Real-Time Flange Bolt Loosening Detection with Improved YOLOv8 and Robust Angle Estimation

Area of Science:

• Engineering
• Computer Science
• Materials Science

Background:

• Flange bolts are critical for structural integrity in civil, mechanical, and aerospace engineering.
• Current bolt loosening detection methods face challenges with accuracy, feature representation, and complex environments.

Purpose of the Study:

• To develop an integrated framework for accurate bolt loosening detection and angle estimation.
• To improve upon conventional methods by leveraging deep learning for enhanced performance and reliability.

Main Methods:

• Utilized a lightweight deep learning network with a MobileViT backbone for balanced feature extraction.
• Incorporated spatial pyramid pooling with attention mechanisms and multi-scale feature fusion for robust detection.
• Employed a training-free pipeline for angle estimation using feature detection, matching, and consensus fitting.

Main Results:

• Achieved high accuracy in detecting small, low-contrast targets with real-time performance.
• Demonstrated reliable angle estimation with success rates of 85-93% and an average error near one degree.
• Showcased stability under varying conditions, including illumination, texture, blur, and viewpoint changes.

Conclusions:

• The proposed framework offers a practical, intelligent, and deployable solution for bolt loosening detection.
• This technology significantly enhances the safety and operational reliability of large-scale equipment and infrastructure.
• The study highlights the potential of deep learning in critical structural monitoring applications.