Enabling Rapid Large-Scale Seismic Bridge Vulnerability Assessment Through Artificial Intelligence
Xin Zhang,C. Beck,A. Lenjani,Leslie Bonthron,A. Lund,Xiaoyu Liu,S. Dyke,J. Ramirez,Prince Baah,Jeremy Hunter
DOI: https://doi.org/10.1177/03611981221112950
IF: 1.7
2022-08-12
Transportation Research Record
Abstract:Departments of transportation (DOTs) throughout the United States maintain vast bridge databases that house information such as bridge services, dimensions, materials, inspection reports, and photographs. These databases are expensive to maintain and have evolved quite gradually over the years. They are meant to be substantial enough, at a bare minimum, to support typical asset management activities and to prioritize maintenance tasks. There is great potential to make use of them to support other decisions. However, these databases often lack certain detailed information related to substructure elements, which is necessary for seismic vulnerability assessment, for example, and would be time-consuming to gather for thousands of bridges in a given region or state. In this study, a technique was demonstrated and validated that reduces the time needed to collect this information, by leveraging artificial intelligence to automate the identification of substructure types using images. We defined categories appropriate for vulnerability assessment task, classifiers were trained to identify visual content, and their performance evaluated. In this paper we illustrate a method to determine whether to use artificial intelligence, human visual confirmation, or a combination of the two, to identify bridge substructure types based on accuracy, cost, and risk tolerance. The technical approach was validated using images from Indiana. This leveraging of artificial intelligence for automated identification of critical bridge characteristics from readily available images could empower asset owners, such as DOTs, to assess their inventory more frequently and with confidence.