International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 11 Issue: 05 | May 2024
www.irjet.net
p-ISSN: 2395-0072
Machine Learning-Based Defect Detection for Printed Circuit Board Einesh Naik1, Prof. Nadine Dias2 1Student, Department of Information Technology and Engineering, Goa College of Engineering, Farmagudi, Goa,
India
2Assistant Professor, Department of Information Technology and Engineering, Goa College of Engineering,
Farmagudi, Goa, India ---------------------------------------------------------------------***--------------------------------------------------------------------Printing boards, arranging components, and soldering are all Abstract - This work presents a deep learning system that
included in PCB production. During these phases, there is a possibility of several potential flaws, such as incorrect solder joins, short circuits, empty or open circuits, and excess solder (which appears as overflowing solder on solder points). These flaws could jeopardize the board's stability or perhaps cause the board to break entirely. Thus, it is imperative to implement a highly effective and precise automatic detection module for the purpose of examining various flaws during the PCB manufacturing process. Figure 2 illustrates examples of surface defects in PCBs.
uses the principle of the You Only Look Once (YOLO) methodology to perform PCB (printed circuit board) quality inspections. Deep learning algorithms have been widely used in many different fields because of their remarkable efficiency and accuracy. Comparably, there has been a lot of interest in the accurate detection of PCB flaws using deep learning techniques, such as the You Only Look Once (YOLO) method. The synthetic dataset from Kaggle was used in the suggested strategy. 1386 pictures representing 6 PCB flaws make up the dataset. Defects include missing holes, mouse bites, open circuits, short circuits, spurious copper, and shorts are present in the dataset. A YOLOv8X model is then trained using the data to identify PCB flaws. With a batch size of 16, the suggested model successfully identified defects in PCBs with 97.9% accuracy. Key Words: YOLO; deep learning; printed circuit board; printed wiring board; missing hole; mouse bite; open circuit; short, spur; spurious copper
1.INTRODUCTION An essential part of electrical devices is the printed circuit board (PCB), sometimes known as the printed wiring board (PWB). PCBs are used in a vast range of electronic products, such as satellites, communication devices, laptops, computers, cellphones, military weapons, and electronic watches. The size of electronic device components has decreased dramatically due to advances in integrated circuit and semiconductor technologies. As a result, the PCBs that hold these parts together have grown delicate and complex. Therefore, in order to satisfy client requests, it is essential to ensure high-quality production. An illustration of a PCB is provided in Figure 1.
Fig -2: Defects in Printed Circuit Board
2.Related works Two datasets were used to train the VGG16 model, which uses both transfer learning and an unsupervised deep learning technique. Four defect categories were identified by this model: abrasion, damaged PCB edge, missing washer/extra hole, and scratches. 60 PCBs chosen at random were tested, and the results showed that the PCB-G dataset had an accuracy of 87.49% while the PCB-1 dataset had an accuracy of 74.12%. The reduced dataset size for PCB-1 is thought to have contributed to the lower accuracy, which highlights the difficulties and costs involved in gathering large numbers of faulty samples [1]. An upgraded convolutional neural network that made use of the MobileNet-v2 model was used. Four types of defects were successfully identified by this model: mouse bite, open, short, and spurs. It accomplished an impressive 92.86% total accuracy. In particular, it achieved 93.33% for spurs, 94.29% for shorts, 98.57% for open, and 88.33% for mouse bites [2].
Fig -1: Printed Circuit Board
© 2024, IRJET
|
Impact Factor value: 8.226
|
ISO 9001:2008 Certified Journal
|
Page 513