Compensating the effects of FES-induced muscle fatigue by rehabilitation robotics during arm weight support

2.1 Experimental setup

Experiments were performed using the setup shown in Figure 1. A stimulator (RehaStim, Hasomed GmbH, Germany), an active cable-driven robotic system (Diego, Tyromotion GmbH, Austria), an EMG Amplifier (Porti 32, TMSi, Netherlands) and a computer running Linux were used. The control algorithms were implemented in Simulink^® (Mathworks, USA) and a modified Linux real-time target was used to generate an executable binary that runs in soft real-time at 25 Hz. Internal variables and parameters were visualized and set using a Python-GUI.

Figure 1

Experimental Setup: The required elbow rope tension u_Te and the recruitment level rλ^ to hold the arm at a shoulder flexion angle θ_u are scaled with the support factors s_r and s_q respectively. The total scale factor s_total = s_r + s_q has to be within [0, 1]. The EMG is recorded and processed to obtain the FES induced muscle recruitment level λ^ and the volitional muscle activity γ^. The recruitment controller calculates the required stimulation intensity q which is then generated by the stimulator using the current I and the pulse width pw. The cable-driven robotics generates the tension vectors s_rT_E and s_rT_W.

The stimulation pulses were transmitted to the muscle via two self-adhesive hydro-gel transcutaneous electrodes overlapping the anterior and medial part of the deltoid muscle. The stimulator generated bi-phasic stimulation pulses of current I ∈ [0, 100] mA and pulse width pw ∈ [100, 500] μs. Thereby I and pw were computed to realize the normalized pulse charge q according to [9].

Two ropes of the robotics system were attached to the forearm’s elbow and wrist joints. The actuators, located above the user, apply a tension to the ropes and attached sensors provided data from which the position of the arm and thus the shoulder flexion angle θ_u were calculated.

2.2 Control idea

Two control algorithms provide adaptive weight support to the user’s arm. The first control algorithm calculates the rope tensions required to hold the arm at the current angle θ_u. A second algorithm computes the needed stimulation intensity to maintain θ_u. Both methods are scaled by factors to determine their amount of weight support provided to the user’s arm. A scale factor controller adapts these two factors if FES-induced muscle fatigue is detected.

2.3 FES weight support controller

The virtual weight compensation using FES is implemented as presented in [10]. The FES induced muscle recruitment level λ^ is estimated from the measured evoked EMG. During an identification routine, the required λ^ for each θ_u ∈ [40, 100]° is obtained. A discrete-time integrating controller with anti-windup strategy is chosen as recruitment controller. It adjusts the stimulation intensity q depending on the current muscle activity λ^ and the required muscle activity rλ^(θ_u), whereby q is limited by the maximum tolerated stimulation intensity q_max of the subject.

As muscle fatigue occurs, the muscle recruitment level λ^ generated by a certain value of q decreases. To achieve a certain θ_u, the recruitment controller increases the stimulation intensity q in order to obtain the muscle activity required to hold the arm at this angle. The patient dependent FES support factor s_q ∈ [0, 1] scales the provided support to an appropriate level.

2.4 Robotics weight support controller

The weight support applied by the robotics to compensate the gravitational forces of the arm is implemented as described in (11). The required rope tension at the elbow ∥TE∥2 is calculated by a static arm model depending on θ_u, where ∥..∥2 denotes the euclidean norm. From experience, the rope tension at the wrist ∥TW∥2 is chosen as 0.55∥TE∥2, which achieves an almost horizontal forearm posture.

In contrast to [11], no additional identification routine is necessary due to an alternative determination of the necessary arm parameters. The weights of the upper arm and forearm are considered to be a percentage of the total body weight [12]. The lengths of both arm segments are measured by means of the robotics at the beginning of each experiment. The patient dependent factor s_r ∈ [0, 1] scales the provided elbow rope tension to uTe=sr∥TE∥2.

2.5 Scale factor controller

Our approach assumes that the user alternates between holding his arm at a lower angle and an upper angle by following a given reference for θ_u. The reference is shown to the subject, who is asked to follow it by using his residual voluntary activity. The scale factor controller updates the scale factors s_q and s_r at the end of every iteration k after the subject’s arm comes down from the upper angle. The mean of q during the phase when the upper angle is reached, is averaged over the last three iterations to yield q¯high[k]. If it exceeds a patient dependent threshold σ_q, the FES scale factor is reduced and coincidentally the robotics scale factor is increased.

To achieve this, the FES scale factor s_q is adapted by an integrating controller as

where K_I is the adjustable gain of the integrator and the error e[k] is defined as

Thereby s_{q, init} is the initial value of s_q at the beginning of the experiment. The total support level s_total = s_r + s_q is maintained by s_r[k] = s_total − s_q[k].

2.6 Experimental procedure

The complete setup was implemented as described in Section 2.1. At first, the maximum tolerated stimulation intensity q_max of the subject was detected by increasing q until the subject reported discomfort. Then q was increased until q_max was reached and afterwards decreased to zero to identify the required λ^ for each θ_u ∈ [40, 100]°. For a patient, the therapist would now manually test with which setting of s_total the patient would just be able to reach the desired upper angle while exerting the maximum of his volitional effort. Thereby, a higher s_q than s_r should be intended. σ_q should be chosen as 90 % of q_max to allow, if necessary, short excursions above σ_q. Nevertheless, this has to be evaluated in experiments with patients.

For the healthy subject, a lower threshold σ_q = 0.75 q_max and at the same time a high s_total = 0.9 were chosen. Thus, it was possible to provoke relatively quickly occurring muscle fatigue. The subject had to complete 20 iterations of a movement, successively raising and lowering the arm to achieve a shoulder flexion angle θ_u that is within two reference bands at 50° ± 5° (lower angle) and 90° ± 5° (upper angle). For this reason, the actual desired reference band was visualized to the subject.