Data-Driven Optimization of Calcined Kaolin Waste for Sustainable Alkali-Activated Binders

Abstract Kaolin beneficiation generates large volumes of fine mineral residues with limited direct industrial application, despite their high aluminosilicate content and potential as reactive precursors. This study investigates the structural transformation of an industrial kaolin waste through thermal activation and its subsequent conversion into a hybrid alkali-activated binder. X-ray diffraction, FTIR, and SEM revealed the progressive collapse of the crystalline kaolinite lattice and the development of a predominantly amorphous aluminosilicate structure after calcination at 750 °C for 4 h. Thermogravimetric analysis of the optimized binder further revealed its thermal behavior up to 900 °C, supporting the formation of alkali-activated binding phases. To address the strong nonlinear coupling among precursor ratio, activator chemistry, and water content, a design of experiments and machine learning framework was applied to 41 metakaolin waste−slag formulations. Gaussian Process Regression outperformed ensemble tree-based models (R2 = 0.962 vs. ≤0.783) and provided predictive uncertainty maps that guided the search toward formulation regions combining high predicted strength with low predictive uncertainty. The identified optimum was experimentally validated, achieving 28.5 ± 0.9 MPa after 7 days of curing at 70 °C, with a prediction error of 1.35%. The optimized binder was further investigated by XRD, FTIR, SEM/EDS, and thermogravimetric analysis. Overall, the results demonstrate an uncertainty-aware, data-driven strategy for optimizing waste-derived alkali-activated binders and provide a transferable framework for materials design based on underutilized mineral residues.