NEURAL NETWORKS TO INCREASE THE AUTONOMY OF INTERPLANETARY NANOSATELLITE MISSIONS

To accomplish more ambitious scientific goals of interplanetary nanosatellite missions, a certain set of technological challenges need to be addressed to enhance systems performance. An area of particular interest is mission autonomy. Increasing the degree of mission autonomy might help overcoming the limitations imposed by the typical low data rate and the simplified ground segment of small missions. This research aims at supporting the development of more autonomous spacecraft by exploiting the potentialities of artificial intelligence. An artificial neural network is proposed to enable a set of autonomous operations, for example to select what payload data are useful for the mission and must be sent to Earth, and what data can be discarded. The algorithm is developed and tested on a case study represented by a CubeSat mission to a near Earth asteroid that requires the autonomous detection of an impact event on the asteroid surface. The proposed algorithm demonstrates the feasibility of a novel training approach based on optimized datasets created directly in-situ using images taken by the spacecraft on-board camera. The validity of the algorithm is demonstrated through several simulations, considering different scenarios and disturbances. The research presented in this paper can be extended to other applications of the artificial neural networks, such as autonomous failure detection and isolation, also in conjunction with other artificial intelligence approaches.