Computational modeling study demonstrates accurate prediction of concrete sulfate resistance, highlighting optimal trade-offs among durability, financial cost, and carbon emissions.
Key Points
To establish an interpretable machine-learning and multi-objective optimization framework for designing durable concrete resistant to sulfate attack while minimizing cost and carbon emissions.
Compiled a literature dataset of 744 records across 21 publications with 18 input features, retaining 549 records following outlier screening.
Trained and evaluated support vector regression, random forest, gradient boosting, and XGBoost using Bayesian hyperparameter optimization and 5-fold cross-validation, applying SHAP values for model interpretability.
Coupled the best-performing predictive model with the NSGA-II algorithm to conduct multi-objective optimization balancing corrosion durability, financial cost, and carbon footprints.
XGBoost achieved the highest predictive accuracy with an R² of 0.96, RMSE of 0.0429, and MAE of 0.0280.
SHAP interpretability analysis identified that magnesium ions erode concrete more aggressively than sodium ions, with durability correlating positively with water-reducing agents and sand, but negatively with coarse aggregate and water-to-binder ratio.
Multi-objective Pareto optimization generated a balanced mixture design yielding a corrosion-resistance coefficient of 1.24, an estimated cost of 312.24 CNY/m³, and carbon emissions of 226.00 kgCO₂/m³.