표제지
목차
Abstract 9
제1장 서론 18
제2장 이론적 배경 23
2.1. 스테인리스강의 부동태 23
2.1.1. 부동태 피막의 성장기구 23
2.1.2. 스테인리스강의 주요 부식 27
2.1.3. 공식에 대한 인자 영향성 30
2.2. 실험계획법 37
2.2.1. 실험계획법 정의 및 목적 37
2.2.2. 실험계획법의 기본원리 38
2.2.3. 요인배치법(factorial design) 41
2.3. Taguchi 방법론 43
2.3.1. 기본개념 43
2.3.2. 삼원 배치법 48
2.4. 기계학습 57
2.4.1. 기계학습의 종류 59
2.5. 회귀분석 62
2.5.1. 회귀분석의 기본개념 62
2.5.2. 단순회귀 적합 64
2.5.3. 중회귀분석 66
2.5.4. 회귀방정식의 정도 71
2.6. 인공 신경망 78
2.6.1. 기본개념 78
2.6.2. 신경망의 기본 구조 81
2.6.3. 단일 계산층: 퍼셉트론 81
2.6.4. 활성화 함수와 손실함수의 선택 85
제3장 실험방법 91
3.1. 부식인자 영향성 평가 91
3.1.1. 재료 및 시편준비 91
3.1.2. 동전위분극곡선 91
3.1.3. 선별설계 93
3.1.4. 완전요인설계 및 분산분석 96
3.2. 기계학습과 통계적 분석을 이용한 공식특성 예측 98
3.2.1. 재료 및 시편준비 98
3.2.2. 전기화학시험(동전위분극시험) 98
3.2.3. 통계적 접근 100
3.2.4. 인공신경망 107
3.3. DL-EPR 시험용액의 매개변수 최적화 및 예민화도 예측 109
3.3.1. 재료 및 열처리 109
3.3.2. Double loop electrochemical potentiokinetic reactivation 시험 109
3.3.3. 다 반응특성 Taguchi 강건설계 112
3.3.4. 데이터 기반 수학적 회귀 모델링 115
3.4. 예민화도와 환경변수(온도, pH)에 따른 부식특성 예측 116
3.4.1. 재료 및 등온열화 116
3.4.2. 미세조직 및 표면 분석 116
3.4.3. 통계분석 및 인공신경망 117
3.4.4. Double loop electrochemical potentiokinetic reactivation 시험 120
3.4.5. Cyclic potentiodynamic polarization 시험 121
3.5. 공식 매개변수 계측을 통한 기계적 강도특성 예측 122
3.5.1. 재료 및 인장시험 122
3.5.2. 정전위 가속부식시험 122
3.5.3. 공식 매개변수 계측 124
제4장 실험결과 및 고찰 125
4.1. 부식인자 영향성 평가 125
4.1.1. 미세조직 특성 125
4.1.2. 동전위 분극실험 125
4.1.3. 통계적 분석 127
4.1.4. 수학적 회귀모델을 통한 공식전위 예측 142
4.1.5. 소결론 143
4.2. 기계학습과 통계적 분석을 이용한 공식특성 예측 148
4.2.1. 동전위분극 곡선 148
4.2.2. 입력변수의 유의성 150
4.2.3. 수학적 회귀모델을 이용한 임계공식전위 예측 156
4.2.4. 인공신경망을 이용한 동전위 분극곡선 예측 157
4.2.5. 소결론 165
4.3. DL-EPR 시험용액의 매개변수 최적화 및 예민화도 예측 168
4.3.1. DL-EPR 곡선 168
4.3.2. Taguchi 설계 170
4.3.3. 개선된 최적조건의 검증 177
4.3.4. 수학적 회귀식을 이용한 DL-EPR 값 검증 181
4.3.5. 소결론 186
4.4. 예민화도와 환경변수(온도, pH)에 따른 부식특성 예측 190
4.4.1. 예민화도에 따른 미세조직 분석 190
4.4.2. 예민화도 측정 193
4.4.3. 예민화도의 영향 193
4.4.4. 온도, pH의 영향 200
4.4.5. 통계적 접근 203
4.4.6. 인공신경망 모델링을 통한 순환분극곡선 예측 207
4.4.7. 소결론 219
4.5. 공식 매개변수 계측을 통한 기계적 강도특성 예측 222
4.5.1. 인공 공식손상 222
4.5.2. 응력-변형률 선도 228
4.5.3. 인장강도 예측을 위한 회귀식 제안 228
4.5.4. 항복비(Y/T)를 이용한 항복강도 예측 234
4.5.5. 다항선형회귀 모델에 대한 성능검증 237
4.5.6. 소결론 237
제5장 결론 240
참고문헌 244
Table 2.3.1. ANOVA table of three-way batch method with 1 repetition number... 49
Table 2.4.1. PlayTennis data 60
Table 2.5.1. ANOVA with regression analysis 74
Table 3.1.1. Chemical composition of the materials 92
Table 3.1.2. Designed factors and their levels for screening design 94
Table 3.1.3. Screen design matrix 95
Table 3.1.4. Full factorial design matrix 97
Table 3.2.1. Chemical composition(wt. %) for materials 99
Table 3.2.2. Experimental independent factor and their levels 101
Table 3.2.3. Full factorial design matrix and experiment results 102
Table 3.3.1. L16 orthogonal array with design factors and their levels[이미지참조] 111
Table 3.3.2. Designed factors and their levels 114
Table 3.4.1. Designed factors and their levels 118
Table 3.4.2. Experiment design matrix and response for CPDP tests in 3.5% NaCl solution 119
Table 4.1.1. The result of ANOVA for full factorial design 138
Table 4.1.2. Results of validation experiment for linear regression 145
