Tech

EPFL and DeepMind use AI to control the plasma for fusion

Plasma in TCV tokamak. Credits: Curdin Wüthrich / SPC / EPFL

EPFL’s Swiss Plasma Center (SPC) has decades of experience in plasma physics and plasma control methods. DeepMind is a scientific discovery company acquired by Google in 2014, working on “solving intelligence to advance science and humanity.” They jointly developed a new magnetic control method for plasma based on deep reinforcement learning and applied it to real-world plasma for the first time at TCV, SPC’s tokamak research facility.Their study has just been published Nature..


Tokamak is a donut-shaped device for conducting research. Nuclear fusion, SPC is one of the few research centers in the world. These devices can use a strong magnetic field to confine the plasma to very high temperatures (hundreds of millions of degrees Celsius, even hotter than the solar core), allowing fusion to occur between hydrogen atoms. The energy emitted from fusion is being studied for use in power generation.

What makes the SPC tokamak unique is that it allows for a variety of plasma configurations. Therefore, its name is Variable Configuration Tokamak (TCV). This means that scientists can use it to explore new approaches to confine and control the plasma. The composition of the plasma is related to its shape and position within the device.

Control substances as hot as the sun

Tokamak forms and maintains plasma through a series of magnetic coils that require careful control of settings, especially voltage. Otherwise, the plasma may collide with the vessel wall and deteriorate. To prevent this, SPC researchers first test the control system configuration in a simulator before using it in a TCV tokamak.

“Our simulator is based on over 20 years of research and is continually updated,” said Federico Felici, an SPC scientist and research co-author. “Still, it takes a long calculation to determine the correct value for each variable in the control system, and that’s where a collaborative research project with DeepMind comes in.”

EPFL and DeepMind use AI to control the plasma for fusion

A 3D model of a TCV vacuum vessel containing plasma, surrounded by various magnetic coils to hold the plasma in place and affect its shape. Credits: DeepMind & SPC / EPFL

DeepMind experts have developed and trained in SPC simulators AI algorithms that can create and maintain specific plasma configurations. This initially involved having the algorithm try many different control strategies in simulation and experience gathering. Based on the experience gathered, the algorithm has generated a control strategy to generate the required plasma configuration. This involved first running the algorithm in several different settings and analyzing the plasma composition that resulted from each. Then the algorithm was asked to work differently. That is, to generate a particular plasma configuration by identifying the appropriate settings.

After training, AI-based systems were able to create and maintain a wide range of plasma shapes and advanced configurations, including those in which two separate plasmas were maintained simultaneously in the vessel. Finally, the research team tested the new system directly on the tokamak to see how it works under real-world conditions.

The collaboration between SPC and DeepMind dates back to 2018, when Felici first met DeepMind scientists at the company’s London headquarters hackathon. So he explained the tokamak magnetic control problem in his research group. “DeepMind was immediately interested in the possibility of testing AI technology in areas such as fusion, especially in real-world systems like tokamak,” says Felici.

Martin Riedmiller, DeepMind’s control team leader and co-author of the study, adds: “Our team’s mission is to research a new generation of AI systems (closed-loop controllers) that can learn from scratch in complex dynamic environments. plasma In the real world, it’s very challenging and complex, but it offers great opportunities. “

Range of various plasma shapes generated by Reinforcement Learning Controller Credits: DeepMind & SPC / EPFL

Collaboration that is mutually beneficial

After talking to Felici, DeepMind suggested working with SPC to develop an AI-based control system for that purpose. Tokamak.. “We quickly agreed with this idea because we saw the great potential of innovation,” said Ambrogio Fasoli, director of SPC and co-author of the study. “All the DeepMind scientists we worked with were very enthusiastic and knew a lot about implementing AI in control systems.” On his side, Felici identified DeepMind. When I was focusing on my project, I was impressed with how quickly it could be done.

DeepMind also draws a lot from collaborative research projects and demonstrates the benefits of taking an interdisciplinary approach for both parties. Brendan Tracey, Senior Research Engineer at DeepMind and co-author of the study, said:

This project should pave the way for EPFL to explore other collaborative R & D opportunities with external organizations. “We are always embracing innovative win-win collaborations, and by sharing ideas and exploring new perspectives, we can accelerate the pace of technology development,” says Fasoli.


Negative Triangle — Positive of Tokamak Fusion Reactor


For more information:
Jonas Degrave et al, Magnetic Control of Tokamak Plasma by Deep Reinforcement Learning, Nature (2022). DOI: 10.1038 / s41586-021-04301-9

Quote: EPFL and DeepMind use AI to control the plasma for fusion (February 16, 2022).

This document is subject to copyright. No part may be reproduced without written permission, except for fair transactions for personal investigation or research purposes. Content is provided for informational purposes only.



EPFL and DeepMind use AI to control the plasma for fusion

Source link EPFL and DeepMind use AI to control the plasma for fusion

Show More

Related Articles

Back to top button