Optimized neural network based carbonation prediction model

Concrete carbonation is one of the major causes of steel corrosion in reinforced concrete structure that can lead to shortened service life. Several carbonation prediction models including mathematical and neural network are available. The mathematical models are simplified and do not take all influential parameters of concrete materials into consideration. Most of the existing neural network based carbonation prediction models do not employ all parameters that influence the microstructural properties of the concrete. They also failed to perform certain essential steps during the model development, which in turn degrade their performance. In this work, novel neural network based carbonation prediction model is proposed. The model selects most relevant parameters, and removes irrelevant and/or redundant features from the original input data to build robust learning models. The performance evaluation of the model shows that the proposed carbonation prediction model predicts reasonably well with increased generalization ability.