Estimation of concrete compressive strength using machine learning

Resumo

Due to its strength and durability, concrete is one of the primary building materials. These properties are crucial for the safety of concrete-based constructions. Although the final concrete strength can only be known after 28 days of production, building companies often employ concrete in the construction process before this period. Consequently, they use concrete strength that is much larger or smaller than safe and necessary. If the companies can predict the strength of the concrete before 28 days, the construction process can be faster, safer, and cost-effective. Thus, one of the main challenges of the construction industry is making the construction process faster without jeopardizing structural safety. A promising approach to this challenge is to predict concrete compressive strength using the concrete industry data. The main goal of this study is to support construction companies in defining the construction process schedule. For this, it investigates using Machine Learning to predict concrete strength before 28 days. For the Machine Learning experiments, the authors collected a dataset with 236,407 examples and 69 attributes, obtained from four Brazilian concrete industries. These industries do not share standards for data collection. Data curation reduced the dataset to 23 attributes and 56,508 examples. The predictive features were collected from many diverse sources. More than just giving a number for the 28-day-old concrete compressive strength, this study uses random forest quantile regression (RFQR) to produce 95 % prediction intervals. By using Random Forest, it is possible to explain how an induced model makes inferences. According to the experimental results, the predictions for the central value (quantile = 0.500) presented a high predictive performance, even when using a small number of the collected features.