Title Page
Abstract
Contents
Comparison table of main symbols 21
Chapter Ⅰ. Introduction 22
1.1. Research Background and Significance 22
1.1.1. Research Background 22
1.1.2. Research Significance 25
1.2. Research Status of PHM Technology at Home and Abroad 27
1.2.1. Current Status of Foreign Research 27
1.2.2. Domestic Research Status 29
1.3. Existing Problems 32
(1) The fault prediction accuracy of key components is low 32
(2) Difficult to predict the maintenance plan in advance 33
(3) Lack of application of big data processing and analysis technology 33
1.4. Main Research Contents and Innovation Points 34
1.4.1. Proposal of Research Content 34
1.4.2. Innovation Points of This Dissertation 36
1.5. Organization Structure of The Dissertation 38
Chapter Ⅱ. Overview of PHM Related Theoretical Algorithms and Technical Architecture Design 40
2.1. PHM Technical Theory System and Method 40
2.1.1. Basic Concepts and Background of PHM 40
2.1.2. PHM Theoretical Method 41
2.1.3. PHM Technology Architecture 44
2.2. Engineering Theory of Fault Reliability Analysis 47
2.2.1. Reliability Index 47
2.2.2. Reliability Analysis 48
2.3. PHM Technology Fault Prediction Algorithm 52
2.3.1. Extreme Learning Machine 54
2.3.2. Support Vector Machine 60
2.3.3. Hidden Semi-Markov Model 64
2.3.4. Neural Network 70
2.3.5. Deep Learning 74
2.3.6. Comparative Analysis of Algorithm Performance 82
2.4. PHM Platform Architecture Design of High-speed EMUs 86
2.4.1. PHM Platform Architecture of High-speed EMUs 86
2.4.2. PHM Platform Subsystem Design 87
2.5. Summary of This Chapter 92
Chapter Ⅲ. Research on Big Data Processing and Analysis Framework of PHM Platform for High-speed EMUs 93
3.1. Big Data Processing Technology in High-speed Railway Field 93
3.1.1. EMUs Data Characteristics 93
3.1.2. Data Batch Processing Technology 96
3.2. Hadoop Big Data Platform Design 97
3.2.1. Distributed File System-HDFS 99
3.2.2. Parallel Computing Framework-MapReduce 100
3.2.3. Stream Computing Framework-Spark 101
3.2.4. Other Relevant Technologies 104
3.3. Processing and Analysis of PHM Source Data for EMUs 107
3.3.1. Data Processing Method 107
3.3.2. Examples of Data Analysis 110
3.4. PHM Big Data Platform Framework of EMUs 111
3.4.1. PHM Big Data Platform Architecture 112
3.4.2. Platform Data Processing Architecture 113
3.5. Summary of This Chapter 115
Chapter Ⅳ. Research on Gearbox Fault Prediction Algorithm of High-speed EMUs Based on HSMM 116
4.1. High-speed EMUs Gearbox Structure and Prediction Method 116
4.1.1. EMUs Gearbox Structure and Function Design 116
4.1.2. Difficulties in Gearbox Fault Diagnosis 119
4.1.3. Overview of Gearbox Fault Prediction Methods 121
4.2. Gearbox Condition Monitoring Data Preprocessing 122
4.2.1. Gearbox Life Prediction Test Plan 122
4.2.2. Time Synchronous Average Technology 126
4.2.3. Vector Auto Regressive Model Technology 133
4.2.4. Fault Data Residual Calculation 136
4.3. Hidden Semi-Markov Model Modeling and Parameter Estimation 139
4.3.1. HSMM Model Building Process 140
4.3.2. Semi-Markov Decision Process 147
4.3.3. HSMM Model Parameter Estimation 149
4.4. Optimal Multivariate Bayesian Control Chart Method Based on HSMM 151
4.4.1. Optimal Bayesian Control Charts 152
4.4.2. Bayesian Control Method 157
4.4.3. Comparison of Different Maintenance Strategies 159
4.5. Remaining Life Prediction Method Based on HSMM 163
4.5.1. Life Prediction Test Results 163
4.5.2. Analysis of Data Comparison Results 166
4.6. Summary of This Chapter 168
Chapter Ⅴ. Research on Maintenance Plan Prediction Method of High-speed EMUs Based on Deep Learning 170
5.1. Research Route of Maintenance Plan 170
5.1.1. Fault Maintenance Problem Raised 170
5.1.2. Research Ideas of Maintenance Plan 174
5.2. Related Prediction Algorithm of Maintenance Plan 175
5.2.1. Application of Deep Learning in High-speed Rail Industry 175
5.2.2. Time Series Forecasting with Deep Belief Networks 176
5.3. Time Model and Mileage Prediction Analysis of Advanced Maintenance Plan 178
