End-to-end deep learning framework for printed circuit board manufacturing defect classification.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Manufacturing Technology

Background:

• Quality control in Printed Circuit Board (PCB) manufacturing is crucial for product reliability.
• Current defect detection methods can be time-consuming and may not achieve optimal accuracy.
• Automating PCB defect detection is essential for increasing production efficiency.

Purpose of the Study:

• To develop and evaluate a novel deep-learning framework for rapid and accurate detection and classification of PCB manufacturing defects.
• To compare the performance of the proposed model against existing state-of-the-art methods.

Main Methods:

• A complete deep-learning framework utilizing a single-step object detection model was designed.
• The model architecture was described and benchmarked against Faster Region Based Convolutional Neural Network (FRCNN), RetinaNet, and You-Only-Look-Once (YOLO).
• Performance was evaluated on a standard PCB defect dataset using both low- and high-resolution images.

Main Results:

• The proposed model achieved a mean average precision (mAP) of 98.1% on low-resolution images.
• It outperformed state-of-the-art models, showing a 3.2% improvement over YOLO V5m (low-resolution) and 1.4% over FRCNN-ResNet FPN (high-resolution).
• The model requires significantly fewer parameters (7.02M) compared to FRCNN-ResNet FPN (23.59M) and YOLO V5m (20.08M).

Conclusions:

• The developed deep-learning framework offers superior accuracy and efficiency for PCB defect detection.
• Implementation in manufacturing lines can significantly enhance production yield and throughput.
• The model's reduced parameter count suggests a more efficient and scalable solution for automated quality control.