Application of Spectral Solution and Neural Network Techniques in Plasma Modeling for Electric Propulsion
Abstract
A solver for Poisson's equation was developed using the Radix-2 FFT method first invented by Carl Friedrich Gauss. Its performance was characterized using simulated data and identical boundary conditions to those found in a Hall Effect Thruster. The characterization showed errors below machine-zero with noise-free data, and above 20% noise-to-signal strength, the error increased linearly with the noise. This solver can be implemented into AFRL's plasma simulator, the Thermophysics Universal Research Framework (TURF) and used to quickly and accurately compute the electric field based on charge distributions. The validity of a machine learning approach and data-based complex system modeling approach was demonstrated. To this end, several multilayer perceptrons were created and validated against AFRL-provided Hall Thruster test data, with two networks showing mean error below 1% and standard deviations below 10%. These results, while not ready for implementation as a replacement for lookup tables, strongly suggest paths for future work and the development of networks that would be acceptable in such a role, saving both RAM space and time in plasma simulations.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 2018
- Accession Number
- AD1063544
Entities
People
- Joseph R. Whitman
Organizations
- Air Force Institute of Technology