Machine learning-based modelling and analysis of carbonation depth of recycled aggregate concrete

Abstract

This paper used machine learning to model the prediction of carbonation depth and the analysis of feature parameters for recycled aggregate concrete (RAC). Specifically, a database containing 579 sets of RAC carbonation test data was developed. Twelve parameters representing material characteristics and environmental conditions were input, along with one output parameter. On this basis, six machine learning models were employed to predict RAC carbonation depth: Artificial Neural Network, Decision Tree, Support Vector Regression, Random Forest, Extreme Gradient Boosting, and Light Gradient Boosting. Different types of analysis, including statistical measures, shapley additive explanations (SHAP) sensitivity analysis, SHAP parametric study, and comparison study, were used to examine the performance of the developed models and the effects of the input parameters on predictions. The results show that the extreme gradient boosting (XGB) model exhibited the highest accuracy with an R² of 0.99 and a mean absolute percentage error (MAPE) of 6.632 on the training set, and an R² of 0.953 and a MAPE of 13.243 on the test set. The variable importance analysis shows that the carbonation depth for RAC was determined by both internal and external factors. The top five factors impacting RAC carbonation depth are exposure time, water-to-binder ratio, CO₂ concentration, coarse aggregate density, and cement content. RAC carbonation depth correlates positively with the former three factors. In contrast, it exhibits a negative correlation with the remaining two variables. In addition, a graphical user interface (GUI) for RAC carbonation depth prediction was designed.