Bachelor Thesis

IMPULSE-2025-00035

http://join2-wiki.gsi.de/foswiki/pub/Main/Artwork/join2_logo100x88.png Application of Machine Learning inEvent-Mode Neutron Imaging

Schmid, D. A. (Corresponding author)MLZ*

2024

Abstract: The application of machine learning and artificial intelligence is a growing field, especially with artificial neural networks (ANN). These tools are helpful for various applications, one of them being the optimization of data processing routines for detectors, from only optimizing classical processing algorithms in speed and efficiency to paving the way to new possibilities. With this perspective in mind, this thesis will apply such ANNs to a novel type of neutron imaging detector: event-mode detection. The goal is to recreate one particular part of the data processing routine. In particular, the reconstructing of scintillator events from scintillation photons. A theoretical overview is provided first to help understand the methods. While trying to recreate the classical algorithm, convolutional neural networks (CNN) were applied. The following performance evaluation of the ANNs did not yield sufficient results to reach the set goal, with several failed attempts and some poorly performing ones. Only a general area of events could be reconstructed by the ANN. The reasons behind the performance are due to the specifics of the chosen approach and necessary simplifications. Therefore, a loose proof of concept is still achieved. Nevertheless, the general idea prevails, and it is suspected that additional changes in the currently tried ANNs or entirely new approaches could solve the problem completely. Thus opening a possibility for further inquiry, in addition to giving a perspective of surpassing the classical approach.

Keyword(s): Instrument and Method Development (1st) ; Instrument and Method Development (2nd)

Note: Bachelorarbeit, TU München, 2024

---

Contributing Institute(s):

1. NECTAR (NECTAR)

Research Program(s):

1. 05K22WO5 - Entwicklung eines Szintillationsdetektors im ns-Bereich mit µm-Auflösung (BMBF-05K22WO5) (BMBF-05K22WO5)

Experiment(s):

1. No specific instrument

---