Damage Scenario Prediction for Concrete Bridge Columns Using Deep Generative Networks

Abstract

Bridges in areas with high seismic risk are constantly exposed to earthquake threats. Therefore, comprehensive bridge damage assessments are essential for postearthquake retrofitting and safety assurance. However, traditional methods of assessing damage and collecting data are time-consuming and labor-intensive. To address this issue, this study proposes a deep generative adversarial network (GAN)-based approach to predict the surface damage patterns of bridge columns. Using visual patterns from experimental tests, the proposed approach can generate surface damage to the synthetic column, such as cracks and concrete splinters. The study also investigates the effects of different data representation schemes, such as grayscale, black and white, and obstacle-removed images, and uses the corresponding damage indices as additional constraints to improve network training. The results show that the proposed approach can offer a reliable reference for bridge engineers to evaluate and repair seismic-induced bridge damage, which can significantly lower the cost of disaster reconnaissance.