Artificial Intelligence (AI) is a potential killer app for fusion power, with demonstrated potential for real-time control and magnetic field optimisation. While powerful, a challenge for AI is the degree to which models are transparent, auditable, and free from flaws. This project develops verifiable AI models of the plasma edge field for real-time field control. The control of such edge fields is critical to abate performance limiting modes in burning tokamak plasmas. The project will extend development of PINN and FNO for the Grad-Shafranov equation, which describes axisymmetric tokamaks, to stellarators. Neural networks will be trained on the SPEC code, a Princeton-ANU developed code designed to describe realistic stellarators. Outcomes include a bounded model checked AI model, validated to experiment, and implementation of an advanced mathematical model describing the emergence of chaotic fields.
Develop a verifiable model artificial neural network model for 3D fusion plasmas