Speaker
Description
Sustainable ocean structures require binders that reduce clinker content without relying on elevated-temperature curing. This study couples machine-learning-guided mix design with experimental verification for ambient-cured fly ash alkali-activated mortars containing ordinary Portland cement (OPC) and silica fume (SF). A database of 331 compressive-strength records compiled from 15 studies and described by 11 mixture and curing variables was used to compare five learning algorithms; XGBoost gave the best test performance (R² = 0.951, RMSE = 3.07 MPa), and SHAP analysis identified calcium- and silica-bearing constituents as the dominant predictors. Coupling the model with a hybrid genetic algorithm–particle swarm optimization procedure, with OPC and SF constrained to 15 wt% of the binder in total, returned an OPC12.5–SF2.5 composition with a predicted 28-day strength of 50.5 MPa. Six mortars spanning the same OPC/SF range were cast and cured under identical ambient conditions; the adjacent composition OPC12–SF3 reached the highest measured mean strength of 49.8 MPa, consistent with the predicted trend. Companion paste analyses showed that chemically bound water and the fine-pore fraction peaked at a higher SF content than strength did, indicating that the calcium–silica balance, rather than pore refinement alone, governs strength development. The approach supports target-specific screening of low-clinker binders for sustainable ocean structures, for which exposure-specific durability verification remains outstanding.
| Topic Area | Sustainable Ocean Structures |
|---|