Research On Motion Intention Recognition Of Upper Limb Exoskeleton Robot And Flexible Actuator Design

Exoskeleton robot is a kind of intelligent equipment worn on human body.It has the advantages of safe in operation and simple in control,which makes it widely used in daily life.Therefore,motion intention recognition and flexible actuator design of exoskeleton robot have become one of the hot topics among of domestic and foreign scholars.At present,there are some problems in the study of motion intention of exoskeleton robot,such as the difficulty of pattern recognition to adapt to different individuals,the use of existing models to estimate continuous motion with certain angular errors,and the lack of flexibility in the study of actuator of exoskeleton robot.In view of the above problems,this paper mainly carries out the following research work.There are significant individuals differences in surface electromyography(sEMG),which lead to the problem of poor adaptability in pattern recognition.That is,a pattern recognition method can only be applied in the same subjects.Once different subjects are involved,and don’t do a lot of training,the accuracy of the pattern recognition will be drop significantly.This paper proposed a back propagation(BP)neural network pattern recognition method which could improve the adaptability of the subjects.It extracted the features from multiple perspectives in time domain,frequency domain and time-frequency domain.The features were weighted and summed to reduce the influence of the subjects’ differences on the final features;then the normalization was used to reduce the impact of different wearing positions on the features and a BP neural network classification model based on subjects’ optional movements was established.The advantages of this model were analyzed by analysis of variance and different combinations of features’ mean classification accuracy.In the experiment,the forearm movements selected by subjects were adopted to verify the applicability and rationality of the proposed algorithm.The established BP neural network classification model could be applied to classify different subjects’ optional movements,but there were some problems such as some individuals’ unstable classification results.To overcome the shortcomings of the BP neural network classification model,this paper introduced attitude sensor signals on the basis of sEMG to improve the adaptability of pattern recognition to different individuals.This paper took the standard deviation and power spectrum of the electromyography(EMG)sensor signals as the first type of features and analyzed the characteristics of the output signals of attitude sensor,then the attitude sensor’s angle,angular acceleration of the three coordinate axes(x,y and z)and the sum of the three angular accelerations were integrated as another kind of features to reduce the impact of individual differences on the accuracy of pattern recognition,improve the robustness of the results and reduce the times of training.In view of the larger error and higher dimension of the obtained features,the features were processed by normalization and elimination abnormal points,and the dimension of features was reduced by principal component analysis.In the experiment,subjects selected five types of forearm movements to verify the rationality of the proposed method.According to the muscle Hill model to estimate the angle of continuous human movement,existing research methods require more parameters,and the parameters are susceptible to the influence of different individuals,so there are often large errors in the estimation results.Therefore,a back propagation neural network model based on the features of sEMG and the angle of human movement was established in this paper.By studying the role of muscles in joint rotation,appropriate muscle tissues were selected to place EMG sensors,and the model of sEMG features and joint angle was established.According to the characteristics of the upper limb movements,appropriate degree of freedom of the joints was determined as the research object.The three degrees of freedom exoskeleton robot was designed and the established model and prototype were used to perform the tracking experiment of upper limb with three degrees of freedom.The results show that the BP neural network model designed in this paper can not only achieve uniform tracking of the upper limb with three degrees of freedom,but also ensure that the angular error is within a reasonable range,which meet the basic requirements for estimating the continuous motion of the human body through sEMG.The upper limb exoskeleton robot lacks flexibility in human-robot interaction control.The existing research mainly uses elastic components to design flexible actuator to obtain flexibility,but the output torque of such flexible actuator changes slowly and the overshoot is too large to adapt to the rapid change of output stiffness and stable operation.In this paper,the characteristics of rapid change in the viscosity of magnetorheological fluid caused by magnetic field changes and the operating mode of magnetorheological fluid were analyzed,and a structural design method of flexible actuator based on magnetorheological fluid was proposed.Based on the basic law of magnetic circuit and Ansoft Maxwell magnetic field analysis method,the input current and output torque models of the flexible actuator were established,and the magnetic field structure characteristics of the flexible actuator were analyzed.The output torque,running follow and variable stiffness performance tests were completed.The experiment results show that the flexible actuator designed has both a large torque output range and the ability to follow in real-time and change the stiffness continuously and rapidly.