Applications of Machine Learning for Enhancing gamma-ray Spectrometry in Nuclear Fusion

Nuclear fusion represents a promising solution to the global energy challenge, offering safe and virtually limitless power. Diagnostics are critical for all life stages of fusion machines, with a core diagnostic being γ-ray spectrometry, used for operation, research, and waste characterisation. This thesis demonstrates machine learning can enhance high-purity germanium γ-ray spectrometry for nuclear fusion applications. Two key applications were identified: reducing the effects of Compton scattering and improving the absolute efficiency calculation process. To achieve these objectives, two novel algorithms that leverage machine learn ing were developed. The first, the Machine Learning-based Compton Suppression Algorithm (MLCSA), employs a convolutional neural network to reduce Compton scattering effects through classification of pulses. The MLCSA reduced the Compton continuum by 97%, outperforming a traditional hardware suppression system by 10%. The second algorithm, the Machine Learning-based Efficiency Calculator (MaLBEC), uses a multilayered perceptron to improve the efficiency calculation process. The MaLBEC predicted absolute efficiency values within 6% and with an improved computational time of 99%, when compared to Monte Carlo N-Particle.