Optimizing Automated Optical Inspection for Printed Circuit Boards Using Computer Vision: A Comparative Study for Beneficial Imagesize Reduction

The increasing digitalization makes printed circuit boards (PCBs) an almost indispensable component as a flexible internal interconnection technology in electronic devices. To ensure reliable operation in the final product, the quality of each manufacturing step must be closely monitored. Quality control can be carried out through visual inspection by trained personnel or via automated inspection systems. Since the range of defects that can occur during the manufacturing process is highly diverse and customer requirements for the boards vary, this presents challenging conditions for automated optical inspection (AOI). Several studies indicate that the effective image area, which meets the resolution requirements for evaluation, as well as the optical limitations of the camera lens and sensor, is around 15 to 20 cm². As a result, multiple images must be taken with a moving inspection head to fully assess PCBs, which typically have a larger surface area. This significantly limits the integrability into a continuous manufacturing process. Furthermore, real-time evaluation necessitates that the detection of defects be synchronised with production speed, which is also influenced by image size. Consequently, a compromise must be reached between image resolution and the level of detail captured in the area to ensure the reliability of the analysis of defects on the PCB. As part of the present study, images of various PCBs were generated, which were tinned using the hot-air leveling process. The evaluation is based on an One-stage detection model, which was trained with PCB images at different resolutions and identifies the regions relevant for defect detection. This comprises a comparative study conducted using diverse image preprocessing techniques, with the objective of optimising the evaluation speed and accuracy. The long-term objective is to implement the model for practical use, thereby enabling its use as an automated inspection method in continuous manufacturing processes.

Keywords: Printed Circuit Boards (PCB), Automated Optical Inspection (AOI), Computer vision, Machine learning, Instance segmentation, YOLO, Robust manufacturing process.