Data Fusion for Machinery Condition Monitoring, Diagnostics and Prognostics
- Alternative Title
- 기계 상태감시, 진단 및 예지를 위한 데이터 융합기술
- Abstract
- 산업의 발전과 더불어 설비관리는 자동차, 조선, 항공, 핵발전, 정유등과 같은 중공업에서 생산성향상, 고장율저하, 안전과 작동신뢰성향상을 가져오는 중요한 부분이다.
그러나 신뢰성 있고 효과적인 설비관리는 외부전자기 노이즈, 기계의 복잡한 구조, 난해한 고장메커니즘 때문에 어려움을 겪고 있다. 최근에 상태기반 정비(CBM) 전략은 효과적이고 정확하여 일반화된 기법으로 사용되고 있다. 대부분의 CBM시스템은 토목이나 군사분야에서 사용하고 있다. CBM의 개방된 플랫폼인 OSA-CBM은 7가지 단계의 표준으로 나누어져 있다. 그것의 핵심적인 기능들은 상태감시, 기계상태평가와 예지로 요약할 수 있다.
상태감시는 예상된 값의 온라인 또는 오프라인 비교검증을 포함한다. 필요에 따라 미리 설정된 경고를 생성할 수 있다. 기계상태평가는 만약 시스템의 상태가 좋지 않다면 결함진단 보고를 한다. 예지진단은 설비의 상태와 잔여유용수명 계산을 수행하게 된다.
최근에 공학적인 적용에 관심을 받고 데이터융합은 원신호 단계, 특징신호 단계, 의사결정단계의 융합이 존재한다. 특히 신호처리기술과 센서의 개발과 함께 CBM의 성능을 향상 시킬 수 있는 상호정보 융합이 가능하게 되었다.
이 연구에서는 상태기반의 정보융합, 결함진단과 예지기술에 대해 연구하였다. 적용된 기술은 정보융합, 신호처리, 인공지능, 통계적 학습, 비선형 예측이 사용되었다. 더불어 몇몇 융합 알고리즘이 개발되고 검증되었다. CBM기반의 데이터 융합은 비용효과적으로 정확하고 신뢰성이 있는 융합기술이 개발되었다.
상태감시를 위해 정해진 시간에 나타나는 잠재적인 결함은 신뢰성 있는 열화 표기와 적당한 알람 설정에 달려있다. 결함은 센서 신호에서 나타내는 다양한 열화의 과정과 상호관련이 있다. 그러므로 센서 융합은 열화 상태감시를 해결하는 간단하고 신뢰성이 있는 기법이다. 이 연구에서는 부분적인 융합 상태감시가 먼저 수행된다. 이것은 SOM 신경망을 특징 값을 융합하는데 사용했다. 그리고 나서 열화경향은 웨이블렛 기법으로 찾아낼 수 있다. 최종적으로 자동 알람 설정기술에 대해 제안했다.
결함진단을 위해 단일 분류 알고리즘은 만족한 결과를 얻을 수 없기 때문에 다양한 인공지능 분류 시스템이 사용되었다. 다양한 분류기의 결과를 통합하는 것이 분류의 정확도를 향상시킬 수 있었다. 이 연구에서 융합 진단 기법이 개발되었고, 이것은 원신호수집, 특징계산, 특징추출, 분류, 데이터융합, 분류기 융합 6 단계로 구분된다.
예지진단을 위해 정확한 잔여유용수명평가는 몇 가지 이유 때문에 어렵다. 왜냐하면 실제 시스템은 훈련하기 위한 정확한 데이터 취득과 예지 모델 생성 어려움, 비정기적인 정비가 존재하게 때문이다. 예지진단을 위한 상호 정보융합은 열화추종, 예측의 정확도를 높이고 불확실성 에러를 감소시킨다. 이 연구에서 새로운 데이터 기반의 예지진단 시스템이 개발되었다. 먼저 시간신호 상태감시 데이터 집합이 재구성되고, 비선형 회귀 모델이 열화예측을 위해 사용된다. 더불어 다른 모델로부터 예측된 값은 신뢰성 향상을 위해 융합된다. 마지막으로 잔여유용수명과 불확실성에 대한 값들이 평가된다.
