Real-time control of industrial manipulator vibration using artificial neural networks

The work reported in this paper addresses the control of robot manipulator vibration, with the specific aim of achieving a greater degree of dynamic accuracy. An overview of existing work on the modelling of robot dynamics and neural control is reported. A model of the dynamics of a two degrees of freedom manipulator inclusive of vibration, is presented and is used to train a time-delay neural network to learn the predicted end-effector vibration. The results are compared with experimental data taken from a PUMA562C industrial manipulator using laser interferometry. Control of an end-effector located, active compensator, based upon on-line training of an artificial neural network controller is discussed and recommendations which form the basis of further investigations, currently being undertaken, are presented