Nonlinear optimal control for active suppression of airfoil flutter via a novel neural-network-based controller

Abstract

This paper proposes a novel nonlinear controller based on neural networks (NNs) for active suppression of airfoil flutter (ASAF). Aeroelastic flutter can damage airfoils if not properly controlled. Existing optimal controllers for ASAF are sensitive to modeling errors while other controllers less prone to uncertainties do not provide optimal control. This study, thus, focuses on solving these problems by deriving a new intelligent model-based control scheme capable of synthesizing nonlinear near-optimal control laws in real time according to a known model and online updated system dynamics. A four-degreesof- freedom aeroelastic model that has nonlinear translational and torsional stiffness and employs leading/trailing-edge control surfaces as control inputs is considered. Optimal control laws for the nonlinear aeroelastic system is synthesized by solving the Hamiltonian-Jacobi-Bellman equation through NN-based value function approximation (VFA) and synchronous policy iteration in a Critic-Actor configuration. A systematic approach based on linear matrix inequalities (LMIs) is proposed for the design of a scheduled parameter matrix for the VFA. An NN-based identifier is also derived to capture un-modeled dynamics online. Extended Kalman filters are employed to tune NN parameters. The proposed controller was tested in wind-tunnel experiments. Comparisons drawn with a linearparameter- varying optimal controller confirms the effectiveness and validity of the proposed control scheme.