Técnicas de aprendizado profundo aplicadas ao monitoramento de integridade estrutural por impedância eletromecânica
Carregando...
Data
2021-07-12
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Federal de Goiás
Resumo
In this work, a study of structural health monitoring (SHM) techniques is carried out, with a
primary focus on the application of the eletromechanical impedance-based SHM method to
characterize structural failures in mechanical systems. This methodology generally combines
modern sensors with intelligent algorithms to identify the presence of damage, observing
changes in the monitored signals and correlating them with physical phenomena. However,
when these variations are sufficiently complex, it becomes necessary to apply more
sophisticated techniques, capable of abstracting information pertinent to the condition of
structural health. In this sense, innovative machine learning tools (especially Artificial Neural
Networks - RNA) have been gaining prominence, being extensively investigated to avoid errors
in structural prognosis and focusing mainly on vibration analysis. Therefore, the present work
aims to contribute to the SHM area, providing an improvement of the electromechanical
impedance technique in terms of structural health analysis, associating it with two neural
architectures (CNN and LSTM) to facilitate the decision-making process. Thus, this contribution
presents a bibliographical review of some of the main concepts associated with this theme, such
as intelligent materials, evaluation methods, the electromechanical impedance-based method
itself, the concepts associated with machine learning techniques and the artificial neural
networks used. Then, four experimental tests were investigated, employing a combination of
the impedance-based method with the respective CNN and LSTM architectures for the
identification and/or prediction of structural failures. Thus, it was possible to validate the
proposed methodologies, verifying the sensitivity of the networks to some environmental
influences (such as temperature and humidity conditions) and operational (operating speed and
unbalanced) influences of the mechanical systems under study. As a result, both proposed
methodologies are efficient in terms of the structural health monitoring, allowing an optimization
of the structural diagnosis.
Descrição
Palavras-chave
Monitoramento de integridade estrutural, Impedância eletromecânica, Aprendizado profundo, Redes neurais vonvolucionais, Long short-term memory, Structural health monitoring, Electromechanical impedance, Deep learning, Convolutional neural networks, Long short-term memory
Citação
REZENDE, S. W. F. Técnicas de aprendizado profundo aplicadas ao monitoramento de integridade estrutural por impedância eletromecânica. 2021. 143 f. Dissertação (Mestrado em Modelagem e Otimização) - Universidade Federal de Goiás, Catalão, 2021.