Predicting Geopolymer Ultra-High-Performance Concrete Strength Using Machine Learning

Abstract

Ultra-high-performance geopolymer concrete (UHP-GPC) can exhibit high to exceptional strength. Given the importance of UHP-GPC’s mechanical properties, prediction of its 28-day compressive strength (fc′) remains insufficiently explored. This study predicts UHP-GPC’s fc′ based on alkali-activated materials, sand, fiber volume, and water-geopolymer binder and alkali activator ratios. Advanced statistical modeling and a spectrum of ensemble machine learning (ML) algorithms including random forest (RF), gradient boosting (GB), extreme gradient boosting (XGB), and stacking are used to predict UHP-GPC’s strength. The derived models reveal the significance of fiber, slag, and sand as the most significant factors influencing the 28-day fc′ of UHP-GPC. All the ML models demonstrate higher precision in forecasting fc′ of UHP-GPC compared to statistical modeling, with R2 peaking at 0.85. Equations are derived to predict the strength of UHP-GPC. This paper reveals that UHP-GPC with superior mechanical properties can be designed for further sustainability.