AI for operations: In a new paper publicized on July 26, researchers at Argonne described how nuclear power plant operators could benefit from diagnostic information filtered through an AI large language model such as GPT-4. The research, funded by the DOE’s Office of Nuclear Energy, saw engineers combine an Argonne diagnostic tool called PRO-AID with a “symbolic engine” that acts as an intermediary between PRO-AID and a large language model to help the model interpret the data accurately.
PRO-AID compares real-time plant data to expected normal behaviors in physics-based models to identify faults. When there’s a mismatch, PRO-AID provides a probabilistic distribution of faults linked to the mismatches. The symbolic engine supplies a structured representation of the fault reasoning process, and this “constrains the output space for the LLM,” ensuring it only provides explanations based on the data and models and avoids producing so-called hallucinations.
The system was tested at Argonne’s Mechanisms Engineering Test Loop Facility (METL), where it diagnosed a faulty sensor in the nonnuclear liquid sodium test facility and explained the issue to operators.
“The system has the potential to enhance the training of our nuclear workforce and streamline operations and maintenance tasks,” said Rick Vilim, manager of the Plant Analysis and Control and Sensors department at Argonne.
Roadmap: The potential of AI in energy and nuclear science and technology goes well beyond reactor operations. On July 16, the DOE announced a road map for its Frontiers in Artificial Intelligence for Science, Security, and Technology (FASST) initiative to “build the world's most powerful integrated scientific AI systems for science, energy, and national security.”
That followed an April announcement of new DOE actions to accelerate AI applications. As part of that broader announcement, the department issued AI and Energy: Opportunities for a Modern Grid and Clean Energy Economy, a report on AI’s near-term potential to support the growth of America’s clean-energy economy, and Advanced Research Directions on AI For Energy, a report developed by Argonne in partnership with researchers from other national laboratories on long-term opportunities and challenges in AI.
As the hosts of some of the world’s most advanced supercomputers and AI systems, DOE national laboratories will have a role to play in AI applications outside of energy, too. The Argonne Leadership Computing Facility is providing access to a suite of AI machines dubbed the ALCF AI Testbed. That testbed and the Oak Ridge Leadership Computing Facility’s Summit supercomputer at Oak Ridge National Laboratory are both available to researchers through the National Artificial Intelligence Research Resource (NAIRR) Pilot, a program launched by the National Science Foundation in partnership with the DOE in January. The NAIRR pilot also offers computational and data resources from other federal agencies and nongovernmental partners, including major tech companies.
FASST facts: FASST goals include transforming the extensive classified and unclassified scientific data produced at DOE user facilities to make it AI-ready and building next-generation supercomputers to analyze the data.
A fact sheet frames it this way: “Certain strategic areas of the U.S. government's artificial intelligence capabilities currently lag industry, while foreign adversaries are investing in Al at scale. If U.S. government leadership is not rapidly established in this sector, the nation risks falling behind in the development of safe and trustworthy Al for national security, energy, and scientific discovery, and thereby compromising our ability to address pressing national and global challenges.”
A visit to a searchable portal listing the DOE’s existing AI tools and partnerships turned up 15 projects related to nuclear science and technology, including these:
.png)
