Investigating surface chloride ingress in concrete: effects of mix parameters and environmental exposure

Abstract

Chloride ingress is one of the main durability concerns for a concrete material which leads to reinforcement corrosion and reduce its service life. This study develops a framework relevant to the data-driven in order to predict surface chloride concentration ( C _Cl⁻ ) using ensemble machine learning (ML) models. A dataset of 638 experimental records was extracted, incorporating material, environmental, and exposure parameters. Three models, Bagging regressor (BR), Random forest (RF), and AdaBoost, were trained and investigated. The results show that the Bagging model achieved the highest accuracy ( R ²  = 0.883, root mean square error [RMSE] = 18.35 kg/m ³ , mean absolute error [MAE] = 10.95 kg/m ³ ), followed by RF ( R ²  = 0.877) and AdaBoost ( R ²  = 0.826). A Cross-validation approach confirmed the consistent performance of the employed models, with mean coefficient of determination ( R ² ) values of 0.85, 0.84, and 0.81 for BR, RF, and AdaBoost, respectively. Feature importance analysis found the parameters such as, annual mean temperature, fine aggregate content, and exposure time as the most dominant parameters which shows that chloride accumulation is governed more by material and environmental conditions than by seawater chloride concentration alone. A graphical user interface (GUI) was created to explore the practical application of the models for the instantaneous prediction. This methodology will help readers to provides an efficient and reliable tool for assessing the durability of marine concrete structures.