Physics-based probabilistic analysis of corrosion initiation in alkali-activated slag concrete assisted by machine learning

With the pressing demand for sustainability in infrastructure construction, alkali-activated slag (AAS) concrete has emerged as a viable green alternative to carbon-intensive cementitious materials. However, the long-term durability prediction of this material, particularly regarding chloride-induced corrosion, remains underexplored, which limits its broader practical applications. To address this gap, this study presents an innovative machine learning (ML)-assisted framework tailored for probabilistic analysis of corrosion initiation in AAS concrete. The framework incorporates an advanced physics-based approach that integrates multi-ionic transport analysis with hydrate-based modeling of chloride binding, thereby delivering a precise chloride ingress evaluation. To overcome computational challenges, a novel ML algorithm, extended support vector regression (X-SVR), is applied to develop metamodels that serve as surrogates for the computationally demanding physical analyses. This metamodeling technique accelerates sampling-based probabilistic analysis, enabling rapid assessment of time-dependent corrosion probability under various uncertainties. The accuracy, efficiency and versatility of the proposed framework are assessed via systematic investigations. The effectiveness of the physics-based model in capturing chloride resistance of AAS concrete is confirmed using experimental data. With accurate metamodels generated by X-SVR, the ML-assisted workflow demonstrates substantial efficiency gains in obtaining detailed statistical information over traditional Monte Carlo simulation, without sacrificing precision. Additionally, the framework proves effective in sensitivity assessments and real-time response updates as new information becomes available. Overall, this study provides a practical approach to leveraging computational materials science for investigating the risk of chloride-induced corrosion in AAS concrete, thereby contributing to the development of sustainable infrastructure systems. • A novel tool for predicting corrosion risk in alkali-activated slag concrete is developed. • The integrated reactive transport model accurately assesses chloride ingress. • The used machine learning algorithm substantially boosts computational efficiency. • The proposed framework allows for efficient sensitivity analysis of random variables. • The developed technique offers real-time response updates under new statistical information.