Transparency in Detecting Defects of a Printed Circuit Board: Harnessing XAI for Improved Quality Control in Electronic Manufacturing Industries

In electronic manufacturing industries, defects emerging from Printed Circuit Boards (PCBs) may cause severe malfunctions of the final product. This paper aims to address this issue by proposing the utilization of artificial intelligence (AI) techniques to detect common defects in PCB through the implementation of deep learning (DL) models. The paper utilizes Convolutional Neural Networks (CNNs) and their derived models, in particular, region-based convolutional neural network (R-CNN), Fast R-CNN, and Faster R-CNN. Analysis of the utilization of these techniques in detecting common defects of the PCB, which are classified as missing hole, open circuit, short, spur and spurious circuit are performed in Google Colab platform using Python, with data retrieved from Kaggle. From the results obtained, the best performance parameters for color input for the defect classes are 95.3%, 83.2%, 89.7%, 85.6% and 82.7% respectively. To better understand the results obtained, explainable AI (XAI) techniques, LIME and Grad-CAM were utilized. The implemented technique successfully shows the decision logic for the defect classes through the interpretation images obtained.