Diagnosis of Filamentary Plasma in MDPX through Convolutional Neural Networks (CNN) and Physics Informed Neural Networks (PINN)

In the Magnetized Dusty Plasma eXperiment (MDPX) operating at magnetic fields, B > 0.75 T, magnetic field aligned structures called "filaments" break the uniformity of the plasma. The appearance of filaments can disrupt both plasma and dust experiments. Because high magnetic fields also interferes with in-situ probe measurements of plasma parameters, a neural network approach is used as an alternative diagnostic tool using the morphology and spatiotemporal dynamics of the filaments to guide the reconstruction of the plasma parameters. The filaments exhibit azimuthal modes resembling Archimedean spirals that can be decomposed into a wave-function-like characterization with CNN. At the same time, the complex spatiotemporal behavior is tracked and supplied into a PINN to extract localized parameters through the moment equations. Preliminary results demonstrate that the PINN can act as a proxy to probe the plasma of localized quantities such as electric field and angular velocity. This presentation discusses the development of the PINN and the resulting dependence of filamentary spatiotemporal behaviors (e.g., rotation, morphology changes, splitting, creation/annihilation) on localized quantities.