Deep Learning Based Printed Circuit Boards Defect Detection Using Multiple Depth 2D X-Ray Image

The miniaturization and increased complexity of components in printed circuit board assemblies present significant challenges in product design and manufacturing. Automated inspection techniques have been developed to enhance efficiency and reduce reliance on skilled labor. However, certain components with hidden solder joints or bubble defects require specialized inspection methods. X-ray imaging has emerged as a valuable tool for detecting defects like bridging, insufficient soldering, and solder voids. Although algorithm parameter fine-tuning in X-ray machines has improved defect detection accuracy, false calls remain problematic, necessitating expert verification. This study proposes a novel approach that combines deep learning and multi-depth X-ray inspection to enhance defect detection accuracy across various defect types. The YOLO object detection model is employed to identify defects in Automatic X-ray Inspection machines. The dataset is prepared using advanced segmentation techniques, with individual solder joints serving as the dataset. Furthermore, image augmentation plays a vital role due to the limited number of examples of defective solder joints. Precise defect detection is achieved through accurate labeling, including optimizing the label box size. This comprehensive approach offers promising potential for automated and reliable defect detection in complex electronic assemblies.