Automatic Soldering Defect Detection for Tantalum Capacitors in Printed Circuit Boards

The automatic detection system proposed in this paper mainly focuses on automatic welding errors detection for tantalum capacitors. Without manual labeling standard PCBs, it can automatically determine the positions and directions of the tantalum capacitors in PCBs and implement automatic labeling by means of segmentation, coarse-fine positioning, direction detection. In the test, the positions and directions of the tantalum capacitors in the PCBs under test can be automatically detected and matched with those in the corresponding standard PCBs, further finding the tantalum capacitors those are missing or in wrong directions in the PCBs under test. On one hand, this method does not need a lot of experienced workers to detect the soldering defects of tantalum capacitors, reducing the probability of defective PCBs being categorized as non-defective due to workers' fatigue or carelessness. It reduces economic cost a lot. On the other hand, unlike deep learning algorithms, it does not need a lot of training examples and performs well on a lot of PCBs whose sizes, shapes and densities of components are unknown before. This improves the productivity and ensures rapid PCB images collection and rapid detection of welding errors. The application shows that the average speed can reach 15 frames per second, and the accuracy is greater than 98%.