Disturbance observer‐based robust motor control enhanced by adaptive neural network in the absence of velocity measurement

Abstract

High‐precision servo motor control has been a challenging problem due to the multiple disturbances such as the friction, the load variation and the wide/narrow‐band torque disturbances. Particularly, the discontinuous and fast time‐varying friction is very difficult to be compensated, which tends to trigger the undesirable limit cycle. In addition, the friction is a function of the angular velocity, which makes the online compensation extremely hard in the absence of velocity measurement. To make the matter worse, the nonvanishing narrow‐band torque disturbances in some applications often cause residual vibrations if they are not sufficiently rejected. To address these problems, a robust motor control method, which can effectively reject the multiple disturbances without velocity measurement, is proposed in this article. To be specific, an adaptive hybrid model, which includes an adaptive neural network and a sign function, is employed to compensate the discontinuous and fast time‐varying friction dynamics. Furthermore, by incorporating the internal models of the narrow‐band disturbances into the plant, a resonant extended state observer is designed to estimate the remaining lumped disturbance. To derive the high‐order derivatives of the angular position, a fixed‐time‐convergent differentiator is employed. Different from the previous studies, elaborately designed filters are introduced in the adaptive laws. In this way, the possible degradation of transient performance and stability robustness caused by the adaptive control can be effectively avoided. The closed‐loop stability can be rigorously proved by using the Lyapunov theory. Finally, extensive experiments are performed on a servo motor to validate the effectiveness of the proposed method.