Design and Realization of an Automatic Optical Inspection System for PCB Solder Joints

Abstract

Recent developments in electronics has led to an increase in fabrication and assembly speeds of printed circuit boards (PCBs). In addition, the size of manufactured PCBs and electronic components (e.g. resistors, capacitors, transistors etc.) are becoming much smaller. By increased demand and production speed, the reliability and thus the inspection of manufactured PCB assemblies became an important issue. In the assembly PCB production process, detection of surface mount technology (SMT) solder defects is made with automatic optical inspection (AOI) devices using image processing methods. Besides these expensive device methods, the method by which the controls are made by the operators visually is a low-cost solution used by most of the companies that manufacture printed circuit board assembly. In addition, circuit defects and solder defects cannot be tested and controlled with 100% accuracy due to human error. This paper proposes to detect solder joint defects with machine learning methods using YOLO algorithm to speed up time and increase accuracy in assembly PCB production line. Approximately 40000 images were obtained from the real production line before training with the YOLOv4 algorithm for high accuracy rate. Detection of solder defects of SMT circuit elements in approximately 5K (4056x3040) images resolution can be achieved with 97% accuracy in around 4 seconds. As a result of the use of the system, this proposed method has been proven with the reports received from the production line and precision-recall curves. Thus, it has been observed that the production speed and accuracy rate are increased.