Modeling of chloride transport in reinforced concrete using Support Vector Regression: a comparison of models for marine environment

Reinforced concrete structures exposed to the marine environment are vulnerable to chloride-induced corrosion. Traditional service life prediction models are commonly based on the solution of Fick's second law, which only adequately describes the diffusion zone of chloride profiles. The objective of this work is to use machine learning to evaluate alternative models that also consider the convection zone. For this purpose, 8 chloride profiles obtained from structures with approximately 50 years of exposure to the marine environment were used. The technique employed was Support Vector Regression (SVR), and in addition to standard kernels, the Fick model and the Modified Holliday Equation were used as custom kernels. The results demonstrated that SVR was able to adequately fit the field data of chloride concentration, with the RBF standard kernel standing out as the one with the best statistical performance. The custom kernels also showed promising results for the complete representation of the profiles. This approach suggests that the SVR technique may overcome the limitations of traditional models, proposing future improvements in the modeling of chloride transport in concrete.