Clusterização automática em seleção de materiais e processos de fabricação utilizando PSO - Particle Swarm Optimization

• Ano de defesa: 2020
• Autor(a) principal: Silva, Yuri Laio Teixeira Veras
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Engenharia Mecânica; Programa de Pós-Graduação em Engenharia Mecânica; UFPB
• Palavras-chave em Português: Otimização por enxame de partículas; Algoritmo evolucionário; Metaheurística; Clusterização; Concreto; Particle Swarm Optimization; Evolutionary algorithm; Metaheuristic; Clustering; Concrete; CNPQ::ENGENHARIAS::ENGENHARIA MECANICA
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/20949

The volume of information to be collected and processed today is greater than human capacity. The amount of data is a reality in the various areas of science, as engineering, industrial sectors, both in manufacturing and in the provision of services. The dynamics of exponential increase in the amount of information needing to be treated has motivated the implementation of computational approaches and tools over the last decades, aiming to develop the ability to handle data automatically and efficiently in organizations. In this sense, the field of data mining science addresses methods and algorithms from different areas of science, such as operational research, metaheuristics and artificial intelligence. The field of materials sciences and engineering, composite materials, such as concrete, have several characteristics that can be taken in a clustering method, in order to make it possible to obtain the best possible groupings per set of characteristics, as the performance metrics stipulated. Such results can bring countless benefits and applications in industrial decisions, mainly related to the production processes and problems of selection of raw materials, being able to significantly improve the operational performance, reduce production costs or even caused environmental impacts, according to the organization’s objectives. In this sense, this study aims to develop an automatic clustering optimization approach that allows the determination of optimal mixtures of components and their respective attributes, so that they meet multiple optimization criteria. In this work, the criteria studied are the chemical and mineralological compositions of rocks used as aggregates for concrete in the Northeast region of Brazil. For this, an evolutionary hybrid metaheuristic was developed in order to achieve an optimal clustering of data related to the chemical and mineralogical compositions of such materials, aiming to optimize the clustering according to the properties and characteristics chosen, considering also constraints and objectives involved in the problem. In view of the results achieved with the method, it was possible to evaluate the attribute configurations that enabled a more efficient clustering in its best solution, determining the most promising and impacting oxides and reasons. In addition, the work also aimed to carry out correlation analyzes between the clusters obtained with their mineralogical compositions and their densities of probabilities of expansion, enabling the establishment of significant relationships between the clusters formed and the analyzed criteria. The results achieved with the method and the studies carried out made it possible to obtain high quality and efficient clusters (as well as the identification of their correlations with multiple criteria), as well as enabling innovative analyzes and insights about the materials and clusters formed, thus being able to serve as a tool for industrial decision-making.