Browsing by Author "Nieuwoudt, Lorelei"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Exploring EMG and EEG signals for the development of a real-time control strategy for a robotic hand
    (Stellenbosch : Stellenbosch University, 2023-12) Nieuwoudt, Lorelei; Fisher, Callen; Perold, Willem; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
    ENGLISH ABSTRACT: Beyond the pronounced physical pain caused by amputation, amputees continue to struggle with their mental health on a daily basis. While being incapacitated as a result of having fewer limbs, navigating a new normal and becoming accustomed to the dependence on others to achieve simple daily tasks yields significant frustration, and deeper emotional turmoil. Furthermore, amputees strive to regain lost functionality in their bodies by seeking out various types of assistive devices. These devices encompass a spectrum, spanning from basic wooden or plastic prostheses primarily focused on aesthetics, to body-powered alternatives, and even to highly advanced and costly active prostheses. The array of available assistive devices for amputees, along with their corresponding price points, unfortunately fails to alleviate the emotional and physical challenges they face. This predicament arises due to financial constraints, often forcing them to settle for basic and less functional prosthetic options. In particular, the most functional and impressive active prostheses are severely unaffordable for those from middle or lower socioeconomic classes. This thesis therefore undertakes an exploration of the efficacy of employing realtime EMG and EEG signals in the development of control strategies for a robotic hand. The focus of this investigation is undertaken by classifying individual finger movements and arm flexion data, through the use of machine learning in real-time. Due to the different natures possessed by EMG and EEG signals, emphasis is placed on different aspects of the design of the two systems explored. In the case of the EMG system, focus is placed on obtaining the minimum number of EMG channels, an optimal window size to collect data, and data preprocessing measures required in order to achieve optimal classification accuracy. Emphasis is also placed on utilising multithreading in order to achieve parallel real-time signal sampling and classification, which is extended into use in the development of the EEG classifier. During the development of the EEG classifier, significant emphasis is directed towards the comparison of various classifier models, coupled with a meticulous fine-tuning of hyperparameters to attain optimal outcomes. The outcome of the developed EMG classifier yielded a real-time classification accuracy of 72.2% for grouped finger movements. The EEG classifier yielded a classification accuracy of 55% for arm flexion signals. As a result of the constraints imposed by the ethical clearance granted by the Research Ethics Committee for Social, Behavioral, and Educational Research (REC:SBER) of Stellenbosch University, the development of this system relied solely on data obtained from a single subject. Embark on a captivating intellectual journey as this thesis intricately unveils the expedition that reveals the meticulous processes developed, resulting in the realisation of the remarkable classifiers presented.