Predictive Maintenance in Dynamic Systems (eBook)

Edwin Lughofer received his PhD-degree from the Johannes Kepler University Linz (JKU) in 2005. He is currently Key Researcher with the Fuzzy Logic Laboratorium Linz / Department of Knowledge-Based Mathematical Systems (JKU) in the Softwarepark Hagenberg. He has participated in several basic and applied research projects on European and national level, with a specific focus on topics of Industry 4.0 and FoF (Factories of the Future). He has published around 200 publications in the fields of evolving fuzzy systems, machine learning and vision, data stream mining, chemometrics, active learning, classification and clustering, fault detection and diagnosis, quality control and predictive maintenance, including 80 journals papers in SCI-expanded impact journals, a monograph on ’Evolving Fuzzy Systems’ (Springer) and an edited book on ’Learning in Non-stationary Environments’ (Springer). In sum, his publications received 4200 references achieving an h-index of 36. He is associate editor of the international journals Information Sciences, IEEE Transactions on Fuzzy Systems, Evolving Systems, Information Fusion, Soft Computing and Complex and Intelligent Systems, the general chair of the IEEE Conference on EAIS 2014 in Linz, the publication chair of IEEE EAIS 2015, 2016, 2017 and 2018, the program co-chair of the International Conference on Machine Learning and Applications (ICMLA) 2018, the tutorial chair of IEEE SSCI Conference 2018, the publication chair of the 3rd INNS Conference on Big Data and Deep Learning 2018, and the Area chair of the FUZZ-IEEE 2015 conference in Istanbul. He co-organized around 12 special issues and more than 20 special sessions in international journals and conferences. In 2006 he received the best paper award at the International Symposium on Evolving Fuzzy Systems, in 2013 the best paper award at the IFAC conference in Manufacturing Modeling, Management and Control (800 participants) and in 2016 the best paper award at the IEEE Intelligent Systems Conference. Moamar Sayed-Mouchaweh received his Master degree from the University of Technology of Compiegne-France in 1999. Then, he received his PhD degree from the University of Reims-France in December 2002. He was working as Associated Professor in Computer Science, Control and Signal processing at the University of Reims-France in the Research center in Sciences and Technology of the Information and the Communication (CReSTIC). In December 2008, he obtained the Habilitation to Direct Researches (HDR) in Computer science, Control and Signal processing. Since September 2011, he is working as a Full Professor in the High National Engineering School of Mines “Ecole Nationale Supérieure des Mines de Douai” at the Department of Computer Science and Automatic Control (Informatique & Automatique). He edited the Springer book Learning in Non-Stationary Environments: Methods and Applications, in April 2012 and wrote two Brief Springer books in Electrical and Computer Engineering: Discrete Event Systems: Diagnosis and Diagnosability, and Learning from Data Streams in Dynamic Environments. He was a guest editor of several special issues of international journals. He was IPC Chair of the 12th IEEE International Conference on Machine Learning and Applications (ICMLA'13), the Conference Chair and IPC Chair of IEEE International Conference on Evolving and Adaptive Intelligent Systems (EAIS2015), and the IPC Chair of the 15th IEEE International Conference on Machine Learning and Applications (ICMLA'16). He is working as a member of the Editorial Board of Elsevier Journal Applied Soft Computing and Springer Journals Evolving Systems and Intelligent Industrial Systems.

