X-ray inspection is widely used in electronics manufacturing for quality assessment of ball grid array (BGA) solder joints, where undetected defects can lead to functional failures and costly rework. Among various defect types, solder bridges are particularly challenging to identify due to their small size, low contrast, and strong visual similarity to surrounding components and package structures in industrial X-ray images. This paper presents a practical deep learning–based inspection framework for automated solder bridge detection in surface-mount technology (SMT) assemblies using real production X-ray data. The proposed approach adopts a coarse-to-fine, two-stage inspection strategy tailored for factory deployment. In the first stage, BGA regions are automatically localized and normalized to suppress background clutter and reduce scale variations across different boards. In the second stage, a lightweight object detection model is applied to identify solder bridge candidates with an explicit emphasis on high recall under limited labeled data. The proposed framework is evaluated on an industrial X-ray dataset collected from a production line. Experimental results on unseen test images demonstrate that the method achieves high defect recall while maintaining stable detection performance, indicating its robustness and suitability for real-world manufacturing inspection scenarios.
Xiran Zhou (Sat,) studied this question.