For decades, researchers have been pursuing fusion energy, an as-yet-unrealized energy dream that could provide massive amounts of clean power. However, fusion plasma is fickle, and disruptions decrease efficiency and render the nuclear reactions unsuitable for power production at scale. Now, researchers at the DIII-D National Fusion Facility have broken new ground by using machine learning to prevent those disruptions in real time.
DIII-D, which is operated by General Atomics, is the largest fusion research facility in the United States. It hosts a “tokamak”: a magnetic, donut-shaped device that is able to hold a fusion reaction in place and prevent it from interacting with surrounding materials (and thus losing heat). In order to increase the efficiency of the reaction, operators stretch the plasma shape vertically – however, this can lead to instabilities in the plasma, causing it to touch the tokamak.
Typically, researchers aim to shut down the reaction just before those instabilities occur. Aiming to render such shutdowns unnecessary, the researchers at DIII-D developed a neural network to estimate the change in instability millisecond-to-millisecond. After training the neural network using General Atomics’ own TokSearch data mining tool and decades’ worth of experimental data from DIII-D, they then used the resulting estimates to constantly adjust the plasma such that it approached – but never exceeded – its stability limit. The researchers also trained the network to err on the side of caution if it detected high uncertainty. After testing, they found that the machine learning-based approach operated a hundred times faster than traditional methods.
“These experiments are quite significant, because they illustrate why the fusion community has been so excited about machine learning,” said DIII-D Director David Hill in a news release from General Atomics. “Although DIII-D has applied machine learning to real-time prediction of instabilities for decades, actual real-time control to prevent disruption using these massive data sets is very novel and exciting.”
The researchers hope that these new approaches will enable critical developments in the path toward scalable fusion energy, putting decades’ worth of data to work.
“We’ve been collecting experimental data at DIII-D since the ’80s, but only recently have we been able to really take advantage of all that data by using modern software and computing hardware,” said Brian Sammuli, co-developer of the machine learning technique. “The answers to some of the hard problems in fusion are just sitting there in the data, waiting to be discovered. We’re now starting to be able to use modern machine-learning techniques to augment our physics understanding, and this allows us to control the plasma more effectively.”
Header image: Brian Sammuli and Jayson Hill, co-developers of the machine learning technique. Image courtesy of General Atomics.