Preface 6
Contents 8
Contributors 10
Prologue: Predictive Maintenance in Dynamic Systems 13
1 From Predictive to Preventive Maintenance in Dynamic Systems: Motivation, Requirements, and Challenges 13
2 Components and Methodologies for Predictive Maintenance 16
2.1 Models as Backbone Component 17
2.2 Methods and Strategies to Realize Predictive Maintenance 19
3 Beyond State-of-the-Art—Contents of the Book 23
References 32
Smart Devices in Production System Maintenance 36
1 Introduction 36
2 State of the Art 39
2.1 Definition of Terms 40
2.2 Physical Devices/Hardware 41
2.2.1 Smartphones and Tablets 41
2.2.2 Smartglasses 41
2.2.3 Smartwatches 42
2.3 Market View 43
2.4 Device Selection and Potentials 44
3 Application Examples in Maintenance 47
3.1 Local Data Analysis and Communication for Condition Monitoring 48
3.2 Remote Expert Solutions 50
3.3 Process Data Visualization for Process Monitoring 52
4 Limitations and Challenges 54
4.1 Hardware Limitations 54
4.2 User Acceptance 55
4.3 Information Compression on Smart Devices 57
4.4 Legal Aspects 58
5 Summary 59
References 60
On the Relevance of Preprocessing in Predictive Maintenance for Dynamic Systems 63
1 Introduction 63
2 Preprocessing 64
2.1 Taxonomy 65
2.2 Data Cleansing 66
2.2.1 Outlier Detection Based on Mahalanobis Distance 67
2.2.2 Outlier Detection Based on ?2 Approximations of Q and T2 Statistics 71
2.3 Data Normalization 73
2.4 Data Transformation 74
2.4.1 Statistical Transformations 74
2.4.2 Signal Processing 75
2.5 Missing Values Treatment 77
2.6 Data Engineering 78
2.6.1 Feature Selection 78
2.6.2 Feature Extraction 80
2.6.3 Feature Discretization 81
2.7 Imbalanced Data Treatment 82
2.7.1 Oversampling 83
2.7.2 Undersampling 84
2.7.3 Mixed Sampling 86
2.8 Models 87
2.8.1 Classification 87
2.8.2 Regression 88
3 Experimentation 89
3.1 Datasets 90
3.1.1 PHM Challenge 2014 90
3.1.2 PHM Challenge 2016 91
3.2 Experimental Schema 92
3.3 Results 93
4 Conclusions 97
References 98
Part I Anomaly Detection and Localization 104
A Context-Sensitive Framework for Mining Concept Drifting Data Streams 105
1 Concept Drifting Data Streams 105
1.1 Concept Drift 106
2 A Novel Framework for Online Learning in Adaptive Mode 107
2.1 Basic Components 107
2.2 Optimizing for Stream Volatility and Speed 108
3 Implementation of a Context-Sensitive Staged Learning Framework 108
3.1 The Use of the Discrete Fourier Transform in Classification and Concept Encoding 110
3.2 Repository Management 114
3.3 The Staged Learning Approach 115
3.3.1 Transition Between Stages 117
3.4 Space and Time Complexity of Spectral Learning 120
4 Empirical Study 121
4.1 Datasets Used for the Empirical Study 122
4.1.1 Synthetic Data 122
4.1.2 Synthetic Data Recurring with Noise 123
4.1.3 Synthetic Data Recurring with a Progressively Increasing Pattern of Drift 124
4.1.4 Synthetic Data Recurring with an Oscillating Drift Pattern 124
4.1.5 Real-World Data 125
4.2 Parameter Values 125
4.3 Effectiveness of Staged Learning Approach 125
4.4 Accuracy Evaluation 128
4.4.1 ARF vs SOL Accuracy of a Concept 130
4.5 Throughput Evaluation 131
4.6 Accuracy Versus Throughput Trade-Off 132
4.7 Memory Consumption Evaluation 132
5 Sensitivity Analysis 133
6 Conclusion 134
7 Future Research 136
References 136
Online Time Series Changes Detection Based on Neuro-FuzzyApproach 138
1 Introduction 138
2 Fuzzy Online Segmentation-Clustering 139
2.1 Probabilistic Approach 141
2.2 Possibilistic Approach 143
2.3 Online Combined Approach 145