Table 4.2.1. Coefficient of determination according to the interaction term of the... 151
Table 4.2.2. Result of ANOVA on the full factorial design 152
Table 4.2.3. Regression models and their determination coefficient for critical... 158
Table 4.3.1. Experimental results on the L16 orthogonal array[이미지참조] 171
Table 4.3.2. The result of delta statistics and ANOVA analysis for DOS 173
Table 4.3.3. The result of delta statistics and ANOVA analysis for Ia[이미지참조] 174
Table 4.4.1. The result of ANOVA for all terms in Table 3.4.1 204
Table 4.4.2. The result of ANOVA for significant terms(confidence level 95%) in... 205
Table 4.4.3. Comparison of performance of ANN structure with hidden neurons... 214
Table 4.4.4. Comparison of the model performance between the ANN and MLR 220
Table 4.5.1. Detailed pit parameters and their responses 229
Table 4.5.2. The result of ANOVA for significant terms in Table 4.5.1 230
Fig. 2.1.1. Schematic diagram of a mechanism for passive film damage 24
Fig. 2.1.2. Schematic diagram of pit formation and growth on Fe surface 29
Fig. 2.1.3. Pitting attack at grain boundaries in austenitic stainless steels 31
Fig. 2.1.4. Time-temperature-sensitisation-pitting (TTSP) diagrams developed for... 32
Fig. 2.1.5. Influence of various alloying elements on the pitting corrosion resistance 34
Fig. 2.2.1. Process of design of experiment 39
Fig. 2.3.1. An example of a Taguchi-type parameter design experiment 45
Fig. 2.4.1. (a) General programming, (b) Machine learning-based programming 58
Fig. 2.6.1. The synaptic connections between neurons: (a) biological neural... 79
Fig. 2.6.2. Basic structure of perceptron: (a) without bias, (b) with bias 82
Fig. 2.6.3. Pre-activation and post-activation values within a neuron 87
Fig. 2.6.4. Various activation functions 90
Fig. 3.2.1. Schematic diagram for ANN architecture used for modeling... 108
Fig. 3.3.1. Microstructure of the AL-6XN 110
Fig. 3.3.2. Schematic diagram for DL-EPR test 113
Fig. 3.5.1. (a) Schematic diagram of a tensile test specimen, (b) configuration... 123
Fig. 4.1.1. Microstructure of the (a) STS 316L and (b) AL-6XN 126
Fig. 4.1.2. Cyclic polarization curves of 316L for partial conditions of Table 3.1.3 128
Fig. 4.1.3. Pareto chart of standardized effect for factors in screening design 129
Fig. 4.1.4. Main effect plot for critical pitting potential in screening design 132
Fig. 4.1.5. Potentiodynamic polarization curves in different factor level: (a) PREN,... 134
Fig. 4.1.6. Potentiodynamic curves with different level of Cl- concentration in (a)...[이미지참조] 136
Fig. 4.1.7. Pareto chart of standardized effect for factors in the full factorial design 139
Fig. 4.1.8. Main effect plot for critical pitting potential in the full factorial design 140
Fig. 4.1.9. Interaction effect plot for critical pitting potential in the full factorial design 141
Fig. 4.1.10. Normal probability diagrams for critical pitting potential 144
Fig. 4.1.11. Potentiodynamic curves for validation test in Table 4.1.2 146
Fig. 4.2.1. Potentiodynamic polarization curves with variable (a) PREN, (b)... 149
Fig. 4.2.2. Pareto chart of the standardized effects of significant factors and... 154
