Research On Lower Extremity Exoskeleton Motion Perception Method For Multiple Motion Modes

The lower extremity exoskeleton is an auxiliary device that assists human body movement and improves human movement ability,and has been widely used in many fields.The basis of precise assisting of the lower extremity exoskeleton is the accurate identification of the wearer’s movement intention,and finally achieve the effect of human-machine cooperative movement.Based on the analysis and data collection of the lower limb movement,this paper screens the main features of the movement information,uses machine learning algorithms to identify and predict the movement pattern and gait phase of the lower limbs,and conducts a simulation based on the exoskeleton platform.Verify experiment.The main work of this paper is as follows:(1)Lower limb movement analysis and data collection.Based on the analysis of the movement mechanism of the lower limbs,the gait phase and movement mode of the lower limbs of the human body are divided to determine the data collection requirements.According to the collection requirements,a motion data collection system is designed to collect data of different motion patterns.(2)Feature selection for lower limb motion information.Aiming at the problem of redundancy among the motion data features extracted based on experience,the FCBF feature selection algorithm is used to select the original features,and the main features of motion pattern and gait phase recognition are screened out to provide morestreamlined information for subsequent motion perception recognition.feature set.(3)Research on lower limb movement pattern recognition methods.Build a random forest classification model to identify and classify motion patterns.For the base classifiers with strong and weak classification capabilities in the traditional random forest algorithm,the shortcomings of using equal voting to calculate the output are added to the decision tree output of the random forest.weight vector,and optimize the parameters of the weight vector through the particle swarm optimization algorithm.(4)Research on the recognition method of lower limb gait phase prediction.On the basis of classifying motion patterns,a PSO-BP neural network model is constructed to identify gait phases.Aiming at the problem that the sensor signal hysteresis leads to the lag of lower extremity exoskeleton control,a multivariate time series prediction model TPA-GRU with gated recurrent neural network combined with temporal pattern attention mechanism is constructed to predict the motion data.The experimental results show that the gait phase classification error of the PSO-BP classification model on the real dataset and the predicted dataset is less than 1%,which verifies the effectiveness of the gait phase prediction recognition method.(5)Finally,based on the exoskeleton prototype platform,verification experiments were carried out on the designed lower limb motion data acquisition system,motion pattern classification model and gait phase recognition prediction method.The experimental results show that the lower extremity exoskeleton integrated with the motion perception system can effectively and comprehensively collect the motion data of the lower limbs.The motion intention recognition method proposed in this paper has an accuracy rate of 98.12%for motion pattern recognition,and a gait phase prediction accuracy rate of 92.41%,which can accurately predict and identify motion patterns and gait phases.