Authorization

 

Journal

Volume 22, №3, Special Issue on the Artificial Intelligence; Where Do We Stand?

Detection of resistance spot welding faults in copper materials by transfer learning method (pp.430-445)
H.İ. Seker1, S. Kacar2,*, O. Castillo3, S. Uzun4, İ. Pehlivan2, Z. Tatlı5
https://doi.org/10.30546/1683-6154.22.3.2023.430


1Department of Electrical-Electronics Engineering, Graduate Education Institute, Sakarya University of Applied Sciences, Sakarya, Türkiye

2Department of Electrical-Electronics Engineering, Technology Faculty, Sakarya University of Applied Sciences, Sakarya, Türkiye

3Division of Graduate Studies and Research, Tijuana Institute of Technology, Tijuana 22414, Mexico, e-mail: ocastillo@tectijuana.mx

4Department of Computer Engineering, Technology Faculty, Sakarya University of Applied Sciences, Sakarya, Türkiye

5Department of Metallurgical and Materials Engineering, Technology Faculty, Sakarya University of Applied Sciences, Sakarya, Türkiye

∗Corresponding author’s e-mail: skacar@subu.edu.tr

 

 

Abstract. Welding quality is a crucial factor that significantly influences the performance, quality, and strength of products. Therefore, determining weld quality and detecting defects are vital processes in industrial production. Welding various materials such as metal, sheet metal, aluminum, and steel, as well as employing deep learning-based defect detection and classification, are common practices in the industrial field. However, there is a lack of deep learning-based studies analyzing welding faults in the copper-to-copper material welding process. This study focuses on detecting welding faults in resistance spot welding machines using transfer learning networks. The dataset consists of a total of 2531 images, with 582 images labeled as faulty and 1949 images labeled as faultless. The performance results obtained demonstrate classification accuracies ranging between 89% and 100%. Thus, this study presents a solution to an industrial problem faced by the manufacturing company, utilizing a dataset obtained under real production conditions.

 

Keywords: Transfer Learning, Pretraining Networks, Welding Faults, Classification, Deep Learning, Copper Welding.

 

AMS Subject Classification: 68.

 



COIA 2027

7-9 July 2027
Baku, Azerbaijan

Search