Continuous sEMG estimation method of upper limb shoulder elbow torque based on CNN-LSTM

Abstract

Aiming at the problems of the assistance efficiency evaluate technology of the upper limb assist exoskeleton, a continuous torque estimation method based on surface electromyography signal (sEMG) and convolutional long short-term memory (CNN-LSTM) was proposed in this paper. Firstly, the dynamic analysis of human upper limb is carried out, and then the sEMG signals directly related to the torque signal of upper limb lifting process are collected, from which the features of muscle are extracted. On this basis, the CNN-LSTM network is used to construct the model of mapping muscle features to multi joint torque of upper limb, and the upper limb lifting experiment was carried out to verily the effectiveness of the network. The normalized root mean square error (NRMSE) of shoulder and elbow joint torque estimated between the estimation value and the real value are 0.1245 and 0.1276 respectively. And the average correlation coefficient (p) are 0.8902 and 0.9071 respectively. The experimental results verify the effectiveness and correctness of the method, which can provide technical support for the evaluation of exoskeleton assist efficiency.