DeepMind keeps plasma fusion reactor under control

This technology promises to enhance the stability and efficiency of fusion reactors by harnessing plasma with artificial intelligence.

Fusion reactors, in which atomic nuclei do not split, but fuse, can provide us with relatively clean energy in the future. Unfortunately, the process is not profitable yet. Researchers at the Technical University of Lausanne in Switzerland and DeepMind may change that. They managed to control turbulent plasma in a so-called tokamak fusion reactor with artificial intelligence.

Read also: AI predicts 3D protein folding

tokamak

Tokamaks are donut-shaped devices used in nuclear fusion research. It’s also possible that future fusion reactors will look somewhat like this — if they become economically viable.

The neutrons capture the energy released when they accelerate the plasma through this circular shape (the so-called phase). These uncharged, energy-rich particles now shoot out and strike a special part of the tokamak where they release their energy into the matter. This energy can then power a turbine in the form of steam.

magnetic fields

The plasma, formed by placing gaseous hydrogen fuel under high temperature and pressure, is controlled by strong magnetic fields. After all, it is important to prevent the plasma from coming into contact with the walls. This will cool the hot plasma, stopping the reaction and thus energy production.

The new study published in the journal natureIn this article, the researchers describe how they control these magnetic fields and thus the plasma with artificial intelligence (AI).

Artificial intelligence

Previously, researchers controlled magnetic fields with 19 different algorithms. one per file. In the new technology, the team uses artificial intelligence developed by DeepMind. This neural network controls the voltage of each coil, and thus the shape of the magnetic fields and hot plasma.

The researchers trained the AI ​​system in simulation, after which they applied the technique to plasma in a real tokamak. The team wrote that the results are promising. In both the simulations and the device, the network was not only able to keep the plasma in check, but was also able to pour red-hot materials into specific shapes and even quench the plasma at the same time (see image above).

Sources: Nature, New World

Photo: Curdin Wüthrich/SPC/EPFL