Robust supervised learning algorithm for predicting long-term sulfate corrosion resistance of reinforced recycled aggregate concrete

The corrosion of reinforced concrete against sulfate attacks is one of the most important factors affecting the durability and the mechanical properties of concrete structures. This study aims to expand the understanding of anti-corrosion strategies in recycled aggregate concrete (RAC) by evaluating its corrosion resistance in different conditions. The results indicated a notable decrease (from -897.11 mA to -1141.62 mA) in the long-term half-cell potential of RAC, with a 100% replacement of natural aggregate (NA) with recycled concrete aggregate (R) under acid rain with pH = 2.5. The incorporation of 6% nano silica (S) resulted in a long-term compressive strength increase of 12.5%-23.7% at pH 2.5-7. Similarly, the electrical resistivity under acid rain condition with pH 2.5 increased about 34.1% with incorporation of 6% S. The sorptivity coefficient showed a significant increase due to capillary absorption. This study presents the development of Random Forest regression model to predict the half-cell potential, compressive strength, electrical resistivity, and sorptivity coefficient of RAC based on experimental database comprising 1920 tested specimens. The developed RF model was used for the sensitivity and parametric analysis. In addition, the RF model was used for studying the long-term corrosion performance of RAC in acid rain simulated and submerging method conditions. The sensitivity analysis illustrated that S, R and pH are among the most influential attributes in mechanical properties and durability. Moreover, the findings indicate that the RF model achieved an R 2 value between 0.9423 to 0.9956 for acid rain condition and 0.9373 to 0.9963 for submerging method. • Long-term corrosion performance of reinforced RAC under sulfuric acid attack was systematically evaluated. • The efficacy of a colloidal silica solution in mitigating long-term corrosion and improving durability was investigated. • Comparative tests under simulated acid rain and submersion conditions revealed distinct deterioration mechanisms. • Half-cell potential, CS, SC, and ER were analyzed to evaluate acid-induced corrosion. • A Random Forest Regression model trained on 1920 data points provided accurate prediction and sensitivity analysis of key corrosion parameters.