Fig. 4.2.3. The main effect plots for the mean of critical pitting potential values 155
Fig. 4.2.4. Comparison of actual and predicted values for critical pitting potential:... 159
Fig. 4.2.5. The main effect plots for the mean of critical pitting potential values 162
Fig. 4.2.6. Comparison of actual and ANN predicted values for log(i): (a) training... 163
Fig. 4.2.7. Experimental and ANN predicted polarization curves for some training... 164
Fig. 4.2.8. Experimental and ANN predicted polarization curves. This conditions not... 166
Fig. 4.3.1. DL-EPR curves for some experimental condition in L16 orthogonal...[이미지참조] 169
Fig. 4.3.2. Main effect plot for S/N ratio: (a) DOS is larger the better, (b) Ia is...[이미지참조] 176
Fig. 4.3.3. DL-EPR test value(DOS) with various conditions 178
Fig. 4.3.4. Surface morphologies after DL-EPR test with different heat-treatment... 180
Fig. 4.3.5. Comparison of IGC morphologies observed after DL-EPR and 10%... 182
Fig. 4.3.6. DL-EPR test value with various temperature and aging time under... 183
Fig. 4.3.7. 3D surface plot diagram for mathematical regression as a function of... 185
Fig. 4.3.8. Comparison between actual and predicted value for DL-EPR test results 187
Fig. 4.3.9. DL-EPR curve for heat-treated AL-6XN during 12 hours at 780℃ 188
Fig. 4.4.1. Result of microstructure analysis for heat-treated specimens at 800 ℃:... 191
Fig. 4.4.2. XRD pattern for the secondary phase of AL-6XN heat-treated for 6... 192
Fig. 4.4.3. DL-EPR curves of AL-6XN with various heat-treatment time in 2M... 194
Fig. 4.4.4. Cyclic potentiodynamic polarization curves of AL-6XN with various... 195
Fig. 4.4.5. Optical microscope images of AL-6XN with various DOS value after... 198
Fig. 4.4.6. A summary of the derived Epit, Erep, and DOS values in the CPDP...[이미지참조] 199
Fig. 4.4.7. CPDP curves with the level variance of specific factors in 3.5%... 201
Fig. 4.4.8. 3D surface plot diagram and actual values vs. predicted values for... 208
Fig. 4.4.9. Pareto chart(a) and interaction effect diagram(b) for significant factors... 209
Fig. 4.4.10. Fitted curve for DOS value 211
Fig. 4.4.11. Experimental and ANN predicted polarization curves for some train... 215
Fig. 4.4.12. Diagram of actual values vs. predicted values for log(i): (a) training... 217
Fig. 4.4.13. Experimental and ANN predicted polarization curves. This conditions... 218
Fig. 4.5.1. Polarization curve for 316L in 3.5% NaCl solution of pH 2 at 60 ℃ 223
Fig. 4.5.2. 3D profile images of specimens applied 0.3 V for 2 hours with... 224
Fig. 4.5.3. Images for corroded specimens at various potential and time in the... 226
Fig. 4.5.4. Optical images of corroded specimen at different applied potentials... 227
Fig. 4.5.5. Stress-strain curves of corroded specimens at various applied potential... 231
Fig. 4.5.6. Pareto chart (a) and interaction effect diagram (b) for significant... 232
Fig. 4.5.7. (a) 3D surface plot diagram for MLR model, (b) diagram for... 235
Fig. 4.5.8. Comparison between σu and σy[이미지참조] 236
Fig. 4.5.9. Validation of experimental and predicted value, Validation test... 238