Prédiction des propriétés à l’état durci du béton léger à l’aide de l’intelligence artificielle

Abstract

Lightweight concrete has increasingly gained importance in civil engineering due to its numerous advantages, particularly the reduction of structural self-weight and the improvement of thermal insulation. However, its complex composition limits the effectiveness of conventional experimental methods. These methods face several constraints, including high costs, long testing durations, and the inability to explore all possible concrete formulations. In this context, this study develops and compares four machine learning models : Artificial Neural Networks (ANN), Random Forest (RF), Support Vector Machines (SVM), and XGBoost. The main objective is to simultaneously predict four key properties of lightweight concrete : compressive strength (CS), tensile strength (TS), dry density (DD), and modulus of elasticity (MOE). The dataset was compiled from international scientific literature and includes twelve input variables. The results highlight the superior performance of the XGBoost model, which achieved a coefficient of determination of R2 = 0.97 for predicting both compressive strength and dry density. The Random Forest model demonstrated the best performance for tensile strength prediction, with an R2 = 0.97. In contrast, the SVM model exhibited the lowest performance among the studied models, with an average R2 value of 0.84. A comparison with findings reported in the literature confirms that the machine learning models investigated outperform traditional prediction approaches. Therefore, these models offer promising opportunities for the optimization and design of new lightweight concrete mixtures with greater efficiency and accuracy.