ARTIFICIAL INTELLIGENCE TECHNIQUES FOR PREDICTING CONCRETE STRENGTH AND DURABILITY: A SYSTEMATIC REVIEW OF CURRENT DEVELOPMENTS

Abstract

Concrete remains the world's most widely used construction material, yet predicting its mechanical properties and durability remains challenging due to the nonlinear relationships between mix design parameters and material performance. Traditional laboratory testing is time-consuming and costly. This systematic review synthesizes peer-reviewed literature on artificial intelligence (AI) and machine learning (ML) methodologies for concrete strength and durability prediction from 2018 to 2025. We analyzed 40 peer-reviewed publications across academic databases, examining algorithm selection, model performance, data requirements, and applications to different concrete types. Findings indicate that ensemble methods, particularly Random Forest and XGBoost, demonstrate superior predictive accuracy (R² = 0.86–0.98) compared to single algorithms. Deep learning approaches, including Convolutional Neural Networks and Transformer-based models, show emerging promise but require larger datasets. For durability prediction, particularly chloride diffusion and carbonation modeling, ensemble and hybrid models utilizing optimization algorithms (Particle Swarm Optimization, Genetic Algorithms) significantly outperform traditional empirical approaches. Critical gaps remain in externally validated models, open-access datasets, and integration of multi-degradation mechanisms. This review identifies best practices for model development, emphasizes the importance of model interpretability using SHAP and LIME techniques, and recommends future directions for advancing AI applications in concrete materials engineering