Aplicação de modelos matemáticos de inteligência computacional na predição da resistência à compressão axial de concreto de cimento Portland

• Autor(a) principal: Tavares, Dennis Santos
• Data de Publicação: 2021
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFLA
• Texto Completo: http://repositorio.ufla.br/jspui/handle/1/48344
• topic: Concreto - Propriedades físicas e mecânicas; Concreto - Resistência à compressão axial; Redes neurais artificiais; Lógica Fuzzy; Concrete - Physical and mechanical properties; Concrete - Resistance to axial compression; Artificial neural networks; Fuzzy logic; Sistemas de Computação

The axial compressive strength is the main property of concrete, the structural material most used worldwide, but there are no empirical equations that provide, easily and quickly, reliable and accurate results for prediction of this important property that is directly related to structural performance and safety of civil construction works. Concrete dosage and compressive strength prediction are obtained through laboratory tests conducted from successive adjustments in pilot batches, which requires time and consumption of materials. The objective of this work is to apply the technologies of computational intelligence, Artificial Neural Networks and Fuzzy Logic for predicting the axial compressive strength of concrete, from a database consisting of 1030 samples with different proportions of constituent materials and age of curing. Several configurations were tested until the choice of an Artificial Neural Network of feedforward architecture of the multilayer-perceptron (MLP) model with one input layer, two hidden layers and one output layer. It was also developed several fuzzy systems with different methods of inference and defuzzification that were statistically evaluated, being possible to verify that the methods of inference and defuzzification adopted influence the final result and the best system was with Mamdani inference and defuzzification center of area (centroid). The models developed with Mamdani inference and centroid, bisector and mom defuzzification, besides Sugeno inference with wtaver and wtsum defuzzification proved to be reliable and capable of providing high precision results, which shows the promise of applying computational intelligence models to concrete technology, contributing to the advancement of the industrialization and automation of civil construction.