Investigating DMD for reduced system models and optimal sensor placement for control

Disruptions and instabilities in tokamak plasmas drastically reduce confinement and can cause damage to the device. Physics-based models can accurately predict some instabilities, but can be challenging or impossible to run in realtime. As a result, reduced-order models for plasma dynamics are often required for control applications. Dynamical Mode Decomposition (DMD) is a technique that identifies a dynamical model from observed data and can be used to build reduced-order models from simulation or experimental data. This presentation will explore the use of DMD to predict the dynamics of Resistive Wall Modes (RWMs), which can be controlled in real time using magnetic sensors and actuators¹. Optimizing the position of these sensors may provide improved identification and control. RWM simulations will be implemented in ThinCurr², which computes inductively-coupled currents in the plasma, walls, and other components of a reactor. DMD’s performance will be evaluated based on the similarity of its computed system to the original ThinCurr system. Use of DMD and related techniques to optimize sensor placement will also be presented.

[1] A. Battey et al., NF 63 066025 (2023)

[2] https://github.com/hansec/OpenFUSIONToolkit