Reinforcement Q-learning based flight control for a passenger aircraft under actuator fault

This paper discusses the design of a flight control system for an airplane whose aerodynamic elevators are damaged. An optimal control utilizing a Q-learning algorithm is designed to achieve the appropriate behavior of the airplane. For this purpose, the linear dynamic model in the cruise flight of the airplane is augmented with the actuator dynamics, incorporating two tunable parameters: the control gain and the actuator time delay into the airplane dynamics. The control coefficients are computed using the Q-learning algorithm for different modes to evaluate the scenarios with healthy and faulty actuator cases. The numerical results in comparison with LQR controller highlight the potential of Q-learning as a practical approach to designing controllers for passenger airplanes, considering fault conditions. Moreover, the key advantage of the proposed algorithm compared to model-based controllers such as LQR is that this Q-learning control strategy is model-free and robust to path tracking, uncertainty, and disturbances.