Plasma fusion

Credit: Pixabay/CC0 Public Domain

In the blink of an eye, uncontrollably, extremely hot Plasma That which drives the fusion reaction can lose its stability and escape the strong magnetic fields that confine it inside the doughnut-shaped fusion reactor. These pathways often spell the end of reactions, presenting a fundamental challenge to developing fusion as a non-polluting, virtually limitless energy source.

But a Princeton-led team of engineers, physicists and data scientists from the university and the Princeton Plasma Physics Laboratory (PPPL) has harnessed its power. Predicting and then avoiding specific plasma problem formation in real time.

In experiments at the DIII-D National Fusion Facility in San Diego, the researchers demonstrated their model, trained only on past experimental data, could predict potential plasma instabilities up to 300 milliseconds in advance. The tearing mode is known as instability.

While this didn’t leave enough time for the humans to blink, it was enough time for the AI ​​controller to change some operating parameters to avoid bursting within the plasma’s magnetic field lines, which upset its equilibrium. Go and start blooming. An escape hatch from the termination of the reaction.

“By learning from past experiences, rather than incorporating information from physics-based models, AI can develop a definitive control policy that supports a stable, high-power plasma system in a real reactor, in real time. “said the leader of the research, Egyman Kolman. , Associate Professor of Mechanical and Aerospace Engineering and Staff Research Physicist at the Andlinger Center for Energy and the Environment as well as PPPL.

This research opens the door to more dynamic control of fusion reactions than current approaches, and it provides a basis for using artificial intelligence to solve a wide range of plasma instabilities, a constant There have long been obstacles to achieving fusion reactions. Published by the team. Their results I The nature On February 21

“Previous studies have generally focused on suppressing or reducing the effects of these explosive instabilities once they occur in plasma,” said first author Jimin Seo, an assistant professor of physics at Chung-Aeng University in South Korea. Professor who did most of the work. While a postdoctoral researcher in Coleman’s group. “But our approach allows us to predict these instabilities and avoid them before they appear.”

Superheated plasma swirls in a donut-shaped device.

Fusion occurs when two atoms – usually lighter atoms such as hydrogen – combine to form a heavier atom, releasing large amounts of energy in the process. This process powers the Sun, and by extension, makes life possible on Earth.

However, fusing the two atoms is difficult, as the two atoms require a large amount of pressure and energy to overcome their mutual repulsion.

Fortunately for the Sun, its great gravity and extremely high pressure at its core allow fusion reactions to proceed. To mimic a similar process on Earth, scientists instead use superheated plasma and super-strong magnets.

In donut-shaped devices called tokamaks—sometimes called “stars in jars”—they struggle to contain plasma in magnetic fields that are hotter than the Sun’s core, more than 100 million degrees Celsius. are

Although there are many types of plasma instabilities that can kill the reaction, the Princeton team focused on solving the tearing mode instability, a disturbance in which the magnetic field lines within the plasma actually bend. breaks down and creates an opportunity for subsequent plasma escape.

“Tearing mode instabilities are a major cause of plasma disturbances, and will become even more prominent as we try to run fusion reactions at the higher powers required to generate sufficient energy,” said Seo. “They are an important challenge for us to solve.”

Combining artificial intelligence and plasma physics

Because instabilities in the tearing mode can set off a fusion reaction in milliseconds and derail it, researchers turned to artificial intelligence to quickly adapt and act in response to new data. can get the ability to

But the process of developing an effective AI controller wasn’t as simple as trying a few things on a tokamak, where time is limited, and the stakes are high.

Co-author Azrakhsh Jalal Wind, a research scholar in Coleman’s group, compared teaching the algorithm to run a fusion reaction in a tokamak to teaching someone how to fly a plane.

“You don’t teach someone by giving them a set of keys and telling them to do their best,” Jalalund said. “Instead, you’ll have them practice on a very complicated flight simulator until they learn enough to try the real thing.”

Similar to developing a flight simulator, the Princeton team used data from past experiments at the DIII-D tokamak to build a deep neural network that predicts the likelihood of future rupture instability based on real-time plasma properties. can estimate.

They used this neural network to train a reinforcement learning algorithm. Like a trainee pilot, a reinforcement learning algorithm can try different strategies for controlling the plasma, learning through trial and error which strategies work and which don’t in the simulated environment. .

“We don’t teach the reinforcement learning model all the complicated physics of a Jalal Wind said. “We tell him what to aim for — maintaining a high-powered reaction — what to avoid — tear-mode instability — and what families he can turn to to achieve those results. . Over time, it learns the best way to achieve a higher power goal while avoiding the torment of instability.”

While the model went through countless artificial fusion experiments, trying to find ways to maintain high power levels while avoiding instability, co-author SangKyeun Kim could observe and refine its actions.

“In hindsight, we can see the intentions of the model,” said Kim, a staff research scientist at PPPL and a former postdoctoral researcher in Coleman’s group. “Some of the changes the model wants are too fast, so we work to smooth and calm the model. As humans, we mediate what the AI ​​wants to do and what the tokamak can do.”

Once they were confident in the AI ​​controller’s capabilities, they tested it during a real fusion experiment in a D-III D tokamak, observing that the controller could avoid the onset of instability in certain tokamaks. Real-time changes to parameters. These parameters include changing the shape of the plasma and reacting beam power.

“Being able to predict instability ahead of time could make it easier to drive these responses than current approaches, which are more passive,” Kim said. “We no longer have to wait for instability to develop and then take immediate corrective action before the plasma is disrupted.”

Powering the future

While the researchers said the work is a promising proof-of-concept showing how artificial intelligence can effectively control fusion reactions, Coleman’s group hopes to advance the field of fusion research. It is one of several next steps already underway.

The first step is to get more proof of operation of the AI ​​controller on the DIII-D tokamak, and then extend the controller to work on other tokamaks.

“We have strong evidence that the controller works quite well on 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 towards something more universal.”

Another line of research involves extending algorithms to handle many different control problems simultaneously. While the current model uses a limited number of evaluations to avoid a specific type of instability, researchers could provide data about other types of instability and give the AI ​​controller access to more knobs to tune. can.

“You can imagine a big reward function that turns a lot of different knobs to control many kinds of instability simultaneously,” said co-author Ricardo Shusha, a postdoc at PPPL and a former graduate student in Coleman’s group. Knowledge that provided support for experiments in DIII- D

And on the way to developing better AI controllers for fusion reactions, researchers may also gain a greater understanding of the underlying physics. By studying the AI ​​controller’s decisions as it tries to contain it. In a way that can be radically different from what traditional methods suggest, artificial intelligence can be not only a tool for controlling fusion reactions, but also a teaching resource.

“Ultimately, it may be more than a one-way interaction of scientists developing and deploying these AI models,” Coleman said. “By studying them in more detail, they may have things they can teach us as well.”

More information:
Egemen Kolemen, Avoiding tearing instabilities in fusion plasmas using deep reinforcement learning, The nature (2024). DOI: 10.1038/s41586-024-07024-9. www.nature.com/articles/s41586-024-07024-9

Reference: Engineers use AI to wrangle fusion power for grid (2024, February 21) https://phys.org/news/2024-02-ai-wrangle-fusion-power-grid February 21, 2024 Retrieved from .html

This document is subject to copyright. No part may be reproduced without written permission, except for any fair dealing for the purpose of private study or research. The content is provided for informational purposes only.