Artificial neural networks applied to structural damage identification using dynamic response and signal processing

• Ano de defesa: 2020
• Autor(a) principal: Reis, Pedro Almeida
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Engenharia mecânica; Redes neurais (Computação); Falha de sistema (Engenharia)
• Link de acesso: https://repositorio.udesc.br/handle/UDESC/18326

In general, all structures can be subject to damage leading to catastrophic failures, causing loss of human life, environmental tragedies, and significant financial losses, due to incremental nature, observed from its emergence. Aiming to ensure reliability and safety, Structural Health Monitoring (SHM) has gained evidence, mainly through the use of tools such as the VibrationBased Model (VBM) and Artificial Neural Network (ANN). VBM uses the premise that some properties are modified when damage is present, such that the dynamic response is modified, making it possible to monitor these changes to identify damage. In this way, this work focuses on the utilize of ANN to identify damage in rolling bearings and composite structures using vibration data. However, it is not feasible to directly use these high dimensional data, since it requires complex models, with a high computational cost. Therefore, it is necessary to use tools capable of processing the data without losing fundamental information for detecting and classifying damages. Thus, strategies as Statistical Parameters (SP), Dislocated-Series (DS), and Principal Component Analysis (PCA) were used to this end. Finally, it is evaluating each one from some case studies considered as damage detection in balls and inner race for rolling bearings of the data set provided by Case Western Reserve University (CWRU) and delamination damage in composite beams. The results found in this work show that the proposed damage detection methodology is an efficient SHM tool