이와 더불어서 최적의 CBM 시스템을 제안했다. 이 시스템은 신뢰성기반 설비유지관리 기법으로 비용효과적인 메커니즘이다. 그리고 설비유지보수 성능 향상을 위해 데이터 융합전략을 소개하였다.
제안된 방법들과 시스템은 프로젝트를 통해 엘리베이터용 유도전동기 결합진단을 수행하고, 메탄 압축기 상태감시 및 예지진단을 수해하였다. 이러한 결과들은 개발된 알고리즘이 CBM의 성능을 비용효과적이고 신뢰성 있게 향상시키는 것을 입증한다.
With the fast development in industry, especially in the areas of heavy industry like automobile, shipbuilding, aircraft, nuclear power and petrochemical etc, maintenance shows increasing importance due to the potential advantages to be gained by improving production availability, reducing downtime cost, enhancing operation reliability and plant safety.
However, carrying a reliable and effective maintenance faces huge challenges due to the outer electromagnetism-noise, the complex inner structure of machine and abstruse failure mechanisms; even say nothing of expensive maintenance cost etc. In recent years, condition-based maintenance (CBM) strategy shows advantages of effectiveness, accuracy, and becomes the popular maintenance approach. Lots of CBM systems are employed and identified in civil or military industry. The open system architecture for CBM organization (OSA-CBM) has divided a standard CBM system into seven different layers, with technical modules solution. The core functions among the architecture can be summarized as condition monitoring, health assessment and prognostics.
Condition monitoring involves comparing on-line/off-line data with expected values; if necessary it should be able to generate alerts based on preset operational limits. Health assessment serves prescribing if the health of the monitored component or system has degraded, and exerting fault diagnostics. Prognostics involve calculating the future health of an asset and report the remaining useful life (RUL).
Currently, data fusion stands for another developing technology containing signal-level fusion, feature-level fusion and decision-level fusion. Applying fusion techniques in engineering practice has been receiving increasing attentions in recent years. Particularly, with the rapid progress of advanced sensor and signal processing technologies, fusing large of mutual information becomes possible, which is expected to bring about enhanced CBM performances.
In this thesis, the data fusion based condition monitoring, diagnostics and prognostics technology is researched. The employed methods, involving fusion techniques, signal processing, artificial intelligence, statistics learning, nonlinear prediction etc., are investigated and analyzed to reach the research aims. In further, several related data fusion subsystems are designed and validated. Finally, a whole data fusion based CBM structure is constructed, which has the advantages of cost-effectiveness, accuracy and reliability.
For condition monitoring, timely indicating potential failures relies on reliable degradation indicator and appropriate alarm setting. Failures can often be attributed to many correlated degradation processes, which could be reflected by multiple degradation indicators extracted from sensor signals. Therefore, fusing multiple indicators would provide a simple and reliable solution to degradation monitoring. In this study, a fusion monitoring subsystem is put forward, which use SOM neural network to fuse multi-indicators (features) into a dominant indicator, then the degradation trend can be picked out by wavelet decomposition methods, finally, an automatic alarm setting technique is suggested to determine the appropriate threshold.
For fault diagnostics, numerous intelligent classification systems have been employed to assist machine fault diagnosis tasks by correctly interpreting the fault data. However, it is imperative to study the comparative performance of a classification of such algorithms since no single algorithm may perform the best on all cases. Integration of different decisions from multiple classifiers can potentially boost the accuracy of recognition. In this research, a fusion diagnostics approach is developed, which consists of six levels: raw data collection, feature extraction, classification, decision fusion, classifiers selection and fusion.
For prognostics, accurate remain useful life assessment is difficult to achieve due to a number of reasons such as suitable data for training, prediction models, uncertainty management etc. Fusing mutual information for prognostics purpose likely improves accuracy of degradation tracking, prediction, and reduces the uncertainty errors. In this research, a novel data-driven prognostics system is developed. At first, the time-series monitoring data sets are reconstructed, then nonlinear regression models are employed to predict future degradation trajectory of machine health state by the way of iterated multi-step-ahead. Furthermore the predicted values and bias from different models are fused to enhance the reliability. Finally, remain useful life and its uncertainty interval can be assessed.
