SDNET 2025- Defective Bolts and Nuts Steel dataset
Description
The dataset focuses on defective steel connections, specifically missing, loose, and corroded bolts and nuts, along with fatigue cracks. It contains over 826 images of defective and intact bolts captured from both real-world steel structures and laboratory setups. The dataset has been annotated with bounding boxes and masks for deep learning applications in automated defect detection.
Files
Steps to reproduce
Data Collection: Images were captured using a Samsung Note 8 camera with a focal length of 8.604 mm. The dataset includes images from North Dakota, Baltimore, Istanbul, Arizona, and Seattle to ensure diversity in environmental conditions. Images were taken from 30 cm to 2 meters away in both natural and artificial lighting conditions. Data Annotation: The Make Sense annotation tool was used to create bounding boxes and polygon annotations. The dataset contains CSV and JSON annotation files. Data Augmentation: To improve model robustness, data augmentation techniques were applied, expanding the dataset to 1,500 images by introducing noise, color variations, and blur. Dataset Organization: The dataset is categorized into four main classes: Fixed Bolts and Nuts Loosened Bolts and Nuts Missing Bolts and Nuts Corroded Bolts and Nuts Each class contains annotated images for deep learning training. Usage and Applications: The dataset is designed for training object detection models such as Faster R-CNN, YOLO, and Mask R-CNN to identify bolt defects. Researchers can use it for infrastructure maintenance, predictive maintenance, and automated inspection tools.