Reversed bond-slip model of deformed bar embedded in concrete based on ensemble learning algorithm

Abstract

A reversed bond-slip relationship that simultaneously considers key factors, such as the concrete cover, stirrups, fiber content, and lateral pressure is required to simulate the seismic response of reinforced concrete structures. In this study, the features of the hysteresis curves with different bond conditions and loading histories were discussed and analyzed, and a corresponding bond stress-slip model based on an ensemble learning (EL) algorithm, XGBoost, was then established. In this model, 10 key factors were selected as the input parameters and 4 reversed bond parameters were selected as the output results. During the training and testing process, a total of 901 sets of experimental data were collected and were randomly split into a training set and testing set at a ratio of 8:2. Compared with the empirical models and two other EL algorithms, the XGBoost method presented a high accuracy to predict the bond parameters with different bond conditions, and the proposed bond stress-slip model correlated well with the test results.