Prediction model for dynamic joint torque of lower limb with surface EMG

Abstract

To achieve the dynamic joint torque decoding from surface electromyography (EMG), the forward biomechanical model of lower limb motor system, which relates the surface EMG and joint torque, is established. The dynamic surface EMG to skeletal muscle activation model is constructed from the perspective of amplitude and frequency. Then the muscle contraction model reflecting physiological structure and micromechanical properties is constructed according to the sliding-filament theory. The force direction and displacement vector of active muscle are determined and the transformation from muscle force to joint moment is realized. The dynamic calibration for the forward biomechanical model using the exact joint torque value obtained with Newton-Eular method is finally put forward. Following the calibration, the flexion/extension (FE) knee joint torque of four objects under different speed walking is predicted. The results show that the forward biomechanical model can capture the general shape and timing of the joint torque, the maximum absolute error is (11.0±1.32) N·m, the mean residual error is (4.43±0.698) N·m, and the linear relationship between predicted and exact knee FE torque reaches 0.927±0.042. This prediction model provides an interface for the study of force interaction pattern in the process of human-machine cooperation in training.