Estimativa da profundidade de carbonatação do concreto com o uso de redes neurais

Abstract

The phenomenon of carbonation as the triggering agent in the process of reinforcement corrosion is particularly important when concrete structures are exposed to urban environments and the atmosphere contamination with gases such as CO2. The concrete carbonation depth control requires the use of tools (mathematical models) that represent the behavior of the variables that interact in the process of concrete carbonation in a clear and objective way to help understand the phenomenon. In this perspective, computer models have been developed to combine complex problems in a simple way. Among those models are the artificial neural networks, which have been inspired on human nervous system and have the ability to learn and generalize, making it possible to solve complex problems. This work studied the application of artificial neural networks like multilayer perceptron, based on backpropagation, supervised learning algorithm in order to obtain a mapping between the input variables of the problem ─ the water/cement (w/c) ratio, body of proof distance from the sea and age of the body of proof ─ and the output variable of interest ─ the depth of concrete carbonation. The results validate that the use of artificial neural networks is an important tool to evaluate concrete carbonation