(ORDO NEWS) — Successfully achieving nuclear fusion promises to provide a limitless, sustainable source of clean energy, but we can only realize this incredible dream if we can master the complex physics that goes on inside a reactor.
For decades, scientists have taken incremental steps towards this goal, but many problems remain. One of the major hurdles is successfully managing unstable and superheated plasma in a reactor, but the new approach shows how we can do it.
In a collaborative effort between EPFL’s Swiss Plasma Center (SPC) and artificial intelligence (AI) research company DeepMind, scientists have used a deep learning (RL) system to study the behavior and control of plasma inside a fusion tokamak, a device that uses a series of magnetic coils placed around the reactor to control and manipulate the plasma inside it.
This is not an easy balance, as the coils require a huge number of fine voltage adjustments, up to thousands of times per second, to successfully hold the plasma in magnetic fields.
Thus, to maintain nuclear fusion reactions, which involve maintaining a stable plasma temperature of hundreds of millions of degrees Celsius, hotter than even the core of the Sun, complex multilayer systems are needed to control the coils.
However, in a new study, the researchers show that the AI system can control the execution of the task itself.
“Using a learning architecture that combines deep learning and a simulated environment, we have created controllers that can maintain plasma stability and be used to precisely shape it into various shapes,” the team explains on the DeepMind blog.
The researchers trained their AI system on a tokamak simulator, in which the system discovered through trial and error how to deal with the complexities of plasma magnetic confinement.
After its training window, the AI moved to the next level – applying in the real world what it learned in the simulator.
By controlling the Variable Configuration SPC (TCV) tokamak, the AI transformed the plasma into various forms inside the reactor, including one never seen before in a TCV: stabilizing “blobs” in which two plasmas coexisted simultaneously inside the reactor.
In addition to the usual shapes, the AI can also create advanced configurations, giving the plasma “negative triangle” and “snowflake” shapes.
Each of these manifestations has a different potential to harvest energy in the future if we can sustain fusion reactions.
According to the researchers, the magnetic prowess of these plasma formations represents “one of the most complex real-world systems to which reinforcement learning has been applied” and could set a radical new direction in the development of real tokamaks.
Contact us: [email protected]