Pilot Study Of Electromyography Analysis Of The Arm Muscle Using Levenberg-Marquadt Back Propagation Neural Network

• Main Author: Yahya, Abu Bakar
• Format: Thesis
• Language: English; English
• Published: 2017
• Subjects: T Technology (General); TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://eprints.utem.edu.my/id/eprint/22412/1/Pilot%20Study%20Of%20Electromyography%20Analysis%20Of%20The%20Arm%20Muscle%20Using%20Levenberg-Marquadt%20Back%20Propagation%20Neural%20Network%20-%20Abu%20Bakar%20Yahya%20-%2024%20Pages.pdf; http://eprints.utem.edu.my/id/eprint/22412/2/Pilot%20Study%20Of%20Electromyography%20Analysis%20Of%20The%20Arm%20Muscle%20Using%20Levenberg-Marquadt%20Back%20Propagation%20Neural%20Network.pdf; http://eprints.utem.edu.my/id/eprint/22412/; https://plh.utem.edu.my/cgi-bin/koha/opac-detail.pl?biblionumber=107365

The study on the EMG signal is useful in providing the information regarding to the force and motion command that can be used in clinical research, rehabilitation and assistive technology. However, it is difficult for one feature parameter to reflect as a unique feature of the measured EMG signals to the force and motion commands perfectly. Moreover, it is challenging to identify and classify the muscle force that exerted by a muscle and the muscle activities according to a specific movement. This research aims to propose EMG signal pattern recognition that based on back propagation neural network approach to identify the force and motion commands. This research focuses on the upper arm muscle, which is the biceps brachii muscle that leads in the improvement of the system of the prosthetic upper arm. The proposed EMG signal pattern recognition is consisting of data acquisition, data processing, data classification and data testing. The data acquisition phase is designed to acquire EMG signal from the subject. Features extraction for the EMG signal is carried out in the data processing phase. In this phase, statistical features such as maximum amplitude, mean and root mean square are computed for the features extraction purpose. In data classification phase, the three extracted time domain features are used as inputs to train the measured EMG signal via Levenberg-Marquadt backpropagation neural network training function. Then the EMG signal is classified via conjugate gradient backpropagation neural network training function. In data testing phases, three additional subjects are selected to follow the proposed EMG signal pattern recognition phases. In this research, muscle force model is used to determine the value of the force exerted by the biceps muscle. The muscle force model is based on the lever system in human body, which is third class lever. The results from the statistical analysis shows that the changes of the amplitude of the EMG signal are changing correlated to the changes of the muscle force exerted by the biceps muscle depending on the size of the loads. The analysis of pattern recognition for the measured EMG signal shows a good performance of the classification. The EMG signal can be classified based on the tasks of different weight of loads and different angle of the arm motion. The analysis of the muscle force model shows that the value of the muscle force exerted by the biceps muscle is different for all subjects. As the conclusion, it is proved that this research has been successfully accomplished and the relevance of the relationship between the changes in the movement of the hand towards the EMG signal changes and the changes of the force exerted by the biceps muscle has been proved. These findings are useful to be applied on the development of the assistive technology in helping the disabled person. These findings also can lead to improve the system of the assistive technology, especially for the improvement of prosthetic arms.