|
CSME 2020/08
Volume 41 No.4
:
401-407
|
|
Detection and Classification of Printed Circuit Board Assembly Defects Based on Deep Learning
Ruifang Yea, Chia-Sheng Panb, Pei-Yuan Hungb, Ming Changb and Kuan-Yu Chenb
aCollege of Mechanical Engineering and Automation, Huaqiao University, Xiamen 361021, Fujian, China.
bDepartment of Mechanical Engineering, Chung Yuan Christian University, Chungli 32023, Taiwan, ROC.
|
Abstract:
In recent years, the booming artificial intelligence technology has been gradually introduced into the defect inspection system with optical images in various industrial production lines, which has improved the precision and yield of products. However, for test images with multiple defects, the fault detection rate is still very high. In addition, the need to identify the processing lack through defect classification is also increasing. In order to enhance the accuracy and intelligence of using optical images to detect and classify the compound defects on printed circuit board assembly (PCBA), this study compared the performance of several deep learning models in dealing with multi-defect images. Through the comparison of test performance, it is suggested to use YOLOv3 model to overcome the challenges of diversity and complexity of PCBA components. Based on YOLOv3, 800 images randomly containing 10 kinds of PCBA defects were trained. Each sample contains an unequal number of defects. The training results show that the mean average precision (mAP) in defect classification is as high as 97.47%. In the test experiment, 60 sample images were inspected and compared with the results of manual inspection. Experimental results show that the error rates of PCBA defect detection and classification are as low as 0% and 2.42% respectively, which indicates that the optimized YOLOv3 model can be applied to industrial production lines to achieve the goal of high-precision detection and classification of composite defects.
|
|
Keywords: printed circuit board assembly, defect detection, defect classification, deep learning.
|
Download PDF
|
|
*Corresponding author; e-mail:
|
©
2020
CSME , ISSN 0257-9731
|