Verification Of Triso Fuel Burnup Using Machine Learning Algorithms

Verification of TRISO fuel burnup using Machine Learning algorithmsCarlos Soto, Odera Dim, Lap-Yan Cheng, Yonggang Cui, Maia Gemmill, Thomas Grice, Joseph Rivers, Warren Stern, and Michael TodosowBrookhaven National LaboratoryAbstractPebble Bed Reactors (PBRs) are fueled with pebbles that are circulated multiple times through the reactor vessel before discharge. This tennis size fuel is formed from TRIstructural ISOtropic (TRISO) fuel particles, which are composed of a minute fuel kernel of uranium dioxide or uranium oxy-carbide surrounded by multiple layers of a highly durable and impermeable coating. Typically, ejected pebbles are returned to the PBR or discharged depending on the fuel burnup and physical condition. Radiation signatures emitted from fission products accumulated in the TRISO particle kernel are used to quantify burnup. Previous research has shown that measurements of fission products, such as 134Cs, 137Cs, 154Eu, etc., can be applied independently or in combination to predict the level of burnup in the fuel. A simple criterion for selecting an isotope for burnup indication is the exhibition of a strong gamma photopeak. However, it remains challenging to measure this complex source due to self-shielding effects, strong radiation background and intervening materials. Another challenge in this measurement is the required high throughput of fuel pebbles in the core over the nuclear reactors’ operational life cycle. Accommodating this throughput necessitates limited measurement time and thus impacts detection efficiency. A high-performing system is therefore required to swiftly and accurately identify patterns in the time-constrained gamma spectrum measurements. Machine Learning (ML) has achieved widespread success in pattern recognition and analysis of varied data types. In particular, modern deep learning approaches have the ability to learn useful features from the raw data via deep neural network architectures. In this work, we apply three proven ML approaches – multilayer perceptrons (MLPs), convolutional neural networks (CNNs), and transformers – to the task of predicting fuel burnup from measured gamma spectra, and we compile a dataset of simulated spectra for training and validation of the ML model. We are using ML method to interpret gamma-ray spectra and predict the burnup values of the pebbles.