In further, an optimal CBM system is put out. This system integrates reliability-centered maintenance as cost- effective management mechanism, and employs data fusion strategy introduced above for improvement of maintenance performance.
The proposed methods and systems are validated respectively by simulation experiments like induction motor fault diagnosis, and industry project cases such as elevator motor diagnosis, low methane compressor condition monitoring and prognostics. The results show that the developed techniques and systems can enhance CBM performance remarkably with the advantages of reliability, accuracy and cost-effectiveness.
- Author(s)
- Niu, Gang
- Issued Date
- 2009
- Awarded Date
- 2009. 2
- Type
- Dissertation
- Keyword
- Monitoring Diagnosis Prognosis Data Fusion
- Publisher
- 부경대학교 도서관
- URI
- https://repository.pknu.ac.kr:8443/handle/2021.oak/10545
http://pknu.dcollection.net/jsp/common/DcLoOrgPer.jsp?sItemId=000001954683
- Alternative Author(s)
- 뉴강
- Affiliation
- 부경대학교 기계공학부
- Department
- 대학원 기계공학부지능기계공학전공
- Advisor
- 양보석
- Table Of Contents
- Ⅰ. Introduction = 1
1. Background and Motivation = 1
2. Research Objectives = 3
3. Tools and Approaches = 4
3.1 Condition Monitoring = 4
3.2 Fault Diagnostics = 5
3.3 Prognostics = 6
4. Thesis Structure = 7
5. Scientific Contribution = 9
References = 11
Publications = 12
Ⅱ. Literature Review = 14
1. Introduction = 14
2. Review of Maintenance = 14
3. Condition Based Maintenance (CBM) = 16
3.1 Standard of CBM = 17
3.2 CBM system technology = 19
3.3 Implementation of CBM = 21
4. Signal Preprocessing and Analysis = 23
4.1 Introduction = 23
4.2 Signal De-noising and Analysis = 24
4.2.1 Time domain techniques: averaging = 25
4.2.2 Frequency domain techniques: FFT = 26
4.2.3 Time-frequency techniques: Discrete wavelet transforms (DWT) = 26
4.2.4 Other techniques = 30
4.3 Feature Extraction = 30
4.3.1 Features in time domain = 31
4.3.2 Features in frequency domain = 33
4.3.3 Auto-regression coefficients = 35
4.4 Feature Selection = 35
4.4.1 Principal component analysis (PCA) = 37
4.4.2 Kernel principal component analysis (KPCA) = 38
4.4.3 Genetic algorithm (GA) = 39
5. Data Fusion = 41
5.1 Introduction = 41
5.2 Fusion Architectures = 42
5.2.1 Signal-level fusion = 43
5.2.2 Feature-level fusion = 43
5.2.3 Decision-level fusion = 44
5.3 Data Fusion Techniques = 46
5.3.1 Voting method = 47
5.3.2 Bayesian belief fusion = 47
5.3.3 Dempster-Shafer Theory = 49
5.3.4 Multi-agent fusion = 51
6. Summary = 54
References = 55
Ⅲ. Data Fusion for Condition Monitoring = 59
1. Introduction = 59
2. A Purposed Fusion Monitoring System = 62
3. Degradation Indicator Using SOM Neural Network Fusion = 64
3.1 Theoretical Background of SOM = 64
3.2 SOM Degradation Detection and Fusion = 66
4. Automatic Alarm Setting Strategy = 68
4.1 Background of Alarm Setting = 68
4.1.1 Alarm = 68
4.1.2 Trips = 69
4.2 Automatic Alarm Setting Method = 70
4.2.1 Criteria based on largest time constant = 70
4.2.2 Criteria based on statistics (6σ) = 71
5. Experiment = 73
5.1 Experimental Setup = 74
5.2 Experimental Results and Analysis = 74
6. Detection Matrix = 80
7. Concluding Remarks = 82
References = 83
Ⅳ. Data Fusion for Diagnostics = 86
1. Introduction = 86
2. Division of Diagnostic Methods = 89
2.1 Statistical Approaches = 89
2.2 AI Approaches = 90
2.3 Model-Based Approaches = 91
