Utilização de inteligência artificial para análise modal de uma viga engastada sob escoamento em túnel de vento

Detalhes bibliográficos
Ano de defesa: 2023
Autor(a) principal: Rodrigues, Pedro Augusto Dias lattes
Orientador(a): Fagundes Neto, Marlipe Garcia lattes
Banca de defesa: Fagundes Neto, Marlipe Garcia, Pena, José Luiz Oliveira, Colherinhas, Gino Bertollucci, Kitatani Júnior, Sigeo
Tipo de documento: Dissertação
Tipo de acesso: Acesso aberto
Idioma: por
Instituição de defesa: Universidade Federal de Goiás
Programa de Pós-Graduação: Programa de Pós-graduação em Engenharia Mecânica
Departamento: Escola de Engenharia Elétrica, Mecânica e de Computação - EMC (RMG)
País: Brasil
Palavras-chave em Português:
Palavras-chave em Inglês:
Área do conhecimento CNPq:
Link de acesso: http://repositorio.bc.ufg.br/tede/handle/tede/12723
Resumo: Wind turbines suffer severe damage due to excessive wind loads or inadequate maintenance conditions, and catastrophic failures often occur causing huge losses. A structure, such as a wind turbine, can be monitored and evaluated through its modal characteristics, where natural frequencies, for example, are characteristics that are independent of operating conditions. They change only in case of damage, i.e. when stiffness and mass change. However, for the application of modal analysis, several sensors distributed in the structure are required, which involves high instrumentation costs. In view of this, it is proposed the use of modal analysis techniques integrated with virtual sensors, which, unlike real/physical sensors, are obtained through models. In this work, the virtual sensors are determined by using an artificial intelligence of the neural network type, which together with the modal analysis allows to obtain the modal characteristics: natural frequencies, modal shapes and damping. For this purpose, it is proposed to study a fixed Euler Bernoulli beam, an approximation model of a wind turbine, where the flow loads are generated through a wind tunnel with a speed controller. The flow velocities analyzed over the beam ranged from 10 to 20 m/s. The virtual sensor for operational modal analysis was modeled using a dynamic neural network where configurations of delay number and number of neurons in the hidden layer were investigated. In sequence, the modal characteristics of the Euler Bernoulli beam are compared using experimental modal analysis, situation in which the input is known and measured, and operational modal analysis, configuration where the input is unknown and not measured. For comparative analysis, the natural frequencies obtained in the different configurations and modal techniques showed good results when compared with the values of the Euler Bernoulli beam. For the modes, the Modal Assurance Criterion (MAC) was used, where when analyzing each independent result, the MAC returns excellent modal results, but when performing a comparative analysis of the different configurations and techniques, the MAC showed low correlation. Finally, the damping ratio showed an increase for higher flow velocities, but further investigations should be carried out in future works using other operational modal analysis techniques.