Fault detection network for X-ray imaging steel cord conveyor belt based on improved YOLOv11

Steel cord conveyor belts are widely used in coal, harbor, and power industries, but harsh environments and high workloads make internal steel cords prone to faults. Undetected faults may lead to serious incidents like belt breakage or longitudinal tearing, threatening production safety. To improve the low accuracy and poor real-time performance of existing methods, this paper proposes WU-YOLO, a fault detection network based on an improved YOLOv11 for X-ray imaging of steel cord conveyor belts. The WTConv module is embedded into the C3k2 structure of the feature fusion module to achieve a near-global receptive field with lightweight parameters, enabling effective capture of local and global features. A P2 detection layer is added to fuse shallow and deep features, improving small target detection. Furthermore, the Unified-IoU loss function is introduced to enhance localization accuracy. The proposed model achieves a balance between speed and accuracy. Experimental results show that WU-YOLO reaches a mAP@0.5 of 99.1% on the self-made X-ray image dataset and 52.5% on the public dataset with similar textures. The detection time is 6.1 ms, demonstrating the model’s capability for real-time, high-accuracy fault detection in steel cord conveyor belts.