Data-driven reduced modeling of streamer discharges in air

Abstract

We present a computational framework for simulating filamentary electric discharges, in which channels are represented as conducting cylindrical segments. The framework requires a model that predicts the position, radius, and line conductivity of channels at a next time step. Using this information, the electric conductivity on a numerical mesh is updated, and the new electric potential is computed by solving a variable-coefficient Poisson equation. A parallel field solver with support for adaptive mesh refinement is used, and the framework provides a Python interface for easy experimentation. We demonstrate how the framework can be used to simulate positive streamer discharges in air. First, a dataset of 1000 axisymmetric positive streamer simulations is generated, in which the applied voltage and the electrode geometry are varied. Fit expressions for the streamer radius, velocity, and line conductivity are derived from this dataset, taking as input the size of the high-field region ahead of the streamers. We then construct a reduced model for positive streamers in air, which includes a stochastic branching model. The reduced model compares well with the axisymmetric simulations from the dataset, while allowing spatial and temporal step sizes that are several orders of magnitude larger. 3D simulations with the reduced model resemble experimentally observed discharge morphologies. The model runs efficiently, with 3D simulations with 20+ streamers taking 4–8 minutes on a desktop computer.