Formal Processes for Inspecting Bridge Deck Drains Using Lidar and Photogrammetry

Abstract

**Please note that the full text is embargoed until 05/04/2025** ABSTRACT: Poor-performing bridge deck drains result in water standing on the bridge deck. The standing water threatens the safety of bridge users and deteriorates bridge structural elements. Bridge deck drains are only inspected biennially, and they do not impact the bridge rating established from inspections. Furthermore, literature shows that visual inspection results are sometimes skewed due to human errors. In this regard, a part of this research was conducting a survey study to help identify the problems with deck drains to help transportation agencies minimize the consequences of poor drainage. Another part of this research was to create a convolutional neural network (CNN) to classify the problems found in bridge deck drains through images in order to improve inspections on deck drains. The results of the survey study shed light on many bridge deck drain problems that were absent in literature. Furthermore, it also highlighted a major issue: most transportation agencies do not keep track of bridge deck drain assets on bridges. Therefore, as-is models were created based on 3D scenes reconstructed from images and a LiDAR sensor. The results of this research show the advantages of tracking bridge deck drain assets through as-is 3D models created based on acquired imagery and laser scans. The advantages lie in the ability to compare the as-is models with as-built models and track future inspections and maintenance operations. Moreover, as-is models allow designers to analyze the hydraulic properties of drains, and easily modify and improve the design should the need arise. The results of the survey also helped in listing all the major and minor problems. The created as-is model accurately resembled the drain on the inspection site. The resemblance includes the different parts of the drain system, their quantities, and their sizes. As-is models provide the opportunity of exporting spreadsheets containing systems’ components with attached images, which enables office teams to conduct a thorough inspection without going through the inconveniences of on-site inspections. The results also showed that using a pretrained CNN model, it was possible to classify problems found in images taken of bridge deck drains. The model reached 96% accuracy in classifying the condition of grate inlets after being trained on a very small dataset. The outcomes of this research offer transportation agencies a robust way of keeping track of bridge deck drain assets and current conditions. The research also provides the basis for automating the inspection of deck drains using images acquired on site, which promotes safer inspections and decreases time spent on inspections.