Active flutter suppression using recurrent neural networks

The application of a recurrent dynamic neural network to the problem of adaptive active flutter suppression is described, A significant feature of such a neural controller is its application to an unstable and nonminimum phase system. Its development and experimental verification are carried out on a wing model with two tip control surfaces, one at the leading edge and one at the trailing edge. It combines a network for the identification of the variables to be controlled to a network carrying out the control task. The output generated by the first network is used to command the deflection of the control surfaces, The controlled variables are two appropriate wing accelerations, at the leading and trailing edges. Adaptivity is obtained by maintaining an online training of both neural networks, Extensive preliminary numerical simulations allow tuning the values of the fundamental design parameters required for an optimal compromise between computational load and system performances. The results of the application of such an approach to a real wing model have demonstrated its effectiveness in adapting to different operational conditions, The flutter free envelope of the wing model has been extended to a speed up to 34% higher than that corresponding to the uncontrolled flutter onset.