Machine learning-driven optimization of compressive and tensile strength in concrete with GGBS, eggshell powder, and waste glass powder.

Abstract

This study investigates the combined use of Ground Granulated Blast Furnace Slag (GGBS), Waste Glass Powder (WGP), and Eggshell Powder (ESP) as sustainable partial replacements in concrete, with optimization guided by machine learning (ML) models. The novelty lies in integrating non-traditional waste materials with ML-driven mix design to simultaneously improve compressive and tensile strength while promoting sustainability. Experimental results showed that the optimized concrete mix, containing 20% GGBS, 4% ESP, and 10% WGP, attained a compressive strength of 24.7 MPa and a split tensile strength of 2.77 MPa, surpassing the control mix (20.4 MPa and 2.33 MPa, respectively). The compressive strength exceeds the minimum requirement of 20.7 MPa (3,000 psi) specified by the International Building Code (IBC), making it suitable for general structural applications, while acknowledging that higher-strength or severe exposure conditions may require adjustments. Among the evaluated ML models, Gradient Boosting provided the most accurate predictions, achieving test R² values of 0.937 and 0.906 for compressive and tensile strength, respectively, with low RMSE values (0.569 MPa and 0.073 MPa). This validates the model's predictive reliability for sustainable mix optimization. Limitations of the study include the absence of microstructural characterization and durability assessments, which are recommended for future research. Overall, the findings establish a framework for applying ML in the design of sustainable concrete mixtures.