Embarking on the journey toward nuclear fusion, the process mimicking the sun's core reactions for energy production, comes with a set of challenges. These hurdles include generating more energy than consumed, finding reactor-resistant materials, ensuring impurity-free reactors, and containing the fuel within the reactor. Princeton University and its Plasma Physics Lab researchers have developed an AI model aiming to tackle the last hurdle: preventing plasma from becoming unstable and escaping magnetic fields within donut-shaped reactors.
In the race for practical fusion reactors, donut-shaped tokamak reactors are front-runners, utilizing magnets to corral plasma particles into a spinning fusion reaction. However, any disruption to the magnetic field lines can lead to catastrophic consequences, jeopardizing the delicate balance that keeps the reaction contained. Plasma physicist Chijin Xiao warns of potential risks, including the release of stored energy that could damage the reactor walls and exert force that might destroy the device.
The researchers at Princeton have introduced an AI model capable of predicting tearing mode instabilities, a type of plasma disruption, 300 milliseconds before occurrence. Though seemingly a short notice, it proves sufficient to regain control over the plasma, as demonstrated in tests on the DIII-D National Fusion Facility in San Diego. The AI system effectively managed power input and adjusted plasma shape to maintain stability.
Co-author Azarakhsh Jalalvand attributes the success of the AI model to its training on real data from past fusion experiments rather than theoretical models. The model learns the optimal pathway for maintaining a high-powered reaction while avoiding instabilities by focusing on goals and outcomes.
While tearing mode instabilities represent a significant challenge, there are various ways plasma can behave unpredictably. Nevertheless, the Princeton team's approach stands out as it enables prediction and avoidance of instabilities before they manifest, providing a unique advantage over previous studies that could only suppress instabilities after they occurred.
Federico Felici, a physicist at the Swiss Federal Institute of Technology, emphasizes the potential of AI in controlling and maintaining fusion reactions, citing its ability to enhance device operation. The researchers acknowledge their work as a proof-of-concept in the early stages of fine-tuning, expressing hope that it could eventually be applied to optimize reactions and energy harvesting in various reactors. However, as Xiao points out, the broad range of scenarios requires cautious consideration, making it a wait-and-see situation despite current experimental progress.