New York, May 19 (IANS): A research team in the US has used Machine Learning (ML) to quickly control the plasma that triggers fusion reactions and pave the way for Earth to receive the clean fusion energy that illuminates the Sun and stars.
Researchers from the US Department of Energy's Princeton Plasma Physics Laboratory (PPPL) use ML to create a model for rapid plasma control – the state of matter composed of free electrons and nuclei or ions.
The sun and most stars are giant plasma spheres that are subject to constant fusion reactions. Here on Earth, scientists need to heat and control the plasma so that the particles fuse and release their energy.
Researchers now show that ML can facilitate such control.
The team led by physicist Dan Boyer trained neural networks – the heart of the ML software – based on the data generated in the first operational campaign of the National Spherical Torus Experiment Upgrade (NSTX-U), PPPL's flagship fusion facility , were created.
The trained model reproduced accurate predictions of the behavior of the energetic particles generated by high-performance NBI, which fires NSTX-U plasmas and heats them up to million-heavy, fusion-relevant temperatures.
"The new ML software reduces the time required to accurately predict the behavior of energetic particles to less than 150 microseconds, allowing the calculations to be done online during the experiment," the results showed.
The fast scores also help operators to better understand the settings between the experiments, which run every 15 to 20 minutes during operation.
"Accelerated modeling capabilities could show operators how to adjust the NBI settings to improve the next experiment," said Boyer, lead author of an article in Nuclear Fusion magazine.