Myoelectric signals (MES) contain information from where data about user movement intention in terms of muscular contraction can be extracted. This information can be detected using surface electrodes (sEMG) [123]. These signals were used in control of prosthetics and rehabilitation devices such a ‘’on-off” control [124], proportional control in position or assistive torque [125], [126], and distinguish between different kind of motion [127], [128].
This section presents the algorithm used to generate the desired reference in posi- tion and torque based on sEMG signals and pressure sensors. The sEMG signals were captured at 1000 Hz frequency with a circuit made in Carlos III University of Madrid. These signals were preprocessed, firstly the raw EMG was filtered with the high-pass filter using a zero-lag fourth-order recursive Butterworth filter to remove movement artifacts, then a full wave rectified was used, and after that in the absolute value was filtered using a Butterworth low-pass filter to cut-off frequency. After this process the algorithm for online calibration overrides the first two seconds (where the used circuit perturbation happens) and uses the next 18 seconds to detect the maximum signal for normalization process. In this time the patient is required to flex the forearm to the maximum. The normalized signal, Enorm, was calculated with the equation 7.1:
Enorm =
Eact− Emin
Emax− Emin
(7.1) where Eact is the actual sEMG signal, Emin is the minimum value of the sEMG
signal in the first 20 seconds and Emax is the maximum value of the sEMG signal in
the first 20 seconds.
Figure 7.8: High level control of the exoskeleton based on sEMG signals types of signals have been used, the sEMG normalized signal and the pressure sensor signal. These two signals were logically compared to detect the intention of move- ment. The binary result in function of the actual position of the joint generates a position reference using two type of increment: one for fast actuation (used when the actual position of the elbow joint is different to the position of the actuator reference) and another increment used to generate the reference pattern following the sEMG normalized signal over the threshold value.
In Fig. 7.8 is presented the high level control algorithm of the exoskeleton based on surface sEMG signals, for active rehabilitation therapies with the reference pattern generated in position.
Here the signals represent:
• Ek represents the sEMG signal of the biceps muscle.
• Pk represents the pressure signal of the sensors which will be placed between
the human arm and the exoskeleton.
• Ep represents the processed sEMG signal. Here the signal is filtered and passed
to absolute value.
• Pp represents processed pressure signal after calibration.
• En represents the normalised sEMG signal after calibration. The calibration
time default is 20 seconds.
• Pn represents the normalised pressure signal after calibration.
• θk is the desired angle for the elbow exoskeleton generated by the sEMG and
7.1. Operating mode 151
Figure 7.9: The generated position reference by the sEmg signal in simulation. • Vk is the PWM control signal for the exoskeleton actuators.
• Yk is the real exoskeleton angular position.
The torque reference pattern consist an assistance torque (a certain percentage of the required torque for mobilization of articulation) from the exoskeleton for the rehabilitation therapy. The reference pattern (torque reference), is calculated in func- tion of the biomechanical model of the human body and the exoskeleton parameters. The biomechanical model of the human body, estimate the necessary torque to mobi- lization the elbow in function of the patient position, weight and height. In function of this torque and the sEMG signal, a percentage of assistance in torque was generate such input reference. This is directly proportional with the sEMG signal, a decision if the patient want to continue with the movement.
Results in simulation
The first results of the two algorithms were tested in the simulation with the SMA actuator model presented in 4.4.2. For the sEMG data acquisition, the electrodes were placed along the biceps muscle fibers and on the midline of the belly of the muscle, considering that the sEMG signals have the greatest amplitude.
In Fig. 7.9 the generated reference in function of the sEMG signal in simulation is shown. Here the red signal is the sEMG signal normalized between 0 and 1. The position reference pattern was generated in function of this sEMG signal and the
Figure 7.10: The generated torque reference by the sEmg signal in simulation. actual position of the actuator which coincides with the angular position of the SMA exoskeleton. The first 20 seconds were ignored for online calibration of the algorithm. In the torque assist movement, the reference was generated in function of the biomechanical model of the human body considering that the rehabilitation is ex- ecuted standing or sitting, and in function of the sEMG signal placed over biceps muscle. A similar idea is presented in [125] but there, they do not take into account the biomechanical structure of the human body.
In Fig. 7.10 was presented the pattern reference in torque assist for one patient with mass of 70 kg and a height of 1.73 m in two cases: the exoskeleton assists the patient with the total torque, 100% (blue signal) and the exoskeleton assist with 50% of the total torque (red signal).