5.3.1. Time Model for Advanced Maintenance Planning 178
5.3.2. Mileage Prediction Analysis 180
5.4. EMUs Mileage Prediction Based on EMD and Deep Learning 183
5.4.1. Empirical Mode Decomposition 183
5.4.2. Optimized Deep Learning Prediction Model 185
5.4.3. EMUs Mileage Prediction Model 193
5.5. Test Verification and Result Analysis 195
5.5.1. Data Description 195
5.5.2. Evaluation Indicators 197
5.5.3. Time Series Data Decomposition 197
5.5.4. Analysis of Experimental Results 200
5.6. Summary of This Chapter 222
Chapter Ⅵ. Summary and Outlook 224
6.1. Summary of Research in This Dissertation 224
6.2. Outlook for Future Work 226
References 228
국문초록 245
〈Table 1-1〉 Comparison of high-speed railway network scale 24
〈Table 2-1〉 Examples of reliability indexes for high-speed EMUs 48
〈Table 2-2〉 FMEA analysis table 49
〈Table 2-3〉 Classification of fault severity 50
〈Table 2-4〉 Algorithm performance comparison table 86
〈Table 3-1〉 Comparison of extraction method 108
〈Table 3-2〉 Types and structures of WTDS data 111
〈Table 4-1〉 Characteristic frequency of EMUs gearbox 120
〈Table 4-2〉 Specific parameters and specifications of gear box of high-speed EMUs 124
〈Table 4-3〉 P-value of normality and independence test 139
〈Table 4-4〉 HSMM parameter estimation results of EM algorithm 149
〈Table 4-5〉 HSMM parameter estimation results of EM algorithm (different initial values) 150
〈Table 4-6〉 Expected average availability under different maintenance strategies 162
〈Table 4-7〉 Residual life prediction and relative error under different models 167
〈Table 5-1〉 Maintenance cycle of CRH EMUs 171
〈Table 5-2〉 Train operation data attributes 196
〈Table 5-3〉 Train maintenance data attributes 196
〈Table 5-4〉 Parameter combination of DBN algorithm 200
〈Table 5-5〉 EMU1 comparison of prediction performance of different methods 215
〈Table 5-6〉 EMU2 comparison of prediction performance of different methods 216
〈Table 5-7〉 EMU3 comparison of prediction performance of different methods 216
〈Table 5-8〉 EMU4 comparison of prediction performance of different methods 216
〈Table 5-9〉 EMU5 comparison of prediction performance of different methods 217
〈Table 5-10〉 EMU6 comparison of prediction performance of different methods 217
〈Table 5-11〉 EMU7 comparison of prediction performance of different methods 217
〈Table 5-12〉 EMU8 comparison of prediction performance of different methods 218
〈Table 5-13〉 EMU9 comparison of prediction performance of different methods 218
〈Table 5-14〉 Comparison of average prediction performance of different methods for 30 EMUs 221
〈Table 5-15〉 High-level maintenance plan of 9 EMUs 222
〈Figure 1-1〉 Research ideas and chapter structure 39
〈Figure 2-1〉 PHM technology architecture 44
〈Figure 2-2〉 EMUs emergency shutdown fault tree structure diagram. 52
〈Figure 2-3〉 Algorithms of PHM fault prediction 54
〈Figure 2-4〉 ELM structure diagram 55
〈Figure 2-5〉 Support vector and interval 62
〈Figure 2-6〉 Markov stochastic process 65
〈Figure 2-7〉 HMM based equipment degradation status 66
〈Figure 2-8〉 Erlang distribution function curve 67
〈Figure 2-9〉 Erlang variable implementation program 68
〈Figure 2-10〉 Structure of HSMM 68
〈Figure 2-11〉 Schematic diagram of neuron 71
〈Figure 2-12〉 RBM structure diagram 76
〈Figure 2-13〉 RBM pseudo code 77
〈Figure 2-14〉 Basic structure of 6-layer DBN network 78
〈Figure 2-15〉 DBN Training Process 79
〈Figure 2-16〉 Example of DBN main program code 81
〈Figure 2-17〉 Example of DBN model training code 82
〈Figure 2-18〉 PHM system architecture of high-speed EMUs 87
〈Figure 2-19〉 Structure diagram of on-board PHM system 89
〈Figure 2-20〉 Ground PHM system architecture 91
〈Figure 3-1〉 Main architecture of Hadoop ecosystem 98
〈Figure 3-2〉 HDFS architecture diagram 100
〈Figure 3-3〉 Execution flow chart of Map/Reduce mode 101
〈Figure 3-4〉 Spark Streaming processing framework 102
〈Figure 3-5〉 Spark reading the program code of bucket data 103
〈Figure 3-6〉 Hive data warehouse architecture 105
