Machine learning for predicting the half cell potential of cathodically protected reinforced cement concrete slabs subjected to chloride ingress

This research work predicts the Half Cell Potential (HCP) values of cathodically protected concrete slabs subjected to chloride ingress using machine learning techniques. Six classes of cathodically protected slabs were cast using centrally placed pure Magnesium (Mg) anodes and AZ91D of dimensions 1000 mm x 1000 mm x 100 mm. The input variables included are: distance of the point in x and y direction (Dist x and Dist y), Relative Humidity (RH), Temperature, Age of concrete in days, and the Class of slab. For each class, three slabs were cast and the average value of HCP was considered as output. Experimental values consisting of 80 HCP values per slab per day were collected for 270 days and were averaged to generate the prediction model. Linear regression, kernel ridge, stochastic gradient descent, support vector machine, decision tree, random forest, gradient boosting, and Light Gradient Boosting Machine (LGBM) machine learning models were used for validating the experimental dataset. Among the eight models tested, the LGBM consistently outperforms the others across all metrics. LGBM provides the highest R-squared score (0.9828), lowest Mean Square Error (0.0015), Root Mean Square Error (0.0386), and Mean Absolute Error (0.0220), a-10 index (0.8007), and a-20 index (0.8987) making it the preferred choice for this specific dataset and problem. The sensitivity analysis performed for the LGBM model showed that the 'Age of concrete' is the most influential input for predicting output, while 'Temperature' and 'RH' show lower significance and rest variables have negligible impact.