표제지
목차
요지 12
ABSTRACT 13
제1장 서론 15
1.1. 연구배경 및 목적 15
1.2. 연구동향 18
1.2.1. 연구동향 분석 18
1.2.2. 선행 연구사례의 한계점 및 시사점 도출 20
1.3. 연구 내용 및 범위 22
제2장 Bayesian optimized DNN 기반 FEM 업데이팅 기법 23
2.1. 모델업데이팅 개요 23
2.2. DNN 절차 25
2.2.1. DNN의 기본 구성 25
2.2.2. 하이퍼파라미터 26
2.2.3. 활성화 함수 27
2.2.4. DNN 학습데이터 29
2.3. 베이지안 최적화 절차 30
2.3.1. 대리 모델 32
2.3.2. 획득 함수 33
2.3.3. 베이지안 최적화 목적함수 34
제3장 수치 시뮬레이션을 통한 BODNN 기반 FEM 업데이팅 검증 36
3.1. PSC-Box 교량 모델 검증 36
3.1.1. PSC-Box 교량 수치해석 모델 및 시뮬레이션 36
3.1.2. 해석 Case 및 손상 시나리오 39
3.1.3. BODDN 기반 FEM 업데이팅 검증 48
3.2. I형 거더 교량 모델 검증 55
3.2.1. I형 거더 교량 수치해석 모델 및 시뮬레이션 55
3.2.2. 해석 Case 및 손상 시나리오 58
3.2.3. BODDN 기반 FEM 업데이팅 검증 68
제4장 계측 교량을 통한 BODDN 기반 FEM 업데이팅 수행 75
4.1. 현장 철도 교량 75
4.1.1. 대상 교량 실험 개요 75
4.1.2. 계측 데이터 기반 동적 특성 분석 결과 77
4.2. 초기 수치해석 및 BODDN 기반 FEM 업데이팅 결과 분석 80
4.2.1. 초기 수치해석 결과 80
4.2.2. BODDN 기반 FEM 업데이팅 결과 83
제5장 결론 88
참고문헌 91
Table 3.1.1. Properties of the PSC-Box bridge 36
Table 3.1.2. Damage severity step in PSC-Box bridge 41
Table 3.1.3. Natural frequency of the PSC-Box bridge in Damage Case1 42
Table 3.1.4. Natural frequency of the PSC-Box bridge in Damage Case2 42
Table 3.1.5. Natural frequency of the PSC-Box bridge in Damage Case3 43
Table 3.1.6. Natural frequency of the PSC-Box bridge in Damage Case4 43
Table 3.1.7. 1st Mode shape of the PSC-Box bridge in Damage Cases 46
Table 3.1.8. 2nd Mode shape Bridge of the PSC-Box bridge in Damage Cases 46
Table 3.1.9. 3rd Mode shape of the PSC-Box bridge in Damage Cases 47
Table 3.1.10. Initial DNN structure 49
Table 3.1.11. Bayesian optimized DNN structure 50
Table 3.1.12. PSC-Box bridge updated model stiffness (1 - 3 mode) 51
Table 3.1.13. Difference in correct stiffness of PSC-Box bridge (1-3 mode) 52
Table 3.1.14. PSC-Box bridge updated model stiffness (1 mode) 54
Table 3.1.15. Difference in correct stiffness of PSC-Box bridge (1 mode) 54
Table 3.2.1. Material properties of I-Girder bridge 55
Table 3.2.2. Damage severity step in I-Girder bridge 60
Table 3.2.3. Natural frequency of the I-Girder bridge in Damage Case1 61
Table 3.2.4. Natural frequency of the I-Girder bridge in Damage Case2 61
Table 3.2.5. Natural frequency of the I-Girder bridge in Damage Case3 62
Table 3.2.6. Natural frequency of the I-Girder bridge in Damage Case4 62
Table 3.2.7. 1st Mode shape of the I-Girder bridge in Damage Cases 66
Table 3.2.8. 2nd Mode shape of the I-Girder bridge in Damage Cases 66
Table 3.2.9. 3rd Mode shape of the I-Girder bridge in Damage Cases 67
Table 3.2.10. Stiffness of updated model of the I-Girder bridge (1, 2, 3 mode) 72
Table 3.2.11. Difference in correct stiffness of I-Girder bridge (training data 100EA) 72
Table 3.2.12. Difference in correct stiffness of I-Girder bridge (training data 200EA) 74
Table 4.2.1. Properties of Railroad bridge 80
Table 4.2.2. Error of 1st natural frequency in initial Railroad bridge 82
Table 4.2.3. MAC of 1st mode shape in initial Railroad bridge 82
Table 4.2.4. Stiffness of updated model of the Railroad bridge 84
