From Mix Design to Strength Prediction: Ensemble Learning Application on the Performance of High-Performance Concrete

Abstract

In the realm of construction, achieving high-performance concrete (HPC) involves incorporating supplementary materials like fly ash and blast furnace slag, along with superplasticizer. The conventional water-to-cement ratio (w/c) concept, established by Abrams in 1918, asserts an inverse relationship between w/c ratio and concrete strength in HPC. However, a critical analysis of experimental data challenges this perspective, revealing that the paste quantity also significantly influences comparable cement strength, introducing complexity to our understanding of HPC and concrete strength dynamics. Furthermore, an exploration of concrete mix models and machine learning algorithms sheds light on variables impacting compressive strength. Surprisingly, blast furnace slag emerges as a predominant contributor, highlighting the significance of water management. Key factors like cement and aggregates play pivotal roles in shaping compressive strength. Notably, the Vote algorithm demonstrates exceptional predictive accuracy with a high correlation coefficient (0.919) and low mean absolute error (4.9166), while RandomForest and AdditiveRegression also exhibit commendable performance, striking a balance between accuracy and efficiency. These insights guide decisions in concrete mix design and machine learning model selection, offering valuable guidance for optimal outcomes across diverse applications in construction.