Integrated machine learning and response surface methodology for comprehensive rheological characterization of low-carbon binders

The development of alkali-activated materials (AAMs) has garnered substantial attention in the construction industry as a low-carbon binder alternative to traditional Portland cement-based concrete. A key factor in the performance of AAMs is their rheological behavior, which determines workability, pumpability, and overall construction feasibility. Accurate prediction of rheological properties is therefore crucial. This study explores the use of machine learning models to predict the rheological properties—yield stress (YS) and plastic viscosity (PV)—of AAM incorporating hydrated lime. Three models—stochastic gradient descent, K-nearest neighbors (KNN), and support vector machine—were developed and compared, with each model exceeding R² values of 0.85. The KNN model exhibited superior accuracy, achieving the highest R² values of 0.98 and 0.96 and the lowest mean squared errors of 1.9 and 958.5 for PV and YS, respectively. The influence of mix constituents on rheology was assessed using response surface methodology, revealing that hydrated lime increased PV while fly ash reduced YS. Optimal mix designs for AAM were identified through optimization, yielding a mix with high YS (700–978 Pa) and low PV (17.5–33 Pa. s), ideal for 3D printing applications. Overall, this research provides a robust framework for predicting and optimizing the rheological behavior of AAM, contributing to the development of sustainable and efficient construction materials.