2.4 Robust Approach 148
3 Robust Forecasting and Faults Detection in Nonstationary Time Series 163
4 Conclusions 172
References 172
Early Fault Detection in Reciprocating Compressor Valves by Means of Vibration and pV Diagram Analysis 174
1 Introduction 174
2 Problem Statement 176
2.1 Reciprocating Compressor Operation 176
2.2 Problem Statement 179
3 Vibration Analysis 181
3.1 Motivation 181
3.2 Feature Extraction 187
3.3 Feature Space 189
4 Analysis of the pV Diagram 190
4.1 Motivation 190
4.2 Feature Extraction 192
4.3 Feature Space 195
4.4 Classification 197
5 Experimental Setup 200
5.1 Compressor Test Bench 200
5.2 Test Runs 201
6 Results 203
6.1 Vibration Analysis 203
6.2 pV Diagram Analysis 206
7 Conclusions 209
References 210
A New Hilbert-Huang Transform Technique for Fault Detection in Rolling Element Bearings 213
1 Introduction 213
2 Minimum Entropy Deconvolution Filter 217
3 The Proposed eHT Technique for Bearing Fault Detection 220
3.1 Brief Discussion of Mathematical Morphology Analysis 221
3.1.1 Structural Element (SE) 221
3.1.2 Dilation and Erosion 221
3.1.3 Closing and Opening 222
3.2 The Proposed Morphological Filter 223
3.3 The Proposed eHT Technique 225
4 Application of the Proposed eHT Technique for Bearing Fault Detection 226
4.1 Experimental Setup and Instrumentations 226
4.2 Performance Evaluation 228
4.2.1 Validation of Morphological-Based Filtering Technique 228
4.2.2 Validation of the Normality Measure 228
4.3 Evaluation of the Proposed eHT Technique 231
4.3.1 Condition Monitoring of a Healthy Bearing 231
4.3.2 Outer Race Fault Detection 232
4.3.3 Inner Race Fault Detection 232
4.3.4 Rolling Element Fault Detection 233
5 Conclusion 234
References 234
Comparison of Genetic and Incremental Learning Methods for Neural Network-Based Electrical Machine Fault Detection 237
1 Introduction 237
2 Electrical Machine Fault Detection 239
3 Genetic Algorithm for Neural Network Learning 243
3.1 Initialization and Parameterization 244
3.2 Phenotype Representation 245
3.3 Recombination Operator 247
3.3.1 Arithmetic Crossover 247
3.3.2 Multipoint Crossover 247
3.3.3 Local Intermediate Crossover 248
3.4 Mutation Operator 249
3.4.1 Gaussian Mutation 249
3.4.2 Random Mutation 249
3.4.3 Post-Processing Based on Local Random Mutation 250
3.5 Fitness Function 250
3.6 Selection Operator 251
3.6.1 Tournament Selection 251
3.6.2 Elitism 251
3.7 Stopping Criteria 252
4 Incremental Algorithm for Neurofuzzy Network Learning 252
4.1 Numerical and Fuzzy Data 253
4.2 Network Architecture 253
4.3 Fuzzy Neuron 255
4.3.1 Triangular Norm and Conorm 256
4.3.2 Neuron Model 256
4.4 Granular Region 257
4.5 Granularity Adaptation 258
4.6 Developing Granules 258
4.7 Adapting Connection Weights 260
4.8 Learning Algorithm 260
5 Results and Discussion 261
5.1 Preliminaries 261
5.2 Genetic EANN for Fault Detection 262
5.3 Incremental EGNN for Fault Detection 266
5.4 Comparative Analyses and Discussion 269
6 Conclusion 271
References 272
Evolving Fuzzy Model for Fault Detection and Fault Identification of Dynamic Processes 275
1 Introduction 275
2 Evolving Fuzzy Model 277
2.1 Fuzzy Cloud-Based Model Structure 277
2.2 Evolving Mechanism 279
3 Fault Detection and Identification 280
3.1 Learning/Training Phase 280
3.2 Fault Detection Phase 280
3.3 Fault Identification Phase 281
4 Description of the HVAC Process Model 282
4.1 Possible Faults on HVAC System 284
5 Experimental Results 285
6 Conclusion 287
References 289
