การประมาณแรงกล้ามเนื้อด้วยคลื่นไฟฟ้ากล้ามเนื้อในการทำกายภาพบำบัดกล้ามเนื้อแขน

Abstract

In today's world, technology plays an important role in every industry as well as in our personal lives. One of the entire applications is a medical technology used for physiotherapy of upper limbs. The rehabilitation activities involve the muscle force stimulation using robot-assisted-exercise-based rehabilitation. However, the development of autonomous robotic trainers always requires multi-axis force sensors, which are so costly. Typically, the force sensors are associated with their specific controller boxes, which are slightly complex systems. Therefore, this research aims to develop EMG-based muscle force estimation using artificial neural networks and support vector machine methods. This technique was successfully implemented in the prototype device of the one-DOF assistive robot for upper limb rehabilitation. Initially, a set of pilot experiments was carried out i.e. (1) the force sensor calibration test, (2) frictional force calibration test (in case of resistance applied against the robot movement), (3) the experiment to identify the appropriate EMG electrode locations by comparing two muscle positions consisting of the forearm muscles and the Biceps/Triceps muscles, and (4) the test to determine the sample sizes for calculating the features of the data set. Afterwards, the substantive experiments have been delivered. There are off-line and real-time muscle force prediction tests based on the artificial neural networks and support vector machine algorithms. According to the off-line EMG-based force estimation, it revealed two preliminary tests consisting of the first test involving to the study of the relation between the qualitative performance of the model estimations and the various frictional forces applied and the second relating to the investigation of the relationship between the model estimation performances and the hand movement speeds. Additionally, the real-time muscle force experiment was introduced to compare the accuracy of model force estimation between the artificial neural network model and support vector machine schemes based on radial basis kernel function. The experimental results show that mathematical models developed based on both methods can be considered acceptable for the EMG-based force estimation in the upper-limb treatment. The efficiency of the off-line force prediction is inversely proportional to the frictional forces and the velocity of the hand movement. In addition, the result of real-time estimation found that the mathematical models developed using the artificial neural network method were superior in the force estimating than mathematical models developed using the support vector machine. Therefore, the EMG-based muscle force estimation using both methods can be further implemented in the one-DOF upper limb rehabilitation robot.