In a groundbreaking study conducted at Purdue University, researchers have made significant strides in utilizing artificial intelligence to enhance the monitoring and control of small modular reactors (SMRs). SMRs are being developed by several countries as a more cost-effective and efficient alternative to traditional nuclear reactors.
Published in Nature’s Scientific Reports, the study focused on employing a machine learning algorithm to predict changes in reactor performance indicators with remarkable accuracy. The algorithm, trained on data from Purdue University Reactor Number One (PUR-1), demonstrated a 99% accuracy rate in forecasting fluctuations in power production stability.
The researchers believe that the implementation of such AI algorithms could revolutionize reactor monitoring and control, leading to improved efficiency and reduced operational costs. The digital instrumentation and controls of PUR-1 make it an ideal testing ground for developing and testing these innovative technologies.
Konstantinos Prantikos, a graduate research assistant at Purdue’s School of Nuclear Engineering and the first author of the study, emphasized the potential of coupling AI systems with nuclear reactors to extract valuable insights for maintenance and optimization. The digital twin system at PUR-1 enables real-time data collection and simulation, allowing researchers to conduct experiments without affecting the actual reactor operation.
The study highlighted the ability of the machine learning algorithm to accurately monitor neutron flux levels in the reactor core. By leveraging a physics model trained on neutron flux data, the algorithm demonstrated an error rate below 1% on average. This seamless integration of learning processes significantly reduces the training time required for developing effective monitoring algorithms.
Collaborations between Purdue University and the U.S. Department of Energy’s Argonne National Laboratory have been instrumental in advancing this research. Through joint appointments and shared facilities, researchers have been able to harness expertise from both institutions to drive innovation in nuclear reactor technology.
Funded by DOE’s Advanced Research Projects Agency-Energy and a generous donation from Goldman Sachs Gives to Purdue’s AI Systems Lab, this research marks a significant step forward in the optimization of SMR performance. With AI-driven monitoring and control systems, the future of nuclear energy production looks promising and economically viable.
For more information:
Writer/Media contact: Kayla Albert, 765-494-2432, wiles5@purdue.edu
Sources have limited availability. Please contact Kayla Albert for interviews.
Reference:
Physics-informed neural network with transfer learning (TL-PINN) based on domain similarity measure for prediction of nuclear reactor transients
Scientific Reports
https://doi.org/10.1038/s41598-023-43325-1