〈Figure 3-7〉 HBase system architecture 106
〈Figure 3-8〉 Kafka Technical Architecture 107
〈Figure 3-9〉 Data cleaning process 110
〈Figure 3-10〉 PHM big data platform architecture of high-speed EMUs 112
〈Figure 3-11〉 Data processing architecture of EMUs PHM 114
〈Figure 4-1〉 Location of the gearbox 116
〈Figure 4-2〉 Gearbox design structure diagram 117
〈Figure 4-3〉 Product drawing of gearbox transmission system 118
〈Figure 4-4〉 High-speed EMUs transmission system test bench 119
〈Figure 4-5〉 High-speed EMUs gearbox test bench 123
〈Figure 4-6〉 Gearbox output torque 125
〈Figure 4-7〉 Gearbox input shaft speed 126
〈Figure 4-8〉 Python program for TSA signal noise reduction 128
〈Figure 4-9〉 Raw data and its signal transformation 132
〈Figure 4-10〉 TSA Signal at 300% Rated Torque 132
〈Figure 4-11〉 Vector Auto Regressive models in Python 133
〈Figure 4-12〉 Residual signal of document 194-231 137
〈Figure 4-13〉 K-S test distance of residual 138
〈Figure 4-14〉 HSMM converted to multi-state Markov chain 144
〈Figure 4-15〉 Schematic diagram of HSMM system state transition 144
〈Figure 4-16〉 Flow chart of HSMM parameters estimated by EM algorithm 146
〈Figure 4-17〉 Gearbox fault detection process based on Bayesian control chart method 152
〈Figure 4-18〉 Flow chart of Bayesian algorithm control 156
〈Figure 4-19〉 Bayesian control chart based on HSMM 158
〈Figure 4-20〉 Bayesian control chart based on HMM 161
〈Figure 4-21〉 Prediction of Average Residual Life of Gearbox 165
〈Figure 5-1〉 Maintenance plan flow of high-speed EMUs 173
〈Figure 5-2〉 Framework of deep learning algorithm system 177
〈Figure 5-3〉 Daily average mileage corresponding to different sample periods 181
〈Figure 5-4〉 Flowchart of DBN prediction model algorithm 189
〈Figure 5-5〉 Pseudo-code of ABC algorithm 190
〈Figure 5-6〉 Flow chart of optimized ABC-DBN algorithm 193
〈Figure 5-7〉 Framework of the EMD-ABC-DBN model 194
〈Figure 5-8〉 EMD three-dimensional time-frequency diagram 198
〈Figure 5-9〉 Original data sequence and EMD decomposition results 199
〈Figure 5-10〉 EMU1 mile comparison of predicted and actual 201
〈Figure 5-11〉 EMU2 mile comparison of predicted and actual 202
〈Figure 5-12〉 EMU3 mile comparison of predicted and actual 202
〈Figure 5-13〉 EMU4 mile comparison of predicted and actual 203
〈Figure 5-14〉 EMU5 mile comparison of predicted and actual 203
〈Figure 5-15〉 EMU6 mile comparison of predicted and actual 204
〈Figure 5-16〉 EMU7 mile comparison of predicted and actual 204
〈Figure 5-17〉 EMU8 mile comparison of predicted and actual 205
〈Figure 5-18〉 EMU9 mile comparison of predicted and actual 205
〈Figure 5-19〉 EMU1 mileage comparison of predicted and actual in different sampling periods 206
〈Figure 5-20〉 EMU2 mileage comparison of predicted and actual in different sampling periods 207
〈Figure 5-21〉 EMU3 mileage comparison of predicted and actual in different sampling periods 207
〈Figure 5-22〉 EMU4 mileage comparison of predicted and actual in different sampling periods 208
〈Figure 5-23〉 EMU5 mileage comparison of predicted and actual in different sampling periods 208
〈Figure 5-24〉 EMU6 mileage comparison of predicted and actual in different sampling periods 209
〈Figure 5-25〉 EMU7 mileage comparison of predicted and actual in different sampling periods 209
〈Figure 5-26〉 EMU8 mileage comparison of predicted and actual in different sampling periods 210
〈Figure 5-27〉 EMU9 mileage comparison of predicted and actual in different sampling periods 210
〈Figure 5-28〉 MSE comparison of 9 EMUs in different sampling periods 212
〈Figure 5-29〉 MAE comparison of 9 EMUs in different sampling periods 212
〈Figure 5-30〉 RMSE comparison of 9 EMUs in different sampling periods 212
〈Figure 5-31〉 Accumulated mileage of 9 EMUs after 12 months in different sampling periods 213
〈Figure 5-32〉 MSE comparison of different prediction algorithms for 9 EMUs 219
〈Figure 5-33〉 MAE comparison of different prediction algorithms for 9 EMUs 219
〈Figure 5-34〉 RMSE comparison of different prediction algorithms for 9 EMUs 220