Recent quantum leaps in the use of artificial intelligence (AI) have paved a way for many applications across the nuclear industry. Nuclear reactor experimental designs are often a balance between competing objectives or goals, and utilizing AI can help put all of those goals into balance.
A research team including doctoral candidate John Pevey, Research Assistant Professor Ondrej Chvala, Assistant Professor Vladimir Sobes, and Department Head, Postelle Professor, Chancellor’s Professor Wes Hines, and graduate student Cameron Salyer explored an evolutionary artificial intelligence algorithm to inform the design of the new subcritical experimental facility, the Fast Neutron Source (FNS), located at the University of Tennessee. Their results can be found in the Annals of Nuclear Energy.
“Using AI offers us a chance to remove biases and open up our thinking to new and better ways to design reactors,” said Hines. “A computer can help us get advanced reactor designs that no one else has thought of, and we’re excited to be able to share this technology with our students.”
The FNS is a flexible-accelerator-driven, fast-spectrum machine, which will provide not just education for students and faculty, but also measurements essential for optimal design of the next generation of advanced nuclear reactors. It will eventually be housed in the newly opened Zeanah Engineering Complex.
The FNS is a unique platform for building targeted integral nuclear cross-section measurement experiments to reduce nuclear data uncertainty on future reactor designs. Using artificial intelligence will help design machine configurations that are best suited to advance particular advanced nuclear reactor design concepts.
The team applied a genetic algorithm using non-dominated sorting, elitism, and crowding distance heuristics to explore the design space of a simplified 3D approximation of the FNS. The data presented correlations in the final suite of designs, which will be the basis of heuristics for more complex optimizations and geometries in the future. Their analysis shows that a multi-objective genetic algorithm can be used to design nuclear facilities with the dual objective of spectrum matching and flux maximization.