Machine Learning Based Efficiency Calculations (MaLBEC) for Improvements to Nuclear Fusion Diagnostics
Diagnostics are critical on the path to commercial fusion machines, since the ability to understand and measure plasma features is important to sustaining fusion reactions. Gamma spectroscopy is a technique in experimental nuclear physics that can be used to aid fusion diagnostics, including providing information in neutron activation analyses to calculate neutron flux and therefore fusion power. In gamma spectrometry, absolute efficiency refers to the ratio of the total number of photons detected by a detector to the number emitted by a radioactive sample, and is dependent on the sample geometry, photon energy, and the sample to detector distance. The efficiency values form part of the activity calculation when measuring nuclear dosimetry reactions used in fusion diagnostics, hence it is imperative that they can be calculated efficiently and accurately. Traditional methods for calculating efficiency values typically require specialist software and training, computing power, and can be time consuming. We present the groundwork to a digital efficiency calculation algorithm, the Machine Learning Based Efficiency Calculation (MaLBEC) algorithm, that uses state-of-the-art machine learning techniques to calculate efficiency values based on sample geometries. The MaLBEC algorithm utilises regression machine learning techniques, where the training data were acquired using a radiation transport code (MCNP) to simulate thousands of potential geometry sizes and positions relative to a detector. To obtain predicted efficiency values for a new geometry, the machine learning approach only requires four input parameters pertaining to the geometry. This novel efficiency calculation algorithm may reduce the computational and physical complexity of creating models for new sample geometries, compared to traditional methods. The algorithm is under development and once completed it will be validated and compared to existing methods using a High-Purity Germanium (HPGe) detector, with a range of fusion-relevant samples containing radionuclides such as Cobalt-60. The MaLBEC algorithm could be trained on several detector geometries and encompass a wide variety of sample types, and so has the potential to be detector agnostic, which will further increase its applicability to other systems. This will have positive implications on any area that uses gamma spectroscopy.