| Autor: |
Момот, А. С., Крецул, В. В., Муравйов, О. В., Галаган, Р. М. |
| Zdroj: |
Technical Diagnostics & Nondestructive Testing / Tekhnicheskaya Diagnostika I Nerazrushayushchiy Kontrol; 2024, Issue 2, p17-24, 8p |
| Abstrakt: |
In the paper, we consider the possibilities of applying the YOLOv5s deep learning model to the task of automating the process of detecting surface defects on printed circuit boards. Modern printed circuit boards are manufactured in large volumes and contain a signifi cant number of elements. The manufacturing process of printed circuit boards is complex, which increases the likelihood of board wiring defects, such as short, open circuits, mouse bites, etc. These defects are superfi cial and can be detected by visual and optical inspection. Compared to other methods, the visual-optical inspection is easier to automate. It is proven that it is promising to use deep learning models to automate the process of detecting objects in images. Modern neural networks can automatically detect surface defects in printed circuit board images with high reliability. The paper considers the YOLO class of models. It is established that the YOLOv5 model has better performance and recognition accuracy than previous modifi cations. In this study, the YOLOv5s model was implemented and trained to test the eff ectiveness of this network in the task of automated detection of surface defects on printed circuit boards. The open «PCB Defects» dataset was used for training. A qualitative and quantitative analysis of the performance of the trained network on a test dataset was carried out. It was found that the network can detect surface defects of printed circuit boards with 92.5 % reliability in terms of mAP50. Additionally, the results of the recognition of diff erent classes of defects are analyzed and recommendations for further improvement of the system are given. In particular, it is promising to apply augmentation of training data and use a more complex architecture of the deep learning model. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
