Predicting carbonation depth in fiber-reinforced ultra-high performance concrete (FR-UHPC) using state-of-the-art machine learning techniques

Carbonation-induced degradation is one of the leading causes of durability loss in concrete structures. Despite advances in conventional concrete carbonation models, predictive models for fiber-reinforced ultra-high-performance concrete (FR-UHPC) remain scarce, given its complex, multiscale behavior. This study presents a new and data-driven analytical framework for predicting the carbonation depth of FR-UHPC using advanced machine learning techniques, including neural operators for modeling physical systems (NOMPS), artificial intelligence-based pipeline search for regression (AIPSR), quantum machine learning (QML), and explainable AI using quantum shapley values (EAIQSV). Analysis of 800 experimental data points identified curing time, temperature, and silica fume content as key determinants of carbonation depth. The models were validated through rigorous statistical analysis and 5-fold cross-validation, with AIPSR outperforming the other models in terms of prediction accuracy (R² = 0.83) and consistency. This framework provides a robust and repeatable method for predicting carbonation in FR-UHPC, while improving interpretability and incorporating quantum-inspired machine learning techniques.