Data-driven assessment of corrosion in reinforced concrete structures embedded in clay dominated soils.

The integration of Artificial Intelligence techniques, particularly Artificial Neural Networks (ANNs), has transformed predictive modeling in structural and durability engineering. This study investigates the use of ANN-based approaches to predict the corrosion rates of mild steel reinforcement embedded in cementitious composites subjected to clay-dominated soil environments. Key environmental parameters, sodium chloride (NaCl) content (0-4%), inhibitor dosage (DOI) (0-5%), and exposure duration (30-180 days), were selected as input variables. Two ANN architectures, Feedforward Backpropagation (FFBP) and Cascadeforward Backpropagation (CFBP), were developed and trained using 72 experimental data points extracted from the literature. The FFBP model outperformed CFBP in terms of predictive accuracy, achieving a correlation coefficient (R) of 0.998, a mean absolute percentage error (MAPE) of 30.43%, and a root mean square error (RMSE) of 0.071 during testing. Sensitivity analysis revealed that inhibitor dosage had the most significant influence on corrosion behavior, followed by NaCl concentration and exposure duration. The findings confirm that ANN models can effectively capture the nonlinear interactions governing corrosion progression, even under complex environmental conditions associated with clayey soils. This research provides a reliable and practical AI-driven framework for assessing corrosion risk, guiding material design, and enhancing long-term infrastructure durability in aggressive subsurface conditions. The study underscores the growing relevance of machine learning in simulating time-dependent deterioration processes in geotechnical and structural materials.