Investigation of the hardening kinetics of alkali-activated binders using machine learning, mechanical waves and rheology

Abstract

The urgent need to reduce CO₂ emissions in the construction industry has driven the search for sustainable alternatives to conventional binders. Alkali-activated binders (AABs) have emerged as a promising solution due to their lower environmental impact and capacity to incorporate industrial wastes. Although extensive research has examined the mechanical performance of AABs, their fresh-state behavior and hardening mechanisms remain insufficiently understood. This gap hinders the optimization of their rheological and microstructural development, which are essential for reliable performance. In parallel, the growing use of machine learning in materials science has opened new perspectives for predicting binder properties; however, its application to AABs remains in its early stages. Existing studies employing non-destructive test (NDT) techniques, particularly ultrasonic pulse velocity, have provided valuable insights into early-age transitions but are limited in capturing the complex viscoelastic behavior of these materials. Impact resonance, in contrast, has demonstrated strong potential for evaluating dynamic properties in hardened cementitious and asphalt materials, yet it has not been adapted for fresh-state analysis. In this context, the main objective of this research is to investigate the hardening mechanisms of AABs through the combined use of machine learning, impact resonance, and microstructural characterization. The methodology is organized into three stages. First, machine learning algorithms will be developed to estimate fresh and hardened properties based on chemical composition. Second, a novel impact resonance test apparatus will be designed and optimized through FEM (Finite Element Method) simulations to enable accurate monitoring of AABs in the fresh state. Finally, the hardening kinetics of AABs will be assessed through impact resonance and rheological time-sweep tests, coupled with thermal and chemical analyses such as FTIR, TGA, and isothermal calorimetry. The integration of these experimental and computational approaches is expected to advance the understanding of hardening mechanisms in AABs and provide a foundation for improved predictive and monitoring tools. The main findings of this research reveal that artificial neural networks (ANN), random forest (RF), and bagging regression (BR) demonstrated the highest predictive performance for estimating setting time and compressive strength of AABs from chemical composition. In parallel, the impact resonance test proved to be a highly effective nondestructive tool for monitoring early-age hardening, showing strong correlations with rheological parameters and calorimetric evolution over time. Integrating impact resonance test with rheology, calorimetry, FTIR, and TGA provides a coherent, multi-scale understanding of early-age hardening in AABs and reveals calcium availability as the dominant factor governing their mechanical and microstructural evolution. This work is carried out in collaboration between the Federal University of Ceará (UFC), Brazil, and ENTPE, France.