Abstract:
Neural networks possess a nonlinear approximation property, so they can be employed in direct adaptive control for flexible manipulators. In this thesis, a rigid model-based neural network control is proposed. The structure of controller is composed of two parts. The first part is designed from the rigid model and the second part, which comprises a neural network, is used to control the flexible part of the flexible manipulator. A major advantage of the neural network adaptive controller design over previous ones is that it does not need the linearity-in-the-parameters assumption. Moreover, model details that are hard to obtain exactly, especially in case of flexible mainpulator, are not necessary. In addition, since the parameter adaption law is obtained from the Lyapunov approach, the neural networks learn on-line in real time with no off-line training needed. Both the computer simulation and the experimental result of single-link flexible manipulator control show that the proposed controller can be used satisfactorily.