AI goes nuclear: INL expo showcases machine learning and artificial intelligence
Published at
IDAHO FALLS – Artificial intelligence is transforming the way the nuclear industry works, and Idaho National Laboratory is leading the way developing applications to streamline processes while improving safety at nuclear power plants.
INL scientists showcased 15 projects on Artificial Intelligence (AI) and Machine Learning at an expo at the Energy Innovation Laboratory in Idaho Falls on Tuesday.
“We’re here to learn about some of the incredible science happening related to artificial intelligence and machine learning,” said Katya Le Blanc, human factors scientist at Idaho National Laboratory. “We’re also developing technologies that can eventually be deployed by the nuclear industry and be used by nuclear utilities.”
According to a lab news release, “computers that mimic cognitive functions and apply advanced algorithms can help researchers analyze and solve a variety of complex technical challenges. This new approach helps everything from improving materials design for advanced reactors to enhancing nuclear power plant control rooms so they become more effective and efficient.”
Technologies on display at the conference included RAVEN – a Risk Analysis Virtual ENvironment that provides “an open source multi-purpose framework for machine learning, artificial intelligence and digital twinning.”
One machine learning technology called Inspection Portal is part of the light water reactor sustainability program that analyzes and aggregates data from human-submitted reports to identify trends and help optimize the operation of nuclear power plants.
The program’s machine learning operation is trained on millions of records from across the industry.
“We can do things here at the INL that no one else can do,” said Brian Wilcken, nuclear science and technology data scientist. “… Utility companies try to do things like this. They can’t touch it. … We have so much data we can train extremely powerful models.”
Other AI systems provide image detection to read gauges, pinpoint anomalies and determine if a valve has been turned, if a screw is corroded or if a fire breaks out in a nuclear plant. These advancements could reduce the need for personnel to perform menial checks at a nuclear power plant and free up manpower for higher-level work and applications.
Additional tools “evaluate the economics of different energy mixes and how to analyze the best cost-benefit and other factors (such as) the reliability associated with energy systems,” Le Blanc said.
These systems can determine the proper output needed from “a nuclear power plant, a hydro plant and a solar facility in order to meet people’s demand for electricity when they need it — while optimizing economic benefit as well,” she said.
Some of the applications utilize existing AI programs, while others were created in-house at Idaho National Laboratory.
“Sometimes, it requires that you develop it. There’s not a model that can do what you need it to do, but sometimes there’s something that already exists that you can adapt,” Le Blanc said. “It varies depending on (the situation), but there’s no reason to start from scratch.”
The Artificial Intelligence and Machine Learning Expo is in its second year.
In the future, organizers hope to expand and collaborate with other experts in the AI space to further share the research occurring at Idaho National Laboratory.
“I read a lot of papers inside scientific journals related to AI,” Le Blanc said. “Seeing how this stuff actually works, being able to mess around with it, play with it, talk to the researchers, … see what they’re doing and get direct access and ask them questions — that’s just exciting!”
