Machine learning-based reactivity evaluation of solid wastes and development of a multi-component all-solid waste cementitious material.

Alkali-activated cementitious materials are recognized as promising low-carbon materials. Preparing all-solid-waste cementitious materials circumvents energy-intensive commercial alkali activators, thereby reducing carbon footprint and energy consumption. The property of all-solid-waste materials is significantly influenced by reactivity of components. Current reactivity evaluation for solid waste relies on empirical mechanical and chemical tests, lacking quantitative activity models based on microscopic features. This study aims to establish quantitative mapping between reactivity micro-characteristics and performance for solid waste. Reactivity parameters of 15 solid wastes, classified by silicate tetrahedron polymerization degree, ionic bond content, and oxygen valence, were acquired via XRF, FTIR, XPS, and TG characterization, then dimensionally reduced into four principal factors. High-precision fitting of compressive strength and principal factors was achieved through support vector regression (SVR) modeling. Subsequently, a multi-solid-waste cementitious material was developed. Results indicates that reactivity parameters of different solid waste types intuitively reflect their roles in all-solid-waste system. Discrepancy in the strength formation mechanism across curing periods stem from phase-dependent hydration kinetics and time-sensitive contributions of constituents. The proposed reactivity evaluation method is applicable to small-sample, nonlinear material property prediction, providing quantitative support for performance regulation.