Lower Limb Torque Prediction And Muscle Fatigue Analysis Based On Cerebellar Model Neural Network

Torque prediction has important applications in many fields,such as sports rehabilitation,clinical medicine research and so on.In the intelligent rehabilitation equipment with biofeedback,it is necessary to predict the status of muscles in real time.Real time monitoring and predicting muscle fatigue state in advance is helpful for doctors and physiotherapists to carry out rehabilitation treatment for patients,and it can be used as an index to judge the prognosis and diagnosis.In addition,predicting muscle activity and fatigue can serve a wider population by preventing individuals from reaching muscle activity levels at risk of injury.Surface electromyography(s EMG)signal is a kind of bioelectrical signal,which contains human motion intention information and contains a lot of motion control information.It is widely used in human-computer interactive rehabilitation robot,intelligent prosthesis and other intelligent rehabilitation equipment.Because surface electromyography(s EMG)contains abundant information such as joint torque and joint movement,this paper uses s EMG signal to predict the related torque and analyze muscle fatigue.In this paper,the ankle joint torque in the lower limb is taken as the research object,the feature extraction method of muscle collaborative analysis is used,the wavelet function is used as the membership function and the fuzzy logic is added to improve the CMNN and RCMNN,and the improved RCMNN is used as the torque prediction model.The torque is predicted in three states.Then,several characteristic parameters of EMG fatigue,which have good effect on the degree of fatigue,are used to predict the characteristic parameters in a specific period of time by using normalized least mean square(NLMS)filter,and then the improved CMNN is used to classify the three fatigue states.The main contents of this paper are as follows:(1)Torque prediction: the surface electromyography(s EMG)signal,velocity,position and torque information of ankle joint were collected,and the EMG signals were preprocessed by filtering,resampling and normalization.Aiming at the problem that the generalization ability of recurrent cerebellar model neural network is relatively poor,fuzzy logic theory is added to the network and combined with wavelet function to obtain faster global convergence speed.Muscle collaborative analysis method is used to extract torque prediction features to improve the practicability of the system.The improved recurrent cerebellar model neural network is compared with other traditional neural network prediction methods in the three states of non fatigue,excessive to fatigue and fatigue.(2)Muscle fatigue analysis: the collected EMG signals were preprocessed by filtering,segmentation and normalization,and the prediction and classification results of muscle fatigue were analyzed.Firstly,the characteristic parameters of EMG,which represent the degree of EMG fatigue,are analyzed,and the characteristic parameters with better performance are selected to form the fatigue characteristic parameters group.In addition,the NLMS algorithm is used to predict the fatigue eigenvector group in a specific time period,and then the fatigue eigenvector group is reduced by principal component analysis.Finally,the improved cerebellar model neural network is used to classify the fatigue state into three categories: non fatigue,excessive to fatigue and fatigue.The experimental results show that the accuracy of the improved recurrent cerebellar model neural network is better than other traditional neural networks in the three states of non fatigue,excessive to fatigue and fatigue.In the analysis of muscle fatigue,the improved classification method of cerebellar model neural network has also achieved good results.