Machine Learning for the Condition Assessment of Concrete Bridges

Abstract

The assessment of concrete bridge conditions is critical for ensuring structural integrity and public safety. Traditional inspection methods, which rely heavily on visual inspections and manual assessments, are time-consuming, subjective, and prone to human error. With the increasing number of aging bridges worldwide, there is a growing need for more efficient and accurate methods to assess bridge health. This thesis aims to explore the application of machine learning techniques for automating the bridge condition assessment process and improving the accuracy and reliability of bridge evaluations. This study investigates the development and implementation of a model consisting of two machine learning algorithms to predict the condition of concrete bridges based on data collected from various public sources. The first algorithm appraises the structural health of a bridge based on bridge rating and the second algorithm assesses the condition of a bridge after a specific failure mechanism. Specifically, this work focuses on using classification algorithms such as Random Forest (RF), XGBoost, and Neural Networks (NN) in both algorithms to achieve their purpose. The results of this study demonstrate that machine learning models can provide a decent performance in predicting bridge conditions. The overall model achieved a testing accuracy of 79%. This research contributes to the field of civil engineering by showcasing the potential of machine learning in infrastructure management. By automating the assessment process, the proposed models can help reduce the time and cost of inspections while providing more accurate data to guide maintenance planning and bridge rehabilitation efforts. Future work will focus on further optimizing the models, incorporating additional data sources, and deploying the system for real-time bridge monitoring.