors; Nuclear Experiment; Physics; Pulse; Shape; Alpha; Shape; Discrimination; Pulse; Beta
• ISSN: 0168-9002; 1872-9576
• DOI: 10.1016/j.nima.2024.169369

The discrimination of α/β-particles based on Liquid Scintillation (LS) is a measurement technique that can be applied for environmental monitoring. This technique relies on the specific feature of the scintillation process yielding two decay components (fast and delayed) of the photon emission following a particle interaction in a LS sample; indeed, the intensity of the delayed component is higher for α-particles than for β-particles. When operating with a photon detector such as a PhotoMultiplier Tube (PMT), the Pulse-Shape Discrimination (PSD) is the traditional approach, relying on low-pass filtered signals allowing pulse-tail analysis for an α-particle identification. This paper presents a new α-particle detection method based on an Artificial Neural Network (ANN) that directly processes fast pulse trains produced by the delayed scintillation component of a single detector. The ANN is trained using an experimental dataset obtained with a single PMT channel connected in a 3-PMTs detection system from which triple coincidences resulting from α-interactions (241Am) in a LS source are used for the single detector labeling. As we target a real-time implementation, the ANN is designed to minimize the computational resources. The classifier can operate with good performances (91.0% of correct classification, 8.5% false negative and 0.5% false positive) on a single-channel detector. [Display omitted] •Focus on the photons train from a scintillator instead of the integrated pulse shape.•α/β discrimination relying on an ANN which analyses the signal from a single PMT.•Training dataset labelization based on coincidence detection.•The coïncidence detection is performed using a 3-PMT TDCR experimental layout.