Explainable Artificial Intelligence-Based Prediction and Optimization of Self-Compacting Concrete Incorporating Manufactured Sand and Silica Fume

Abstract

High depletion of natural river sand along with increasing interest towards clean construction material have led to the utilization of manufactured sand (M-sand) and silica fume (SF) in self-compacting concrete (SCC). This paper analyses the mechanical properties of SCC with different contents of M-sand and silica fume proportioned to 10 percent, and explores the possibilities where machine learning (ML) models applied in predicting concrete properties. Fresh concrete properties were assessed using slump flow, V-funnel, and L-box tests, while hardened concrete was evaluated through compressive strength, split tensile strength, flexural strength, and reinforced concrete beam load–deflection behaviour. The experimental results showed that the SCC mixture containing 60% M-sand and 10% silica fume exhibited the best overall performance, achieving a 28-day compressive strength of 33.63 MPa, split tensile strength of 1.62 MPa and flexural strength of 9.50 MPa, were found remarkably higher than the conventional mixes. Other machine learning models such as Linear Regression, Decision Tree, Random Forest, and XGBoost are proposed for its prediction of mechanical property of SCC. The best prediction accuracy of XGBoost (99.0%) was attained with a R² value of 0.99, MAE of 0.04, and RMSE of 0.07, indicating excellent agreement between experimental and predicted results. The findings demonstrate that the combined use of M-sand, silica fume, and machine learning provides an effective approach for developing sustainable, high-performance SCC while reducing the dependence on natural sand and minimizing experimental effort through accurate AI-based prediction.