Leveraging Bridge and Environmental Features to Analyze Coating Conditions of Steel Bridges in Florida Using Neural Network Models

Abstract

The safety, integrity, and longevity of steel bridge elements are affected by various environmental factors, such as moisture, atmospheric pollutants, and temperature. Protective coatings of steel bridge elements are especially sensitive to the service environments (e.g., atmospheric environments, water environments) of bridges, as many environmental stressors may accelerate premature failures of coatings. By leveraging the data on both bridge-related features (e.g., bridge age, type of service under bridge) and environmental features (e.g., chloride, moisture, atmospheric pollutant, temperature), this study focuses on bringing a data-driven understanding of steel bridge coating deterioration patterns and how the service environments of bridges may impact such patterns. Steel bridge coating performance prediction (SBCPP) models were built based on multilayer perceptron (MLP)–based artificial neural network (ANN) algorithm to predict and assess the coating conditions of girder elements of steel bridges in Florida. The results show that the SBCPP model with the best performance can precisely predict steel bridge coating conditions with a mean absolute error (MAE) of 0.0807. As compared to the models that do not account for environmental features, the performance of the proposed SBCPP models was significantly improved. Shapley additive explanations (SHAP) analysis was further conducted to interpret and analyze the influences of input features on the performance of the SBCPP model. The study offers an effective decision-making tool that has the potential to benefit state transportation agencies by allowing for easier and more efficient analysis or prediction of steel bridge coating performance. This study proposes data-driven models that analyze and predict the coating conditions of steel bridge elements based on the data of bridge-related features (e.g., bridge age, average daily traffic, type of service under bridge) and environmental features (e.g., chloride, moisture, atmospheric pollutant, temperature), thus providing knowledge and insights on corrosion-induced coating degradation for steel bridges in Florida. The proposed models serve as the foundations for an automatic coating performance assessment and prediction tool that can help state transportation agencies and other bridge owners easily evaluate the coating conditions of their steel bridges and identify those bridges that are in critical maintenance needs. Inspection of coating conditions traditionally requires significant manual efforts, which are expensive, time-consuming, and cause safety concerns. An automatic tool based on the proposed models can quickly and easily analyze the element-based coating conditions while considering the impacts of the service environments. By using the tool, a bridge owner can quickly identify those bridges that require more attention or funding support and decide which bridges should be given priority in terms of inspection, maintenance, and/or repair. Hence, acting as the foundation of such a decision support tool, the proposed work has the potential to reduce field-level coating inspection costs, efforts, and risks and enhance the efficiency of maintenance and repair-related decision-making.