An AI controller promises to predict what can go wrong in a nuclear fusion reaction 300 milliseconds before it actually happens.
Researchers at Princeton University have developed an AI controller that can predict plasma instabilities in a nuclear fusion reaction and adjust the plasma conditions to suit the ongoing reaction.
This innovation can make nuclear fusion more feasible and scalable than ever.
Currently, all man-made nuclear power plants produce energy via nuclear fission, a process involving the splitting of atoms and producing long-lasting radioactive waste.
The Sun and stars on the other side are naturally powered by nuclear fusion, another type of atomic reaction where two nuclei fuse into one.
Compared to nuclear fission, fusion reaction produces more energy, very less (and short-lived) radioactive waste, and doesn’t require uranium.
Many scientists believe that nuclear fusion could also serve as the cleanest and greenest source of energy on Earth. In their new study, Princeton University researchers reveal how the new AI controller can help humans achieve this goal.
The challenge with stable nuclear fusion
https://youtube.com/watch?v=29R8sKCsxEk%3Ffeature%3Doembed
In December 2022, for the first time, scientists successfully achieved net energy production from a nuclear fusion reaction. However, a very small amount of energy (just enough to boil a kettle) was produced during that reaction.
This is because although scientists successfully achieved nuclear fusion, they didn’t know how they could sustain the reaction for long.
To achieve a stable fusion reaction, one also needs to sustain plasma, a state of matter comprising extremely hot gases and free electrons (at temperatures reaching millions of degrees Celsius).
This ionized high-energy plasma state is necessary to facilitate the fusion of two nuclei into one.
However, at such high temperatures, anything can go wrong. Even a small change in the plasma could destabilize the reaction.
Scientists have previously tried preprogrammed controllers to maintain the plasma conditions, but those tools failed, according to the researchers.
Until now, there was no way to keep track of the plasma in real-time and keep it stable.
The AI controller solves the plasma problem
To solve the problem and prevent plasma instabilities, the researchers created an AI controller that utilizes data from multiple highly advanced diagnostic tools operational at DIII-D, a national fusion facility in San Diago.
“DIII-D is equipped with high-resolution real-time plasma diagnostic systems. These diagnostic tools allow for the real-time profiling of electron density, electron temperature, ion temperature, ion rotation, pressure, current density, and safety factor,” the researchers note.
The AI processes this real-time data and allows researchers to predict how the plasma will behave under different conditions. With the help of the AI, they can predict any instability 300 milliseconds before it occurs.
Meanwhile, based on the real-time information and their proposed outcomes, they can test various methods that can be used to control and stabilize the plasma by manipulating its conditions.
While 300 ms may sound like no time to a normal person, the study authors claim that this duration would be enough to make any changes to ensure the stability of the reaction. They even tested the AI controller to back their claims.