Prediction model for carbonation depth of concrete subjected to freezing-thawing cycles

Development of hybrid machine learning-based carbonation models with weighting function

“…According to past publications, carbonation models can be classified into four categories based on how the relationship between carbonation depth and its influencing parameters is determined [1]: (1) empirical models, where the relationship between the carbonation depth and its determinants is derived from actual experiments [2][3][4][5][6][7][8][9], such as the square root model [3]; (2) statistical models, where the dependent and independent variables are related by mathematical functions, such as multiple linear regression [10]; (3) numerical models, which consider several physiochemical equations, including chemical reaction rates, mass conservation in gas-liquid two-phase flow, diffusion and dispersion of CO2 in water, energy conservation in porous media, and solubility of CO2 in water, and simulate the phenomena by computer software [11][12][13][14]; (4) machine learning (ML)-based models, which has been applied in recent simulations to find complex nonlinear relationships during the carbonation process [15][16][17][18]. However, all these models have their drawbacks.…”