Neural network-based friction and nonlinear load compensator

In this paper, the possibilities of using off-line trained neural networks as estimators and compensators of servo system nonlinearities such as friction and gravitation-dependent load have been considered. Experiments have been performed on the positioning servo system controlled by a P controller. After applying the neural network-based compensator transient responses of the system have been considerably improved. Some problems which may deteriorate the effectiveness of compensation have been observed and discussed.