Otimização e previsão de parâmetros para definição do Comportamento cisalhante de descontinuidades rochosas por meio De métodos meta-heurísticos e redes neurais artificiais

• Ano de defesa: 2021
• Autor(a) principal: Mendes, Guilherme Anderson Rodrigues
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Descontinuidades rochosas; Algoritmo genético; Resistência ao cisalhamento; Perceptrons; PSO; Funções de Base Radial
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/61233

The use of test methods that more realistically consider boundary conditions acting allowed the development of analytical formulations that take into account the main factors that govern the shear behavior of the present discontinuities in rock masses, as an example the analytical model of Oliveira and Indraratna (2010). In this particular case, the application of the analytical formulation becomes difficult due to the need to determine constants necessary to adjust the results of the model to the experimental data, since there is no clear and established for such constants. Therefore, this work aims to present a optimization methodology and prediction of the adjustment parameters of the proposed analytical model by Oliveira and Indraratna (2010) using meta-heuristic methods (genetic algorithms and PSO) associated with radial basis functions (RBF) to determine the adjustment parameters of the analytical model from large-scale direct shear test data, and subsequently applying perceptron-type neural networks to establish relationships between the adjustment parameters and the characteristics of the rock discontinuities. From the definition the idealized shear mechanism and the variables that characterize the discontinuities rocks, an experimental database was established based on the results of 110 large-scale direct shear tests performed on discontinuities with and without filled and tested under CNL and CNS conditions. Due to the particularities of the model analysis of Oliveira and Indraratna (2010), which presupposes the determination of the dilation angle, Radial basis functions were applied in the representation of shear displacement curves. versus dilation. Through the computational implementation of an optimization algorithm in open source software, the adjustment constants of the analytical model to from an appropriate combination of variables that guaranteed the minimum error of to the experimental data. Subsequently, prediction models based on the optimal values ​​found using artificial neural networks of the perceptron type, which were defined as a function of normal boundary stiffness, initial normal stress, coefficient of roughness, simple compressive strength of intact rock, basic friction angle, relationship between fill thickness and height of roughness and internal friction angle of the filling material. The results obtained showed that the use of the methods meta-heuristics and artificial neural networks allowed an interpolation in the tests performed satisfactory results of experimental data. In addition, the application of the methodology developed in this work made it possible to obtain geotechnical solutions - from limit equilibrium analysis - compatible with other available mathematical models. This shows that the establishment of a prediction model developed based on input data obtained from a robust optimization process allows obtaining the constants of the analytical model of simple form, contributing to its implementation and use in satisfactory representation of the shear behavior of rock discontinuities.