OPTIMIZED NEURAL NETWORK BASED CONTROLLER FOR A NONLINEAR AEROELASTIC SYSTEM

Attenuation of vibratory response is an important design consideration in many aeroelastic systems, and active methods of vibration reduction have been extensively studied in this context. The synthesis of active controllers requires that a good analytical model of the system be available. In those problems where the aeroelastic system is inherently nonlinear, a robust control scheme is difficult to implement, particularly in the presence of large uncertainties in the model. The present paper explores the use artificial neural networks (NN), with on-line learning capabilities, as an approach for developing robust control strategies for such problems. In particular, the use of neural networks to mimic the behavior of a modified LQG controller that is applicable to nonlinear systems, is presented in the paper. The helicopter rotor blade is a classic example of an aeroelastic system where vibration reduction is an overriding concern, and where the plant is both nonlinear and contains uncertainties. A simplified 2-D representation of this aeroelastic system, consisting of an airfoil with a trailing-edge flap, is considered in the present work; both structural and aerodynamic nonlinearities are included in the problem. The optimal design is approached as a dual structural-control design, where the aeroelastic system is optimized in conjunction with the development of the optimal active control system.