An Online RFID Localization in the Manufacturing Shopfloor 292
1 Introduction 292
2 RFID Localization System 295
3 eT2QFNN Architecture 296
3.1 Input Layer 298
3.2 Quantum Layer 298
3.3 Rule Layer 298
3.4 Output Processing Layer 299
3.5 Output Layer 299
4 eT2QFNN Learning Policy 300
4.1 Rule Growing Mechanism 301
4.2 Parameter Adjustment 303
4.2.1 Fuzzy Rule Initialization 304
4.2.2 Winning Rule Update 305
5 Experiments and Data Analysis 309
5.1 Experiment Setup 309
5.2 Comparison with Existing Results 310
6 Conclusions 312
References 313
Part II Prognostics and Forecasting 315
Physical Model-Based Prognostics and Health Monitoring to Enable Predictive Maintenance 316
1 Introduction 316
2 Challenges in the Field of Predictive Maintenance 317
2.1 Combining Diagnosis and Prognosis 317
2.2 System Versus Component Level 318
2.3 Monitoring of Usage, Loads, Condition or Health 319
2.4 Interpretation of Monitoring Data 320
2.5 Data-Driven or Model-Based Prognostics 320
2.6 Selection of Most Suitable Approach and Technique 321
2.7 Data Quality 322
3 Structural Health and Condition Monitoring 323
3.1 Sensors 323
3.2 Vibration and Vibration-Based Monitoring 325
4 Physical Model-Based Prognostics 328
4.1 Relation Between Usage, Loads and Degradation Rate 329
4.2 Developing a Prognostic Method 330
4.3 Comparison to Data-Driven Approaches 334
5 Decision Support Tools 335
5.1 Guidelines for Selecting Suitable Approach 335
5.2 Critical Part Selection 338
6 Case Studies 341
6.1 Maritime Systems 341
6.2 Railway Infrastructure 344
6.3 Wind Turbines 345
7 Conclusions 351
References 353
On Prognostic Algorithm Design and Fundamental Precision Limits in Long-Term Prediction 357
1 Introduction 357
2 Cramér–Rao Lower Bounds 358
2.1 Bayesian Cramér–Rao Lower Bounds 359
2.2 BCRLBs for Discrete-Time Dynamical Systems 359
3 Methodology for Prognostic Algorithm Design 361
3.1 Conditional Predictive Bayesian Cramér–Rao Lower Bounds 363
3.2 Analytic Computation of MCP-BCRLBs 366
4 Case Study: End-of-Discharge Time Prognosis of Lithium-Ion Batteries 367
4.1 State-Space Model 367
4.2 Prognostic Algorithm 369
4.3 Avoiding Monte Carlo Simulations in EoD Prognostic Algorithms 370
4.4 Prognostic Algorithm Design: Known Future Operating Profiles 371
4.5 Prognostic Algorithm Design: Statistical Characterizations of Future Operating Profiles 377
5 Conclusions 380
Acronyms 380
References 381
Performance Degradation Monitoring and Quantification: A Wastewater Treatment Plant Case Study 382
1 Introduction 382
1.1 Energy Consumption on WWTPs 383
1.2 Energy Savings Through Maintenance 384
2 Methodology 386
3 Results 388
3.1 External Recirculation Pumping System 388
3.1.1 Experimental Setup 389
3.1.2 Application of the Methodology 389
3.1.3 Results and Discussion 390
3.2 Plant Input Pumping System 392
3.2.1 Experimental Setup 393
3.2.2 Application of the Methodology 393
3.2.3 Results and Discussion 394
3.3 Aeration System Blowers 396
3.3.1 Experimental Setup 396
3.3.2 Application of the Methodology 397
3.3.3 Results and Discussion 399
4 Conclusions and Future Works 400
References 401
Fuzzy Rule-Based Modeling for Interval-Valued Data: An Application to High and Low Stock Prices Forecasting 403
1 Introduction 403
2 Interval Arithmetic 407
3 Interval Fuzzy Rule-Based Model 408
3.1 Interval Participatory Learning Fuzzy Clustering with Adaptive Distances 409
3.2 Rules Consequent Parameters Identification 411
3.3 iFRB Identification Procedure 412
4 Computational Experiments 413
