Title Page
Contents
ABSTRACT 10
Ⅰ. Introduction 12
1.1. Background 12
1.2. Problem statement 13
1.3. Objective and scope of the study 14
1.4. Research significance 14
1.5. Organization of the dissertation 15
Ⅱ. Literature Review 17
2.1. Background 17
2.2. Finite Element Modeling of RCB 17
2.3. Metamodeling and Bayesian Additive Regression Trees (BART) 20
2.4. Seismic performance of RCB structures 21
2.5. Sensitivity analysis of shear wall structures 22
2.6. Sensitivity analysis using metamodel. 23
2.7. Concluding Remarks 24
Ⅲ. Methodology 25
3.1. Background 25
3.2. Development of FE-model 25
3.3. EDP selection 29
3.4. Uncertain variables 29
3.5. Development of BART model 32
3.5.1. Methodology 32
3.5.2. Verification of BART method 40
3.6. Sensitivity analysis 45
3.7. Concluding Remarks 47
Ⅳ. Seismic Damage Sensitivity of the Containment Building 48
4.1. Background 48
4.2. Description of the Model and input parameters 48
4.3. Pushover analysis and EDP estimation 49
4.4. Dataset preparation 51
4.5. Application of BART model 54
4.6. Sensitivity analysis 58
4.7. Concluding remarks. 60
Ⅴ. Seismic Demand Sensitivity of the Containment Building 62
5.1. Background 62
5.2. Description of the model and input parameters 62
5.3. Selection of Ground Motion and Nonlinear time history analysis 64
5.4. Selection of EDP and dataset preparation 65
5.5. Application of BART model 73
5.6. Sensitivity analysis 83
5.7. Concluding remarks. 91
Ⅵ. Conclusions and Recommendation 93
6.1. Summary and conclusion 93
6.2. Recommendation 95
References 96
Abstract (in Korean) 108
Table 3-1. COV of material parameters from previous studies 30
Table 3-2. Selected random variables. 31
Table 3-3. BART algorithm 37
Table 3-4. Explanations of different performance metrics 39
Table 3-5. Interpretation of MAPEs 39
Table 3-6. Statistical description of input and output variables of deep beam 41
Table 4-1. Performance metrices of BART model. 55
Table 5-1. Details of input motions 64
Table 5-2. Significantly correlated input parameter to the EDP 73
Fig. 3-1. Configuration of the RCB structure. 26
Fig. 3-2. Configuration of RCB 27
Fig. 3-3. Formation of BTM 28
Fig. 3-4. Application of equalDOF in BTM 28
Fig. 3-5. Concrete02 and Steel02 model. 29
Fig. 3-6. Demonstration of BART model. a) a single tree (Tj), b) Ensemble of trees.[이미지참조] 34
Fig. 3-7. Flow chart of BART model development 38
Fig. 3-8. Random data split for developing the BART model. 40
Fig. 3-9. Correlation among the parameters of deep beam 42
Fig. 3-10. Batching for stability check of the BART model for deep beam. 43
Fig. 3-11. BART performance comparisons with other AI approaches 44
Fig. 3-12. Benchmark with the results and comparisons to BART 45
Fig. 3-13. Sensitivity analysis framework 47
Fig. 4-1. Beam-truss model of reactor containment building (Panel size: 1x1m) 49
Fig. 4-2. Damage states estimation. 50
Fig. 4-3. Scatterplot of the dataset 52
Fig. 4-4. Number of observations of base shear at different DSs 53
Fig. 4-5. Batching and Histogram of BART model for DS2 55
Fig. 4-6. Effect of number of samples to the mean of MAPE 56
Fig. 4-7. Effect of number of samples to the COV of MAPE 57
Fig. 4-8. Effect of number of posterior samples and number of trees 58
Fig. 4-9. Sobol' sensitivity indices. 60
Fig. 5-1. Beam-truss model of reactor containment building (Panel size: 3x3m) 63
Fig. 5-2. Validation of the 3x3 model 63
Fig. 5-3. Input motions 64
Fig. 5-4. Location of monitoring nodes 66
Fig. 5-5. Application of ground motion 66
Fig. 5-6. Dataset formation 67
Fig. 5-7. Correlation between input and output parameters in the dataset 72
Fig. 5-8. Random train-test split for BART algorithm. 75
Fig. 5-9. Statistics of Random train-test split for BART (EDP-PFA). 76
Fig. 5-10. Statistics of Random train-test split for BART (EDP-PFD). 78
Fig. 5-11. Histogram of MAPE of BART for selected nodes (EDP-PFA) 79
Fig. 5-12. Histogram of MAPE of BART for selected nodes (EDP-PFD) 80
Fig. 5-13. Effect of the size of dataset 82
Fig. 5-14. Effect of the posterior samples and number of trees 83
Fig. 5-15. Sobol' indices at different nodes for all both EDPs of the RCB. 87
Fig. 5-16. Sobol' sensitivity indices for different nodes and EDPs 90