Improvement of aeroelastic vehicles performance through recurrent neural network controllers

Aeroelastic systems have the peculiarity of changing their behavior with flight conditions. Within such a view, it is difficult to design a single control law capable of efficiently working at different flight conditions. Moreover, control laws are often designed on simple linearized, low-fidelity models, introducing the need of a scheduled tuning over a wide operational range. Obviously, such a design process can be time-consuming, because of the high number of simulations and flight tests required to assure high performance and robustness. The present work aims at proving the high flexibility of neural network-based controllers, testing their adaptive properties when applied to typical fixed and rotary-wing aircraft problems. At first, the proposed control strategy will be used to suppress the limit cycle oscillations experienced by a rigid wing in transonic regime. Then, as a second example, a controller with the same structure will be employed to reduce the hub vibrations of an helicopter rotor with active twist blades.