Intelligent Vision-Based Pose Estimation with the Turn-of-Nut Method for Bolt Monitoring

In the hidden architecture of the modern world, from the steel girders of skyscrapers to the intricate frames of aircraft, countless metal fasteners hold everything together. These are bolts, simple threaded cylinders that rely on a hidden force called preload to stay secure. When a bolt is tightened, it stretches slightly, creating a powerful clamping force that locks the joined parts in place. However, the relentless vibration of engines, the shifting weight of a bridge, or the rhythmic pounding of machinery can slowly erode this tension. As the bolt relaxes, the joint loosens, potentially leading to catastrophic failure. For decades, ensuring these connections remain tight has been a labor-intensive task, often requiring human inspectors to strike bolts with hammers to listen for a change in sound or to use heavy wrenches to check if they have turned. While sensors can be embedded into bolts to monitor them, these devices are expensive, require complex wiring, and often fail in harsh environments. The challenge has long been finding a way to "see" the invisible tension inside a bolt without touching it or altering the structure.

Researchers at Harbin Engineering University and Bahauddin Zakariya University have developed a new approach that treats the bolt and its nut as a visual puzzle to be solved by a camera. Instead of relying on physical sensors or human ears, they built a system that uses two cameras to watch a bolt and its nut simultaneously. The core idea is simple yet precise: if a bolt is loosening, the nut and the bolt head will rotate relative to each other. By measuring this tiny rotation with extreme accuracy, the system can calculate exactly how much tension has been lost. To make this work, the team trained a sophisticated computer vision model, a type of artificial intelligence known as YOLOv11-pose, to recognize specific points on the metal surfaces. They taught the software to identify five distinct landmarks on both the bolt head and the nut: four fixed points that define the orientation of the metal piece, and one moving marker that acts as a reference for rotation.

The process begins with the cameras capturing images of the assembly. Because the cameras are not always perfectly aligned with the bolt, the images can be distorted by perspective, much like how a straight pole looks bent when viewed from an angle. To correct this, the system applies a mathematical transformation that flattens the image, creating a perfect, head-on view of the bolt and nut faces. This allows the computer to measure the angle of the moving marker with high precision. By comparing the angle of the marker on the bolt to the angle of the marker on the nut, the system calculates the relative rotation between the two. This measurement is then fed into a mechanical model based on the "Turn-of-Nut" method, a standard engineering principle that relates how far a nut is turned to how much the bolt stretches. By knowing the stiffness of the metal parts and the initial tightness, the system can translate that tiny angle of rotation into a specific force value, telling engineers exactly how much preload remains.

To test this method, the researchers set up a controlled experiment using a single large bolt, specifically an M20 size, made of stainless steel. They tightened the assembly and then used a digital load cell, a highly sensitive scale for measuring force, to record the actual tension. They then compared the load cell's readings with the estimates generated by their camera system. The results were promising. When measuring the angle of rotation, the system was remarkably accurate, with an average error of less than one degree and a maximum error of just 2.4 degrees. When translating this angle into a force measurement, the system estimated the preload with an average error of 0.63 kilonewtons. In the context of industrial safety, where bolts often carry forces of hundreds of kilonewtons, this level of precision suggests the method is viable for detecting significant loosening.

However, the study also revealed the limitations of relying on sight alone. The researchers found that the system is sensitive to lighting conditions. When they reduced the brightness of the environment, the accuracy of the measurements dropped. In dimmer light, the system struggled to pinpoint the exact location of the markers, leading to larger errors in the force calculation. At a 75% reduction in brightness, the average error in estimating the force rose to nearly 3 kilonewtons. This indicates that while the technology works well in controlled settings, it requires clear visibility and consistent lighting to function reliably. The researchers also noted that surface conditions like rust, oil, or paint could interfere with the camera's ability to see the markers, suggesting that future versions of the system would need to be robust enough to handle dirty or degraded industrial environments.

The significance of this work lies in its potential to replace invasive and expensive monitoring methods with a non-contact, low-cost solution. By combining the visual detection of rotation with a proven mechanical model, the researchers have demonstrated a way to monitor the health of critical structures without drilling into them or attaching wires. While the current study was limited to a single type of bolt in a laboratory setting, the framework provides a clear path forward. The system successfully proved that a dual-camera setup could track the relative movement of a bolt and nut with enough precision to estimate the internal forces holding a structure together. As the technology matures and is tested on a wider variety of bolts and under more challenging real-world conditions, it could become a standard tool for ensuring the safety of bridges, wind turbines, and aircraft, keeping the invisible forces of the built world under constant, watchful observation.