Thesis Type: Postgraduate
Institution Of The Thesis: Yildiz Technical University, Graduate School Of Natural And Applied Sciences, Turkey
Approval Date: 2023
Thesis Language: English
Student: METE BERBEROĞLU
Supervisor: Tülay YıldırımAbstract:
Limb loss is a serious problem which can impact a persons life in different aspects. Requirement of a solution for this problem created the field of prosthetics. Despite being a field which has been studied excessively for decades, the field still has a great length to go. The field took advantage of today’s developments on electronics, robotics and material sciences, but the prosthetic devices which are being widely produced today still lack the functionality and intuitiveness of the real limb they try to imitate. Even the commercially produced prostheses can only imitate basic gestures like grasping and pointing with limited variety. Some of the most improved experimental prosthetic devices can be controlled directly with motor neurons, but these kind of prostheses require surgical intervention and are very expensive. Even if they prove to be an optimal solution, they won’t be widely available because of their costs for a long time. In the meantime, the electronic prosthetic devices which utilize surface electromyography and machine learning try to fill the gap. Even though they are very advanced, these kind of prosthetic devices still cannot satisfy the requirements of the people in need. Furthermore, they are too expensive and they lack the functionality so much so that some of the users still prefer mechanical prosthesis. The aim of this thesis is to create a relatively cheap and more functional control method which addresses to solve the cost and functionality problems. Proposed control method takes advantage of the intuitiveness of sEMG signal and artificial intelligence. In this thesis, classification of the finger movements was studied. In order to achieve this goal, Myoware muscle sensor v1 was used for signal acquisition and Raspberry Pi Model 4 B was used for feature extraction, classification and motor control.Three different but similar methods were tested in this thesis in total. The first method's test resulted with 74% accuracy rate. The second method's test was carried out to improve the results of the first test and resulted with 76% accuracy rate. The third method's test was carried out to evaluate the performance of the methods and compare it with the literature and resulted with 84% accuracy rate. Despite being promising, the control system needs further development for a more robust and reliable control. Proposed methods were practically tested and resulted similarly. As a result of this thesis, the second method's test showed that a higher number of movement sets can be reached with a low cost and the third method's test showed the validity of proposed methods in comparison with other methods proposed in the literature.