For decades, the United States has sought a permanent disposal solution for spent nuclear fuel and high-level radioactive waste.
Artificial intelligence may be able to help.
Engineering Assistant Professor Ankita Shukla, an expert in this field, is exploring how AI can help scientists reason over decades of scientific and regulatory information used in nuclear waste repository safety assessments. Her research project, Data to Decision, was funded this summer through the Department of Energy’s Genesis Mission, a national initiative to use AI to accelerate scientific discovery.
“What interested me about this problem is that it is not just a data problem,” Shukla said. “It is a reasoning problem. There are decades of scientific studies, models, regulations and technical reports, and experts have to connect all of that information to make and document a safety argument.”
The project brings together researchers with expertise in artificial intelligence, nuclear engineering, geoscience and repository safety from the ·¬ÇÑÊÓƵ; the Desert Research Institute; Sandia National Laboratories; and industry partners.
Shukla’s proposal was one of 278 projects selected from more than 5,000 applications nationwide and announced at the July 22 Genesis Mission Summit in Washington, D.C.
“It was exciting and honestly, a little surreal,” Shukla, who attended the Summit, said. “It was also exciting to see an idea that we developed here at UNR become part of a much larger national effort.”
Professor Shamik Sengupta, head of the College of Engineering’s School of Computer Science, Electrical and Biomedical Engineering, echoed that sentiment.
“Being selected for the inaugural Genesis Mission cohort is a testament to the quality of Dr. Shukla's work,” Sengupta said. “This recognition underscores how Nevada Engineering faculty are contributing innovative solutions to some of the nation’s most critical challenges.”
Yucca Mountain is the ‘ground-truth benchmark’
More than 95,000 metric tons of spent nuclear fuel are stored across 79 sites in more than 30 states, according to the DOE. The Nuclear Waste Policy Act of 1982 directs the DOE to find a geologic repository for permanent disposal of high-level nuclear waste, and a lot of work goes into that.
Experts must make a ‘safety case’ for a potential repository site, Shukla said, bringing together geology, hydrology and waste information, engineering analyses, scientific models, regulations and technical documentation.
Yucca Mountain has been extensively studied as a proposed geologic repository, resulting in decades of scientific, technical and regulatory documentation. This makes Yucca Mountain a “ground-truth benchmark,” Shukla said.
“Yucca Mountain has one of the most extensive scientific and regulatory records developed for a proposed geologic repository in the United States,” Shukla said. “There are decades of site characterization, models, technical analyses and regulatory review. This makes it very useful for us as a benchmark.”
Basically, Shukla’s team can provide the AI with selected underlying scientific and regulatory evidence used in the Yucca Mountain safety assessment and then compare the system’s reasoning and outputs with established expert analyses.
“For an AI project, that is extremely valuable because it gives us a way to test whether the system is actually reasoning from the evidence rather than producing something that just sounds convincing,” Shukla said.
The point is to determine whether AI can identify the relevant evidence for a repository safety-case workflow, reason over that information correctly and document its conclusions so that scientists can review and verify the results.
Shukla is working to demonstrate this during the nine-month Phase 1 of her Data to Decision Genesis Mission project.
“Nine months is a short time for a problem this large, so we are being very focused,” Shukla said. “We are not trying to solve the entire repository problem in Phase 1. We want to show that an AI system can carry out a carefully bounded part of the safety-case workflow in a way that is accurate, source-traceable and useful to experts.”
If the team can demonstrate that Data to Decision can reliably reproduce parts of an established safety-case workflow while keeping its reasoning traceable to the underlying evidence, the framework could eventually be extended to other repository concepts, geologic settings and safety-assessment workflows.
“It’s really about helping scientists and engineers work through the very large amount of information that goes into developing a safety case for long-term nuclear waste disposal,” she said.
