Machine Learning-Based Prediction of Corrosion Rate in Reinforced Concrete Incorporating Corrosion Inhibitors

Abstract

Corrosion of reinforcing steel is a major durability concern in reinforced concrete, particularly under chloride-rich exposure conditions. This study develops a machine-learning framework for predicting the corrosion rate of reinforced concrete incorporating calcium nitrite and diethanolamine as corrosion inhibitors. Experimental corrosion-rate data were developed using sodium chloride concentration (NaCl), inhibitor dosage (DOI), and exposure duration (T) as input variables, with corrosion rate expressed in mm/year as the target variable. An 80:20 training–testing strategy was adopted to develop Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Regression (SVR) models. Five-fold cross-validation with grid-search optimization was employed to determine the optimum model parameters. Model performance was evaluated using the correlation coefficient (R), coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and scatter index (SI). The optimized RF model exhibited the highest predictive capability on the independent test dataset, achieving R = 0.999628, R² = 0.999208, RMSE = 0.000014 mm/year, MAE = 0.000006 mm/year, MAPE = 3.21%, and SI = 0.053698. XGBoost also demonstrated strong predictive performance, with R = 0.992982 and R² = 0.980223, whereas optimized SVR achieved R = 0.960408 and R² = 0.788177. The corresponding five-fold cross-validation values were 0.994667, 0.933424, and 0.852174 for RF, XGBoost, and SVR, respectively. Feature-importance and SHAP analyses consistently identified inhibitor dosage as the most influential predictor, followed by exposure duration and NaCl concentration. The results demonstrate that machine learning, particularly RF, can effectively predict corrosion rate within the investigated experimental domain and provide a useful data-driven approach for assessing corrosion behaviour in inhibitor-modified reinforced concrete.