The Development of Novel Pulse Shape Analysis Algorithms for AGATA



Holloway, Fraser
(2022) The Development of Novel Pulse Shape Analysis Algorithms for AGATA. PhD thesis, University of Liverpool.

[img] Text
201041730_Mar2022.pdf - Unspecified

Download (48MB) | Preview

Abstract

In the field of Nuclear Physics the use of large-scale γ-Ray Tracking (GRT) for arrays like the Advanced Gamma Tracking Array (AGATA) is critical in pushing the envelope of our understanding of the complex phenomena that govern our universe. GRT allows for AGATA to track γ-rays across crystals within the array, allowing for unrivalled Doppler correction and Compton add-back. In order for GRT to function effectively, the interaction position and energy depositions of γ-rays within the array must be effectively determined using Pulse Shape Analysis (PSA). Within AGATA, optimisation-based PSA methods are used to localise γ-ray interactions by comparing experimental detector signals against a simulated basis. A simulated basis has been produced for the A005 AGATA detector crystal, which was used to underpin the development and evaluation of novel PSA methods. Machine Learning was also utilised to perform signal discrimination, compression, correction & regression. Graph-Accelerated k-Nearest Neighbour techniques for PSA were profiled and found to offer significant improvements to execution rate and accuracy. An extensive investigation into the performance of the PSA algorithms with respect to noise level, timeshifting and embedded dimensionality was performed to determine to the most effective algorithm of PSA for AGATA. By utilising the GPU & graph-accelerated algorithm Facebook AI Similarity Search (FAISS) on a principal component analysis reduced 100D embedding, comparable accuracy to the accepted standard was found with an ∼ 43, 000% increase in execution rate. The mathematical framework for the efficient precomputation of the responses of γ-rays that interact multiple times across the crystal (High-Fold) is proposed that should allow the augmentation of Fold-1 kNN search to work on High-Fold with minimal penalty to execution rate. It has also been demonstrated that FAISS can successfully reconstruct a variety of experimental data acquired with AGATA detector crystals.

Item Type: Thesis (PhD)
Divisions: Faculty of Science and Engineering > School of Physical Sciences
Depositing User: Symplectic Admin
Date Deposited: 24 Mar 2022 12:46
Last Modified: 18 Jan 2023 21:10
DOI: 10.17638/03150791
Supervisors:
  • Harkness-Brennan, Laura
  • Kurlin, Vitaliy
URI: https://livrepository.liverpool.ac.uk/id/eprint/3150791