A DNN-based algorithm for multi-constraint intelligent reentry guidance techology for hypersonic gliding vehicle

A hypersonic vehicle is an aircraft with a flight speed exceeding 5 times the speed of sound, combining the characteristics of both spacecraft and aircraft, with significant military and economic potential. The thesis focuses on a re-entry hypersonic vehicle, which is a spacecraft that travels through space and re-enters the atmosphere of a planet (e.g., Earth). When landing, a safe re-entry is needed. The entry guidance system is crucial for ensuring a successful atmospheric re-entry, particularly for these class of vehicles which face multiple constraints during this critical flight phase. Traditional entry guidance approaches, such as reference trajectory-based guidance (RTG) and numerical predictor–corrector guidance (NPCG), have been widely employed in past missions. However, the high non-linearity and non-convex path constraints of hypersonic vehicles demand more advanced solutions. This thesis introduces an intelligent multi-constraint entry guidance approach that integrates a deep neural network (DNN) with the NPCG algorithm for better real-time performance and ensure safety during atmospheric re-entry. The DNN is designed to approximate the relationship between flight states and range and flight time, enabling rapid and accurate trajectory predictions. The developed DNN-based predictor significantly improves the NPCG algorithm by replacing traditional propagation-based predictions, offering robust solutions with real-time computational capability. The structure of the thesis is organised as follows: initially, a global overview about hypersonic vehicles and related works is established. Then, the first section involves the development of the DNN using Python, with training, testing, and validation phases. Subsequently, the DNN is integrated into the NPCG algorithm, where the bank angle and the angle of attack of the vehicle are used as control variables. The guidance algorithm consists of longitudinal control, in which the bank angle and angle of attack amplitudes are determined, for each guidance cycle, by constructing a parametrized height profile, in order to satisfy the range constraint, and a velocity profile, for meeting the time constraint. Then, lateral control is elaborated, for the managment of the sign of the bank angle, for each cycle. Finally, the thesis presents the results of the proposed approach in terms of the trajectory, longitude, and latitude achieved by the vehicle at the end of the controlled flight phase. The analysis demonstrates the advantages of incorporating artificial intelligence into the guidance algorithm, offering improved real-time decision-making capabilities and overall performance.