Physics-informed machine-learning to investigate spatiotemporal dynamics of coherent filamentary structures at high magnetic field in MDPX

Dust-free plasmas in the Magnetized Dusty Plasma eXperiment (MDPX), produce magnetic field aligned structures called “filaments” that break the uniformity of the plasma which disrupts typical dust experiments. These filaments exhibit complex spatiotemporal dynamics (e.g., rotation, translation, splitting, merging and changes in morphology) that may give insight to the underlying instabilities that generate the filaments. This work showcases various ways in which physics-informed machine-learning can bridge the gap between plasma parameters and filament dynamics. A customizable tracking code utilizing Python, OpenCV, and a search algorithm is developed to capture the fast-changing filament dynamics as well as capturing morphology information for each filament. The morphology information can be decomposed into unique azimuthal modes similarly to those of an Archimedean spiral through a Fast Fourier Transformation (FFT) or neural networks. Lastly, all the filament information is converted to a database that can be analyzed to correlate plasma parameters to trends of filament dynamics.