Artificial neural network model to estimate the long-term carbonation depth of concrete exposed to natural environments

Carbonation influences the performance and reliability of reinforced concrete structures during its lifecycle. Predicting Carbonation Depth (CD) in concrete requires understanding the complex relationship between different variables. The DURACON project (Environment Influence on Concrete Durability) has been active since 2000 to characterize the durability of exposed concrete in Ibero-American urban/rural environments (10 countries participated with 24 natural test stations installed); where carbonation is one of the main mechanisms that induces corrosion of steel reinforcement. In this publication, the data obtained from the DURACON project during the first nine years of exposure was used to generate an Artificial Neural Network (ANN) model that was trained to predict the CD of reinforced concrete in different natural (non-accelerated) circumstances and environmental situations. 8420 ANN structures were constructed to find the most precise and efficient model that predicts long-term (up to 10 years) CD. The results show a significant nonlinear relationship between the selected variables and the optimized ANN model, which has satisfactory accuracy in predicting long-term CD. In terms of environmental parameters, the ANN model demonstrated that even though initially the Relative Humidity (RH) has a strong influence on the CD, over Time (t), the Temperature (T) and Accumulated Precipitation (APP) dominate the concrete carbonation process. Nevertheless, Calcium Oxide (CaO) and capillary absorption (k), which represent the concrete quality, have the most influence on the CD. Furthermore, the performance of the ANN model was compared to other predictive models, such as Decision Tree (DT) and Multiple Linear Regression (MLR) and was found to provide more accurate CD predictions than the other models.