Publication: Clustering solder joint of gullwing on x-ray images using deep convolutional neural network
Loading...
Date
2020-07-01
Authors
Ravi Kumar, Kupenthyran
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Soldering of electronics component on a circuit board can sometimes cause defect on the component’s joint. This defect can be identified either manually through inspection from
a human or automatically by using a computer program. Although the human eye is a sophisticated tool, for some cases false inspection problem could occur due to human fatigue. Hence, automated classification of solder joint using its X-ray image could be effective for assisting the process of detecting defects in the joint. This project proposes to implement
deep convolutional neural network (DCNN) to classify solder joint of gullwing on X-ray images to overcome the weaknesses of the manual classification method. This research will further address the advantages of automating the classification process rather than using the previous manual method. In this research, the solder joint X-ray image dataset is provided by ViTrox Technologies along with JavaScript Object Notation (JSON) file which contains information of the image dataset such as the Region of Interest (ROI) of joint in each image and the labels of each joint i.e. good or bad. Each joint is labelled manually by human. Then, the dataset is separated for training and testing processes. Finally, a DCNN model is trained and optimized to classify solder joints. The results show that the developed model has a testing accuracy of 99.87%. Furthermore, the time taken to classify a single joint image averages to 2.4 milliseconds which outperforms the manual classification method in term of speed. Hence, the proposed method provides a reliable solution for the development of automated defect detection in solder joint of gullwing.