Artificial Intelligence Can Predict and Prevent Plasma Instabilities in Milliseconds

A team of engineers, physicists, and data scientists from Princeton University and the Princeton Plasma Physics Laboratory (PPPL) have used artificial intelligence (AI) to predict, and then avoid, the formation of a specific type of plasma instability in magnetic confinement fusion tokamaks.

The researchers built and trained a model using past experimental data obtained from operations at the DIII-D National Fusion Facility in San Diego, California.

Using real-time experiments, their model can predict the so-called tearing mode instabilities up to 300 milliseconds in advance, which is enough time for the AI controller to adjust operating parameters and avoid a plasma tear.

Prior to this discovery, scientists could only adjust for instabilities, but in the future, the latest approach will expand the reactor’s capabilities. The results demonstrate that dynamic plasma control using AI can take magnetic confinement fusion one step further in scientific progress.

Instability in tearing mode is a phenomenon of plasma disturbances in which magnetic field lines in the plasma actually tear, creating the possibility of its leakage.

The Princeton team used data from past experiments at the DIII-D tokamak to build a neural network that can predict the likelihood of future instability. Using the neural network, they trained an auxiliary algorithm to test the plasma control strategy in a simulated environment to learn by trial and error which strategies worked.

“We don’t teach the reinforcement learning model all the complex physics of a fusion reaction,” said researcher Azarakhsh Jalalvand.

“We set the goal of the algorithm – to maintain a powerful reaction, what to avoid – the instability of the tearing mode – and how the algorithm can influence to achieve these results. Over time, it learns the optimum path to achieve the high power goal while avoiding the penalty of instability.”

The next steps of the team will be to test the AI controller on the DIII-D tokamak and then work on its functionality.

According to Nuclear Newswire