Physics-based surrogate model for reinforced concrete corrosion simulation

Abstract

Computational models have been used in conjunction with field data, such as in reinforcement concrete corrosion monitoring. It typically involves evaluating the physical model of the corrosion process. However, computation of the numerical solution for the PDE can be highly time intensive. In this paper, a surrogate model is developed to substitute the Finite Element (FE) model of reinforced concrete corrosion. The surrogate model is trained to learn from the FE model by querying the boundary conditions configurations that represents corrosion profile on the steel rebar. A corrosion profile describes the length and the number of the corroded parts of steel rebar in one case. The results demonstrate that the surrogate model is in good agreement to the FE model with a standard deviation less than 0.02. The example cases tested include corrosion profiles with various numbers and lengths of the corrosion. The computing time of the FE model is approximately 360 ms, while the surrogate model is 88 ms, a four-fold time improvement. The surrogate model can be used to improve computational model of corrosion for inverse analysis or optimization.