Friction compensation in servo motor systems using neural networks

Compensation of negative effects caused by friction in high precision servo control systems is an important and challenging problem. Conventional compensation methods often rely on an explicit friction model, which is difficult to acquire accurately in practice. We propose a neural network-based compensation scheme to cope with this problem. The visible disturbance resulting from friction is first identified by a BP neural network. The friction compensator is constructed by cascading this neural identifier with the inverse model of the motor system. It is shown that our approach has the advantages of simplicity and generality. Moreover, no prior information concerning the friction is needed. Simulations are carried out to demonstrate the efficiency of the proposed method in compensating for deterministic as well as nonlinear friction.