Research On Hysteresis Modeling Of Flexible Joints For Industrial Robot

Industrial robots have the advantages of high degree of automation,high repeatability,and high versatility.With the wide application of industrial robots,various application fields put forward higher requirements for various indicators of industrial robots.In recent years,robots such as lightweight robots and collaborative robots have adopted harmonic reducers as their transmission devices.The complex nonlinear hysteresis characteristics of flexible joints composed of motors and harmonic reducers directly affect the repeat positioning accuracy of industrial robots,and severely restrict the development and application of industrial robots such as light robots and collaborative robots in high-precision fields.Therefore,modeling the nonlinear hysteresis characteristics of flexible joints,from the perspective of control,based on the hysteresis model,reduces the adverse effects of the hysteresis characteristics through compensation control,which has become an important research topic.Aiming at the complex nonlinear hysteresis characteristics of the flexible joints of industrial robots,such as light robots and collaborative robots,this paper proposes the following two modeling methods.(1)Hysteresis model based on improved LSTMCombining with the memory characteristics of hysteresis characteristics,the LSTM hysteresis model of the flexible joints of industrial robots is constructed by using Long ShortTerm Memory(LSTM)with memory characteristics.In view of the LSTM hysteresis model's amplitude and phase errors when describing hysteresis characteristics,in order to further improve its model accuracy,the LSTM is improved,and an RBF neural network is connected in series after the LSTM hysteresis model to compensated the amplitude and phase errors of the LSTM hysteresis model.According to the built-up industrial robot experimental platform,the data collected under different flexible joint operating speeds are modeled and verified,and the improved LSTM hysteresis model has higher accuracy and stronger generalization ability.(2)Neural Network Hybrid Hysteresis ModelIn order to accurately describe the hysteresis characteristic curve of the flexible joints of industrial robots,a neural network hybrid hysteresis model combining Elman and RBF was constructed.First,the Elman dynamic neural network with local memory is used to construct a curve that has a smaller error with the hysteresis curve of the flexible joint,and then the RBF neural network is connected in series to compensate for the error in the Elman dynamic model.Experimental data modeling and verification results prove that,compared with the Elman dynamic model,the neural network hybrid hysteresis model has higher accuracy and stronger generalization ability.LSTM can store long-term or short-term information.Elman has local memory.In this paper,combined with the memory characteristics of hysteresis,based on these two neural networks,a hysteresis model is established.The accuracy of the LSTM hysteresis model is higher than that of the Elman model,but there is a certain error with the hysteresis characteristic curve of the flexible joint.After RBF compensation,the proposed improved LSTM hysteresis model and neural network hybrid hysteresis model can accurately model the hysteresis characteristics of the flexible joints for industrial robots.However,Elman's training parameters are few,and the model structure of the neural network hybrid hysteresis model is simpler and easy to implement.