Elemental Design of Alkali-Activated Materials with Solid Wastes Using Machine Learning.

Understanding the strength development of alkali-activated materials (AAMs) with fly ash (FA) and granulated blast furnace slag (GBFS) is crucial for designing high-performance AAMs. This study investigates the strength development mechanism of AAMs using machine learning. A total of 616 uniaxial compressive strength (UCS) data points from FA-GBFS-based AAM mixtures were collected from published literature to train four tree-based machine learning models. Among these models, Gradient Boosting Regression (GBR) demonstrated the highest prediction accuracy, with a correlation coefficient (R-value) of 0.970 and a root mean square error (RMSE) of 4.110 MPa on the test dataset. The SHapley Additive exPlanations (SHAP) analysis revealed that water content is the most influential variable in strength development, followed by curing periods. The study recommends a calcium-to-silicon ratio of around 1.3, a sodium-to-aluminum ratio slightly below 1, and a silicon-to-aluminum ratio slightly above 3 for optimal AAM performance. The proposed design model was validated through laboratory experiments with FA-GBFS-based AAM mixtures, confirming the model's reliability. This research provides novel insights into the strength development mechanism of AAMs and offers a practical guide for elemental design, potentially leading to more sustainable construction materials.