Enabling Rapid Large-Scale Seismic Bridge Vulnerability Assessment Through Artificial Intelligence

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.

Machine-Aided Bridge Deck Crack Condition State Assessment Using Artificial Intelligence

“…In the previous section, a cost function was proposed to calculate the cost due to the limitations of this AI-based method. Since the primary attraction of implementing an AI method is to reduce the costs associated with a human's work [39], it is important to develop a method to quantitatively analyze whether the inspector should trust any automated method. For every reinforced concrete bridge deck, an inspector must decide to either accept the automated result and its uncertainty, with the risk of overlooked degradation, or ignore the results in favor of manual inspection.…”

“…The cost of manual inspection should be considered in this decision. Previous research considered how to make this decision for a very difficult problem involving data entry [39]. Thus, this method assumes that all classification steps prior to the semantic segmentation step are 100% accurate and all risk in this workflow stems from the semantic segmentation step.…”

Numerical Assessment of the Seismic Vulnerability of Bridges within the Italian Road Network

“…For instance, Ozsarac et al [59] explored the use of simulated records in the assessment of R.C. bridges; Biondini et al [60] proposed and implemented a procedure based on a two-level structure-network risk assessment for bridge prioritization, while the work of Zhang et al [61] introduces a technique that employs artificial intelligence for the large scale vulnerability assessment of bridge structures.…”

Machine-Supported Bridge Inspection Image Documentation Using Artificial Intelligence

“…Researchers have explored the application of AI in different fields. At this point in its development, AI is not able to perform most tasks independently, and humans still need to participate in the decision-making process ( 7 , 8 ). In the medical field, doctors use AI to assist with interpreting medical images, and making diagnoses with the help of these results ( 9 ).…”

“…Liu et al also developed a method of automating the localization of images collected from structures to support building reconnaissance rapidly within a large-scale area, and to support the classification of the damage state of a building based on the images ( 19 , 20 ). For applications in the bridge engineering field, Zhang et al explored the use of convolutional neural networks (CNNs) to rapidly extract bridge substructure information for coupling with a typical database to perform an automated seismic vulnerability analysis of typical bridge networks in regions with low-to-medium seismicity such as the American Midwest ( 8 , 21 ). Drones and robots are also being used in collaboration with AI to aid civil engineering researchers or engineers in their work.…”