Intelligent Detection of Bolted Connection Damage in Steel Truss Structures Based on LBD-YOLO

<title>Abstract</title> <p>Bolted connections are critical load-transferring joints in steel truss structures. Damage such as loosening, missing, and rusting directly threatens the service safety of various steel infrastructures, including towers, bridges, and crane systems. However, due to complex structural backgrounds, small target sizes, and weak visual features, existing detection methods struggle to simultaneously achieve satisfactory accuracy, inference efficiency, and edge-deployment capability. To address these challenges, we propose LBD‑YOLO, a lightweight bolt damage detection method built upon an improved state‑of‑the‑art object detection framework. We first construct a dedicated dataset of 3,600 images covering four bolt conditions (normal, loosening, missing, and rusted). We then design a collaborative feature enhancement strategy that integrates GSConv and ADown modules to reduce computational redundancy while preserving fine‑grained features, embeds a BoTNet module to enhance global context perception, and adopts a BiFPN structure to optimize multi‑scale feature fusion. Experimental results show that LBD‑YOLO achieves 97.7% mAP@50 and 61.0% mAP@50‑95 on the self‑built dataset. Compared with the baseline, it reduces parameters by 16.1% and computational cost by 19.2%, while demonstrating strong robustness under various complex scenarios. The proposed method provides an effective technical solution for intelligent high‑precision damage detection of bolted connections in steel truss structures.</p>