A study of neural network and its application in robot manipulator control
Date of Issue1999
School of Electrical and Electronic Engineering
This thesis focuses on the study of the neural network (NN) and its application to robot tracking control. Firstly, a neural network tracking controller and a robust NN weight-tuning algorithm are proposed for a class of discrete-time multi-input multi-output (MIMO) nonlinear system. This scheme uses a multi-layer neural network to reconstruct a certain required nonlinear function and incorporates with a proportional controller. The dead-zone strategy is employed in the weight-tuning algorithm to train the neural network on-line. Thus, the controller exhibits a learning-while-functioning feature. Theoretical investigation shows that such weight tuning mechanisms guarantees the convergence of both the NN estimation error and the control system tracking error in the presence of disturbance. We also prove, through a Lyapunov's approach, that selection of a smaller dead-zone leads to a smaller estimate error of the neural network, in turn, a smaller tracking error of the NN tracking system. In addition, there is no linear approximation in our convergence proof to deal with the nonlinear activation function in the NN hidden layer.
DRNTU::Engineering::Electrical and electronic engineering::Control and instrumentation::Robotics