The study of burning plasmas offers fusion scientists and engineers the opportunity to better understand and control sustained, ongoing fusion reactions. Tokamak experiments are aimed at investigating the physics, engineering, and technologies associated with self-heating plasma — issues critical to using self-heating plasma reactions to produce more power than they consume, a massive step toward commercial fusion power production. Equilibrium reconstruction is the determination of both the shape of the plasma and its pressure and current profile information given a set of diagnostic signals. Equilibrium reconstructions are key in understanding data interpretation, plasma control, disruption control, and code and physics model validation.

An Argonne team is working with collaborators at General Atomics, Tech-X Corporation, and Lawrence Berkeley National Laboratory to develop EFIT-AI — a modern, advanced equilibrium reconstruction code for tokamak experiments and burning plasmas — by adopting machine learning and artificial intelligence to enhance equilibrium reconstructions, making them more efficient and accurate. This is done by automating and maximizing the information extracted from measurements and leveraging physics-informed ML models constructed from experimental and synthetic solution databases to guide the search for the solution vector.

Key ML elements of the EFIT-AI framework include (1) improved optimization and data analysis capabilities using an ML-enhanced Bayesian framework, (2) a model order reduction (MOR) version of the two-dimensional Grad-Shafranov equation solver and three-dimensional perturbed equilibrium reconstruction using information-theoretic Bayesian deep learning.

These developments will enhance the applications of the EFIT tokamak equilibrium reconstruction tool at DOE user facilities. With collaborators at General Atomics, we will leverage the extensive discharge databases from the DIII-D tokamak, as well as other tokamaks around the world, to achieve real-time construction of 2D and 3D plasma equilibria for tokamak applications — enabling improved control and optimization of tokamaks as a fusion reactor.