Reinforced concrete (RC) structures strengthened with steel plates, such as bridges and circular pipe culverts, play a vital role in traffic and civil engineering. These structures often face the risk of damage such as corrosion, cracks and other related years and environmental factors. In order to detect and prevent these potential damages in time, a damage detection system based on deep learning (DL) is proposed. In this paper, various sensor systems are deployed to collect real-time data of structures, such as strain, vibration and corrosion data. These data are input into DL model after preprocessing. The model combines the characteristics of convolutional neural network (CNN) and recurrent neural network (RNN), which can not only analyze the spatial characteristics of data, but also capture the dynamic changes of time series. In addition, for special structures such as circular pipe culverts, the model is specially designed to identify tiny cracks and corrosion within them, further refining the accuracy of damage detection. The significant reduction in the structural damage rate (from 15 to 5%) illustrates the effectiveness of the system in monitoring and preventing potential structural problems. With this approach, the study not only significantly improved the accuracy of damage detection in circular pipe culverts, it provided engineers with a window of time to respond quickly and significantly improved public safety and security.