Multiphysics modeling of mesoscale concrete cracking under environmental conditions using phase field method and artificial neural network

Corrosion of reinforcement in concrete structures is a primary cause of deterioration, significantly affecting durability and safety. This study investigates corrosion-induced cracking in reinforced concrete under environmental conditions of humidity and temperature using the phase-field method. A novel mathematical model of hygrothermal-electrochemical-mechanical framework, combined with a Monte-Carlo sampling method for aggregate generation, is adopted to represent this physical phenomenon. The numerical scheme is first validated against experimental results and is then applied to further computations. Uniform and non-uniform boundary conditions are applied for marine exposure conditions, including submerged, atmospheric and splash zones. Furthermore, we integrate multi-layer artificial neural networks (ML-ANNs) to predict the corrosion-driven fracture in the reinforced concrete. The numerical results are presented in terms of rust thickness, corrosion depth, cathodic current density, metal loss, crack width, concentration contours and Sherwood number. It is observed that temperature strongly affects the cracking process, resulting in earlier crack initiation and larger crack width. Degradation of rebar and rust thickness are increasing functions of diffusion coefficient and rise exponentially with time. The undulation number greatly changes the concentration distribution and the greatest crack width is obtained under uniform boundary conditions. Fluctuations in Sherwood number (Sh) indicates diffusion disruption due to aggregates, shows the importance of multi-phase concrete over homogeneous one. The proposed ML-ANN methodology predicts crack width with a coefficient of determination above 0.9971501 (99.71 %) and a mean squared error (MSE) below 10−7, signifying high accuracy and generalization.