Enhancing Bridge Management: A Data-driven Approach for Accurate Forecasting of Concrete Bridge Condition

Ahmed Assad and Ahmed Bouferguene

Abstract

Highway bridges are vital components of infrastructure systems that support effective vehicle flow throughout the transportation network. Nevertheless, several factors, including the age of the bridge, operational characteristics, and exposure to the climate threaten the continuous function of such bridges. Because of this, it may be challenging to confidently anticipate the condition of bridges and prioritize needed maintenance tasks. In this article, we suggest a smart data-driven methodology for forecasting the condition of bridges based on a range of structural and operational aspects. In addition, several climatic factors were considered to assess the impact of environmental exposure on bridge conditions.

Different machine learning algorithms including neutral network, support vectors machine, random forest, and others were trained utilizing historical bridge inspections in the U.S. Feature engineering and hyperparameter tuning techniques was used to identify the factors that have the most influence on the condition. With a mean relative error of 3.8%, GBT produced the most promising results. Additionally, the research demonstrated the predictive significance of some climatic parameters, particularly the freezing index. The developed model provides an accurate and timely assessment of their condition which can be leveraged to prioritize maintenance and renewal activities.