A Flight-Information-Perceived Approach to Parameter Determination of Giant Asteroids: Revisiting Dawn Mission

The internal structures of asteroids are estimated iteratively via a classic gravity-field-inversed method on the ground computer in the post-mission phase due to a large amount of gravity data. This paper develops a flight-information-perceived approach that has a superb capacity for constructing the interiors of a differentiated celestial body, for example, giant asteroids, utilizing the flight paths of a spacecraft. A kind of highly efficient invertible neural networks (INNs) are adopted as the fundamental framework. Compared with the time-consuming classic method, the data-prediction process of the INNs is completed in seconds and their training is also very efficient (∼0.1 hr in Liang et al. https://doi.org/10.1093/mnras/stac3389, MNRAS). This novel approach challenges the classic method from the perspectives of observation data, optimization algorithm, and error treatment and thus is proved to remarkably accelerate the autonomous and real-time guidance and control of spacecraft in deep space missions. To verify its scientific and engineering effectiveness, the Dawn mission that adopted the classic method is revisited as an example. To guarantee fair competition, the INNs are trained using ground observations as a priori information. Once a single real flight path of the Dawn spacecraft is input, this flight-information-perceived approach immediately constructs the internal structures of Ceres and Vesta using an on-board computational device. It is demonstrated that the on-site constructions match fairly well the Dawn team's scientific conclusions on the total mass, the chemical composition, and the observed gravity field. Importantly, new knowledge on Ceres and Vesta are discovered by this novel method.