A PCB soldering joint defect recognition system using convolutional neural network
Document Type
Conference Item
Publication Date
1-1-2021
Abstract
In the recent years, the implementation of artificial intelligence in various industry has increased significantly. This is due to the Industrial Revolution 4.0 (IR4.0) where the industry needs to move towards a smart industry. This paper discusses the development of a model to detect multiple types of soldering defects using the Convolutional Neural Network (CNN) approach. The scopes of this project include developing a database of PCB solder defects. The total images used to train the models were 3121 images including the PCB orientation, good, bridge and missing solder images. We propose YOLOv2 network with the feature extraction of ResNet-50 to train the models to detect the solder joint defects. The accuracy of the models achieved 88.56 for the good solder, 90.47 for the bridge solder and 87.86 for the missing solder, respectively. The effect and relationship of epochs, learning rate, drop rate factor and different angles of datasets to the accuracy of the training model are discussed in this paper. © 2021 IEEE.
Keywords
Convolution, Convolutional neural networks, Deep learning, Defects, Feature extraction, Organic pollutants, Printed circuit boards, Soldering, Convolutional neural network, Deep learning, Defect recognition, Features extraction, Industrial revolutions, Industry needs, Recognition systems, Resnet-50, Solder defects, YOLOv2, Polychlorinated biphenyls
Divisions
sch_ecs
Funders
None
Publication Title
3rd IEEE International Virtual Conference on Innovations in Power and Advanced Computing Technologies, i-PACT 2021
Event Title
3rd IEEE International Virtual Conference on Innovations in Power and Advanced Computing Technologies, i-PACT 2021
Event Location
Virtual, Online
Event Dates
27 November 2021
Event Type
conference