Análise de padrões de fadiga muscular usando inteligência artificial e sinais de eletromiografia durante exercício dinâmico

• Ano de defesa: 2024
• Autor(a) principal: Schiaber, Pamela de Souza
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Cornelio Procopio; Brasil; Programa de Pós-Graduação em Engenharia Elétrica - Uel/Utpfr; UTFPR
• Palavras-chave em Português: Padrões de desempenho; Fadiga; Eletromiografia; Performance standards; Fatigue; Electromyography; CNPQ::ENGENHARIAS; Engenharia Elétrica
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/35328

The study of muscle fatigue is essential for preventing injuries, and numerous investigations have focused on static muscle contractions. When the contraction is dynamic (involves body movement, muscular relaxation, and contraction), the muscle behavior is different, mainly when muscle fatigue occurs. There are not enough studies on muscle fatigue analysis in dynamic contractions, and even the best features and measures to apply in the signal to identify muscle fatigue are known. Because of that, this work aims to study muscle fatigue patterns of a pedaling exercise (dynamic contraction) using different features from the signal, applying the Independent Component Analysis (ICA) in a signal pre-processing, Paraconsistent Annotated logic of two Values (PAL2v) to study the stages of fatigue and doing a classification between the classes fatigue and no-fatigue. First will be extracted different groups of features from the signal; the groups will be compared in a classification performance, and the best group will be applied in an analysis of feature impact and also in a study of the best configuration of EMG windowing and using or not the parts of muscular inactivity in the signal. After that, will be applied the PAL2v to a better understanding of the muscle behavior, analyzing a possible intermediary stage of fatigue. Using PAL2v will also define the best label propagation in the stages of fatigue. In the last step, will be made a classification between the two classes and observed the importance of the features. Results show that the best configuration uses the whole signal, the best features are from the spectral group (from the frequency domain), and the use of an intermediary stage of fatigue needs more signal processing and studies to be applied in a simple muscle fatigue classification. The best performance classification got an accuracy score of 0.84 using the XGBoost classifier and the KSWIN change detector.