For the first time, artificial intelligence has been deployed to control the hot plasma in a nuclear fusion reactor. This can increase the stability and efficiency of such reactors.
Nuclear fusion reactors provide cheap and clean energy – if we can get them to work. Fusion researchers from the Swiss technology institute EPFL have now joined forces with artificial intelligence firm DeepMind to create a step closer a useful fusion reactor. They used artificial intelligence (AI) to generate the plasma in their reactor, the so-called TCVreactor, to be kept under control.
Magnetic Coils
A fusion reactor uses magnetic fields to keep the hot plasma inside it away from the walls. If the plasma hits the walls of the reactor, it can cool down. This could bring the nuclear fusion reaction to a standstill, which is harmful to the reactor.
ALSO READ
When the clock is ticking back
The TCV reactor uses nineteen magnetic coils to contain its plasma. Previously, each coil was driven by its own algorithm that monitors the inside of the reactor thousands of times per second with sensors. DeepMind designed a single neural network that monitors all coils simultaneously. The network automatically learns what voltages are needed to keep the plasma on board.
The team first trained the AI in a digital test situation before moving on to experiments with the real reactor. In the end, it turned out to be possible to control the plasma for about two seconds. This is close to what the theory says is the highest achievable: the TCV reactor can only be on for three seconds. The reactor must then cool down for fifteen minutes.
New shapes
The AI could do more than simply keep the plasma inboard. It could also make the plasma move or change shape through the reactor. It could even control two different pieces of plasma at the same time.
New plasma forms can be more efficient and stable. That would be useful in, for example, the ITER, a reactor under construction in France. When completed in 2025, it will be the largest nuclear fusion power plant in the world.
Unexpected method
There are theoretically many different ways to push plasma into a particular shape. Still, most researchers keep choosing the same tried-and-true strategy, says Federico Felicia† He participated in the study on behalf of EPFL. The artificial intelligence surprised the team by arriving at the same plasma shapes in a different, yet unknown way.
‘This AI algorithm chose to use the TCV coils in a completely different way, which produces more or less the same magnetic field,’ says Felici. ‘As expected, it made the same plasma, but it used the magnetic cores completely differently. The researchers who monitored how the electric current developed through the coils were quite surprised.’
Gianluca Sarric, professor of plasma physics at the Queen’s University of Belfast, says artificial intelligence is the future for control systems in fusion reactors. But so far, it has never been possible to create a fusion reaction that produces more energy than it consumes.
“Once we can do that, it won’t be the end of the story,” Sarri says. ‘Then you still have to build a power plant. And this artificial intelligence, in my opinion, is the only way forward. There are so many variables, while a small change in one of them can already make a big change in the outcome.’ It is therefore almost impossible to manually try out what the correct method is. AI can do this job.
beta value
To make fusion reactors efficient and usable energy sources, physicists need to increase ‘beta value’, says physicist Howard Wilson from the University of York, in England. That means physicists must increase the pressure of the plasma relative to the strength of the magnetic field.
“The plasma wiggles and twists and tries to escape the clutches of magnetic fields,” he says. ‘As the plasma pushes the beta value up, the magnetic field has to work harder and harder to keep the plasma under control. The more you put pressure on the plasma, the more likely you are to lose control.’
According to Wilson, these artificial intelligence experiments hold great promise for controlling plasma in ‘extreme forms’. That paves the way for experiments with different plasma shapes that could increase stability or efficiency. ‘It expands the border area in which it is safe to work. In addition, we are opening a new playing field that we can explore.’