A Deep Learning System for PCB Trace Defect Detection: Implementation and Evaluation

Abstract

This paper outlines the design, implementation, and assessment of an automated anomaly detection system for Printed Circuit Board (PCB) production lines utilizing the YOLOv11x deep learning model. The primary aim is to surmount the constraints of manual inspection, which is susceptible to human mistake and inefficiency, by creating a system proficient in properly detecting six prevalent forms of anomalies: Short Circuit, Open Circuit, Spur, Spurious Copper, Mouse Bite, and Missing Hole. The research technique encompasses dataset preparation via data augmentation, system architecture design, training of the YOLOv11x model for 150 epochs, and performance evaluation employing standard quantitative metrics. The testing findings indicate that the system attains state-of-the-art performance, achieving a mean [email protected] of 97.6% and an exceptional inference speed of 140.8 FPS. Subsequent investigation indicates that maximal detection efficacy is attained in the Missing Hole and Short Circuit categories, but the primary difficulty resides in identifying the Open Circuit class due to its nuanced visual characteristics. The findings validate that the YOLOv11x model provides an exceptional equilibrium of accuracy, speed, and efficiency, rendering it a highly effective and promising alternative for enhancing automated quality control operations in the electronics manufacturing sector.