Prediction of Carbonation Depth for Concrete Containing Mineral Admixtures Based on Machine Learning

This study developed a prediction model of the carbonation depth of concrete containing mineral admixtures based on an intelligent algorithm. A carbonation test database of mineral admixture concrete was established considering the influence of 17 parameters. The intelligent algorithm and three existing carbonation depth prediction models were analysed based on the database. The evaluation results indicated that the prediction accuracy of the back-propagation neural network is higher than that of the support vector machine, and the prediction accuracies of the two intelligent algorithms are higher than those of the existing numerical prediction models for carbonation depth. A variable importance analysis indicated that the content of fly ash in mineral admixture has a relatively large influence on the carbonation depth, and the carbonation time is the most critical factor affecting the carbonation depth of concrete containing mineral admixture.