Table 4.2.5. Parameter updating results of Railroad bridge 84
Table 4.2.6. MAC and error of 1st mode in updated Railroad bridge 84
Table 4.2.7. Max displacement in the initial Railroad bridge 87
Table 4.2.8. Max displacement in the updated Railroad bridge 87
Fig. 1.1.1. Morandi bridge collapsed portion 15
Fig. 1.1.2. Status of bridges in Korea 16
Fig. 1.1.3. Safety index of bridges in Korea 17
Fig. 1.2.1. Process of FEM model updating 21
Fig. 2.1.1. Bayesian optimized DNN-based FEM updating flowchart 24
Fig. 2.2.1. Basic architeture of DNN 25
Fig. 2.2.2. Hyperparameters of DNN 26
Fig. 2.2.3. Activation function 27
Fig. 2.2.4. Grouping the data set 29
Fig. 2.3.1. Bayesian optimization process 31
Fig. 2.3.2. Gaussian process surrogate model 32
Fig. 2.3.3. Expected Improvement acquisition function 33
Fig. 3.1.1. PSC-Box bridge elevation 37
Fig. 3.1.2. Modeling of the PSC-Box bridges using 3D Frame 37
Fig. 3.1.3. Segmented span considering shear span ratio 37
Fig. 3.1.4. Natural frequency and mode shape of the PSC-Box bridge 38
Fig. 3.1.5. Sensor locations in PSC-Box bridge 39
Fig. 3.1.6. Damaged case in PSC-Box bridge 41
Fig. 3.1.7. Natural frequency reduction based on damage location (PSC-Box) 42
Fig. 3.1.8. Trends in PSC-Box bridge's 1st and 2nd mode shape changes 44
Fig. 3.1.9. Trends in the PSC-Box bridge's 3rd mode shape changes 45
Fig. 3.1.10. Training data random sampling process in PSC-Box bridge 48
Fig. 3.1.11. Initial DNN training loss 50
Fig. 3.1.12. Bayesian optimized DNN training loss 50
Fig. 3.2.1. I-Girder bridge design 56
Fig. 3.2.2. Modeling I-Girder bridges using plate-frame element 56
Fig. 3.2.3. Segmented girder 56
Fig. 3.2.4. I-Girder bridge natural frequency and mode shape 57
Fig. 3.2.5. Mode shape extraction location in I-Girder bridge 58
Fig. 3.2.6. Damaged case in I-Girder bridge 60
Fig. 3.2.7. Natural frequency reduction based on damage location (I-Grider) 61
Fig. 3.2.8. Trends in I-Girder bridge's 1st mode shape changes 63
Fig. 3.2.9. Trends in I-Girder bridge's 2nd mode shape changes 64
Fig. 3.2.10. Trends in I-Girder bridge's 3rd mode shape changes 65
Fig. 3.2.11. Training data random sampling process in I-Girder bridge 69
Fig. 3.2.12. Minimum objective value based on the number of training data 69
Fig. 3.2.13. Member stiffness prediction error by damage scenario (training data 100EA) 73
Fig. 3.2.14. Member stiffness prediction error by damage scenario (training data 200EA) 74
Fig. 4.1.1. Overview of Railroad 40m PSC-Box bridge 76
Fig. 4.1.2. Instrumentation installation 76
Fig. 4.1.3. Sensors layout 76
Fig. 4.1.4. Measured data 77
Fig. 4.1.5. FFT of Railroad bridge measured acceleration data 78
Fig. 4.1.6. Measured 1st mode shape 79
Fig. 4.2.1. Modeling Railroad bridge using 3D Frame 81
Fig. 4.2.2. Connection of Railroad bridge 81
Fig. 4.2.3. Train load application 86
Fig. 4.2.4. Triangular pulse load 86
Fig. 4.2.5. Max displacement as a function of train speed in Railroad bridge 87