EPFL and DeepMind use AI to control plasmas for nuclear fusion
Plasma inside the TCV tokamak. Credit: Curdin Wüthrich /SPC/EPFL

The Swiss Plasma Center has decades of experience in the field. DeepMind is a scientific discovery company that was acquired by the internet giant, and they have developed a new magnetic control method based on deep reinforcement learning. Their study was published in Nature.

One of the few research centers in the world that has a tokamak is the SPC. Nuclear fusion can occur between hydrogen atoms because of the magnetic field used by these devices. The fusion energy is being studied to see if it can be used to generate electricity.

Variable-configuration tokamak is the name of the tokamak that allows for a variety of configurations. Scientists can use it to investigate new approaches to controlling and controlling the plasmas. The shape and position of the device are related to the configuration.

A substance is as hot as the sun.

The settings of the magnetic coils that form and maintain the tokamaks must be controlled carefully. The vessel walls could be damaged if the plasma collides with them. To prevent this from happening, researchers at the SPC first test their control systems configurations on a simulation before using them in the TCV tokamak.

The simulation is based on more than 20 years of research and is updated continuously, according to the co-author of the study. Our joint research project with DeepMind comes in.

EPFL and DeepMind use AI to control plasmas for nuclear fusion
3D model of the TCV vacuum vessel containing the plasma, surrounded by various magnetic coils to keep the plasma in place and to affect its shape. Credit: DeepMind & SPC/EPFL

DeepMind's experts created and trained an artificial intelligence system that can create and maintain specific plasma configurations. The first thing the algorithm had to do was try many different control strategies. The control strategy was generated based on the collected experience. The first thing that needed to be done was having the algorithm run through a number of different settings and analyze the configurations that came from each one. The formula was used to identify the right settings and produce a specific configuration.

After being trained, the system was able to create and maintain a wide range of shapes and configurations, including one where two separate plasmas are maintained simultaneously in the vessel. The research team tested their new system on the tokamak to see how it would perform under real-world conditions.

Felici first met DeepMind scientists at a Hackathon in London in the summer of 2018. He explained that his research group was interested in testing their technology on a real-world system like a tokamak.

Martin Riedmiller, control team lead at DeepMind and co-author of the study, says that their mission is to research a new generation of closed-loop controllers that can learn in complex dynamic environments. In the real world, controlling a fusion plasma is extremely challenging and complex.

Range of different plasma shapes generated with the reinforcement learning controller Credit: DeepMind & SPC/EPFL

A win-win collaboration.

After speaking with Felici, DeepMind offered to work with the SPC to develop an artificial intelligence-based control system for its tokamak.

DeepMind got a lot out of the joint research project, showing the benefits to both parties of taking a multidisciplinary approach. The collaboration with the SPC pushes us to improve our reinforcement learning algorithms, and as a result can accelerate research on fusion.

The project should lead to more joint R&D opportunities with outside organizations.

More information: Jonas Degrave et al, Magnetic control of tokamak plasmas through deep reinforcement learning, Nature (2022). DOI: 10.1038/s41586-021-04301-9 Journal information: Nature Citation: EPFL and DeepMind use AI to control plasmas for nuclear fusion (2022, February 16) retrieved 17 February 2022 from https://phys.org/news/2022-02-epfl-deepmind-ai-plasmas-nuclear.html This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.