Four-dimensional and vision-based framework for infrastructure maintenance management

Billions of dollars are spent each year to address the ever-increasing demands of maintaining large civil infrastructure assets in the United States. Rapid and reliable updating of the status of tens of thousands of structures and making the information readily accessible for maintenance engineers has become a big challenge for infrastructure management.;Four-dimensional (4D) and computer vision technology are two relatively new and promising technologies in the field of civil engineering. By linking construction schedules with 3D geometric models, 4D has shown its strength in managing and visualizing time-related data in construction planning. However, the capability of 4D tools to manage maintenance information through the lifecycle of infrastructure has not yet been much explored.;A 4D and vision-based framework to acquire and manage the maintenance information of infrastructures is proposed to fully explore and combine the advantages of both 4D modeling and computer vision technology. In the framework, 4D modeling is used to manage and provide easy access to maintenance data through multiple interfaces. Computer vision technology is used to automate the process of acquiring cracking from reinforced concrete structures and provide timely updates about the conditions of structures for maintenance prioritization.;The developed 4D-based bridge/infrastructure maintenance information management system was evaluated with a group of maintenance engineers and engineering students. Experiment results showed that with the help of the 4D-based system, access to maintenance information was made easier and the engineers were able to complete assigned tasks with more accuracy, efficiency and consistency than using traditional paper-based reports.;A series of computer vision-based techniques were developed for acquiring condition information, in this case, cracking from pictures of reinforced concrete structures. The techniques developed include a self-adaptive segmentation and registration algorithm for reinforced concrete beams, crack-detection enhancing algorithms for reinforced concrete, a method to deduce load configuration from crack patterns using probabilistic inference, a method to recover cracks that are not captured in the images and a technique to estimate the remaining capacity of reinforced concrete structures using the obtained crack data. The developed techniques were evaluated with images and data from three laboratory tests of reinforced-concrete beams. Experiment results showed that the segmentation algorithm worked consistently and successfully with all 159 images taken during the beam tests. The quality and speed of crack detection was improved with the enhancing algorithms. Based on the obtained crack pattern, the algorithm was able to successfully recover cracks that were not captured in the images, deduce the underlying load configuration and estimate the sustained plastic damages of the reinforced concrete beams.