SDNET 2025- Annotated dataset of steel bolted connection evaluation

Steel structures, particularly bridges and high mast poles, must be carefully inspected for their connection integrity, mmissing connection parts (missing or loose bolts and nuts), holes, or other defects negatively impact the structure's ability to safely carry loads, potentially leading to catastrophic failures on transportation infrastructure. Traditional inspection methods are time-consuming, labour-intensive, unsafe, and costly-especially for tall structures. The use of Unmanned Aerial Systems (UAS) for visual imaging, combined with object detection, to identify defective bolts in steel structures has recently been implemented for structural condition assessment. However, there is a lack of publicly available datasets for this type of defect detection and model development, limiting adaptation of AI for autonomous bolted connection evaluation. This paper presents a dataset that includes multiple types of bolt and nut defects, such as missing bolts, loosening bolt.The dataset included 324 annotated images for loosen bolt, 200 images with missing bolt, and 302 images with fixed bolt. The dataset was compiled to meet the need for automated condition assessment of defective bolted connections in steel structures using deep learning techniques. Therefore, the authors aimed to gather as much data as possible to create a diverse set of 826 images, capturing various types of defects. To make the data more generalizable, it was collected through visual inspections of street structures from multiple cities in the world. Additionally, to increase the dataset size, laboratory data were created with bolts exhibiting various defects, and images were captured from these samples. All images in this dataset are annotated with bounding boxes and masks, making them suitable for any type of object detection algorithm.