Predição da resistência mecânica de concretos com materiais alternativos através de técnicas de inteligência artificial

Abstract

The scarcity of natural resources and the problems caused by the accumulation of mining residues are mainly responsible for the development of studies on the reuse of these materials in the concrete composition. Mechanical resistance is one of the most important and crucial key performance parameters in concrete structures design. However, the experimental study of these concretes has high cost and is time consuming in the process of making and characterizing specimens, in addition to the intrinsic complexity and heterogeneity of the material components used in the concrete formulation. In this sense, this work proposes the study of the prediction of compressive and tensile strength of concretes produced with the waste from enclosing rocks (also known as sterile) from a niobium mining, as a substitute for coarse aggregate, through different machine learning techniques and other statistical methods (Artificial Neural Networks, Support Vector Machine, Decision Tree, Random Forest, Gradient Boosting, XGBoosting, Ada Boost, K-Near Neighbor, Multiple Linear Regression, Polynomial Regression, LASSO, Ridge, Bayesian, and Ordinary Least Squares). This approach could minimize production costs and material waste, caused by a large number of mixing dosing trials. The methodology consisted of obtaining experimental data of concrete specimens and mechanically characterized with 0, 25, 50, 75 and 100% replacement of coarse aggregate for sterile, water/cement ratio (w/c) of 0.5 and 0.7, in curing periods of 7, 14, 28 and 56 days. The mechanical properties of the concrete specimens with sterile were also compared with other concretes data from the literature. The found results indicate that the machine learning techniques RNA, Multiple Linear Regression, Bayesian, Ordinary Least Squares, Random Forest, Gradient Boosting and XGBoosting were efficient for predicting the mechanical strength of the concrete produced with waste. The models also present the materials parameters used in the concrete formulation that have the greatest influence on the mechanical behavior of these specimens.