Scientists at the Department of Energy’s Argonne National Laboratory in Illinois have developed a machine learning system that could “transform nuclear reactor operations,” according to Interesting Engineering.
What’s going on: Researchers created an ML system to monitor and detect anomalies in a “sodium-cooled fast reactor … a type of nuclear reactor that employs liquid sodium as a coolant for its core … enabl[ing] it to efficiently generate electricity without producing carbon emissions.”
- While SFRs are not in widespread commercial use in the U.S., many experts see them as a path toward a more sustainable energy mix.
The challenge: To prevent corrosion and system clogs, SFR technology requires a high level of purity in its liquid sodium coolant.
The fix: That’s where the Argonne National Laboratory ML model comes in. It continuously monitors the cooling system, analyzing data from 31 sensors at Argonne’s Mechanisms Engineering Test Loop facility.
- The system has successfully detected “operational irregularities quickly and correctly.”
A caveat: However, the model has significant limitations, such as the possibility of false alarms produced by random spikes or sensor inadequacies. “Currently, [it] sends an alert when a spike exceeds a predetermined threshold”—even though not all spikes indicate anomalies.
What’s next: “The team plans to refine the model to distinguish between genuine process anomalies and random measurement noise,” according to a press release from the lab.
The NAM’s take: “Innovation like the kind going on at Argonne is exactly what we need for a secure, sustainable energy future,” said NAM Vice President of Domestic Policy Brandon Farris. “And manufacturers are at the ready to produce the infrastructure needed to power that future.”