Adaptive Neuro-Fuzzy Friction Compensation Mechanism to Robotic Actuators

This paper presents a non-linear friction compensation mechanism using a combination of neural network (NN) with fuzzy system (neuro-fuzzy compensator), applied to harmonic-drive robotic actuators. The friction compensation torque is constituted by NN output, which is trained off-line. Since the friction changes significantly over time, temperature and equipment operational conditions, the NN loses its performance. To recover this performance, a fuzzy algorithm is proposed to deal with the variation friction parameters. The output of the fuzzy algorithm is a gain that multiplied by the NN output will adjust the friction compensation torque. Experimental results have shown the efficiency of the proposed mechanism.