Detection of Missing Bolts for Engineering Structures in Natural Environment Using Machine Vision and Deep Learning.

Area of Science:

• Engineering
• Computer Science
• Artificial Intelligence

Background:

• Accurate detection of missing bolts is critical for structural integrity and safety.
• Existing methods may lack efficiency, accuracy, or adaptability to varied conditions.

Purpose of the Study:

• To develop an automated, accurate, and efficient method for detecting missing bolts in engineering structures.
• To leverage machine vision and deep learning for enhanced bolt detection.

Main Methods:

• A comprehensive dataset of bolt images was created for model training.
• Compared deep learning models (YOLOv4, YOLOv5s, YOLOXs), selecting YOLOv5s for bolt target detection.
• Implemented a missing bolt detection approach using perspective transformation and Intersection over Union (IoU).

Main Results:

• The YOLOv5s model achieved high average precisions for bolt heads (0.93) and nuts (0.903).
• The proposed method accurately identified bolt targets (>80% confidence) under diverse conditions (distance, angle, light, resolution).
• Successfully detected missing bolts on a real footbridge structure, even from 1 meter.

Conclusions:

• The developed method offers a low-cost, efficient, and automated solution for monitoring bolted connections.
• The approach demonstrates feasibility and effectiveness in real engineering applications.
• Enhances safety management by providing reliable missing bolt detection.