Robust torque control of harmonic drive systems

A harmonic drive is a compact, light-weight and high-ratio torque transmission device which has almost zero backlash. Its unique performance features has captured the attention of designers in many industrial applications, especially in robotics. Despite widespread industrial application of harmonic drives, the torque control of this system has not been fully addressed. In this thesis the robust torque control of harmonic drive system is examined in detail.In order to measure the transmitted torque of a harmonic drive and for the purpose of a torque feedback scheme, an intelligent built-in torque sensing technique is developed in this thesis. Specifically, strain-gauges are mounted directly on the flexspline and therefore no extra flexible element is introduced into the system. To have maximum sensing accuracy, four Rosette strain gauges are employed using an accurate positioning method. Kalman filter estimation is employed to cancel the torque ripples, oscillations observed on the measured torque and caused mainly by gear teeth meshing. A simple forth order harmonic oscillator proved sufficient to model these torque ripples. Moreover, the error model is extended to incorporate any misalignment torque. By on-line implementation of the Kalman filter, it is shown that this method affords a fast and accurate way to filter torque ripples and misalignment torque from torque measurements.Based on experimental and theoretical studies, a systematic way to capture and rationalize the dynamic behavior of the harmonic drive systems is developed next in this thesis. Simple and accurate models for compliance, hysteresis, and friction are proposed, and the model parameters are estimated using least-squares approximation for linear and nonlinear regression models. A statistical measure of variation is defined, by which the reliability of the estimated parameter for different operating condition, as well as the accuracy and integrity of the proposed model is quantified. By these means, it is shown that a linear stiffness model best captures the behavior of the system when combined with a good model for hysteresis. Moreover, the frictional losses of harmonic drive are modelled at both low and high velocities. The model performance is assessed by comparing simulations with the experimental results on two different harmonic drives.Finally, robust Hinfinity -based torque controllers are designed and implemented for harmonic drive systems under constrained- and free-motion applications. A nominal model for the system is obtained in each case from ensembles of experimental frequency response estimates of the system, while the deviation of the system from the model is encapsulated by a multiplicative uncertainty. Robust Hinfinity -based torque controllers are designed using this information, and the controllers are implemented on two different setups using the Kalman filtered torque as an integral part of the torque-feedback loop. Exceptional performance results are obtained from the tune and frequency response of the closed-loop system, especially for the constrained-motion system. The further improve the performance of the free-motion system, a model-based friction-compensation algorithm is implemented in addition to the robust torque control. It is shown that the friction-compensation shrinks the model uncertainty at low frequencies. Hence, the performance of the closed-loop system is for tracking signals with low-frequency content