Concrete Carbonation Analysis using Neural Network Algorithm and Change in Pore Structure

Carbonation on concrete structures is one of the major causes of deterioration in concrete structures. For quantitative evaluation of carbonation, a physico-chemo modeling for reaction with dissolved CO2 and hydrates is needed. Even the amount of hydrates and CO2 diffusivity coefficient are very important to evaluate behavior of carbonation, it is limited to obtain a various CO2 diffusivity coefficient from experiment due to time and cost. In this study, a numerical technique to predict carbonation depth in concrete using neural network algorithm for proportions of mixture design and change in porosity is developed. In order to obtain the comparable data set of CO2 diffusivity coefficient, existing experimental results are utilized. Data for mixture proportion and relative humidity are selected as neurons and learning for neural network using the so-called back-propagation algorithm is carried out. The results from neural network are in good agreement with experimental data obtained for different water to cement ratios (42%, 50%, and 58%) as well as different relative humidity (10%, 45%, 75%, and 90%). The mercury intrusion porosimetry (MIP) is also performed to evaluate the change in porosity during the carbonation. Finally, a numerical technique, which is based on micro modeling for hydration and pore structure (multi component hydration heat model and micro pore structure formation model), is developed using diffusivity coefficient from neural network algorithm and porosity change from the MIP test. The technique is verified by comparing the numerical results and the experimental results for carbonation depth, and also expected to be more reasonable technique with more various experimental data