Data-driven inverse mix design for sustainable alkali-activated materials

Abstract

Alkali-activated materials (AAMs) are promising alternatives to ordinary Portland cement (OPC), but standardized mix design approaches are limited. This study introduces a machine learning-based framework for inverse mix design of AAMs, predicting optimal mixes based on target properties and sustainability. The model considers eight key factors, including precursor reactivity, activator properties, and liquid-to-binder ratio. It operates in four steps: (i) constructing forward prediction models, (ii) generating new mix designs, (iii) evaluating workability and strength, and (iv) filtering for performance and sustainability. Experiments validated the model, targeting fluidity >100% and compressive strengths of 40 MPa (A40) and 50 MPa (A50). The A40 and A50 mixes achieved fluidities of 147.7% and 170.0%, and 28-day compressive strengths of 41.5 and 57.5 MPa. Both had lower environmental impacts (GWP, AP, and EP) compared to OPC-based mortars. This framework enhances AAM design efficiency and sustainability for broader applications.