Graphing the Neural Networks of Nuclear Reactor O&M: Digital Twin Work at Argonne National Lab
The future of small modular reactor (SMR) nuclear power plants is wide open, which means they could happen or not depending on regulators, financiers and the electricity infrastructure industry.
It’s a big risk but might be the best option to meet expanding future load growth with dependable, flexible power while also reducing carbon emissions.
Automation and artificial intelligence are huge consumers of energy, but they also will play a huge role in managing operations of both SMR and conventional nuclear. The U.S. Department of Energy’s Argonne National Laboratory, based in Illinois, has set its nuclear researchers on developing digital twin technology which can accelerate improvement in efficiency, reliability and safety of nuclear reactors, according to the lab.
Argonne’s expert engineers are deploying advanced computer modeling and AI to predict reactor behavior, thus taking predictive maintenance to a potentially new level of precision.
Digital twin technology essentially creates a virtual model of power plant operations and maintenance, utilizing machine learning to evaluate and predict how power plant equipment will function over operational time periods. The team at Argonne applied their methodology to create digital twins for two types of nuclear reactors: the now-inactive Experimental Breeder Reactor II (EBR-II) and a generic type of fluoride-salt-cooled high-temperature reactor (gFHR).
Argonne is not alone in this work as other researchers examine elements of future advanced reactors. A collaboration between advanced-reactor designer Natura Resources and Abilene Christian University is studying a molten-salt cooling through a future 1-MW on-campus test reactor. Other universities doing SMR research include Texas A&M and the University of Illinois Urbana-Champaign.
At the Argonne National Lab, the core of the digital twin research is graph neural networks (GNNs), a type of AI that processes data as graphs to indicate interconnected components. GNNs are designed to excel at recognizing complex patterns and connections, thus creating what are believed to be highly accurate replicas of real systems.
Argonne contends that its GNN-based digital twins evaluate faster and predict reactor behavior, including modulations in power and cooling systems, more quickly than previously studied.
No SMR has been built in the U.S. yet, but the nation’s Nuclear Regulatory Commission is considering construction permits for projects from Dow and X-energy, NuScale Power and the Tennessee Valley Authority. SMR are envisioned as more efficient, less expensive nuclear power which can offer baseload and carbon-free electricity to meet future load growth from data centers and industrial electrification.
Outside the U.S., Canada, Russia and China are all making inroads on SMR technologies.
