Deep Learning AI Code for Successful Prediction of Disruptions that Occur in Fusion Reactors

For many years, researchers have been making efforts to generate clean, unlimited energy by re-creating on Earth the conditions found at the center of the sun. However, for nuclear fusion to be practical for generating electricity, it is necessary for that enormous, fierce power to be controlled.

“When the plasma in a fusion experiment becomes unstable, it can escape confinement and touch the wall of the machine, causing severe damage and sometimes even melting or vaporizing components,” stated Julian Kates-Harbeck, a physics PhD student and a Department of Energy (DOE) Computational Science Graduate Fellow. “If you could predict those escapes, or ‘disruptions,’ you could mitigate their effects and build in safety protocols that would cool the plasma down gently and keep it from damaging the machine.”

In a new study reported in Nature and guided by the U.S. DOE’s Princeton Plasma Physics Laboratory (PPPL), Kates-Harbeck and his team developed a “deep learning” artificial intelligence (AI) code to successfully predict disruptions in fusion reactors. The predictions of the new technique can be applied over different types of machines, rendering it an important step forward for international fusion energy initiatives.

The Harder They Come, the Harder They Fall

In the sun, lighter elements fuse into heavier ones as extremely hot plasma, thereby generating energy. Researchers use a tokamak to re-create fusion: a donut-shaped, building-sized machine that includes hot plasma using a powerful magnetic field.

Some of the biggest nuclear fusion machines in the field—hundreds of tons of solid steel—can jump a centimeter up in the air when a disruption happens. That gives you an idea how much power is released. You really don’t want this to happen.

Julian Kates-Harbeck, Physics PhD Student, Harvard University

It is crucial to predict the disruption since a bigger machine relates to a bigger disruption. It is expected that the disruptions will be severe in the $25 billion Iter tokamak being constructed in France: The volume and energy of the machine is over eight times that of the Joint European Torus (JET), the next largest magnetically confined plasma physics experiment in operation, as well as less surface area to capture it.

We don’t have good strategies for completely avoiding these disruption events yet. The best thing we can do is to predict when they are going to happen so we can avoid most of their adverse effects. That might be, for example, by injecting neutral gas that cools the plasma before it smashes into the wall. But you can’t mitigate anything if you don’t know it’s coming.

Julian Kates-Harbeck, Physics PhD Student, Harvard University

Summer Research

Through his DOE fellowship, Kates-Harbeck worked at Princeton for a summer, where he teamed up with Bill Tang, principal research physicist at PPPL, professor in the Department of Astrophysical Sciences at Princeton, and the senior author on the study. Tang was only beginning a new study using machine learning to overcome disruption prediction. It was the ideal combination for Kates-Harbeck.

“My background is in physics and AI, so being able to combine both while working on a problem as meaningful as fusion energy—a topic I had always wanted to work on—was like hitting the jackpot,” he stated. “People have been using classical machine learning on disruption prediction for years, but I’d just come out of my [computer science] master’s program and got really excited about applying deep learning to the problem. Bill was very open and supportive of the idea, and that’s how we got started.”

According to Tang, artificial intelligence is “the most intriguing area of scientific growth right now,” and “to marry it to fusion science is very exciting. We’ve accelerated the ability to predict with high accuracy the most dangerous challenge to clean fusion energy.”

Deep-Learning AI

It is necessary for machine learning to be trained to learn. The researchers trained their code using massive, diverse streams of measurement data from previous experiments. Their new algorithm, called the Fusion Recurrent Neural Network (FRNN), looks for patterns in the data that occur before a disruption. FRNN studies these patterns, thereby enabling it to make disruption predictions.

Although this is not the first research to use AI for disruption prediction, FRNN was based on deep learning, making a big difference. Deep learning is a unique field in AI with the ability to handle considerably more complexity compared to other techniques.

Classical machine learning is good at analyzing simple sources of data, such as the average density or current in the plasma. But we have access to data that is much more complex, like the temperature of electrons as a function of radius in the plasma. That’s high-dimensional data, and you need deep learning to make sense of it.

Julian Kates-Harbeck, Physics PhD Student, Harvard University

High-dimensional data might include considerably more information related to what is happening in the plasma. This is precisely what enables the deep-learning algorithm to make predictions so much better compared to other AI techniques.

When provided with adequate data, deep learning possesses immense generalization power; in other words, what it learns from one structure can be used on another. Generalization is crucial for fusion prediction, in which disruptions become worse in larger machines.

Scaling Up

The Iter nuclear fusion experiment is a collaboration between 35 nations. Researchers believe that in the decades to come, the plasmas in the reactor will travel well over the energy break-even point—the moment when it liberates at least the energy required to heat it.

At 150 million degrees Celsius and 23,000 tons, “Iter is going to be so big and powerful that it can’t afford more than a handful of disruptions in its entire lifetime,” stated Kates-Harbeck. “But you need thousands of disruptions to train a machine-learning algorithm. That’s why we need to train on one machine that can afford the disruptions, and test the predictor on another.”

Two facilities offered the data for the study: the DIII-D National Fusion Facility, operated by General Atomics for the DOE in California; and JET in the United Kingdom, managed by the EUROfusion consortium. The researchers demonstrated that the FRNN code trained on the smaller DIII-D could generalize to the larger, considerably more powerful JET.

This holds great promise for the potential of the code to forecast disruptions on Iter.

This research opens a promising new chapter in the effort to bring unlimited energy to Earth. Artificial intelligence is exploding across the sciences and now it’s beginning to contribute to the worldwide quest for fusion power.

Steve Cowley, Director, PPPL

“This has been a really exciting first step,” stated Kates-Harbeck, “but of course simply mitigating an oncoming disruption is a rather crude approach. Next, we’ll need to shift focus from predicting disruptions to actively controlling the plasma in the tokamak. The ultimate goal is to learn how to adjust the plasma to avoid disruptions in the first place.”

This study was supported by the Department of Energy Computational Science Graduate Fellowship Program of the DOE Office of Science and National Nuclear Security Administration, Princeton University’s Institute for Computational Science and Engineering, and Laboratory Directed Research and Development funds provided by PPPL.

Source: https://www.harvard.edu/