Analysis of carbonation behavior in concrete using neural network algorithm and carbonation modeling

Carbonation on concrete structures in underground sites or metropolitan cities is one of the major causes of steel corrosion in RC (Reinforced Concrete) structures. For quantitative evaluation of carbonation, physico-chemo modeling for reaction with dissolved CO2 and hydrates is necessary. Amount of hydrates and CO2 diffusion coefficient play an important role in evaluation of carbonation behavior, however, it is difficult to obtain a various CO2 diffusion coefficient from experiments due to limited time and cost. In this paper, a numerical technique for carbonation behavior using neural network algorithm and carbonation modeling is developed. To obtain the comparable data set of CO2 diffusion coefficient, experimental results which were performed previously are analyzed. Mix design components such as cement content, water to cement ratio, and volume of aggregate including exposure condition of relative humidity are selected as neurons. Training of learning for neural network is carried out using back propagation algorithm. The diffusion coefficient of CO2 from neural network are in good agreement with experimental data considering various conditions such as water to cement ratios (w/c: 0.42, 0.50, and 0.58) and relative humidities (R.H.: 10%, 45%, 75%, and 90%). Furthermore, mercury intrusion porosimetry (MIP) test is also performed to evaluate the change in porosity under carbonation. Finally, the numerical technique which is based on behavior in early-aged concrete such as hydration and pore structure is developed considering CO2 diffusion coefficient from neural network and changing effect on porosity under carbonation.