Innovative Artificial Intelligence and Machine Learning Applications at SRNL
“Models are needed for AI to ingest and process cyber data. They live, eat and sleep… on data.” – Glenn Fink
Illustration by Alphonzo James and Elyssa Burgess, SRNL.
AI Changes the Landscape
The previous century saw a multitude of technological revolutions that changed the way we all experience the world – how we live, work, travel and even defend our nation. Among these were the splitting of the atom, the invention of the transistor and microchip, the development of vaccines, and humans and satellites in space. All of these advancements were accomplished thanks to the magic of human intelligence.
Human intelligence, however, has developed sophisticated artificial intelligence to augment the mind and, in some cases, fully take its place. While rudimentary AI goes back several decades, the last few years have seen a true revolution in this technology. AI platforms, such as ChatGPT, Microsoft Azure, Google Cloud and several others, have permeated our society. The average person can use these tools to accomplish everything from constructing an email to writing music. In addition, proprietary AI platforms have been developed and deployed by private companies to greatly increase efficiency and the speed of product invention and improvement.
It’s critical that AI be used also to defend our nation and solve its biggest problems, and Savannah River National Laboratory has stepped up to meet the challenge. SRNL has recognized the value of AI, as well as related machine learning, and is using the latest in AI/ML capabilities to keep the U.S. on the forefront of discovery. In doing so, the lab is answering the call from the Secretary of Energy to prioritize AI/ML development and lead the world in this developing technology.
AI has been incorporated into all manner of relevant projects and use cases throughout the lab. All three SRNL mission areas – national security, environmental stewardship and energy resilience – are making use of AI/ML tools. Each of the lab’s three directorates are contributing to this revolution.
Leveraging AI/ML for Environmental Stewardship
SRNL’s Environmental Management and Legacy Management directorate is employing AI to solve challenging environmental problems. This includes the AI Accelerated Strategies and Solutions in Environmental Technology, or AI-ASSET, initiative. AI-ASSET is based on an existing project known as the Advanced Long-Term Environmental Monitoring Systems, or ALTEMIS, which has been used to predict the behavior and movement of contaminated groundwater plumes. Sensors embedded around the plume use geochemical, hydrological and geophysical data to passively monitor and predict contaminant movement. In doing so, long-term monitoring, which is a substantial portion of the GAO-estimated $550 billion liability for the DOE-EM cleanup mission, can be performed at a fraction of the cost.
ALTEMIS, which SRNL developed over the course of about fifteen years, has been used at a contaminated groundwater plume at Savannah River Site. AI-ASSET uses AI/ML technology to make ALTEMIS usable at other contaminated sites. These sites have their own specific environmental conditions and unique contaminant data, which creates applicability challenges.
“What we want to do is shorten that time span such that a site could get ALTEMIS up and running within a year, if not under a year,” said SRNL Scientist Tom Danielson. To achieve this, AI is being brought to bear. AI/ML tools such as generative AI and large language models are being used to tailor the approach on a site-specific basis.
“We are leveraging many existing models in our approach,” Danielson continued. “And it’s our responsibility as a national lab to determine which models can be deployed in, and then improved for, AI-ASSET. So, we are using existing models but being very innovative in how we put it all together.”
The AI-ASSET team hopes to deploy the system at the Moab site in Utah and expand into different sites around the Department of Energy Complex. The AI/ML software being developed by SRNL Scientist Alejandro De La Noval can hopefully transition the technology from groundwater projects to monitoring contamination at physical facilities themselves. This includes a building with legacy mercury contamination at Oak Ridge National Laboratory. International sites are also targets for AI-ASSET deployment in the near future, making it applicable across the globe.
AI/ML to Help Efficiently Dispose of Excess Nuclear Material
The Weapons Production Technology directorate is using a novel AI/ML approach in the disposal of some of the nation’s legacy plutonium surplus. SRNL engineers Corey Hopper and Nicolas Issa are operationalizing the concept.
The plutonium is first downblended to eliminate its potential misuse. Then it must be packaged safely and effectively before it can be transported to permanent storage. An outer storage container used for the downblended plutonium is called a Criticality Control Overpack, or CCO, drum and the inner container is called a Criticality Control Container, or CCC. Then, a canister inside the CCC holds the material proper.
“You can think of it like a Russian nesting doll,” said Issa. “In that there are three layers to the ultimate container.”
The CCO drums are received onsite and must be inspected for quality control and security. When approved, a tag is placed on the container signifying the drum is safe for use. Approved CCOs are transported to a holding pad until they are ready for packing.
Nicolas Issa stands next to a CCO drum and robotic arm in an SRNL engineering research lab. (Photo: LJ Gay, SRNS)
The inspection and readiness activities are tedious and time consuming when done manually. Thus, ML techniques are used to greatly improve the speed and accuracy of the process. A robotic arm, cameras and laser scanners open and manipulate the drum to look for anomalies. The automation learns exactly how the drum should look, and what defects and damage should look like, before approval.
“The first ML algorithm is an anomaly detection algorithm that takes a picture of the inside of the CCO and compares it to what it knows is a good CCO,” said Issa. “If the CCO inspection goes well, it does the same thing to the CCC, which should be empty at that point. The next ML algorithm is called OCR, Optical Character Recognition model, and it reads the date, model number, serial number and the like on the CCC to verify it’s all as it should be.” From there, the containers are ready for use to dispose of the nuclear material.
A Role for AI/ML in Cybersecurity
SRNL’s Global Security Directorate is also rapidly moving forward with AI/ML technologies. The development of secure, AI-enabled cybersecurity in the national interest is the focus of a major initiative at the lab. The project is known as the Threat Hunting Representations for Embedded-system Anomaly Tracking, or THREAT, effort. THREAT’s focus is cybersecurity associated with energy and nuclear infrastructure. It is supported by an integration of investments in high-performance computing infrastructure, workforce development and key external stakeholder collaboration.
Work is being conducted via partnerships with Sandia National Laboratories, Idaho National Laboratory and several universities in the southeast. In late 2024 SRNL deployed the lab’s first Graphics Processing Unit, or GPU computer, which is named Raptor. It will soon be moved to SRNL’s new Advanced Manufacturing Collaborative facility. This unique computer is being used to develop AI models for cybersecurity. It will play a pivotal role in enabling large language model research.
“Models are needed for AI to ingest and process cyber data,” said Glenn Fink, SRNL researcher. “They live, eat and sleep, so to speak, on data. Raptor gives us a powerful tool to process those data with GPU-based computing.”
While Raptor can develop the necessary models and store the data, it is being utilized also to train researchers on AI-enabled skills. Thus, the THREAT project is preparing a cybersecurity workforce that is able to leverage AI.
“We need people and computers that can detect data anomalies caused by criminals, nation states and other bad actors that could affect our electric power grid,” said Fink. “We are using large language models to ingest cyber data the same way ChatGPT has ingested trillions of words of human languages.”
In doing so, SRNL can significantly advance cybersecurity to serve the nation.
Tom Danielson (top) and Alejandro De La Noval are creating software that uses AI/ML to monitor environmental contamination. (Photo: LJ Gay, SRNS)
AI/ML at SRNL in the Future
The projects described above are highlights from a broad swath of AI/ML applications across the lab. SRNL is rapidly bringing AI/ML to bear, but it’s still in its infancy. The lab will relentlessly continue to determine how and where this technology can best be used to serve its mission areas.
While private citizens use AI tools for everyday tasks, they can rest assured that SRNL is on the leading edge of AI/ML applications to keep them safe, protect the environment and enhance their lives.