Influence of Silica Fume on Performance of Alkali-Activated Cementitious Materials and Prediction via BP Neural Network

Alkali-activated slag–fly ash cementitious materials (AAMs) are low-carbon and mechanically favorable, but suffer from pronounced drying shrinkage and high brittleness. Silica fume (SF), with its large specific surface area and strong pozzolanic reactivity, offers a promising route to mitigate these drawbacks. This study systematically investigates how SF dosage affects workability, mechanical properties, drying shrinkage behavior, chloride resistance, and microstructural evolution in alkali-activated slag–fly ash mortars. The optimal baseline mixture is determined as slag/fly ash = 3:1, activator content 15%, and water/binder ratio 0.26. At an SF replacement level of 10%, compressive and flexural strengths peak at 3, 28, and 90 days, with 90-day compressive strength reaching 83 MPa, about 24% higher than the SF-free control. Concurrently, SF exerts a marked retarding effect on drying shrinkage; at 15% replacement, 90-day shrinkage is reduced by more than 43%. Microstructural analyses reveal that SF’s micro-filler and pozzolanic reactions promote additional C-S-H/C-A-S-H gel formation, substantially densifying the matrix. However, when SF exceeds 10%, workability deteriorates, and strength slightly declines; moreover, the chloride migration coefficient increases with SF content, implying excessive SF may adversely affect chloride durability. We developed a BP neural network model to quantify nonlinear relationships between SF content and compressive/flexural strengths, shrinkage, and the chloride migration coefficient, achieving high predictive accuracy (training R2 = 0.90763, testing R2 = 0.81832) and providing a reliable data-driven tool for mix proportion optimization and performance prediction. In summary, proper SF dosage markedly enhances mechanical and shrinkage properties, yet careful consideration of trade-offs with workability and durability is required.