Optimized AI-Based Real-Time Fault Detection in Printed Circuit Boards

Printed Circuit Boards (PCBs) are the backbone of contemporary electronic devices, where even slight defects can cause disastrous functional failures. In this paper, we introduce an effective and scalable defect identification system based on the YOLOv11n (You Only Look Once - version 11n) object detection model. It is trained on a synthetic dataset with six different types of PCB defects: missing holes, open circuits, mouse bites, shorts, spurs, and spurious copper traces. Our approach consists of thorough preprocessing and annotation parsing with Pascal VOC format followed by the implementation of a compact neural architecture optimized for accuracy as well as inference speed. To support real-world deployment, the model is transformed into various formats (ONNX, TFLite) with the use of a set of model optimization tools, enabling deployment on edge devices like mobile phones, GPUs, and embedded systems. Experimental results show that the introduced framework supports high detection accuracy at significantly low computational overhead, which makes it perfectly suitable for real-time industrial inspection systems. This paper showcases the feasibility of using small deep learning models within fault detection pipelines, particularly in power- or computationally constrained environments.