Quantitative characterization of reinforcement cross-sectional roughness and prediction of cover cracking based on machine learning under the influence of pitting corrosion

The roughness characteristics caused by pitting corrosion on the reinforcement surface have an important influence on cover cracking. This study proposes two new indicators, R_MPC and C_MPC, for quantitatively evaluating reinforcement roughness and concavity. Then a novel approach to predicting crack volume was introduced based on ML. Results show that, R_MPC is more applicable than commonly used morphological indicators for reinforcement roughness evaluation. The dry-wet cycle corrosion produces more severe section roughness and concavity than the applied current corrosion, up to about 2.4 times. When the corrosion level exceeds 3%, average R_MPC of the dry-wet cycle samples are consistently higher. When the corrosion level is less than 1%, the cross-section is typically concave. The introduction of roughness indicators significantly improves the accuracy of crack volume prediction, increasing R² value from 0.646 to 0.956. Machine learning prediction models using ensemble learning algorithms demonstrate superior accuracy and stability compared to non-ensemble models.