Intelligent Detection of Bolted Connection Damage in Steel Truss Structures Based on LBD-YOLO
This paper proposes LBD-YOLO, a lightweight and high-precision object detection framework that integrates GSConv, ADown, BoTNet, and BiFPN modules to effectively identify bolt damage in steel truss structures, achieving 97.7% mAP@50 while significantly reducing computational costs for efficient edge deployment.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine a world where giant steel skeletons hold up our bridges, power lines, and mountain cable cars. These structures are like the bones of our modern infrastructure, and just like our bones, they need to be strong and connected. The "joints" holding these steel bones together are bolts. But over time, wind, rain, and heavy loads can make these bolts loosen, rust, or even disappear entirely. If a bolt fails, it's like a loose screw in a chair leg; eventually, the whole thing could wobble or collapse. For decades, checking these bolts has been a dangerous job for humans who have to climb high into the sky, squinting at tiny metal nuts to see if they are safe. It's slow, risky, and relies heavily on a human's tired eyes.
To solve this, scientists have started using "computer vision," which is basically teaching computers to "see" and understand images just like we do. A popular tool for this is called YOLO (You Only Look Once), a type of artificial intelligence that acts like a super-fast scanner, spotting objects in pictures in a split second. However, spotting a tiny, rusty bolt against a complex, messy background of steel beams and clouds is like trying to find a specific grain of sand on a beach while wearing sunglasses. The computer often gets confused, misses the small targets, or gets tricked by the background. This is the puzzle the researchers in this paper set out to solve: how do we make a computer smart enough to spot these tiny, dangerous bolt problems quickly, accurately, and without needing a super-computer to do the work?
The researchers, a team from Jishou University, have built a new, super-smart detective named LBD-YOLO. Think of LBD-YOLO as a high-tech security guard for steel bridges. Instead of just using a standard camera, they gave this guard a special set of tools to make its eyes sharper and its brain faster. They started by creating a massive photo album of 3,600 images, showing bolts in four different states: happy and tight (normal), wiggly (loosening), gone entirely (missing), and covered in orange rust (rusted). They even used a cool drone that flies like a first-person-view video game character to snap these photos from tricky angles, ensuring the computer learns what a bolt looks like in the real world, not just in a perfect lab.
To make their detective better, the team swapped out the old, heavy parts of the computer's brain with four new, lightweight upgrades. First, they used GSConv, which is like giving the detective a pair of high-definition lenses that can see tiny details (like a rust spot) without getting tired. Second, they added ADown, a clever filter that makes the image smaller for the computer to process but refuses to throw away the important edges of the bolt, ensuring the "grain of sand" doesn't get lost. Third, they installed BoTNet, a "global awareness" module that helps the detective understand the whole picture, so it knows a missing bolt is missing because it sees the hole in the context of the whole bridge, not just a random spot. Finally, they used BiFPN, a smart organizer that mixes information from different distances, helping the computer spot both huge bolts and tiny ones equally well.
When they tested this new detective, the results were impressive. On their custom photo album, LBD-YOLO correctly identified bolts 97.7% of the time (a score called mAP@50) and was very precise even with strict rules (61.0% mAP@50-95). But the real magic was in its efficiency. Compared to the standard version it was built on, this new model used 16.1% fewer brain cells (parameters) and 19.2% less energy to think, yet it was actually better at finding the bolts. It's like upgrading a car engine to be both faster and more fuel-efficient at the same time.
The team didn't stop at the computer screen. They wanted to see if this detective could work in the real world, so they put it on a smartphone (an OPPO Find X8). The app was tiny (only 11.4 MB) and could check bolts in real-time, processing about 20 images every second. To prove it wasn't just a lab trick, they took the drone and the app to a real, working power transmission tower in Zhangjiajie. Even though this tower looked different from the ones in their training photos and the weather was messy, the detective still found 91.6% of the bolts correctly. It didn't get fooled by the sky, the trees, or the glare of the sun.
The paper suggests that this method is a strong step forward for keeping our steel structures safe without putting humans in danger. While it isn't perfect—rusty bolts are still a bit tricky to spot with high precision, and extreme shadows can sometimes confuse it—the results show that a lightweight, smart AI can now do a job that used to require a human climber. The researchers conclude that this approach offers a practical, high-precision way to check our infrastructure, and with a little more training to handle even tougher weather, it could become a standard tool for keeping our bridges and towers standing tall.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.