Data-theory driven prediction approach for the residual axial load carrying capacity of RC columns under salt freeze-thaw cycles

This study aims to develop a data-theory driven approach to predict the residual axial load carrying capacity of reinforced concrete (RC) columns of bridges subjected to salt freeze-thaw cycles (SFTCs). The proposed approach integrates a data driven calculation model and a theory driven simulation model to simulate SFTCs, damage inheritance, and the residual axial load carrying capacity. The calculation model for the residual axial load carrying capacity of RC columns under SFTCs employs a Bi-directional Long-Short-Term Model (BiLSTM) as the primary structure, Convolutional Neural Networks (CNN) for data preprocessing, and Squeeze-and-Excitation Networks (SE) to enhance the data preprocessing capability of CNN, yielding reference values for the residual axial load carrying capacity. Additionally, the simulation model is constructed in ABAQUS to support the data-theory driven prediction of the residual axial load carrying capacity of RC columns under SFTCs. The results demonstrate that the data-theory driven approach accurately predicts the failure modes of RC columns under SFTCs and achieves a prediction accuracy exceeding 93 % for the residual axial load carrying capacity. Consequently, this approach offers an efficient solution for predicting the degradation of the residual axial load carrying capacity of RC columns under SFTCs, reducing experimental costs and enhancing efficiency.