3. Classifier Selection = 91
3.1 2-Classifier Correlation Analysis = 92
3.2 Multi-Classifier Correlation Analysis = 93
4. A purposed Fusion Diagnostic System = 94
4.1 Level 1 to Level 3 - Preparation for Decision Fusion = 95
4.2 Level 4 - Data Fusion at Decision Level = 96
4.3 Level 5 - Classifier Selection = 97
4.4 Level 6 - Decision Fusion = 97
5. Experiments = 97
5.1 Self-designed Test Rig Motor Diagnostics = 97
5.1.1 Data Acquisition = 97
5.1.2 Feature extraction and classification = 98
5.1.3 Classifiers selection and fusion = 100
5.1.4 Classifiers fusion comparison = 102
5.2 Elevator Induction Motor Diagnostics = 102
5.2.1 Data acquisition = 105
5.2.2 Feature extraction and classification = 105
5.2.3 Classifiers fusion = 109
5.3 Decision-level Fusion Diagnostics Using Transient Current Signal = 110
5.3.1 Experiments and data acquisition = 111
5.3.2 Signal preprocessing and wavelet transform = 112
5.3.3 Features calculation and classification = 117
5.3.4 Fusion performance evaluation = 118
6. Fusion Diagnostic Toolbox 2. 0 Introduction = 119
7. Concluding Remarks = 120
References = 120
Ⅴ. Data Fusion for Prognostics = 123
1. Introduction = 123
2. A Review of Prognostics Approaches = 124
2.1 Data-driven Approaches = 127
2.2 Model-based Approaches = 128
3. A Purposed Data-driven Fusion Prognostic System = 129
4. Time Series Prediction = 131
4.1 State Space Reconstruction = 131
4.1.1 Determining the delay time τ : C-C method = 133
4.1.2 Determining the minimum embedding dimension m: FNN method = 135
4.2 One-step-ahead / Multi-step-ahead Prediction = 137
4.3 Time Series Model = 140
4.3.1 Regression analysis = 140
4.3.2 Dempster-Shafer regression model = 141
4.3.3 Least squares support vector machines = 144
5. Experiments = 147
5.1 Performance Degradation Prediction = 147
5.1.1 State space reconstruction and parameter selection = 148
5.1.2 Multi-step ahead time-series prediction = 150
5.1.3 Performance evaluation = 152
5.2 Enhanced Data-driven Prognostics = 157
6. Prognostics Performance Matrix = 161
7. Fusion Prognostics Toolbox Introduction = 162
8. Concluding Remarks = 163
References = 164
Ⅵ. Development of An Optimal CBM System = 169
1. Introduction = 169
2. New Development of Maintenance Theory = 170
2.1 CBM Plus (CBM+) = 170
2.1.1 The concept of CBM+ = 171
2.1.2 Infrastructure of CBM+ = 172
2.2 Reliability Centered Maintenance (RCM) = 175
2.3 Data Fusion = 176
3. A Framework of Cost-effective and Accurate CBM System = 177
3.1 Integrating CBM and RCM: Cost-Effective Maintenance = 178
3.1.1 Structured steps of applying RCM = 178
3.1.2 Relationship between CBM and RCM = 181
3.2 Integrating CBM and Data Fusion: Accurate Maintenance = 181
3.2.1 Purposed CBM system based on data fusion = 182
3.2.2 Fusion diagnostics subsystem = 183
3.2.3 Fusion monitoring and prognostics subsystem = 184
4. Concluding Remarks = 187
Reference = 187
Ⅶ. Conclusions and Future Work = 189
1. Conclusions = 189
1.1 Data Fusion for Condition Monitoring = 190
1.2 Data Fusion for Diagnostics = 190
1.3 Data Fusion for Prognostics = 191
1.4 Optimal CBM Architecture = 191
2. Future Work = 192
Appendix A: Technical Standards = 194
Appendix B: = 200
국문요약 = 203
Acknowledgements = 206
- Degree
- Doctor
-
Appears in Collections:
- 대학원 > 기계공학부-지능기계공학전공
- Authorize & License
-
- Files in This Item:
-
Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.