Classificação de níveis de desbalanceamento de um aerogerador em escala utilizando máquina de vetores suporte

• Ano de defesa: 2021
• Autor(a) principal: Palácio, Gilderlanio Barbosa Alves
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Aerogeradores; Análise de vibração; DFA; SVM
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/59769

Every rotating system, such as wind turbines, suffers from unbalance problems due to bad weather, degradation and problems during operation that can cause a decrease in its performance. This dissertation analyzes vibration signals captured in a scaled turbine with different levels of unbalance. Such unbalances were caused by the addition of masses at the tips of the blades. Masses of 0.5 g, 1.0 g and 1.5 g at the tip of one or two blades were used. The rotations chosen for the test were 900, 1200 and 1500 rpm. The vibration signals went through a DFA (Extended Fluctuation Analysis), which recognizes properties and correlations of nonstationary time series. The DFA output data were classified using the support vector machine method SVM (Support Vector Machine) into groups of 3 and 7 classes. In the training stages of the support vector machine implemented in the group with three classes, there was a small reduction in the success rate with the increase in rotation speed, within the range of values from 100% to 98.80%. For the testing stage, the hit rates were slightly lower, still showing the phenomenon of a drop in the hit rate with the increase in rotation. For the set with seven classes, the rotation with 1200 rpm has the lowest success rates, in relation to the rotation level of 900 rpm and 1500 rpm, and the test stage in relation to the training stage, there are falls, sometimes sensitive , others more significant depending on the class analyzed in the three levels of rotation. Comparing with other methods implemented to classify the same data, it is clear that the SVM stands out, as it manages to maintain relatively high standards of correct answers in the classification.