Prediction of chloride content in concrete using ANN and CART

Chloride-induced corrosion of concrete structures in marine areas is a serious problem and is generally affected by several factors. Chloride concentration is an important parameter for estimating the corrosion state of concrete. In this research, first chloride concentration at various depths of concrete specimens was measured using the accelerated chloride penetration test method under laboratory conditions, simulating a marine environment after 4.5 and 9 months. Then the obtained experimental dataset of 162 in 9 months of exposure was used to develop classification and regression trees (CARTs) and an artificial neural network (ANN) as subsets of artificial intelligence methods. Environmental condition, penetration depth, water-to-cementitious material ratio and silica fume mass were considered as input parameters, and chloride concentration was taken as the output parameter. Finally, results for the two methods were compared with the experimental observations to evaluate their accuracy in phases of training and testing. As a further aspect to the study, prediction of chloride concentration as a function of the exposure time and unavailable testing parameters was carried out. The results showed that ANN and CART have good ability and accuracy for predicting the chloride concentration in concrete under marine environment conditions. In the present research, the ANN method showed more accuracy.