Impact of Environmental Conditions on Predicting Condition Rating of Concrete Bridge Decks

Abstract

Highway agencies prioritize maintaining bridge infrastructure through bridge management systems amid budget constraints. The premature deterioration of reinforced concrete (RC) bridge decks caused by more frequent and increasingly heavy truck load spectra coupled with aggressive environmental conditions has become a critical concern. Despite the prevalence of conventional models and the emerging popularity of machine learning (ML) models in bridge deterioration predictions, they fall short in feature selection and handling of climate conditions, leading to suboptimal accuracy. To address these gaps, this study presents a data-driven framework utilizing ML-based techniques to predict the condition rating of RC bridge decks with a focus on identifying the influencing factors that affect the deck condition. The framework employs the XGBoost algorithm for model development, encompassing comprehensive datasets that include structural, geographical, and climate variables from across the U.S. Furthermore, the Shapley additive explanations approach is applied to identify the explanatory variables with the most impact. Age emerged as the most crucial factor, followed by freeze-thaw cycles and truck traffic, as indicated by the average daily truck traffic. Rainfall also plays a substantial role in deck deterioration. Based on feature importance and monotonicity, this study recommends a series of bridge classifications for transportation agencies to incorporate into their deterioration models. Overall, this research enhances understanding of the primary causes of bridge deck deterioration, enabling more informed decisions about funding allocation and bolstering bridge performance against environmental challenges.