Redes neurais profundas para análise de fissuras em revestimentos argamassados com diferentes tipos de acabamento superficial

• Ano de defesa: 2023
• Autor(a) principal: Garcia Sobrinho, Renner de Assis
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Engenharia Mecânica
• Palavras-chave em Português: Inteligência artificial; Artificial intelligence; Redes profundas; Deep learning; Classificação de fissuras; Crack classification; Revestimento argamassado; Mortar coating; Rede neural convolucional; Convolutional neural network; Acabamento superficial; Surface finish; CNPQ::ENGENHARIAS::ENGENHARIA CIVIL::CONSTRUCAO CIVIL::MATERIAIS E COMPONENTES DE CONSTRUCAO; Engenharia mecânica; Redes neurais (Computação); Argamassa
• Link de acesso: https://repositorio.ufu.br/handle/123456789/38049; http://doi.org/10.14393/ufu.te.2023.260

The application of technologies such as artificial intelligence (AI) in production processes has been optimizing several industrial realities. In civil construction, AI can be used in different approaches, one of them is in building inspection. It is observed studies that applied AI for the classification of cracks in homogeneous materials such as concrete and asphalt paving, however, the literature lacks applications in materials such as mortar coating. The studies that have been carried out with this type of material are focused on the smooth and painted coating, not considering the different types of surface finish that exist. In view of this, this research aimed to evaluate the performance of the application of AI, through deep networks, for the classification of cracks in mortar coating with different types of surface finish. For this, a public database of images of cracks in mortar coating was created considering different types of surface finishes, namely: smooth type, scrapped type and rough type. A bank was created with 33088 images that went through a systematic labeling process based on classes defined in the study. Network training was carried out through learning transfer using the VGG16 in different groupings of finishes. It was found that the training accuracy varies according to surface finish and data balancing. From this, points were raised in relation to the characteristics of the database elaborated in this research and verified hypotheses. The bank will be available for researchers to implement studies in the area. It was found that the graffiti type finish was the one that presented the lowest performance in the assertiveness measures, while the smooth and chapiscado had similar values.