Análise de desbalanceamento de massa, aerodinâmico e erosão no bordo de ataque em pás de aerogeradores utilizando aprendizado de máquina

• Ano de defesa: 2022
• Autor(a) principal: Rosa, Leonardo Dias da
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Engenharia Elétrica; UFSM; Programa de Pós-Graduação em Engenharia Elétrica; Centro de Tecnologia
• Palavras-chave em Português: Manutenção; Aeroelasticidade; FAST; Inteligência artificial; Maintenance; Aeroelastic; Artificial intelligence; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: http://repositorio.ufsm.br/handle/1/28494

Studies and surveys demonstrate that maintenance is responsible for up to 30% of a wind turbine project total cost. This raises the necessity of improving the maintenance policies in the wind energy industry. To improve the policies, it is necessary to take advantage of condition monitoring systems (CMS) in order to asses wind turbines health, predict faults, and optimize maintenance. However, the CMS relies on different subsystems, such as data acquisition, treatment, processing. The detection of problems and faults in order to avoid downtime or expensive maintenance requires the CMS to be reliable enough to detect and identify problems in advance and accurately. Considering the fact that wind turbines have a Supervisory Control and Data Acquisitiom (SCADA), this data can be used to develop and improve CMS. Among wind turbines faults, the literature demonstrate that three of the most commons wind turbines maintenance problems are rotor mass imbalance, pitch error, and leading-edge erosion in the blades. Those problems can be mitigated with the use of efficient CMS, and the early detection of them poses great value to avoid further complications or even catastrophic failures. This work proposes a methodology to analyse all of the three mentioned problems using data in SCADA. To this end, the data is obtained through numerical simulations with FAST. This data is used to train and test two machine learning (ML) algorithms, the support vector machines (SVM) and decision trees. Since SCADA data often is abundant in terms of variables available, a mathematical and numerical description of the problems are presented in order to define the most relevant variables to detect the aforementioned faults. With the defined variables, further data analysis is carried out to define the best range of operation of the wind turbine for detecting the problems. After training and testing both algorithms, the SVM achieved better results, with high accuracy, demonstrating the numerical and data analysis effectiveness.