There is increased interest in using AI to increase the efficiency of nuclear power plants, which is being fueled in part by the data centers’ insatiable demand for AI.
Senior nuclear engineer Richard Vilim of the Energy Department’s Argonne National Laboratory, located in Lemont, Illinois, and renowned for its work on nuclear reactors, said the lab has created an AI-based tool that can help with reactor design and assist operators in operating nuclear plants.
In a discipline that peaked in the last part of the 20th century, Argonne’s tool, known as the Parameter-Free Reasoning Operator for Automated Identification and Diagnosis, or PRO-AID, represents a technological advance.
According to Vilim, “the nuclear plants are kind of dinosaurs when it comes to technology because they were built more than 30 years ago.”
Nearly 20% of the electricity produced in the United States is still produced by the 94 nuclear reactors that are currently in operation, all of which have had their licenses extended. The U.S. Energy Information Administration estimates that they are about 42 years old on average.
Although Argonne intends to provide PRO-AID to new, technologically advanced nuclear projects, it is also keeping an eye on the so-called dinosaurs, some of which are being brought back to life by corporations such as Microsoft and Amazon to power their AI data centers. The International Energy Agency warned Thursday that the global push for AI is set to drive a dramatic increase in the demand for power, with data center consumption predicted to more than double by the end of the decade.
The Wall Street Journal has revealed that the owners of about one-third of the nuclear reactors in the United States are in negotiations with tech corporations to supply electricity for those data centers.
When something appears wrong at a plant, PRO-AID uses generative AI in conjunction with large language models to do real-time monitoring and diagnostics, alerting and guiding staff. According to Vilim, it also adopts a type of automated reasoning that mimics how a human operator would ask questions and learn about how the plant is running by encoding knowledge in AI systems using mathematical logic.
Additionally, the technology can help increase the productivity of the staff required to run a nuclear facility, according to Vilim. As more senior workers retire, that becomes even more crucial. When someone retires, you don’t have to replace them if we can transfer some of these lower-level functions to a machine, he said.
The technology at nuclear reactors lags well behind that of gas-powered plants, which are more modern and automated with digital monitoring systems, Vilim continued. As is now the case with gas-powered facilities, a utility’s nuclear reactors’ monitoring staff will be consolidated in one central place, which will contribute to the efficiency of updating technology.
“It’s not always clear that the technology is worth the cost for legacy nuclear plants,” Vilim remarked. Although it can be licensed to a nuclear supplier or software developer, PRO-AID has not yet been implemented in a commercial nuclear facility.
Some power plants may wish to improve their technology, but it can be difficult to remove power from the grid for a long time to do so, according to Bob Johnson, an analyst with the market research and IT consulting firm Gartner.
“The utilities must decide if there is enough benefit to adding this. Or do we just get to the finish line, which could be in 20 years, with what we have? Vilim said.
According to the company, TerraPower, one of the more recent nuclear enterprises created by Bill Gates, has been designing its reactor technology since the early 2000s using sophisticated computer modeling. “From conception to commercialization, TerraPower’s Natrium reactor will be the first to be designed and modeled in a completely digital environment,” stated Chris Levesque, president and CEO of the firm.
AI has been used by Oklo, a nuclear firm financed by Sam Altman, to analyze reactor designs. According to Jacob DeWitte, co-founder and CEO of Oklo, those technologies can assist cut down on the time required to run high-fidelity simulation situations. However, he noted that the full integration of AI into commercial nuclear power will take time. It represents a significant productivity boost. He believes that nuclear is still very early in this process.
