Deep learning-based automated damage assessment for RC double-column piers

Reinforced concrete (RC) double-column piers, essential bridge substructures, are highly susceptible to earthquake damage. Traditional damage assessment methods primarily depend on visual inspection and structural analysis, which are often subjective and inefficient. This study proposes a Hybrid Structural-Visual Damage Evaluation (HSVDE) framework integrating structural analysis and deep learning-based computer vision. The structural analysis provides an initial classification of performance levels using material strain and drift ratio. To enhance evaluation accuracy and enable rapid post-earthquake assessment, a modified DeepLabv3+ model is employed to identify concrete spalling and exposed rebar. Finite element analysis was utilised to determine drift ratio thresholds for each performance level. The modified DeepLabv3+ model significantly improved rebar detection accuracy, achieving an IoU of 42.80% compared to 33.37%, with only a slight decrease in spalling detection accuracy. The proposed HSVDE framework enhances the accuracy, reliability, and efficiency of seismic damage evaluation, supporting timely emergency response and recovery.