Analysis of dynamic compressive fracture of carbon composites by digital image correlation and machine learning

Detalhes bibliográficos
Ano de defesa: 2019
Autor(a) principal: Cidade, Rafael de Azevedo
Orientador(a): Não Informado pela instituição
Banca de defesa: Não Informado pela instituição
Tipo de documento: Tese
Tipo de acesso: Acesso aberto
Idioma: eng
Instituição de defesa: Universidade Federal do Rio de Janeiro
Brasil
Instituto Alberto Luiz Coimbra de Pós-Graduação e Pesquisa de Engenharia
Programa de Pós-Graduação em Engenharia Metalúrgica e de Materiais
UFRJ
Programa de Pós-Graduação: Não Informado pela instituição
Departamento: Não Informado pela instituição
País: Não Informado pela instituição
Palavras-chave em Português:
Link de acesso: http://hdl.handle.net/11422/13532
Resumo: This thesis proposes the use of digital image correlation technique on the analysis of dynamic compressive fracture of carbon fibre reinforced composites (IM7-8552). Impact tests images of double edge notched specimens were analysed. A method for determining the subset size base on morphological aspects of the speckle pattern is presented. The dynamic fracture toughness was defined as the the energy release rate at the peak load, during the test, and was calculated by an extended J-integral over an area domain. The comparison between area and contour domain integrals results is also presented and shows that the area domain is a more robust method regarding the method parameters, showing a maximum of 14% discrepancy amongst results. An artificial neutral network was used to calculate the relative influence of the parameters. The results did not present satisfactory consistency amongst specimen of the same types, yet show, in a certain level, agreement with those found by the research group by using the size-effect law methodology. In addition, a novel methodology is proposed, utilising convolutional neural networks to predict the fracture initiation by the stress fields. A case study for type III specimens shows that a methodology including convolutional neural networks can be developed as an alternative to the peak load assumption.