Neural control of a flexible-link manipulator

This paper addresses control of a flexible one-link manipulator whose dynamics are only partially known. Control and identification is initiated as a two stage process in which the identification of the unknown part of the system is done using a neural net in a supervised training mode. Control is accomplished using a model-based predictive control scheme for the known dynamics and an unsupervised neural net-based control scheme to provide “corrective” control of the unknown system dynamics. Simulations of a one-link flexible manipulator are performed first for a constant payload, then for a variable payload, showing the ability of the neural net controller and identifier to adapt to a changing environment