Modern Quality Control: Integrating Computer Vision in Inspection of PCB Elements

This study introduces a computer vision application for the automated detection, localization, and inspection of microcontroller boards and Single Board Computers, generally referred to as Printed Circuit Boards (PCBs), as part of the quality control process. Given the key role of quality control in product pricing for systems with automated and robotized production, there is an imperative to enhance its speed and cost-effectiveness. To address this, we propose the utilization of computer vision tools, particularly deep neural networks. The proposed system encompasses several modules: PCB detection and component verification using YOLOv8 convolutional neural network (CNN), targeted image extraction of components, solder joint quality assessment via adaptive thresholding and optical character recognition (OCR) on selected images. The overreaching goal is to devise a practical industry application of computer vision for adaptable quality control. This system's development involved PCB dataset creation, neural network training, and the development of final implementation. The trained YOLOv8 neural network achieves Mean Average Precision (mAP) of 99.5%, the proposed inspection system scores 96.1% precision and 97.8% recall. The OCR model scores 95.6% true positive character predictions.