Inspection Data-Driven Machine Learning Models for Predicting the Remaining Service Life of Deteriorating Bridge Decks

The bridge deck is more vulnerable to deterioration than other structural components. This is due to its direct exposure to environmental factors such as vehicular loads, chloride ingress, and freeze–thaw cycles. The resulting accelerated degradation often results in a serviceability life that is shorter than the intended design life. However, the absence of standardized condition assessment methods coupled with clear definitions of remaining service life has limited the establishment of rational guidelines for repair and strengthening. In a bid to address this lack, this study focuses on PSC-I type bridges in South Korea, utilizing long-term field inspection data to analyze environmental, structural, and material factors—including reinforcement corrosion, chloride diffusion, and freeze–thaw actions. Environmental zoning was applied based on regional conditions, while structural zoning was performed according to load characteristics, thereby allowing the classification of deck regions into moment zones and cantilever sections. Machine learning models were employed to identify dominant deterioration mechanisms, with the validity of the zoning classification being evaluated via model accuracy and SHAP value analysis. Additionally, a regression-based approach was proposed to estimate the remaining service life of the bridge deck for each corrosion phase, thereby providing a quantitative framework for durability assessment and maintenance planning.