A physics-guided generative framework for parameter-conditioned seismic damage visualization of bridge piers

Traditional seismic damage assessment of reinforced concrete (RC) bridge piers relies on costly finite element (FE) simulations and subjective manual inspections. This study proposes a physics-guided conditional generative adversarial network (PG-CGAN) for continuous and physically consistent damage visualization. The slenderness ratio, longitudinal reinforcement ratio, transverse reinforcement ratio, and drift ratio are encoded into a unified normalized condition vector, enabling interpolation across the design space. A multi-task regression discriminator jointly performs adversarial discrimination and physical-parameter regression to enforce consistency between generated damage maps and prescribed structural conditions. Trained on 933 FE-derived damage contour maps from 11 parametric pier models, the PG-CGAN achieves an SSIM of 0.861 and FID of 39.81 on the test set, with a drift-ratio inversion R² of 0.8130 for unseen configurations. The generated damage maps are further used to compare seismic damage distributions under different design schemes, supporting rapid performance-based seismic design.