Physics-Informed Machine Learning Approach to Satellite Attitude Dynamics and Control

This thesis explores a novel Physics-Informed Machine Learning (PIML) approach to solve the minimum-time rest-to-rest reorientation problem in satellite attitude dynamics and control. The problem of satellite reorientation consists in maneuvering a satellite from an initial rest state to a desired rest state within the shortest possible time. It is critical for a wide range of space missions, including Earth observation, scientific research, and defense applications. While traditional optimization methods are effective, they are computationally intensive and sensitive to initial conditions, posing challenges for real-time applications. To address these limitations, this research investigates a Physics-Informed Neural Network (PINN) architecture specifically designed to solve the minimum time reorientation problem. The PINN integrates the governing physical laws of satellite motion directly into the training process of the neural network, enhancing its ability to predict optimal trajectories compared to purely data-driven models. This research aims at bridging the gap between traditional control methods and emerging Machine Learning (ML) approaches, laying the groundwork for a promising new methodology for solving complex control problems in space missions.