Edge-Based Deep Learning System for Automated PCB Component Inspection

Automatic Optical Inspection (AOI) is essential in Switched Mode Power Supply (SMPS) PCB manufacturing, where missing components can lead to functional failure and safety risks. This paper presents a low-cost embedded deep learning-based AOI system for real-time component verification using a Raspberry Pi 5 as the edge computing platform and a YOLOv11 object detection model to identify critical components such as the fuse, capacitor, MOV, and optocoupler. A dataset of approximately 500 PCB images was collected under varying lighting and orientation conditions and annotated for supervised training. The trained model was optimized and deployed in NCNN format to enable efficient inference on resource-constrained hardware. The complete system integrates conveyor-based PCB handling, Region of Interest (ROI) extraction, and a Finite State Machine (FSM) to ensure stable and synchronized operation. Operating at a 640 x 640 resolution, the system achieves 5-10 FPS with an average inspection time of approximately 3 seconds per PCB. Based on detection results, automatic PASS/FAIL classification and actuator-based sorting are performed.