Motivation and Background

Hall-effect thrusters are prominent part of space industry over the past decades. With the advent of mega constellations in next few years the demand for them is likely to surge beyond what was observed at any point in history. Despite this wide adoption, the understanding of physics of the thrusters is still limited and computational models of high complexity are required for the prediction of performance. This limits applicability of the modelling in design process and thus drives qualification efforts up. Problem is even further amplified in the field of new propellants, where dedicated numerical models need to be developed from scratch often rendering physical understanding less applicable. This can be rectified by applying deep learning techniques to capture underlying physics based on the simple experimental data, thus enabling predicting performance of new thruster during the design phase.

Methodology

Current approach to the problem is focused on applying Universal Differential Equations as a method of representing the evolution of the system state, with closure terms trained to represent either experimental data or high fidelity simulations reliably. This should facilitate development of simulation models capable of capturing interactions of the thruster with both environment and other spacecraft subsystems and support the design process. In further stages of the project other methods such as reservoir computers will be investigated as means of capturing the unknown states of the system in areas which will prove to be beyond applicability of UDE approach.

Main Contact

Piotr Fil