CHAPTER 4

Conclusions and Suggested Research

After the research team conducted the UAS-based inspections at the selected bridge sites and analyzed the element-level inspection data, it was apparent that UAS can be incorporated into routine bridge inspections with element-level data collection, but their effectiveness depends on the inspection requirements, the type of UAS employed, remote pilot skill, and the data review process. If the appropriate UAS is paired with the appropriate structure type and used to inspect certain elements, then incorporating the UAS can provide higher-quality data, time savings, a safety improvement, or all three. Not all inspections may require or benefit from UAS, but the technology can be considered as another tool for bridge inspectors to use at their discretion.

Regarding the two UAS form factors used in this research, both had their advantages and disadvantages. The FF1 UAS was compact, portable, and had advanced obstacle avoidance capabilities. This meant the airframe could access more areas of structure, particularly in girder bays above the level of the bottom flange. However, the small size meant the airframe was more susceptible to wind gusts and vortex shedding (as around cables), was less visible at farther distances, and was more sensitive when flown around slender elements. When flown close to the bridge elements, the lower-resolution camera was adequate. However, the zoom capabilities were insufficient to inspect the inboard face of opposite cables, and the remote pilots reported that the exposure adjustments on the single-operator controls were cumbersome to access in flight.

Conversely, the FF2 UAS had dual operator controls, which meant the workflow could be efficiently split between a remote pilot and a sensor operator. The higher-resolution interchangeable prime and zoom lenses were also capable of collecting higher-quality video and imagery, even when the UAS needed to be flown at farther standoff distances because of its larger size and limited obstacle avoidance functions. With the FF2 airframe, the UAS easily inspected the inboard face of opposite cables without flying over traffic. The larger size also meant the airframe was more stable and less susceptible to wind gusts and vortex shedding but less portable and less accessible in tight spaces. For example, the FF2 UAS rarely flew above the bottom flange of steel girders. If multiple airframes were available to inspectors, they may find it practical to use an FF2 UAS for an initial scan and then an FF1 UAS for a closer inspection of targeted locations with limited access.

If a UAS is incorporated into a Routine Inspection, its usefulness depends on the structure type and current inspection methods. For example, if a Routine Inspection of a common steel or concrete multi-girder structure is typically conducted by telescopic-aided (binoculars) visual observation from the ground below and by walking the bridge deck, UAS may improve the quality of inspection data but take more time. The UAS may provide enhanced access, better viewing angles, and more thorough imagery, but the time to set up the UAS and collect, review, or process the inspection data may exceed the time typically spent on a Routine Inspection. However, if a structure of this type was typically inspected using an aerial lift or UBIT, the UAS could improve the quality, efficiency, and safety of the inspection.

Using a UAS in this manner may reduce or eliminate the need to mobilize heavy equipment, set up traffic control, or work in traffic. UAS would also provide time savings if used to inspect non-fracture-critical elements that were typically inspected using rope access, such as the free length of cables on arch, suspension, or cable-stayed structures. In one example, incorporating UAS into an in-depth inspection of the stay cables of Bridge 2 saved 5 days when compared with an inspection conducted solely by rope-access methods. In another inspection, using UAS to inspect the free length of the main span cables on Bridge 7 saved 4 days of work compared with a previous rope access inspection.

Much of the data review process for the inspection video and imagery collected in this research involved reviewing the videos to determine what elements and deficiencies were visible and to rank their detectability. Detectability was based on access, sensor adjustments, and remote pilot comfort. This work was done by a team member who was present when the video was collected, but a small subset of data was sent to three inspectors who were not present on site. The inspectors’ comments and review demonstrated that a desktop review of UAS imagery was adequate for report generation but was limited by the inspectors’ inability to see all facets of an element and more importantly, their inability to immediately follow up with an in-depth inspection while still on site. One reviewer noted a bent cross frame and a poor-quality weld that warranted a closer inspection, which would have required remobilizing to the site. For these reasons, using a desktop review as a substitute for an on-site inspector is not suggested. Instead, it is suggested that an inspector conduct a desktop review on site using a larger screen with enhanced resolution to determine locations of concern across elements.

In addition to the desktop review by three inspectors, two companies performed AI-based defect detection services for some data from Bridge 3 approach spans. After each company reviewed the results, it was apparent that the data set used to train the AI model was important to the applicability of AI-based inspection. For example, the AI model used by Company 2 was trained mainly to detect flexible pavement cracking and was therefore ineffective when used to detect cracking in prestressed concrete. This highlighted the need to develop AI models specifically for bridge inspections, using data sets that encompass a range of structural systems, materials, and deficiencies representative of the bridges found across the country. Most AI models and research currently focus on crack and spall detection in concrete elements, with only limited research on crack detection in steel materials. Additional research is needed to advance the use of AI models for bridge inspections. AI applications hold promise for bridge inspections, but bridge owners need to treat the technology as another tool to assist inspectors in identifying a limited subset of elements and condition states present in a Routine Inspection with element-level data collection.