LIGHTWEIGHT VIBRATION CONTROL USING NEURAL NETWORKS

A decentralized neurocontroller for motion control and vibration suppression for flexible structures is proposed for Phase I. This controller uses forward dynamics and a limited look ahead horizon to evaluate the effect of each control command. Thus the controller is more effective for systems having periodic behavior than the usual horizon-of-one controllers. This controller is also more stable than the commonly used inverse dynamics controller. The neurocontroller gives high performance, with tracking errors below 7x10-4 radians, and over 90% vibration reduction. The decentralized architecture allows for fast and accurate control. The controller is superior to conventional controllers in its ability to learn on-line, thus it is more suitable for deployment where autonomous adaptation is imperative. Preliminary research do date has resulted in one simple conclusion: The use of a neural network controller for active vibration suppression has been shown to be both feasible and realistic.