Document Type : Original Research Paper
Authors
Isfahan University of Technology. Department of Mining Engineering. Isfahan 8415683111, Iran
Abstract
This study investigates the damage of surface aesthetic properties of polished building stones under the coupling action of freeze-thaw cycles and acid rain. Laboratory experiments were conducted on seven types of carbonate building stones to evaluate changes in surface gloss and roughness. Two damage indices, namely gloss damage (GD) and roughness damage (RD), were introduced for quantifying the degradation of surface aesthetic properties. A database containing 168 datasets was obtained based on coupling action experiments. Using this database, several boosting-based machine learning models, including GradientBoost, AdaBoost, XGBoost, CatBoost, LightGBM, and NGBoost, were developed to predict GD and RD. Four input parameters including acid type, pH, cycle number, and Leeb hardness were used for constructing the models. The models were optimized through 5-fold cross-validation and the Optuna framework to avoid overfitting of models and to enhance their performances. The results indicated that CatBoost achieved the best performance for GD prediction (with R² = 0.945), while GradientBoost showed superior accuracy for RD prediction (with R² = 0.975). Shapley additive explanation (SHAP) analysis revealed that acid type and cycle number are the most influential parameters on surface degradation. The findings demonstrated that the proposed approach provides an effective and reliable framework for predicting the surface aesthetic properties of polished building stones under complex environmental conditions, offering practical insights for stone selection in cold and polluted regions.
Keywords
- Carbonate building stones
- Surface damage
- Coupling action
- Aesthetic properties
- Boosting-based algorithms
Main Subjects
)