The project will use AI to help speed up the safety analysis required for new nuclear energy reactors.

Reducing the time and cost to safely analyze the inner workings of a nuclear power plant leading to accelerated deployment is the thrust behind a UNLV College of Engineering project — one of just 278 from across the country — selected for the Genesis Mission, a historic national initiative designed to double America’s scientific productivity.

Led by WooHyun Jung, professor of mechanical engineering at UNLV, the $750,000, nine-month project was chosen for the “Delivering Nuclear Energy that is Faster, Safer, Cheaper” challenge, one of over 20 key research areas identified by the Department of Energy (DOE) to accelerate breakthroughs in energy, discovery science, and national security.

The projects “represent the very best of our nation’s scientific enterprise,” U.S. Secretary of Energy Chris Wright said in a press release.

“The remarkable number of high-quality proposals we received demonstrates that America’s innovation pipeline is strong, and it points to even greater opportunities for future investment and continued expansion of the Genesis Mission portfolio,” Wright added.

In recent years, nuclear energy technology has evolved to incorporate newer and smaller designs, such as advanced small modular reactors (SMR). But, in order to license, build, and operate these nuclear power plants, they need to be verified as safe under a broad range of accident conditions. And that safety analysis — including severe accident simulations for seismic events, tsunamis, fires, and more — is labor intensive, consuming substantial expert time and significantly delaying licensing and deployment schedules for nuclear energy projects. 

Interest in these smaller, modular reactors has grown as electricity demand rises, driven in part by AI data centers. But against the backdrop of disasters like Fukushima and Chernobyl, safety is a top priority for U.S. nuclear regulators.

Simulating Severe Accidents in Nuclear Power Plants

MELCOR is a simulation tool developed by Sandia National Laboratories — one of Jung’s project collaborators — for the Nuclear Regulatory Commission (NRC) to model severe accidents in nuclear power plants for licensing.

To run the MELCOR code simulating real nuclear power plants accurately, detailed specifications including blueprints, reactors, pipings, valves, and exact dimensions are needed. 

“Converting those design documents into thousands of tightly constrained input cards that MELCOR can understand takes more than several months of expert effort,” said Jung. “My target is to cut the model-development process down to a period of several weeks while maintaining the underlying safety requirements.”

Jung’s project introduces modern AI techniques to automate and enhance the preparation of safety and regulatory documentation, enabling rapid exploration of reactor design and licensing processes to reduce cost, schedule, and regulatory uncertainty. 

For the first phase of the project, Jung's team will focus on a single design from NuScale Power — SMR developer and project collaborator — and will develop a LLM-based agentic AI to generate a MELCOR input deck automatically from the design documents, checking the results against a safety analysis Jung’s team has already completed manually for that same design through another DOE program.

“We hope this project will make the nuclear power plant design and licensing processes remarkably efficient and agile, while also saving millions of dollars,” Jung said.

The project is multi-institutional and interdisciplinary, also bringing in Junggab Son, a UNLV computer science professor and LLM expert, Sandia National Laboratories, NuScale Power, and the University of Wisconsin-Madison's nuclear engineering department.

If this first phase proves the concept works, Jung can apply for the Genesis Mission's second phase: a three-year, multimillion-dollar award to expand the tool to more reactor designs.

Project details

Title: LLM-based agentic AI assistant to automatically generate the input deck for severe accident system codes
Lead PI: WooHyun Jung (UNLV)
Collaborators: Junggab Son (UNLV), Juliana Duarte and Ben Lindley (University of Wisconsin, Madison), Brandon Alexander De Luna (Sandia National Laboratories), Kent Welter (NuScale Power)