Learning-Based Output Feedback MPC and Q-Learning for Flutter Suppression

This paper presents the use of model predictive control and Q-Learning for active flutter suppression for a pitch and plunge apparatus. The main problem with the flutter phenomenon is that the dynamics have a high dependency on the airspeed. The proposed model predictive control has the advantage that the model can be dynamically adjusted continuously to match the current state of the system, allowing one controller to be applicable for all airspeeds. Whereas Q-Learning is a model-free reinforcement learning algorithm that keeps learning and adapting itself to changes in the dynamics. Both controllers have been adapted for the typical aeroelastic section, and the effectiveness of the controller is shown. While the MPC can handle higher flutter speeds than with Q-Learning, it also needs more computational load.