Desenvolvimento de diagramas de vida constante probabilísticos de compósitos utilizando RNA modular

• Ano de defesa: 2017
• Autor(a) principal: Diniz, Bruno da Cunha
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Brasil; UFRN; PROGRAMA DE PÓS-GRADUAÇÃO EM ENGENHARIA MECÂNICA
• Palavras-chave em Português: Compósitos; Fadiga; Diagrama de Goodman; Distribuição de Weibull; Redes neurais artificiais; CNPQ::ENGENHARIAS::ENGENHARIA MECANICA
• Link de acesso: https://repositorio.ufrn.br/jspui/handle/123456789/23313

In mechanical designs, in addition to static loads, another type of loading very frequent is the cyclic loading in which the load varies over time. Structures and equipment, when subjected to such loads, must undergo a large number of mechanical tests for their characterization. However, due to the time and cost of the tests to characterize the behavior due to cyclic loading, the ideal situation would be to obtain it with a minimum number of samples and, because of that, it developed in previous works mathematical models that represent the fatigue behavior of composite materials with a minimum quantity of experimental data. Although the results are satisfactory in the vast majority of cases, these models always consider a deterministic behavior of the material, disregarding a factor of great importance in the fatigue, the dispersion of the results. As is known in the literature, the dispersion of results always requires a probabilistic analysis and, in most cases, a Weibull probability distribution is used, obtaining an S-N probability curve. The aim of this work is to develop an Artificial Neural Network (ANN) with modular architecture and verify if it is able to model the fatigue probabilistic behavior of the laminated composites with only three S-N curves as input data, developing an algorithm capable of analyzing any value of Probability of failure, using the Weibull distribution equation, using two methodologies to obtain its parameters (methods 1 and 2), which are considered constant for all material, only after the network training, which was performed with deterministic data. From the obtained results, it can be concluded that the robustness of the algorithm was perceived for the deterministic data and a good repeatability occurred in the obtained answers. In order to evaluate the generalization capacity of probabilistic ANN, the constant life diagrams (Goodman Diagrams) were created for the analyzed materials and they were compared with the values obtained by the S-N probability curves, where satisfactory results were obtained.