Detection and Classification of Faults in Printed Circuit Boards Using Hybrid Deep Learning Model

Printed circuit boards (PCB) are very compilable and has the ability to rework, therefore, it can be applied for different applications and this enables the electronic information equipment develop in new paths. But, identification of defects can be difficult in PCBs and their manufacturing and operation. This paper provides a research based on a better defect detection method in PCBs manufacturing via the combination of the deep learning. Contrary to the existing methods the proposed hybrid deep learning model Principal Component Analysis- Bi-Long Short Memory Network (PCA-Bi-LSTM) capitalizes on Speeded Up Robust Features (SURF) extraction as a robust feature from the defective PCB images and subsequently Principal Component Analysis (PCA) for dimensions' reduction. Secondly, after all the extracted and selected feature vectors are fed into Bi-Long Short Memory Network (Bi-LSTM), it discovers from its hidden layers that defect types can be mouse bite, spurious copper, short, spur, missing hole, and open circuit. The experiment result and the performance provide superior ability in PCB defect detection that has a high accuracy of 99.86% and efficiency, when compared to existing Improved You Only Live Once (YOLOv7) and Improved Fully Convoluted Network (FCNN). From the proposed approach, one can find whether the product is a promising one that provide the enhancing PCB production quality process.