AI can now forecasts nuclear fusion reactor instabilities in real time

These are caused by the interaction of current and pressure gradients in the plasma, leading to magnetic islands that disrupt confinement.

While nuclear fusion shows great promise, significant technical and engineering challenges remain to be overcome before fusion power can be realized on a commercial scale. 

One of the challenges being that the plasma inside a fusion reactor can lose stability. Plasma instability can lead to disruptions, which are sudden and large-scale events that can cause the loss of confinement and termination of the fusion process.

“Tearing mode instabilities are one of the major causes of plasma disruption, and they will become even more prominent as we try to run fusion reactions at the high powers required to produce enough energy,” said Jaemin Seo, first author of the study. “They are an important challenge for us to solve.”

At the DIII-D National Fusion Facility in San Diego, scientists showed that their AI model, trained using old data, could predict tearing mode instabilities up to 300 milliseconds beforehand. 

That’s 30 percent of a second, just how long a person’s takes to read another person’s expression. But that’s plenty for the AI to adjust how the reactor operates. 

Their experimental study has the potential to help prevent damage to the magnetic field lines in the plasma, which could stop the reaction.

“Previous studies have generally focused on either suppressing or mitigating the effects of these tearing instabilities after they occur in the plasma,” added Seo. “But our approach allows us to predict and avoid those instabilities before they ever appear.”

A fusion of AI and plasma physics

But creating an AI tool turned out to be as difficult as teaching someone to fly a plane. Using past data from the DIII-D tokamak, they constructed a deep neural network to predict future instabilities and trained a reinforcement learning algorithm to control the plasma. 

Through simulated experiments, the AI learned optimal strategies for maintaining high power while avoiding instabilities. Once refined, the AI controller successfully prevented instabilities during a real fusion experiment by adjusting tokamak parameters in real time. 

This proactive approach contrasts with current methods, which react to instabilities as they occur.

“Being able to predict instabilities ahead of time can make it easier to run these reactions than current approaches, which are more passive,” said SangKyeun Kim, co-author of the study. “We no longer have to wait for the instabilities to occur and then take quick corrective action before the plasma becomes disrupted.”

“We have strong evidence that the controller works quite well at DIII-D, but we need more data to show that it can work in a number of different situations,” said first author Seo. “We want to work toward something more universal.”

The study was published in Nature.