PSI - Issue 17

Al-Salih/ Structural Integrity Procedia 00 (2019) 000 – 000

2

Hayder Al-Salih et al. / Procedia Structural Integrity 17 (2019) 682–689

683

Peer-review under responsibility of the ICSI 2019 organizers.

Keywords: fatigue; distortion-induced fatigue; cross-frame; digital image correlation; bridge inspection.

1. Introduction and Background

1.1. Fatigue Cracking and Inspection of Bridges Distortion-induced fatigue cracks account for almost 90% of fatigue cracks in aging steel bridges in the United States (Connor and Fisher 2006). Older steel bridges were often designed such that no connection was provided between the flanges and connection plates. When a bridge experiences traffic loading, it undergoes differential deflection between the girders. This allows cross-frames to push or pull on girder webs resulting in secondary out-of-plane stresses applied to the weak web gap regions causing distortion-induced fatigue. In order to mitigate the impact of distortion-induced fatigue, aging bridges are required to undergo regular inspections, and often require repairs and retrofitting. Bridge inspections are typically performed on a 24-month cycle (FHWA 2004), and visual inspection is the most common approach for detecting fatigue cracks. One of the challenges with visual bridge inspection is that fatigue cracks are initially small, and therefore difficult to detect through visual inspections. Undetected cracks, however, can propagate to a size that has the potential to compromise the structural integrity of the bridge. Although these inspections are required to improve the safety of bridge infrastructure by identifying and monitoring cracks over time, manual visual inspections are expensive, time consuming, and dangerous for bridge inspectors and drivers. Additionally, successfully identifying realistic fatigue cracks has been shown to be extremely difficult (Whitehead 2015; Zhao and Haldar 1996). Researchers in various fields have examined technologies to detect and monitor cracks, but many of the approaches are dependent on sensors or other physical attachments. This prevents the detection methods from effectively monitoring the various fatigue susceptible regions on steel bridges. A vision-based, non-contact approach that does not require physical attachment would allow for large areas of bridges to be surveyed in a much safer and efficient manner. While research on vision-based crack detection methods have been conducted, testing has primarily occurred under idealized conditions looking only at in-plane fatigue loading or at cracks in non-metallic materials. Few research programs have evaluated vision based crack detection methods on out-of-plane loading with the complex geometries found on steel highway bridges. In this paper, the performance of vision-based crack detection is being examined using digital image correlation on an out-of-plane test setup with a geometrically complex crack. 1.2. Computer Vision Computer vision refers to technology that uses optics and computer algorithms to collect information from pictures and videos. Various forms of computer vision have been used in engineering and material science disciplines to characterize mechanical parameters. Many researchers have evaluated the potential of using computer vision in the context of crack detection for various materials. For example, researchers have used edge detection to find edge-like features in digital images, leading to the detection and localization of cracks in concrete surfaces (Abdel-Qader et al. 2003). There has been work to develop algorithms to remove short, thick, or exceedingly linear edges, in the hopes of detecting cracking in concrete and asphalt pavement (Yamaguchi and Hashimoto 2010; Zou et al. 2012; Cha et al. 2017). These materials typically have large crack openings, as well as high contrast between cracked and uncracked regions; due to differences in the material, the application of edge detection in evaluating cracks is challenging for steel bridges. In metallic materials, there is a high rate of false positives when using edge detection to identify cracks. The false positives occur from inadvertent detection of corrosion, surface textures, defects, and component boundaries. Kong and Li developed a computer vision strategy to detect fatigue cracks by tracking structural surface motion in a short video, but identification of the crack tip remains a challenge (2018). 1.3. Digital Image Correlation Digital image correlation (DIC) is a subcategory of computer vision that uses image analysis to generate surface displacement measurements. The full-field displacement measurements can then be used to develop three-dimensional strain fields. DIC software can be utilized for both two-dimensional and three-dimensional analysis, depending on the number of cameras used when testing. DIC