Gamma spectrometry for forward and reverse problems in radio-isotope identification

Turner, Anthony Nicholas (2021). Gamma spectrometry for forward and reverse problems in radio-isotope identification. University of Birmingham. Ph.D.

Preview

Turner2021PhD.pdf
Text - Accepted Version
Available under License All rights reserved.
Download (6MB) | Preview

Abstract

This work seeks to develop methods for the automated and rapid identification of sets of unknown radioactive sources via gamma-ray spectrometry. The development, testing, and integration of various tools into a simulator enabled high-fidelity modelling of radioactive sources, shielding, and detection systems. Taking full advantage of this simulator, it served to supply a rapid development environment for testing the reverse, where Radio-Isotope IDentification (RIID) algorithms analyse multi-isotope energy spectra to make predictions. Historically, algorithms fail due to unexpected or large perturbations in the profile of these spectra. For example, fluctuations in the detector gain from drifts in temperature or voltage will change the calibration. Transient effects that alter spectra over time can be extremely difficult to explicitly code, so a machine learning approach was taken to address such challenges. A Convolutional Neural Network (CNN) was designed and investigated to meet this need in the RIID problem space. With a representative set of training data, these models were seen to perform extremely well for making predictions on uncalibrated, multi-isotope gamma spectra from NaI detectors perturbed by different source activities, environmental background, low energy thresholds, and extremes in gain shifts. These models were also applied to the more challenging RIID conditions of heavy shielding and close geometries with equally promising results.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):