Machine learning applied to fusion research: Predicting and avoiding disruptions
In order to obtain net power from controlled thermonuclear fusion, matter (in the form of plasma) must be confined at temperatures hotter than the sun's core. Using a system of strong magnetic fields arranged in particular configurations, the conditions for net fusion energy production have just about been achieved, albeit over short timescales. While the state of this plasma system is too complicated to understand completely from first principles, machine learning can now be used to elucidate some of its integral processes, such as energy flow, state transitions, and other processes.
Magnetically-confined, high-energy density plasmas also tend to be close to stability limits. A particularly virulent instability is the ‘disruption,’ which can have catastrophic consequences. This talk will concentrate on efforts to apply machine learning to develop a real-time warning of impending disruptions, which could possibly be used to avoid them, or at least mitigate their consequences.
Robert Granetz has been working on the Alcator tokamak fusion experiments at the MIT Plasma Science and Fusion Center for more than forty years. His principal areas of research encompass magnetohydrodynamic instabilities and disruptions, including their prediction and mitigation. Granetz is an active contributor to the ITER Project, a multinational collaborative aiming to build the first fusion device that will produce net energy. He spent several years in Europe as a visiting scientist at the Joint European Torus, the world's largest tokamak. Granetz has also taught graduate student courses in plasma physics and fusion at MIT.