A reactive-phase-driven predictive model for compressive strength of solid waste-based alkali-activated materials

The inherent variability in reactivity among industrial solid wastes hinders the standardized mix design of alkali-activated materials (AAMs). To address this, a novel reactive-phase-driven predictive model based on optimized clustering of amorphous aluminosilicate phases is proposed. Using backscattered electron and energy dispersive spectroscopy analysis, amorphous aluminosilicates in solid wastes, e.g., fly ash, ground granulated blast-furnace slag, and circulating fluidized bed ash, were classified into 42 groups via atomic ratios (Si + Al)/Ca (1, 1.5, 2.3, 4, 9, ∞) and Si/Al (0.3, 0.7, 1, 1.5, 2.3, 4, ∞). Furthermore, these groups were grounded into nine grades (H20, H30, H40, H50, H60, H70, H80, H90, H130) based on microhardness of alkali-activated synthesized phases. Results revealed that medium-hardness grade phases (H40–H70) exhibited the highest reaction degrees (70.1–95 %) and strong positive correlation with compressive strength (σc) (R = 0.62–0.81), while low/high-hardness phases negatively impacted performance. A grey wolf optimizer-enhanced backpropagation neural network model, utilizing phase content as input, achieved exceptional prediction accuracy, significantly outperforming conventional solid-waste-dosage-driven models in terms of scientific rigor and cross-waste applicability. This study establishes a more universal amorphous aluminosilicates classification framework and ultimately enabling intelligent design of multi-source solid waste-based AAMs.