Date of Award
6-8-2026
Document Type
Thesis
Publisher
Santa Clara : Santa Clara University, 2026
Departments
Electrical Engineering; Electrical and Computer Engineering; Mechanical Engineering
First Advisor
Fatemeh Davoudi-Kakhki
Abstract
Printed circuit boards (PCBs) are foundational components in modern electronic systems and play a critical role in consumer electronics, industrial automation, power systems, and emerging artificial intelligence hardware. As PCB designs continue to increase in complexity and component density, reliable and scalable defect inspection becomes increasingly important to ensure product quality, manufacturing efficiency, and operational safety. Traditional manual inspection and conventional automated optical inspection (AOI) methods can be limited by human error, inspection fatigue, rigid rule-based configurations, and reduced adaptability to varying board layouts and defect characteristics. This project presents a computer-vision assisted robotic workcell for automated PCB defect detection and sorting. The proposed system integrates a Universal Robots UR3e collaborative robot, a Robotiq 2F-140 gripper, a close-range industrial microscope camera, and a YOLOv8-based deep learning object detection framework to perform end-to-end PCB inspection. A custom dataset was fabricated using 60 physical PCBs containing representative manufacturing defects, including excessive soldering, lifted components (tombstoning), missing components, and surface scratches. A total of 300 raw images were collected under five lighting conditions and expanded through augmentation to generate more than 2,000 training images for model development. Multiple YOLOv8 architectures were evaluated, with a YOLOv8m model selected as the final implementation due to its favorable balance between detection performance and computational efficiency. Experimental results demonstrated substantial improvements over the initial baseline model. The final model achieved average class accuracies of 96.5% for excessive solder defects, 84.0% for lifted-component defects, 91.0% for missing-component defects, and 93.3% for surface-scratch defects. Recall values exceeded 0.90 at practical confidence thresholds, reducing the likelihood of missed defects and supporting quality-critical manufacturing applications. The integrated robotic system successfully executed a complete pick–inspect–sort workflow, automatically classifying boards as defect-free, defective, or requiring further inspection based on computer vision predictions. The results demonstrate the feasibility of combining collaborative robotics and deep learning-based computer vision into a flexible laboratory-scale automated inspection platform. The proposed workcell provides a practical proof-of-concept for intelligent quality inspection and highlights the potential of robotic vision systems for adaptive manufacturing environments and future smart factory applications.
Recommended Citation
Li, Shangyuan and Murugasu, Elise, "Computer-Vision Assisted Robotic Workcell for PCB Defect Detection" (2026). Interdisciplinary Design Senior Theses. 103.
https://scholarcommons.scu.edu/idp_senior/103
