Prediction model for corrosion resistance of carbonated recycled aggregate concrete based on machine learning

Jian Wang, Zhiyong Yang, Yuanyuan Song,
Prediction model for corrosion resistance of carbonated recycled aggregate concrete based on machine learning,
Structures,
Volume 82,
2025,
110521,
ISSN 2352-0124,
https://doi.org/10.1016/j.istruc.2025.110521.
(https://www.sciencedirect.com/science/article/pii/S2352012425023367)
Abstract: Carbonated recycled aggregate concrete (CRAC) exhibits broad application prospects, but its durability has not been thoroughly investigated. In this paper, an accelerated corrosion test was conducted on CRAC specimens, the effects of the carbonation of recycled coarse aggregates (CR), the replacement ratio of recycled coarse aggregates (r), the carbonation of recycled fine aggregates (CF), the replacement ratio of recycled fine aggregates (f), short-term load level (l), and corrosion duration (t) were considered, and prediction models for corrosion depth of steel bars and width of corrosion-induced cracks based on machine learning were established. Six machine learning models, including ANN, GBDT, SVR, RF, XGBoost, and CatBoost, were developed based on experimental data, and on the basis of the optimal CatBoost model, the relationships between the corrosion depth of steel bars, the width of corrosion-induced cracks, and their influencing factors were investigated using SHAP method, respectively. The results showed that the CatBoost model was determined to be optimal for predicting both the corrosion depth of steel bars and the width of corrosion-induced cracks of CRAC, exhibiting strong generalization capability with reduced outliers, the distribution of sample points was observed to closely align with the line y = x, while exceptional predictive performance was demonstrated at local extreme points. The relative importance of influencing factors on the corrosion depth was determined to follow the order: l > CF > r > f > CR, and the positive correlations were observed between the corrosion depth and the factors r, l, and f, whereas the factors CF and CR showed negative correlations with the corrosion depth. Similarly, for the width of corrosion-induced cracks, the importance of influencing factors was ranked as: r > t > l > CF > f > CR, and the positive correlations were identified between the width of corrosion-induced cracks and the factors r, t, l, and f, while the factors CF and CR maintained negative correlations with the width of corrosion-induced cracks. A new approach for accurately assessing the corrosion resistance of CRAC is offered by the findings of this paper.
Keywords: Recycled aggregate concrete; Carbonated recycled aggregates; Short-term load; Corrosion depth; Corrosion-induced cracks; Machine learning