Neural Net Control for Electric Propulsion in 3-Body Orbits

The proposed innovation uses the powerful function approximation capabilities of neural networks (NNs) to enable real-time trajectory correction for spacecraft with electric propulsion (EP). We use a NN to ldquo;learnrdquo; the relationship between state and control near a reference trajectory, then use the NN to optimally follow the reference path in the presence of uncertainty. This innovation applies recent advancements from the field of artificial intelligence to spacecraft guidance and control. NN technology for EP automation has two key benefits: 1) the accuracy and optimality of running an onboard sophisticated program, and 2) a low computational requirement similar to legacy linear control architectures.nbsp; The innovation enables increased spacecraft autonomy and makes spacecraft robust to large errors or large changes in target trajectory. Legacy onboard control algorithms are incapable of maintaining a spacecraft in highly sensitive orbits such as Earth-Moon libration point orbits (including NRHOs). Spacecraft with EP systems currently rely on frequent ground contacts in order to get updated thrust instructions. The proposed innovation will enable spacecraft with EP to autonomously follow a nominal path or rendezvous with another spacecraft in sensitive regimes, without significant onboard computation. The innovation benefits a wide variety of NASA projects, particularly the Lunar Orbiting Platform - Gateway, which will operate in this sensitive regime with an electric propulsion based Power and Propulsion Element.nbsp; Example applications and benefits of the proposed innovation include: reducing operational costs for constellations of spacecraft, enabling EP spacecraft to perform transfers which are too sensitive for ground-in-the-loop control, autonomous stationkeeping in sensitive orbits, and ground-based Monte Carlo analyses of sensitivity and/or contingency cases. nbsp;