محمدرسول كنكاشور

Satellite formation flying (SFF) has garnered significant attention in recent years owing to its advantages in executing complex space missions that are unfeasible with a single satellite. The control of such formations presents substantial challenges due to the inherently nonlinear dynamics, uncertainties, and perturbations pervasive in the space environment. Moreover, space missions demand stringent reliability and fault-tolerance levels. Traditional control methodologies predominantly rely on precise system models, which may not always be available in practical scenarios. In this dissertation, novel methodologies for optimal adaptive control of satellite formation flying, leveraging reinforcement learning, are proposed. Employing reinforcement learning algorithms, particularly Q-learning, controllers are developed that can autonomously learn optimal control policies without necessitating exact system models. This approach enables the controllers to adapt to system uncertainties and parametric variations, thereby maintaining robust performance in the presence of perturbations and faults. The proposed control architectures utilize an actor-critic framework, with their stability rigorously validated using Lyapunov theory. The cost functions are designed in a generalized manner to encapsulate multiple criteria, including fuel optimization, positional accuracy, and formation geometry preservation. This generalization facilitates the simultaneous optimization of multiple performance metrics. Additionally, fault-tolerant controllers based on state and output feedback mechanisms are designed to sustain optimal system performance amidst sensor and actuator faults. A fault detection, isolation, and recovery (FDIR) algorithm is developed to enhance the system's reliability and safety. These controllers exhibit the capability to adaptively accommodate system faults, preventing performance degradation. The proposed methodologies are implemented and eva‎luated within the context of satellite formation flying. Simulation results demonstrate that employing reinforcement learning in the design of optimal and fault-tolerant controllers yields significant performance enhancements. The controllers not only maintain the desired formation but also ensure system functionality under potential faults and perturbations. This research contributes to advancing practical and innovative strategies for future space missions. By integrating reinforcement learning into the design of optimal adaptive controllers, it addresses the complexities and uncertainties inherent in space systems. It is anticipated that the findings of this study will enhance the reliability, safety, and efficiency of satellite formation flying missions.

آرايش پروازي ماهواره‌ها , يادگيري تقويتي , كنترل تطبيقي بهينه , كنترل تحمل پذير عيب , يادگيري-كيو , سيستم جند عاملي , تشخيص، شناسايي عيب و بازيابي

Satellite Formation Flying , Reinforcement Learning , Optimal Adaptive Control , Fault-Tolerant Control , Q-learning , Multi-Agent Systems , Fault Detection, Isolation, and Recovery

Mohammadrasoul Kankashvar

Hossein Bolandi