Automated Detection and Classification of Defects in PCB Using Deep Learning Techniques

Printed Circuit Boards (PCBs) are crucial in daily electronics. In 2018, the global single-sided PCB market was projected to reach $79 billion by 2024, directly impacting electronic device performance. Rigorous inspection of bare PCBs is essential to reduce manufacturing costs related to defects. Visual defect inspection, often done manually, is challenging, time-consuming, and has a low error tolerance. Emerging techniques like CNNs can aid in this process. The challenge here compared to other object detection tasks is that the defects was very small in size which makes detection little hard.
We compared different DL architectures and found that YOLO demonstrated better accuracy and performance, particularly in detecting small defects like mouse bites and open circuits compared to other models. Excellent work done by my student Dhruv Patel.