A data-driven approach for predicting interface bond strength between corroded reinforcement and concrete

Abstract

The deterioration of interface bond strength of reinforcement and concrete in a corrosive environment is the primary cause of premature degradation that affects reinforced concrete structures. Thus, it is crucial to establish a highly reliable bond strength model of corroded reinforced concrete. For this study, we collected data from previous experimental investigations on the interface bond strength between corrosion reinforcement and concrete, and developed a BP-ANN model on the interface bond strength based on a data-driven approach. The model utilized input variables for instance corrosion level, reinforcement diameter, concrete cover to reinforcement diameter ratio, reinforcement diameter to bond length ratio, reinforcement yield strength and concrete compressive strength. The output target was interface bond strength, and the model was trained and tested by using 166 groups of pull-out test datas for corrosion reinforcements and concrete with varying corrosion rates. The outcomes demonstrate that the predictions of the BP-ANN model match well with actual values. Furthermore, the predictive efficiency for the BP-ANN model was verified by comparison of the predicted values with the actual values. A comparative analysis of the applicability and computational error of the BP-ANN model indicates that a data-driven approach to predict bond strength between reinforcement and concrete in an erosion environment has better prediction accuracy than the empirical or semi-empirical bond strength models used in previous studies. Additionally, a sensitivity analysis of the primary influencing factors on bond strength of corroded reinforced concrete is conducted, and it is quantitatively confirmed that the corrosion level was the most significant factor on bond strength, while the influence of reinforcement yield strength is comparatively small.