Machine learning prediction of carbonation depth in recycled aggregate concrete incorporating SCMs

Abstract

Abstract While recycled aggregates and supplementary cementitious materials have often been hailed for enhancing concrete sustainability, their effects on the resistance of concrete to carbonation has been controversial. Thus, deploying robust machine learning tools to overcome the lack of understanding of the implications of incorporating such sustainable materials is of paramount importance. Accordingly, this study proposes a gradient boosting regression tree (GBRT) model to determine the carbonation depth of recycled aggregate concrete incorporating different mineral additions, including metakaolin, blast furnace slag, silica fume, and fly ash. For this purpose, a database comprising 713 pertinent experimental data records was retrieved from peer-reviewed publications and used for model development and testing. Furthermore, predictions of the GBRT model were compared with calculations of available mathematical formulations to determine the carbonation depth in concrete. The results demonstrate that the machine learning methodology outperformed all the mathematical models considered in this study. The GBRT proved to be a robust tool that could be used to provide an insight into the resistance of concrete to carbonation and could be extended to predicting other features of concrete incorporating diverse recycled materials.