SPR 접합 품질 분류를 위한 CNN 기반의 딥러닝에 관한 연구Study of Convolution Neural Network Based Deep Learning to Classify the Quality of Self-Piercing Riveting Joint
- Other Titles
- Study of Convolution Neural Network Based Deep Learning to Classify the Quality of Self-Piercing Riveting Joint
- Authors
- 김민규; 이태현; 이승환; 김철희; 감동혁
- Issue Date
- Dec-2022
- Publisher
- 대한용접접합학회
- Keywords
- Convolution Neural Network(CNN); Classification; Self-Piercing Rivet(SPR); Deep learning; Abnormal process conditions
- Citation
- 대한용접접합학회지, v.40, no.6, pp.502 - 511
- Indexed
- KCI
- Journal Title
- 대한용접접합학회지
- Volume
- 40
- Number
- 6
- Start Page
- 502
- End Page
- 511
- URI
- https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/185977
- DOI
- 10.5781/JWJ.2022.40.6.6
- ISSN
- 2466-2232
- Abstract
- The SPR(Self-Piercing Riveting) process is a mechanical joining process that is mainly applied to assembling multi- material parts to reduce the weight of the car body. Because the quality of SPR joints is mainly evaluated through cross sectional inspection, which is a type of destructive inspection, it is expensive and time-consuming. Machine learning technology is being proposed as a non-destructive testing because it can predict the quality based on the signals generated during the process. However, research result on the quality prediction modeling of SPR joints have not yet been reported. In this study, the prediction accuracy according to the signal combination was compared and evaluated by applying the CNN algorithm using the displacement and load signals generated during the SPR process and the acoustic signal obtained from the acoustic sensor. The materials used in the experiment were SGAFC 1180Y, CFRP, and SPFC 590 and the thickness were 1.4 mm, 1.8 mm, and 1.0 m respectively and a three-layer SPR process was performed. After evaluating joining was performed by selecting the abnormal process conditions, with 12 con- ditions that may occur during the process. Consequently, in the case of the first model in which the CNN algorithm was based on displacement and load signals, the quality prediction accuracy was estimated to be 90.0%. In the case of the second model in which the CNN algorithm added acoustic signals to the displacement and load signals, the quality prediction accuracy was estimated to be 77.5%.
- Files in This Item
-
Go to Link
- Appears in
Collections - 서울 공과대학 > 서울 기계공학부 > 1. Journal Articles
![qrcode](https://api.qrserver.com/v1/create-qr-code/?size=55x55&data=https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/185977)
Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.