ESA GNC Conference Papers Repository

The sloshing effect of liquids occurring within spacecraft’s fuel tanks is a determinant factor in the pointing performance of space missions with high-reactivity needs. Accurately modelling this phenomenon allows anticipating its detrimental effects on satellite control, and thus possibly improve the tranquillisation time. This paper examines a modelling trade-off between the quick but low performance analytical approaches and the slow but accurate CFD simulations. Based on the hypothesis that the problem presents underlying regularity characteristics, and considering that dataset can be generated thanks to CFD simulations, Machine Learning techniques are explored, from a Proof of Concept approach to a fully representative agile space mission. Using Machine Learning techniques such as Multi-Layer Perceptron (MLP), Long-Short Term Memory (LSTM) or Convolutional Neural Network (CNN), this study demonstrates high-accuracy results whilst remaining efficient in terms of computation time. Furthermore, its industrialization capability for future missions is assessed, providing promising results. For instance, it is shown that a relatively small dataset is needed to reach high representativeness and robustness. Besides, a methodology is developed to implement these models into a Guidance, Navigation and Control (GNC) closed loop simulator based on Matlab/Simulink.