A Feature-Enhanced Physics Informed Neural Network for Trajectory Simulation of Charged Particles
28 Pages Posted: 23 May 2025
Abstract
Physics-Informed Neural Networks (PINN) have shown great potential in Computational Electromagnetics. This paper proposes a Feature-Enhanced PINN (FE-PINN) framework for the simulation of charged particle dynamics, improving computational efficiency by elimination of mesh discretization and node mapping. For electromagnetic fields with evolving internal sources and complex geometries, Feature-Enhanced PINN eliminates the need for network retraining under varying space charge distributions, and applies targeted enhancement in collocation strategies and network structure to improve convergence and accuracy in intricate field domains. Built upon FE-PINN electromagnetic field solutions, particle trajectory simulation is achieved by iteratively solving the Poisson’s equation and particle motion equation under a fixed magnetic field. The proposed method is validated using the magnetron injection gun of an 800 GHz gyrotron with results compared with finite difference method . Numerical results indicate that the proposed simulation method achieves higher computational efficiency while maintaining the same level of accuracy.
Keywords: Physics-Informed Neural Networks, charged particle dynamics, Electromagnetic Field Computation, Space Charge Effect
Suggested Citation: Suggested Citation