CNN 딥러닝을 이용한 셀프 피어싱 리벳팅의 비파괴 품질 모니터링에 관한 연구 : On-Machine 시스템으로 작동하는
A Study on On-Machine Non-Destructive Quality Monitoring of Self-Piercing Riveting Using CNN Deep Learning
- 주제어 (키워드) 셀프 피어싱 리벳팅(SPR) , 합성곱 신경망(CNN) , 다중 클래스 분류 , 하중 곡선 분석 , 온-머신 품질 예측 , 사용자 인터페이스(GUI) 모니터링 , 자동화 시스템
- 발행기관 서강대학교 일반대학원
- 지도교수 김동철
- 발행년도 2025
- 학위수여년월 2025. 8
- 학위명 석사
- 학과 및 전공 일반대학원 기계공학과
- 실제 URI http://www.dcollection.net/handler/sogang/000000082133
- UCI I804:11029-000000082133
- 본문언어 영어
- 저작권 서강대학교 논문은 저작권 보호를 받습니다.
목차
1. Introduction 1
1.1 Background 1
1.2 Previous research 6
1.3 Research purpose 10
2. Experimental method 11
3. Deep Learning model selection 15
3.1 Deep Neural Network 16
3.1.1 Deep Neural Network input data 17
3.1.2 Deep Neural Network model design 22
3.1.3 Development of On-Machine system based on Deep Neural Network 24
3.1.4 Results of applying Deep Neural Network On-Machine system 26
3.2 Convolution Neural Network 28
3.2.1 Material combination #1: Defining error mode and setting defect criteria 30
3.2.2 Convolution Neural Network input data #1 35
3.2.3 Material combination #2: Defining error mode and setting defect criteria 42
3.2.4 Convolution Neural Network input data #2 48
3.2.5 Convolution Neural Network model design 53
4. Improving classification accuracy of Convolution Neural Network 61
4.1 Optimization of Hyperparameter for enhanced classification accuracy 62
4.1.1 Epoch optimization 62
4.1.2 Batch size optimization 64
4.1.3 Learning rate optimization 65
4.1.4 Final structure of the CNN model 67
4.2 Selecting the input data learning range to improve classification accuracy 68
4.2.1 Alignment of load curve start time for joining 69
4.2.2 Load curve analysis for feature segmentation in SPR 70
4.2.3 Model reliability verification using K-Fold cross-validation under limited data 72
5. Results and discussion 73
5.1 Development of CNN-based On-Machine quality prediction algorithm 73
5.1.1 On-Machine quality prediction algorithm flow for CNN-based real-time prediction 74
5.1.2 Development of GUI for On-Machine SPR monitoring and quality prediction 77
5.2 Material #1 & #2 error mode CNN classification results 78
5.3 On-site component unit testing 80
5.3.1 Field equipment optimization 80
5.3.2 Limited field custom error mode selection 81
5.3.3 Field evaluation using aluminum door parts 82
6. Conclusion 85
Reference 88
