Intelligent Print Circuit Board Defect Detection

Abstract

During the manufacturing process of electronic products, one of the most critical stages is the soldering of components onto the printed circuit board (PCB). Even a minor defect in the PCB can lead to significant issues in the final product. Therefore, a rigorous\ndefect inspection process is employed during manufacturing, which can be categorized into manual inspection and automated optical inspection (AOI). Manual inspection suffers from drawbacks such as slow speed and the expenditure of manpower and costs. Hence, most manufacturers opt for automated optical inspection to expedite production. However, most current automated optical inspection systems rely on traditional optical inspection algorithms. These computational methods are susceptible to variations in lighting conditions caused by slight differences in the placement of PCBs or the amount of solder. Consequently, these variations often result in misjudgments, where qualified PCBs are mistakenly categorized as defective, leading to high false positive rates in AOI systems.\nThis paper presents a deep learning based method for PCB defect detection. The proposed approach involves the creation of two separate models for classifying defective components individually. Once both models demonstrate a basic recognition capability, they are combined into a master model using the method proposed in this study to enhance overall recognition accuracy. The model has\nbeen trained on datasets for capacitors and resistors, and the experimental results indicate an accuracy of over 99% for both component types.