Predictive modeling of abrasion resistance of hydraulic concrete using supplementary cementitious materials

Many hydraulic concrete structures are subjected to severe abrasion from sediment-laden flows, necessitating durable and sustainable material solutions. Conventional concrete, with its high cement content, contributes significantly to CO₂ emissions; thus, there is a growing trend of using fly ash and silica fume as eco-friendly Supplementary Cementitious Materials (SCMs) in concrete mixtures to partially replace cement, which not only reduces environmental impact but also enhances durability and strength. Despite this, there remains a lack of research on prediction of abrasion resistance of such concretes in hydraulic structures. To assess their durability, abrasion resistance was predicted through abrasion depth modeling, based solely on experimental abrasion test data obtained using standardized Underwater Method (ASTM C1138), ensuring relevance to field hydraulic conditions. A dataset of 241 test results reported in previous studies was collected and statistically analyzed. Various models were applied and compared to predict the abrasion depth of hydraulic concrete, with Extreme Gradient Boosting (XGB) achieving the highest accuracy (R 2 = 0.93, RMSE = 0.415, MAE = 0.230, SI = 0.202). SCMs showed a weak correlation with abrasion depth (Pearson r < 0.4) and had limited influence on model performance, as confirmed by sensitivity analysis. This suggests that SCMs had only a slight positive impact on abrasion resistance, especially when compared to more influential variables such as testing time and concrete age. However, their inclusion contributes to CO 2 reduction while having an almost neutral effect on abrasion resistance in hydraulic concrete. The developed predictive models provide: i) a cost-effective, data-driven approach to evaluating abrasion resistance in hydraulic concrete, ii) minimizing a need for extensive laboratory testing, and iii) supporting resilient, low-carbon hydraulic infrastructure and sustainable construction. • Predictive modeling of abrasion depth as key parameter for durability assessment. • Different prediction modeling techniques were comparatively reviewed. • The Extreme Gradient Boosting model exhibited highest accuracy (R 2 = 0.93). • Developed predictive models serve as a cost-effective alternative to experiments. • Supplementary cementitious materials reduce CO₂ without affecting concrete durability.