Motion Information Collection And Motion Pattern Recognition Of Lower Limb Exoskeleton Robot

Exoskeleton robot,as a human-machine integrated wearable intelligent device that combines human intelligence and robot power.The roboy can assist or enhance human physiological functions,such as walking,and weight-bearing.It can also greatly improve the user's working ability.Exoskeleton robots can be used in military,medical,tourism,disaster relief and other fields.In the process of optimizing exoskeleton robots,human-machine collaboration is the most important evaluation criterion.That means the exoskeleton recognizes and predicts the wearer's movement intention and follows or assists the wearer in exercising.In order to achieve the goal of human-machine collaboration,the lower limb movement data must be collected first,and then the data is quickly preprocessed according to the collected data.Finally,we can use the pattern recognition method to get the wearer's movement pattern.Then the exoskeleton robot uses the corresponding assistance strategy according to the obtained movement pattern to complete the help or protection.The specific research content of this thesis includes the following aspects:(1)Set up the exoskeleton simulation platform to collect motion data.Perform kinematic analysis on the human body structure.Then analyze the degrees of freedom and selection range of the joints of the lower extremities,and design a simple simulated exoskeleton for data collection.Select the appropriate motion data signal as the data set for pattern recognition.Based on the required data,choose the motion sensor and design the corresponding data acquisition system.Complete the multi-sensor synchronous acquisition of the key joint information of the lower limbs.Finally,we can obtain the angle and angular velocity information data of the hip joint and knee joint,and the acceleration data of the big and small leg links.(2)Motion signal preprocessing.Analyze the noise source of the collected data,and design the corresponding wavelet filter for denoising according to the noise characteristics.After the noise is removed from the signal,calculations are performed for different feature extraction methods.Select the features with large changes and better separation as preliminary features.Then use the PCA method for feature fusion to further reduce the input data dimension,reduce the training time of the sports model,and improve the accuracy.(3)Recognition of movement mode and transition stage.We use KNN algorithm,GA-BP network,ELM algorithm,SVM algorithm among the many pattern recognition methods to analyze the data.By compareing the experimental results,we choose SVM algorithm for movement modes recognition.Then we analyze the movement characteristics and signal trends of the movement mode conversion stage.According to the analysis result,we can use the trained model to identify the transition stage.This paper accomplish the preliminary discrimination of the movement transition stage and provide the basis for the prediction of the movement mode.