Identification of chloride diffusion coefficient in concrete using physics-informed neural networks

The corrosion of steel caused by chloride is one of the important causes of the deterioration of reinforced concrete structures. In order to describe the transport behavior of chlorides in concrete and to make quantitative durability designs based on it, it is necessary to determine the chloride diffusion coefficient correctly. However, the commonly used analytical method is not suitable for the free chloride diffusion coefficient considering the nonlinear chloride binding capacity of concrete. In this study, by comparing the results of solving forward problems involving chloride transport equations using finite element method (FEM) and solving corresponding inverse problems using physics-informed neural networks (PINNs), it is verified that PINNs can be applied to identify the chloride diffusion coefficient in non-steady-state immersion tests and accelerated chloride migration tests considering no chloride binding, Langmuir binding, or Freundlich binding. The robustness of PINNs is verified by training with 5% and 10% noise data as observed data. The results of interpolation and extrapolation using PINNs trained successfully show its excellent generalization performance.