Modeling The Hysteresis Characteristics Of Harmonic Reducers In Industrial Robot Joints Based On The Memristor Model

With the improvement of manufacturing automation and intelligent production level,industrial robots are playing an increasingly important role in manufacturing.Harmonic reducer plays an important role in reducing the output speed of the manipulator in the robot and increasing the output torque of the manipulator due to its simple structure,high transmission ratio and stable operation,and occupies an important position in the field of industrial robots.In the transmission process,there is a non-linear relationship between the output torque and the output torsion angle due to the influence of lost motion,transmission error,insufficient torsional stiffness and friction,etc.,which leads to reducer showing hysteresis characteristics.As an inherent property of the harmonic reducer,the hysteresis characteristic seriously weakens the transmission accuracy of the transmission system and limits the execution accuracy of the target positioning of the industrial robot.Accuracy compensation of the reducer is one of the effective ways to improve the robot repeatability accuracy,and reducer modeling is a prerequisite for achieving compensation control.Based on the improved memristor hysteresis model and neural network,this paper establishes three models for the hysteresis characteristics of the harmonic reducer.(1)Memristor hysteresis modelThe volt-ampere characteristic of memristor has nonlinear memory characteristic.Based on the volt-ampere characteristics,the memristor hysteresis model is modified to describe the basic output characteristics of the hysteresis of the harmonic reducer by combining memristor model similarity with the output characteristics of the harmonic reducer.(2)Hybrid hysteresis model of harmonic reducer with memristor hysteresis model and neural network in seriesThe memristor hysteresis model can only approximate the hysteresis characteristics of the harmonic reducer.To further describe the hysteresis characteristics of harmonic reducer accurately,a hybrid hysteresis model of harmonic reducer is constructed by connecting the memristor hysteresis model with neural network in series.The hysteresis model of the harmonic reducer is completed by connecting the hysteresis model with the RBF neural network in series to realize the nonlinear mapping from torque to torsion angle.The experimental data results show that the constructed series hybrid hysteresis model is more accurate than the memristor hysteresis model.(3)Hybrid hysteresis model of harmonic reducer with memristor hysteresis model in parallel with neural networkThe model accuracy of the hybrid hysteresis model varies with different connection methods.In order to compare the model accuracy of the hybrid hysteresis models with different connection methods,a hybrid hysteresis model is constructed in which the memristor hysteresis model and the neural network are connected in parallel.The difference between the hysteresis output of the harmonic reducer and the output of the memristor hysteresis model is compensated by the RBF neural network with nonlinear fitting capability.The hysteresis modeling is completed by the summation of the memristor hysteresis model and the output of the neural network.The experimental data validation results show that the constructed parallel hybrid hysteresis model can reflect the abrupt and non-smooth characteristics of the harmonic reducer hysteresis more accurately than the series structure hybrid hysteresis model.The memristor hysteresis model and the hybrid hysteresis model with neural networks connected in series and in parallel,have different structures and differences in model accuracy.The results show that the accuracy of the two hybrid hysteresis models is higher than that of the memristor hysteresis model,and the accuracy of the parallel hybrid hysteresis model is higher than that of the series hybrid hysteresis model.