Direct artificial neural network control of single link flexible joint

The currently existing methods for control of a flexible joint manipulator have not been desirable precision because of existence of uncertain and nonlinear dynamics. In this paper, we propose an online neural network controller for single-link flexible joint manipulator. The nonlinear dynamical model of the manipulator is determined by Lagrange's approach, and its order is four; however, the model is uncertain. A multi-layer perceptron neural network consists of three layers with one hidden layer as a controller. The weights of the proposed controller are regulated by back propagation error online approach. The learning rate is carefully selected so that the output response has a minimum oscillation. The control objective is high precision tracking the angular displacement trajectory. The simulation results show the effectiveness of the controller.