Automated Industrial Surface Defect Detection on Bolts Using Deep Learning

English (OpenAlex)

Defects in the head region of bolts used as fasteners directly affect load transfer, the proper transmission of tightening torque, and the safety of on-site maintenance and repair operations. In mass production lines, monitoring such defects primarily through operator inspection and infrequent sampling is both prone to human error and insufficient for highvolume manufacturing. In this study, a computer vision and deep learning-based inspection system capable of automatically monitoring bolt head defects in a production environment is designed and evaluated. The proposed prototype is based on capturing bolt head images as they pass over a rotating platform under a fixed-position monochrome industrial camera and homogeneous LED illumination. An adaptive region of interest (ROI) is defined to process only the head region, thereby suppressing background noise and automating the data acquisition and labeling process. A DETR (DEtection TRansformer) model with a ResNet-50 backbone is adopted as the baseline method on the dataset labeled as defective and nondefective, and its performance is compared with YOLOv8x, YOLO11x, and YOLO12x architectures under identical training settings. Experimental results indicate that YOLObased models achieve balanced and high performance in terms of mAP@0.5 and mAP@0.5:0.95, while the DETR-ResNet50 model, in addition to comparable accuracy levels, significantly reduces the false negative rate by decreasing the number of missed defective bolts.