Neural network model for predicting the carbonation depth of slag concrete

Concrete containing supplementary cementitious materials is vulnerable to carbonation. To know whether carbonation-initiated corrosion is a risk within the life span of the concrete structure, the carbonation depth should be predicted. This paper introduces an artificial neural network model for predicting the carbonation depth of slag concrete. A database was selected that consists of several subsequent research results, including nine variables as input parameters (binder content (B), slag percentage (BFS), water-to-binder ratio (W/B), slag acid index (AI), slag fineness (F), CO 2 concentration (CO 2 ), relative humidity (RH), curing time (Cure) and exposure time (Age)). The carbonation depth was considered as the output parameter of the model. The learning and validation results were found to be reliable with a correlation of 0.98. Moreover, the sensitivity check of the model agrees well with the literature. On the other hand, accelerated carbonation tests were elaborated on concretes containing 0, 20, 40 and 60% slag with W/B ratios of 0.4, 0.5 and 0.6, for 7, 28 and 90 days. The carbonation depth rises with raising slag substitution, water-to-binder ratio and time of exposure. By substituting 20–40% cement with slag, the carbonation depth is the same as that of concrete without additives for water-to-binder ratios not exceeding 0.5. Experimental validation of the developed model showed that the mean relative error is very low (MRE = 7.65%) which proves the accuracy of the ANN model. Besides, the proposed model can be used as a basic tool for service life prediction of concrete structures.