A Structural–Visual Integrated Evaluation Framework for Seismic Damage Assessment of RC Double-Column Piers

Abstract

The post-earthquake damage state of reinforced concrete (RC) double-column piers directly affects bridge traffic capacity and emergency response efficiency. To improve the interpretability of damage assessment, this study proposes a Structural–Visual Integrated Evaluation (SVIE) framework that combines structural response analysis with image-based damage evidence. Structural responses from quasi-static tests are used to define four baseline damage states: intact-to-slight, moderate, severe, and critical damage. An improved DeepLabv3+ model is then applied to 315 global-scene images for end-to-end semantic segmentation of background, concrete spalling, and reinforcement exposure. The extracted visual evidence is used to verify its consistency with the baseline structural states. On the test set, the model effectively identified concrete spalling regions, achieving an IoU, F1-score, Precision, and Recall of 78.86%, 88.18%, 89.93%, and 86.49%, respectively. For reinforcement exposure, although IoU and Recall were relatively low because of sample scarcity and small-target characteristics, Precision reached 70.40%, indicating that detected regions can provide supplementary evidence for severe local damage. The consistency analysis showed that the morphology of visual damage was generally compatible with the progression of structural damage states. The results provide a laboratory-based proof of concept for a mechanically grounded and visually interpretable framework for rapid post-earthquake assessment of RC double-column piers.