Enhanced X-Ray Inspection of Solder Joints in SMT Electronics Production using Convolutional Neural Networks

The electronics production is prone to a multitude of possible failures along the production process. Therefore, the manufacturing process of surface-mounted electronics devices (SMD) includes visual quality inspection processes for defect detection. The detection of certain error patterns like solder voids and head in pillow defects require radioscopic inspection. These high-end inspection machines, like the X-ray inspection, rely on static checking routines, programmed manually by the expert user of the machine, to verify the quality. The utilization of the implicit knowledge of domain expert(s), based on soldering guidelines, allows the evaluation of the quality. The distinctive dependence on the individual qualification significantly influences false call rates of the inbuilt computer vision routines. In this contribution, we present a novel framework for the automatic solder joint classification based on Convolutional Neural Networks (CNN), flexibly reclassifying insufficient X-ray inspection results. We utilize existing deep learning network architectures for a region of interest detection on 2D grayscale images. The comparison with product-related meta-data ensures the presence of relevant areas and results in a subsequent classification based on a CNN. Subsequent data augmentation ensures sufficient input features. The results indicate a significant reduction of the false call rate compared to commercial X-ray machines, combined with reduced product-related optimization iterations.