Abstract:
In order to accurately assess the structural bearing capacity of corroded reinforced concrete (CRC) structures under sudden fire incidents, there is an urgent need for research on a unified predictive method for the high-temperature bond strength of CRC. However, the degradation mechanism of bonding is complex, and there are numerous bonding factors. Experimental methods cannot consider the influence of all the complex relationships between bonding factors. Based on a large amount of existing experimental data, Machine Learning (ML) methods can effectively establish regression relationships between input and output features through data. In this study, two ML algorithms, artificial neural network (ANN) and extreme gradient boosting (XGB), were used to establish a unified predictive model for the high-temperature bond strength of CRC. The model was trained and tested upon 612 sets of experimental research data on high-temperature CRC. The results show that the predictions of the ML model are in a good agreement with the experimental results. In addition, to address the "black box" problem inherent in ML algorithms, the SHAP method was used to solve the interpretability problem during the prediction of high-temperature bond strength of CRC. Moreover, the calculation outcomes of the ML model were compared with those of three theoretical calculation formulas, and the ML model demonstrated clear advantages. The newly constructed hybrid ML model is likely to become a new choice for accurately assessing the extent of damage to CRC structures after exposure to high temperatures.