Machine Learning-Based Spatiotemporal Prediction of Corrosion Potential in Cathodically Protected Reinforced Concrete under Chloride Exposure

The durability and service life of reinforced concrete structures are substantially impacted by steel corrosion, particularly in the presence of chloride and under varying environmental conditions. This study predicts half-cell potential (HCP) in chloride-induced-corroded reinforced concrete slabs using machine learning. HCP measurements were taken on reinforced concrete slabs with AZ91D sacrificial anodes over 280 days, yielding 1134 data points. Spatial coordinates (X, Y), concrete age, temperature, and RH were input factors. The performance of five regression techniques was evaluated, namely, Random Forest, Gradient Boosting, Support Vector Regression, Artificial Neural Network, and Linear Regression. Correlation analysis demonstrated that concrete age has a strong positive correlation with HCP (r ≈ 0.82), whereas relative humidity has the strongest negative correlation (r ≈ − 0.88), suggesting that moisture has the most significant impact on corrosion activity. Ensemble learning models showed excellent predictive power, with R² = 0.91, RMSE ≈ 26.6 mV, and MAE ≈ 19.5 mV during spatial validation. Feature importance analysis showed relative humidity to be the best predictor, followed by temperature and concrete age. The nonlinear progression of corrosion processes over time reduced temporal validation prediction accuracy. The results show that experimental monitoring and machine learning can predict spatial corrosion and assess reinforced concrete infrastructure durability.