ABSTRACT
Coastal natural draft cooling towers face accelerated aging from reinforcement corrosion and concrete spalling, threatening seismic safety in thermal power infrastructure. This study establishes an integrated framework combining drone/laser-based corrosion mapping, numerical simulation, and a Bayesian Backpropagation Neural Network (BBNN) to enhance seismic fragility analysis. Field scans reveal non-uniform corrosion concentrated at throat and top regions. Numerical simulations show service-time-dependent compressive damage migration toward the throat, though corrosion in these areas exhibits limited impact on seismic safety margins. These findings demonstrate that localized corrosion in critical zones does not proportionally compromise global seismic performance, providing insights for risk-prioritized maintenance of aging cooling towers. The BBNN model effectively optimizes incremental dynamic analysis efficiency when trained with preprocessed data that mitigates outlier influences. The emerging integration of physics-informed neural networks (PINNs) strategically combines data-driven approaches with mechanics-based principles, offering theoretical advantages for predictive robustness even with sparse datasets.