4.1 Performance Assignment 414
4.2 Empirical Results 416
5 Conclusion 421
References 422
Part III Diagnosis, Optimization and Control 425
Reasoning from First Principles for Self-adaptive and Autonomous Systems 426
1 Introduction 426
2 Example 428
3 Model-Based Reasoning 431
3.1 Model-Based Diagnosis 433
3.2 Abductive Diagnosis 438
3.3 Summary on Model-Based Reasoning for Diagnosis 441
4 Modeling for Diagnosis and Repair 442
5 Self-adaptation Using Models 447
6 Related Research 454
7 Conclusions 455
References 456
Decentralized Modular Approach for Fault Diagnosis of a Class of Hybrid Dynamic Systems: Application to a Multicellular Converter 460
1 Learning from Data Streams 460
1.1 Basic Definitions and Motivation 460
1.2 State of the Art 461
1.3 Contribution of the Proposed Approach 462
2 Proposed Approach 463
2.1 System Decomposition 463
2.2 Discrete Component Modeling 466
2.3 Residual Generation Based on System Continuous Dynamics 468
2.4 Enriched Local Models Building 470
2.5 Local Hybrid Diagnoser Construction 471
2.6 Equivalence Between Centralized and Decentralized Diagnosis Structures 472
2.7 Computation Complexity Analysis 474
3 Experimental Results 475
4 Conclusion 479
References 481
Automated Process Optimization in Manufacturing Systems Based on Static and Dynamic Prediction Models 483
1 Introduction 483
1.1 Our Approach 485
2 Problem Statement 486
2.1 Process Optimization Based on Parameters 486
2.2 Process Optimization Based on Process Values Trends 488
3 Establishment of Predictive Models 492
3.1 Iterative Construction of Static Predictive Mappings (Parameters Quality) 492
3.1.1 Expert Knowledge Initialization 493
3.1.2 Hybrid Design of Experiments (HDoE) 493
3.1.3 Predictive Mapping Models Construction 495
3.2 Time-Series-Based Forecast Models (Process Values Quality) Learning and Adaptation 496
3.2.1 Learning by a Nonlinear (Fuzzy) Version of PLS (PLS-Fuzzy) 497
3.2.2 On-Line Model Adaptation with Increased Flexibility 500
4 Process Optimization with Predictive Models 504
4.1 Static Case (Mappings as Surrogates) 504
4.1.1 Evolutionary Algorithms for Solving Many-Objective Optimization Problems 505
4.1.2 A New Efficient Method for Multi-Objective EA (DECMO2) 506
4.2 Dynamic Case (Time-Series-Based Forecast Models as Surrogates) 507
4.2.1 Optimization Strategies 507
4.2.2 Reducing Dimensionality of the Optimization Space 509
5 Some Results from a Chip Production Process 510
5.1 Application Scenario 510
5.2 Experimental Setup and Data Collection 511
5.3 Results 514
5.3.1 Static Phase (Based on Process Parameter Settings) 514
5.3.2 Dynamic Case (Based on Time-Series of Process Values) 519
6 Conclusion and Outlook 524
References 526
Distributed Chance-Constrained Model Predictive Control for Condition-Based Maintenance Planning for Railway Infrastructures 530
1 Introduction 530
2 Preliminaries 532
2.1 Hybrid and Distributed MPC 532
2.2 Chance-Constrained MPC 533
3 Problem Formulation 534
3.1 Deterioration Model 534
3.2 Local Chance-Constrained MPC Problem 535
3.3 Two-Stage Robust Scenario-Based Approach 536
3.4 MLD-MPC Problem 538
4 Distributed Optimization 538
4.1 Dantzig-Wolfe Decomposition 539
4.2 Constraint Tightening 540
5 Case Studies 541
5.1 Settings 541
5.2 Representative Run 543
5.3 Computational Comparisons 544
5.4 Comparison with Alternative Approaches 545
6 Conclusions and Future Work 547
Appendix 547
Parameters for Case Study 547
Cyclic Approach 548
References 549
Index 552