|
CSME 2025/04
Volume 46 No.2
:
205-210
|
|
Automated Detection of Arc Welding Defects by Using a Convolutional Neural Network Model
Shu-Hsien Huang a, Ting-En Wu b and Chia-Hung Lai c
aDepartment of Information Management, National Chin-Yi University of Technology, No.57, Sec. 2, Zhongshan Rd., Taiping Dist., Taichung 411030, Taiwan (R.O.C.)
bDepartment of Industrial Education and Technology, National Changhua University of Education, No.2, Shi-Da Road, Changhua City50074, Taiwan (R.O.C.)
cDepartment of Intelligent Automation Engineering, National Chin-Yi University of Technology, No.57, Sec. 2, Zhongshan Rd., Taiping Dist., Taichung 411030, Taiwan (R.O.C.)
|
Abstract:
Arc welding has low costs and offers high welding efficiency. Visual inspection, the easiest welding inspection method, is prone to human error. This research proposes a convolutional neural network (CNN)-based method for classifying welding defects into the following categories: ‘no defect’, ‘blow hole’, ‘lack of fusion’, ‘incompletely filled groove’, and ‘undercut’. The weld bead was positioned under a camera, and the captured images were transmitted to a computer, which recognised defects in the images. After recognition, the computer saved the images and added them to a defect detection dataset to ensure that the dataset was continuously updated. The proposed defect recognition model achieved an accuracy as high as 97.2%. A comparison was conducted for different numbers of images and training iterations. We recommend collecting a minimum of 800 images when a CNN model is to be trained from scratch to detect welding defects.
|
|
Keywords: Arc Welding, Defect Detection, Convolutional Neural Network (CNN), Welding Quality Control
|
|
|
*Corresponding author; e-mail: chlai@gm.ncut.edu.tw
|
©
2025
CSME , ISSN 0257-9731
|