New AI-enhanced tools detect electrical arcing, reduce wildfires

Researchers at the Department of Energy’s Oak Ridge National Laboratory have developed advanced tools to detect abnormal power grid conditions that lead to wildfires, equipment damage and blackouts. The approach incorporates artificial intelligence to rapidly analyze grid data and automatically alert a utility to dangerous grid behaviors requiring immediate response. 

Researchers tested the platform against a small utility data set and are now validating it against five years of field-collected data at Southern California Edison (SCE). The partnership will scale detection algorithms from the lab to the grid, strengthening its reliability and resilience in wildfire-prone regions.

“The faster we realize what’s happening, the faster we can respond,” said Ali Ekti, leader of the ORNL project and the lab’s Grid Communications and Security Group. “This tool is designed to provide utilities with a continuous pathway from signals to analytics to decisions.”

The approach involves an analytics pipeline to identify and characterize seven different types of faults — disturbances that cause abnormal current or voltage. Notably, it detects dangerous arcing faults, which happen when electricity jumps through an air gap between a conductor, like a power line, and a poor conductor, like the ground. Low conductivity prevents any noticeable current increase, allowing electrical arcs to persist undetected by conventional sensors without triggering circuit breakers.

Arcing can ignite wildfires that endanger lives, damage property and infrastructure, and cause cascading outages. Examples include the devastating 2018 Camp Fire in California, which killed 85 people and caused $16.5 billion in damages, according to the National Oceanic and Atmospheric Administration; and the 2023 Maui wildfire, which killed more than 100 people and caused $5.5 billion in damages, according to the U.S. Fire Administration.  

Amplifying waveforms, from the lab to the grid

Although SCE’s digital fault recorders (and soon its smart meters) monitor the grid, today it has no automated way to rapidly process the data, said SCE senior engineer Michael Balestrieri. “We want to know not only if an anomaly has occurred, but exactly what kind it is, so we can take action to prevent a fire or an outage,” he said. “Having more insight into the specific meaning of these signals will allow us to approach issues like arcing with a sense of urgency, so we know when we need to get a crew of first responders on the scene as soon as possible.”

Balestrieri said the framework will also flag signs that equipment might need repair or replacement before overheating or failure causes outages or fires.

The approach begins with applying advanced signal processing to waveform data — visual representations of shifts in voltage, current and frequency on the grid. Arcing faults are too subtle to be visible in waveforms, but ORNL’s AI-assisted algorithms highlight and amplify these signals. In a test using real utility data, waveform signal strength increased from only 6 percent to 72 percent  using the ORNL algorithms, revealing previously hidden grid disturbances.

The platform was trained using waveform data from the Grid Event Signature Library, a web-based repository of over 5,700 waveform signatures hosted online by ORNL for the DOE Office of Electricity.

Researchers validated the system against SCE’s historical outage records, finding strong correlation between detected grid disturbances and real events. The team is building a more robust version trained on SCE data and will test its accuracy, speed and sensitivity on an SCE demonstration circuit. The final step will integrate the detection codes into an internal data analysis platform under development at the utility, Balestrieri said.

“ORNL brings expertise in signal processing and analytics to tackle this very complicated use case,” Balestrieri said. “It takes collaboration from people across research fields to accomplish this.” 

Data analytics tools beyond arcing

Beyond arcing, the ORNL platform can detect and classify overcurrent faults, recloser operations, blown fuses, short-lived faults, capacitor switching, motor starts and line switching.

Other utilities have also expressed interest in the ORNL detection and classification framework. Additionally, sensing companies such as GridVisibility are collaborating with ORNL to explore incorporating its detection algorithms into their grid sensor products. 

The ORNL analytics platform will soon be available in a new open-source “data analytics toolbox” for use with waveforms in the Grid Event Signature Library, Ekti said. The toolbox of statistical signal processing methods and machine learning approaches will be available for free use by utilities and researchers.

Other ORNL staff contributing to the project include Ozgur Alaca, Ali Boyaci, Aaron Wilson and Omer Aziz. The research was supported by the DOE Office of Electricity through the Grid Event Signature Library, Grid Data Analytics, Low-Current Arcing Detection and Wildfire Prevention, and Arcing Detection and Warning for Advanced Resiliency and Electrification for Wildfire Prevention projects.

UT-Battelle manages ORNL for the Department of Energy’s Office of Science, the single largest supporter of basic research in the physical sciences in the United States. The Office of Science is working to address some of the most pressing challenges of our time. For more information, please visit energy.gov/science. — S. Heather Duncan