| Unacceptable levels of vibration continue to be a major technological challenge in the design of rotary-wing vehicles. An important source of vibration may be traced to the structure-aerodynamic interactions in the rotor system. Passive means to limit the magnitude of vibrations include design of blade stiffness and mass distribution, and the use of formal optimization methods have been explored in this context with varying degrees of success. Additional reductions in vibration levels are possible if an active control system is deployed to eliminate or minimize the vibration at its source-higher harmonic control (HHC) and individual blade control (IBC) are two concepts that have received considerable attention in the literature. The design of such a controller would require a model of the plant, which in this case is inherently nonlinear. While linear control systems have been extensively studied, their performance is inadequate for the time-varying, uncertain, and nonlinear dynamic systems. Artificial neural networks, with on-line learning capabilities provide the potential for developing robust nonlinear control strategies for these class of problems. In particular, the use of neural networks to model the behavior of an {dollar}Hsb{lcub}infty{rcub}{dollar} controller, and to modify a classic linear quadratic Gaussian (LQG) controller for nonlinear systems is explored in this study.; A simplified two-dimensional representation of the aeroelastic system, consisting of an airfoil with a trailing-edge flap, is first considered in the present work. The model allows for the inclusion of both structural and aerodynamic nonlinearities and may be extended to represent a three-dimensional helicopter blade system. Analytical models available for predicting dynamic response of helicopter blades are somewhat inadequate, and give rise to uncertainties that must be taken into account when selecting control strategies. Linear models based on system identification are ineffective in the presence of large uncertainties, and render the task of implementing robust control schemes for such systems difficult. The present study proposes the use of neural network based controllers, with on-line and off-line training implementations, to provide an alternative approach to solve these complex problems. The study also shows that such an adaptive neurocontroller performs better when it is synthesized in conjunction with a design of the rotor blade for optimized system response. In such an approach, in addition to the design variables of the control system, design variables and constraints are also drawn from disciplines such as structures, aerodynamics, and multibody dynamics. The resulting coupled multidisciplinary optimization problem is complex, involving a mixed variable design space that is also multimodal. The use of a genetic algorithm (GA) based optimization procedure is an effective tool to locate the optimal design. The study shows that an effective use of this search strategy can only be implemented in conjunction with a global function approximation scheme. |
