Printed Circuit Boards Fault Detection using Deep Learning Techniques

Le résumé fourni par la source

Printed Circuit Boards (PCBs) are key elements in electronic systems, and manufacturing defect detection at an early stage is necessary to make them reliable and minimize downstream failures. Conventional inspection methods like Automated Optical Inspection (AOI) and X-ray imaging fail to provide speed, accuracy, and scalability for current highthroughput environments. This paper discusses the use of deep learning with YOLOv5, YOLOv8, and ResNet50 models to detect faults in PCBs in real-time. These models were trained on a high-resolution PCB defect dataset with typical fault types of mouse bites, open circuits, short circuits, and missing components. Preprocessing consisted of data augmentation, normalization, and structured annotation. Precision, recall, F1-score, and mean Average Precision (mAP) were the evaluation metrics used to ensure model validation. YOLOv8 obtained the best performance of 93.5% mAP and the minimum inference time of 15ms per image, both outperforming the traditional approaches and existing deep learning-based models. The outcome validates the applicability of deploying such models in real-time industrial environments with scalability, cost-effectiveness, and less human involvement. The research makes a contribution to the development of intelligent PCB inspection systems according to Industry 4.0 standards.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.