A comprehensive analysis of surface electromyography for control of lower limb exoskeleton / Abdelhakim Deboucha

Faculty of Manufacturing Engineering

Thesis (Doctor of Philosophy) -- Universiti Malaysia Pahang – 2016

Bibliography: p. 112-121

The development of exoskeleton robotic device (ERD) is one of the most applicable devices for rehabilitation purposes and human-assistance. Unlike other control methods applied to industrial robotic systems in the sense of giving specific trajectory to be tracked, ERD physically interacts alongside with the user. To attain high cognitive interaction and safe human-machine system, there is a need to detect the user‘s movement intention. One of the bio-signals that have been found to reflect directly the individual‘s motion intention is the Electromyography (EMG). Although these signals are to some extent insulated by myelin, with the remarkable advancement in bio-sensors technology and standard recommendations in signal acquiring processing, it becomes affordable to acquire, analyze, interpret and use them to control robotic devices. Surface Electromyography (sEMG) signal measured by surface electrodes has become of great interest among researchers in both clinical and engineering aspects. To ensure high cognitive user-robotic system, sEMG signal is implemented as control command for ERD. However, this signal is highly sensitive to noises and exhibits additional measurements (crosstalk) contaminated on the signal of interest. In order to add to this area of knowledge, recording and analyzing these signals may give an optimum and safe control performance for ERD. Particular experiments were conducted on the rising from a chair and walking tasks. The experiments were conducted on five subjects where the sEMG signals were recorded over four major muscles of the lower limb (Biceps Femoris (BF), Rectus Femoris (RF), Gastrocnemius (Gas) and Soleus (Sol) muscles) along with the kinematics recordings. A novel algorithm to determine the overlapped crosstalk recordings was developed along with a modified low pass filter that adaptively removes these recordings. A parametric model based on Hill Muscle Model (HMM) to estimate the knee joint moment is developed for both experiments protocols. The parametric model involves the mapping of the sEMG signals to the knee joint moment. Obviously, selecting four muscles to attain a full joint moment and motion is not sufficient, therefore we introduced the net joint moment obtained from the inverse dynamics to optimize the predicted joint moment. Initial estimate of the model is obtained from literature review while the Levenderg-Marquardt (LM) method is applied to solve the nonlinear least squares optimization problem. Results showed that the filter parameters selection could significantly affect the amplitude of the sEMG as well as it may conceal the exact onset/offset time of the signal. The developed algorithm for the crosstalk recordings detection shows ability in determining the presence of the overlapped measurements period. The results of the modified Butterworth filter showed good suppression of the crosstalk and brought the signal of interest into its baseline state. This will increase and ensure the safety of the users of the ERD. For both experiment protocols, the R2 between the net and the predicted joint moment showed good agreement in the chair-rise protocol (0.99), while the in the walking task the R2 was (0.91). The RMSE for both protocols were relatively low varying between 6.88 and 8.31. This means the model can accurately predict the knee joint moment.The development of exoskeleton robotic device (ERD) is one of the most applicable devices for rehabilitation purposes and human-assistance. Unlike other control methods applied to industrial robotic systems in the sense of giving specific trajectory to be tracked, ERD physically interacts alongside with the user. To attain high cognitive interaction and safe human-machine system, there is a need to detect the user‘s movement intention. One of the bio-signals that have been found to reflect directly the individual‘s motion intention is the Electromyography (EMG). Although these signals are to some extent insulated by myelin, with the remarkable advancement in bio-sensors technology and standard recommendations in signal acquiring processing, it becomes affordable to acquire, analyze, interpret and use them to control robotic devices. Surface Electromyography (sEMG) signal measured by surface electrodes has become of great interest among researchers in both clinical and engineering aspects. To ensure high cognitive user-robotic system, sEMG signal is implemented as control command for ERD. However, this signal is highly sensitive to noises and exhibits additional measurements (crosstalk) contaminated on the signal of interest. In order to add to this area of knowledge, recording and analyzing these signals may give an optimum and safe control performance for ERD. Particular experiments were conducted on the rising from a chair and walking tasks. The experiments were conducted on five subjects where the sEMG signals were recorded over four major muscles of the lower limb (Biceps Femoris (BF), Rectus Femoris (RF), Gastrocnemius (Gas) and Soleus (Sol) muscles) along with the kinematics recordings. A novel algorithm to determine the overlapped crosstalk recordings was developed along with a modified low pass filter that adaptively removes these recordings. A parametric model based on Hill Muscle Model (HMM) to estimate the knee joint moment is developed for both experiments protocols. The parametric model involves the mapping of the sEMG signals to the knee joint moment. Obviously, selecting four muscles to attain a full joint moment and motion is not sufficient, therefore we introduced the net joint moment obtained from the inverse dynamics to optimize the predicted joint moment. Initial estimate of the model is obtained from literature review while the Levenderg-Marquardt (LM) method is applied to solve the nonlinear least squares optimization problem. Results showed that the filter parameters selection could significantly affect the amplitude of the sEMG as well as it may conceal the exact onset/offset time of the signal. The developed algorithm for the crosstalk recordings detection shows ability in determining the presence of the overlapped measurements period. The results of the modified Butterworth filter showed good suppression of the crosstalk and brought the signal of interest into its baseline state. This will increase and ensure the safety of the users of the ERD. For both experiment protocols, the R2 between the net and the predicted joint moment showed good agreement in the chair-rise protocol (0.99), while the in the walking task the R2 was (0.91). The RMSE for both protocols were relatively low varying between 6.88 and 8.31. This means the model can accurately predict the knee joint moment.