Machine learning-based active flutter suppression for a flexible flying-wing aircraft

It is challenging to synthesize controller parameters for high-dimensional aeroservoelastic systems, such as a flexible aircraft, so that the controller cannot work effectively. This paper presents a novel design approach of machine learning-based control law for the problem of active flutter suppression. The approach is able to automatically tune the controller parameters via machine learning and avoid the conventional and tedious procedure of manual tuning. As such, the approach leads to a controller with better performance synthesized. The paper deals with a case study of active flutter suppression for a flexible flying-wing aircraft and demonstrates the control performance and efficiency of the machine learning-based control scheme in expanding the flutter boundaries. Based on the environment/agent interface of reinforcement learning, the proposed approach takes the closed-loop aeroservoelastic system as an environment and the actor-critic neural networks as an agent. The approach trains the policy of synthesizing the optimal controller parameters through the interaction between the environment and the agent. In the numerical simulation, with the active flutter suppression controller synthesized via the well-trained policy automatically, the critical flutter speed of the closed-loop aeroservoelastic system increases by about 36.6% compared to the open-loop system robustly. Moreover, the stability and the robustness of the closed-loop aeroservoelastic system designed via the proposed approach are better than that with a conventional robust H_∞ controller.