Fusion energy development has reportedly made its biggest stride yet thanks to artificial intelligence (AI), after a team led by Princeton University and the Princeton Plasma Physics Laboratory (PPPL) used AI to forecast - and then prevent - the genesis of a particular plasma problem in real time. Effectively providing a solution to one of the biggest challenges in creating a fusion power that has virtually limitless clean energy.

AI was reportedly used by the researchers to predict and avoid the likely tendency of erratic, superheated plasma that powers a fusion reaction to become unstable in an instant and break free from the powerful magnetic fields holding it inside the reactor, an escape that often signals the end of the reaction and thus the end of the limitless energy source as well.

This likely tendency is an obvious and fundamental obstacle to the development of fusion as a nearly infinite, non-polluting energy source. 

(Photo: Janos Kummer/Getty Images)
The UK government recently unveiled its ambitious plans for the "biggest expansion of nuclear power for 70 years."

The researchers showed their AI model, trained purely on historical experimental data, could predict probable plasma instabilities known as ripping mode instabilities up to 300 milliseconds ahead of time in experiments conducted at the DIII-D National Fusion Facility in San Diego.

This was reportedly plenty of time for the AI controller to alter some operating parameters in order to prevent what would have developed into a tear within the plasma's magnetic field lines, upsetting its equilibrium and creating the possibility of a reaction-ending escape.

What's more is that the AI model could reportedly help prevent these escapes in real time and in a real reactor by learning from previous recorded instances instead of using data from physics-based models. 

The AI was allegedly able to create a final control policy that maintained a stable, high-powered plasma regime according to Egemen Kolemen, a staff research physicist at PPPL and an associate professor of mechanical and aeronautical engineering at the Andlinger Center for Energy and the Environment. 

Read Also: AI Data Centers Restricted Worldwide Due to Energy Consumption 

AI Prediction Training

However, creating a successful AI controller required more than just experimenting a little on a device with limited time and huge stakes. Kolemen's group research scholar Azarakhsh Jalalvand, a co-author, likened learning an algorithm to control a fusion reaction to learning how to fly an airplane.

According to Azarakhsh Jalalvand, teaching involves more than just giving someone a set of keys and telling them to try their hardest. Rather, one would have them practice on a complex flight simulator until they were proficient enough to try out the real thing.

Consequently, the Princeton researchers built a deep neural network using data from earlier tests, just like they would when creating a flying simulator, with the ability to forecast, using real-time plasma parameters, the probability of a future ripping instability. 

Confidence in Limitless Energy

The study's findings were reportedly lauded by Koleman, claiming that it is a step forward adding that these kinds of disruptions and instabilities would be very problematic, so developing solutions as found by the study increases their confidence that these machines can run without any issues.

Koleman claims that fusion energy instabilities are one of the biggest obstacles or disruptions in any reactor before it can operate 24/7 for years without any problem. 

Related Article: EU Seals Clean Energy Tech Deal to Boost the Bloc's Green Industry