Deep model for classification and Detection of Defects in electronic Printed Circuit Boards (PCBs)

Abstract

defect-free production is essential for the function and reliability of electronic devices. Tradi tional manual inspection methods are no longer enough due to the complexity and reduced size of the boards. In the context of the project, deep learning is applied to the automated recog nition of the defects, where YOLOv11 object detection architecture is the algorithm used to f ind the six most frequent defects in the PCB images. The data acquisition is the main stage of the project, and it is followed by the preprocessing, annotation transformation, and data aug mentation, the processes which contribute to increasing the training diversity. The model has been presented, using a carefully selected dataset, as a good candidate to have high precision, recall, and mean Average Precision (mAP) metrics. Additionally, the mobile application was launched with the model proving its feasibility and being ready for server-side inference in real time defect detection. The experimental results validate the prototype system as a solution to the defect detection problem in electronics. The system is also a scalable and practical quality control mean in electronics production besides being accurate and reliable.