This AI can control hot plasma from the sun in a nuclear fusion reactor
Curdin Wuthrich /SPC/EPFL
- DeepMind used artificial intelligence to power magnets for nuclear fusion reactors.
- The deep reinforcement learning algorithm redistributes the power to match the plasma.
- Different forms of plasma have advantages and disadvantages to aid in the development of fusion.
Google’s artificial intelligence powerhouse, DeepMind, has made news again using artificial intelligence (AI) to help shape the stream of solar-heated plasma contained within a tokamak nuclear fusion reactor. The DeepMind team has been in the news for years using increasingly sophisticated AI to master iconic games like chess, shogi and go, as well as solve “real world” problems. The London-based task force was formed in 2010 and bought by California-based Google in 2014.
In new research published in the peer-reviewed journal Nature, the DeepMind team explains how they used deep reinforcement learning, a subfield of machine learning in which a system can learn from its own decisions, to help magnetically control hot plasma, typically iron. hydrogen, in a tokamak reactor. Plasma is a state of matter usually mentioned in fourth place here on Earth, but it’s actually the most common state because it’s what powers virtually every star in the universe. It is a cloud of highly charged ions, usually very hot and often fluid. For this reason, plasma often conducts electric current.
☣️ You like nuclear. U.S. too. Let’s discuss it together.
Thus, in a tokamak reactor, the plasma is heated more and more until its charged ions begin to fuse. It is fusion in the term nuclear fusion. But the plasma in these reactors reaches millions of degrees Celsius, so it can only be contained by magnetic fields that create a buffer of gas rather than solid material. (Even without direct plasma contact, tokamaks still need to be lined with specially designed high-temperature materials like carbon and tungsten.)
“One of the main challenges is shaping and maintaining a high temperature plasma in the tokamak vessel. This requires high-dimensional, high-frequency closed-loop control using magnetic actuator coils, further complicated by the varying requirements of a wide range of plasma configurations,” the DeepMind team explains in their paper. .
So what’s wrong with the magnets in a typical tokamak race? Well, nothing really – the magnets do a good job, and the tokamaks usually turn off because their parts have overheated. Recently, the Joint European Torus ran for a record five seconds while producing 59 megajoules of energy before overheating too much to continue. When there is millions of degrees of plasma swirling inside a physical structure that would not contain it, you can see why it is important to have a very effective shutdown process.
But that doesn’t mean magnets can’t be upgraded for tokamaks. Basically, the DeepMind team took the reins from the magnetic coils themselves and used deep machine learning to dynamically adjust the coded behaviors of each electromagnet. This means researchers might be able to explore newer shapes, including the similar stellarator’s donut-shaped reactor filled with helical plasma. We speak of “negative triangularity and ‘snowflake’ configurations,” the researchers write.
Negative triangularity is a specialized form of plasma formed by essentially flipping the cross section inside the tokamak. The result is a “back D” instead of a front “D” when viewing the section view (see above). In a snowflake configuration, hot plasma is magnetically deflected into a snowflake shape that better distributes heat and prevents material destruction. And, of course, helical plasma simply requires a much more complex magnetic field than a “simple” million-degree donut.
Plasma ignition – the industry term for a reactor that produces more energy than it takes to operate – is years away for even the most advanced fusion reactors being built. But DeepMind is working to make more interesting and potentially safer options available to researchers before those blueprints are set in stone, which could help lead to designs that work better in a shorter time. Sure, it’s nice to have the option.
This content is created and maintained by a third party, and uploaded to this page to help users provide their email addresses. You may be able to find more information about this and similar content on piano.io