Self-Aware Deployable RF Antennas With Integrated Shape Sensing Using Neural Networks and Temperature Sensors

Antennas are growing in size to satisfy the demands of faster communication, connectivity, and improved remote sensing in space. However, maintaining precision in large space antennas remains a challenge due to electronic and solar driven surface deformations, which result in degraded performance. To ensure optimal RF performance while on-orbit, emerging antennas can carry the capability to correct for deformations via in situ electronic phase adjustments and mechanical actuation, but deformations need to be known a priori for real-time feedback. This work addresses this need by demonstrating a low-cost shape metrology technique capable of predicting accurate deformations in real-time from a discrete number of temperature sensors using a neural network. Experiments show the neural network can predict deformations of a 610 mm x 406 mm x 3 mm flat panel to within 0.1 mm for a variety of thermal loading scenarios unseen during experimental training. The current technique satisfies the surface precision requirements of large-scale RF arrays operating at V band wavelengths or longer at a panel level. This paper describes the framework used for training the neural network, which allowed for a low number of experimental training cases and can be extended to other space missions. Finally, the neural network architecture, test setup used to validate the neural network, and results will be discussed.