Control Strategy for Upper Limb Rehabilitation Robot Based on Muscle Strength Estimation

Abstract

This paper proposes a control strategy for the active training mode of upper limb rehabilitation robot based on muscle strength estimation aiming at the control problem of active interaction of rehabilitation robots for patients with hemiplegia. Firstly, the sEMG signal is preprocessed by filtering and notch methods and extracted by three time-domain eigenvalues of root mean square value, absolute value mean value and variance after completing the collection of the surface electromyography (sEMG) and muscle strength signals of the upper limbs. Secondly, the muscle strength estimation model is evaluated by relative root mean square error and R-squared after proposing a muscle strength estimation algorithm based on Adaboost improved BP neural network (BPNN). Relative root means square error value and R-squared value decrease by 0.0340 and -0.4143 on average respectively compared to the BP model. The result shows that the effect of muscle strength estimation has been significantly improved. Lastly, the simulation of the patient's elbow flexion task shows the feasibility of the controller after introducing the patient's motion intention moment into the force impedance controller.