Reactor Core Optimization is a constantly advancing field of study within the field of nuclear engineering. Meta-Heuristic algorithms such as Simulated Annealing and Genetic Algorithms have become staples of reactor core optimization. As the need for reactor core optimization increases, new optimization techniques are developed to fulfill demand. Parallel Simulated Annealing (PSA) is a still largely unexplored form of meta heuristic. By implementing parallel simulated annealing into reactor core optimization problems, we can start to analyze the effectiveness of the algorithm, and we can start to see where the algorithm stands in reactor core optimization.

Fault Diagnostic Approach for Centrifugal Pump Bearings using Deep Neural Networks

Over the years, the amount and quality of data extracted from industrial systems has increased due to the advancement of sensing technologies. This data can be used by Machine Learning Models (MLMs) for diagnostics to recover valuable information about the state of a system. These models are an integral part of the predictive maintenance (PdM) framework that uses real time sensor measurements to support maintenance operations for advanced reactors by optimizing the type and frequency of maintenance actions to decrease overall maintenance cost. In this study, fault characterization associated with centrifugal pump bearings were investigated, with more than 40% of motor failures being attributed to bearing faults. Since publicly available measurement data of such faults are scarce, a synthetic data generation procedure was created. To perform the diagnostics, the Convolutional Neural Network (CNN) architecture was leveraged to attain a 96.28% characterization accuracy on measurement data when trained on synthetic data; an accuracy that shows the synthetic data representing the measurement data well. The PdM framework can be beneficial to reactor system maintenance teams by providing diagnostics when appropriate to decrease overall maintenance costs, and the synthetic data generation procedure can be leveraged to build an initial database that can be continuously improved once reactor system specific measurement data is available.

Correlating the Radiative Emission of PBR Fuel to Burnup with Machine Learning

A unique feature of Pebble-Bed Reactor (PBR) designs is the opportunity for continuous multi-pass fuel circulation and reloading. An accurate knowledge of the fuel exposure history is necessary to inform the reloading decision during continuous fuel circulation in a pebble-bed High Temperature Gas Reactor (HTGR) to ensure that no fuel pebble will exceed the exposure limit following reinsertion. Additionally, it enables the estimation of the composition of the pebble fuel element and can serve both as a non-invasive method for pebble identification and a protection against proliferation concerns. The objective of this work was to develop a Machine Learning (ML) model to accurately estimate the fuel exposure of an ejected fuel pebble based on gamma emission detection data. The developed ML model identifies and leverages the multivariate correlations between the radiative emissions of a HTGR Tri-structural Isotropic (TRISO) Fuel Pebble and the exposure, or burnup, of the irradiated pebble. The modelling of the TRISO Fuel Pebble in Serpent 2 for pebble irradiation and gamma emissions detection is presented. The generation of preliminary training data and ML training performance evaluations are discussed in this work.