sEMG Signals and Prony Method for intelligent Control of Rehabilitation Robots

This paper suggests a new strategy for intelligent control of rehabilitation robots. A bidirectional neural network (BRNN) is developed to estimate the nonlinear model describing the relationship between the joint angles of lower limbs and the surface electromyography (sEMG) signals. The most effective features of the sEMG signals are extracted by the Prony method. The features with the highest amplitude are selected as most effective ones. In this case, the calculation and estimation time is reduced because the most valuable part of the sMEG signals is used. On the other hand, the most common methods for removing noise from the sEMG signals are no longer needed. The model accepts the features as the input and generates the estimated joint angle of knee. A variable impedance controller (VIC) improved by the features and human force corrects the reference trajectory of the rehabilitation robot. The tracking error between the robot trajectory and the reference trajectory is minimized by a Lyapunov neural network based model reference adaptive controller. Laboratory results prove that the suggested method is able to decrease the trajectory tracking error and correct the reference trajectory synchronously with the intention of moving patients.

کلیدواژه ها:

joint angle estimation ، rehabilitation robots ، sEMG signals ، bidirectional recurrent neural network