A New Wrist-Forearm Rehabilitation Protocol Integrating Human Biomechanics and SVM-Based Machine Learning for Muscle Fatigue Estimation

Abstract

In this research, a new remote rehabilitation system was developed that integrates an IoT-based connected robot intended for wrist and forearm rehabilitation. In fact, the mathematical model of the wrist and forearm joints was developed and integrated into the main controller. The proposed new rehabilitation protocol consists of three main sessions: the first is dedicated to the extraction of the passive components of the dynamic model of wrist-forearm biomechanics while the active components are extracted in the second session. The third session consists of performing continuous exercises using the determined dynamic model of the forearm-wrist joints, taking into account the torque generated by muscle fatigue. The main objective of this protocol is to determine the state level of the affected wrist and above all to provide a dynamic model in which the torque generated by the robot and the torque supplied by the patient are combined, taking into account the constraints of fatigue. A Support Vector Machine (SVM) classifier is designed for the estimation of muscle fatigue based on the features extracted from the electromyography (EMG) signal acquired from the patient. The results show that the developed rehabilitation system allows a good progression of the joint's range of motion as well as the resistive-active torques.

Keywords: SVM classifier; human robot interaction; machine learning; robotic rehabilitation; wrist–forearm biomechanics.

Figure 1

Movable RoM of the hand. (a) Wrist joint: extension and flexion, (b) Wrist joint: radial and ulnar deviation, (c) Forearm movement: supination and pronation.

Figure 2

Three-dimensional design of the robot: (1) Inner plate; (2) Forearm outer plate; (3) Arm outer plate; (4) Giant scale servo motor-HS-805BB; (5) Nema stepper motor; (6) Screw-nut system; (7) Servo-motor HS-755HB.

Figure 3

Robot kinematics.

Figure 4

Control architecture overview.

Figure 5

Wrist model and resulting torque.

Figure 6

Loaded robot control architecture.

Figure 7

Muscle fatigue estimation approach.

Figure 8

Representative (a) nonfatigue and (b) fatigue segment of sEMG signals.

Figure 9

Median value extracted from (a) MNF and (b) MNP.

Figure 10

Mean value extracted from (a) MNF and (b) MNP.

Figure 11

Operating system flowchart.

Figure 12

Iot platform.

Figure 13

Setup sequences: (a) measuring passive wrist components; (b) measuring wrist active components; (c) continuous exercises.

Figure 14

Data base interface: (a) add new subject; (b) save exercise; (c) generate report.

Figure 15

Generated reports: (a) patient 1; (b) patient 2; (c) patient 3.