Neural network modeling of cold-gas thrusters for a spacecraft formation flying test-bed

This work presents a neural network based modeling strategy to precisely identify the thrusts of cold-gas thrusters deployed in a nano-satellite experimental test-bed developed at the Canadian Space Agency (CSA). Eight thrusters are used to control the planar motion of an emulated free-floating spacecraft supported by air-bearing. Calibration experiments conducted on these thrusters revealed that the generated thrusts are highly nonlinear with respect to their inputs, the digital openings and the air pressure. Motivated by the learning and approximation capabilities of artificial neural networks (ANNs), an ANN is used to model the nonlinear thruster behavior using experimental data. The performance of the proposed strategy is satisfactory and clearly demonstrated by the resulting high precision model.