SDNET 2025- Annotated dataset of steel bolted connection evaluation

Area of Science:

• Structural Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Traditional inspection of steel structures like bridges is labor-intensive, costly, and unsafe.
• Defective bolts compromise structural integrity and can lead to catastrophic failures.
• Existing methods lack sufficient data for AI-driven autonomous bolt defect detection.

Purpose of the Study:

• To present a comprehensive, annotated dataset for training AI models to detect bolt and nut defects in steel structures.
• To address the scarcity of public datasets for automated bolted connection evaluation.
• To facilitate the development of AI for enhanced structural condition assessment.

Main Methods:

• Compilation of 826 annotated images featuring various bolt defects (missing, loose, fixed).
• Data collection through real-world visual inspections of street structures globally and controlled laboratory samples.
• Annotation of images with bounding boxes and masks for object detection algorithms.

Main Results:

• A diverse dataset of 826 images, including 324 for loose bolts, 200 for missing bolts, and 302 for fixed bolts.
• The dataset is suitable for various object detection algorithms due to bounding box and mask annotations.
• The data aims for generalizability through diverse collection methods.

Conclusions:

• The presented dataset is crucial for advancing AI-based automated inspection of steel structures.
• It enables the development of more robust and accurate models for detecting defective bolted connections.
• This resource supports improved safety and efficiency in structural health monitoring.