Abstract
Limestone calcined clay cement (LC(3)) is a promising low-carbon construction material in terms of its comparable mechanical performance to ordinary Portland cement (OPC) but a much less embodied carbon footprint. Previous literature have demonstrated that the large-scale implementation of LC(3) can reduce embodied CO(2) emissions associated with OPC production by at least 30%. This study proposes a hybrid framework combining machine learning (ML) and multi-objective optimization (MOO) to design cost-effective and eco-friendly LC(3) mixtures. A dataset of 387 LC(3) specimens was constructed to develop ML models for predicting compressive strength. Multivariate Imputation by Chained Equations-Extreme Gradient Boosting (MICE-XGBoost) model achieved the highest accuracy of R(2) = 0.928 (± 0.009). SHAP analysis identified key factors influencing strength, including water-to-cement/binder ratio, and kaolinite content. The local range of each feature showing more significant contributions was also identified. Non-dominated Sorting Genetic Algorithm-II was employed for MOO, generating Pareto fronts to minimize cost and embodied carbon while meeting strength requirements. A minimum balanced reduction in cost by 13.06% and embodied carbon by 14.83% was obtained. Inflection points on Pareto fronts were identified to guide decision-making for low-medium grade mixtures. A table of optimal mix designs is provided, offering practical solutions for selecting sustainable LC(3) formulations.