A Hybrid Approach using Deep Learning and Plasma Fluid Modeling for Simulating High-Power Microwave Breakdown

High-Power Microwave (HPM) breakdown at high pressures primarily comprises two major phenomena occurring on different time scales: microwave-plasma interaction and spatio-temporal evolution of plasma. Classical 2D/3D simulations of such breakdown coupling Maxwell's equations with plasma fluid equations is computationally very expensive. To address this, we propose a hybrid approach that combines Deep Learning (DL) for handling the microwave-plasma interaction, thereby accelerating the simulation and a Differential Equation (DE)-based model for plasma evolution to ensure accuracy in simulating plasma dynamics. We employ a U-Net based DL model, trained with data from our in-house plasma-fluid solver.

We validated the hybrid approach for simulating microwave streamers which focuses on the formation of a plasmoid due to HPM breakdown and its subsequent expansion into a single streamer, confined at the antinode between two incident linearly polarized waves with opposing wave vectors [1]. The evaluation includes a quantitative comparison between results obtained from classical and hybrid approaches in terms of the rate of streamer growth, shape of the streamer, and the overall spatio-temporal evolution of plasma density. Our initial results demonstrate that the proposed hybrid approach can serve as a novel and fast alternative for investigating the computationally challenging multiscale multiphysics simulation of HPM breakdown.

[1] Chaudhury et al 2011 Journal of Applied Physics 110 113306.