Tested and proven strategy to develop optimal automated process fault analyzers Process fault analyzers monitor process operations in order to identify the underlying causes of operational problems. Several diagnostic strategies exist for automating process fault analysis; however, automated fault analysis is still not widely used within the processing industries due to problems of cost and performance as well as the difficulty of modeling process behavior at needed levels of detail. In response, this book presents the method of minimal evidence (MOME), a model-based diagnostic strategy that facilitates the development and implementation of optimal automated process fault analyzers. MOME was created at the University of Delaware by the researchers who developed the FALCON system, a real-time, online process fault analyzer. The authors demonstrate how MOME is used to diagnose single and multiple fault situations, determine the strategic placement of process sensors, and distribute fault analyzers within large processing systems. Optimal Automated Process Fault Analysis begins by exploring the need to automate process fault analysis. Next, the book examines: Logic of model-based reasoning as used in MOME MOME logic for performing single and multiple fault diagnoses Fuzzy logic algorithms for automating MOME Distributing process fault analyzers throughout large processing systems Virtual SPC analysis and its use in FALCONEER? IV Process state transition logic and its use in FALCONEER? IV The book concludes with a summary of the lessons learned by employing FALCONEER? IV in actual process applications, including the benefits of "intelligent supervision" of process operations. With this book as their guide, readers have a powerful new tool for ensuring the safety and reliability of any chemical processing system.

Table of Contents:
Foreword xiii Preface xv Acknowledgments xix 1 Motivations for Automating Process Fault Analysis 1 1.1 Introduction 1 1.2 CPI Trends to Date 1 1.3 The Changing Role of Process Operators in Plant Operations 3 1.4 Methods Currently Used to Perform Process Fault Management 5 1.5 Limitations of Human Operators in Performing Process Fault Management 10 1.6 The Role of Automated Process Fault Analysis 12 1.7 Anticipated Future CPI Trends 13 1.8 Process Fault Analysis Concept Terminology 14 References 16 2 Method of Minimal Evidence: Model-Based Reasoning 21 2.1 Overview 21 2.2 Introduction 22 2.3 Method of Minimal Evidence Overview 23 2.3.1 Process Model and Modeling Assumption Variable Classifications 28 2.3.2 Example of a MOME Primary Model 31 2.3.3 Example of MOME Secondary Models 36 2.3.4 Primary Model Residuals’ Normal Distributions 39 2.3.5 Minimum Assumption Variable Deviations 41 2.3.6 Primary Model Derivation Issues 44 2.3.7 Method for Improving the Diagnostic Sensitivity of the Resulting Fault Analyzer 47 2.3.8 Intermediate Assumption Deviations, Process Noise, and Process Transients 48 2.4 Verifying the Validity and Accuracy of the Various Primary Models 49 2.5 Summary 51 References 52 3 Method of Minimal Evidence: Diagnostic Strategy Details 55 3.1 Overview 55 3.2 Introduction 56 3.3 MOME Diagnostic Strategy 57 3.3.1 Example of MOME SV&PFA Diagnostic Rules’ Logic 57 3.3.2 Example of Key Performance Indicator Validation 67 3.3.3 Example of MOME SV&PFA Diagnostic Rules with Measurement Redundancy 71 3.3.4 Example of MOME SV&PFA Diagnostic Rules for Interactive Multiple-Faults 74 3.4 General Procedure for Developing and Verifying Competent Model-Based Process Fault Analyzers 79 3.5 MOME SV&PFA Diagnostic Rules’ Logic Compiler Motivations 80 3.6 MOME Diagnostic Strategy Summary  83 References 84 4 Method of Minimal Evidence: Fuzzy Logic Algorithm 87 4.1 Overview 87 4.2 Introduction 88 4.3 Fuzzy Logic Overview 90 4.4 MOME Fuzzy Logic Algorithm 91 4.4.1 Single-Fault Fuzzy Logic Diagnostic Rule 93 4.4.2 Multiple-Fault Fuzzy Logic Diagnostic Rule 97 4.5 Certainty Factor Calculation Review 102 4.6 MOME Fuzzy Logic Algorithm Summary 104 References 105 5 Method of Minimal Evidence: Criteria for Shrewdly Distributing Fault Analyzers and Strategic Process Sensor Placement 109 5.1 Overview 109 5.2 Criteria for Shrewdly Distributing Process Fault Analyzers 109 5.2.1 Introduction 110 5.2.2 Practical Limitations on Target Process System Size 110 5.2.3 Distributed Fault Analyzers 112 5.3 Criteria for Strategic Process Sensor Placement 113 References 114 6 Virtual SPC Analysis and Its Routine Use in FALCONEERTM IV 117 6.1 Overview 117 6.2 Introduction 118 6.3 EWMA Calculations and Specific Virtual SPC Analysis Configurations 118 6.3.1 Controlled Variables 119 6.3.2 Uncontrolled Variables and Performance Equation Variables 120 6.4 Virtual SPC Alarm Trigger Summary 123 6.5 Virtual SPC Analysis Conclusions 124 References 124 7 Process State Transition Logic and Its Routine Use in FALCONEERTM IV 125 7.1 Temporal Reasoning Philosophy 125 7.2 Introduction 126 7.3 State Identification Analysis Currently Used in FALCONEERTM IV 128 7.4 State Identification Analysis Summary 131 References 131 8 Conclusions 133 8.1 Overview 133 8.2 Summary of the MOME Diagnostic Strategy 133 8.3 FALCON, FALCONEER, and FALCONEERTM IV Actual KBS Application Performance Results 134 8.4 FALCONEERTM IV KBS Application Project Procedure 136 8.5 Optimal Automated Process Fault Analysis Conclusions 138 References 139 Appendix A Various Diagnostic Strategies for Automating Process Fault Analysis 141 A.1 Introduction 141 A.2 Fault Tree Analysis 142 A.3 Alarm Analysis 143 A.4 Decision Tables 143 A.5 Sign-Directed Graphs 144 A.6 Diagnostic Strategies Based on Qualitative Models 145 A.7 Diagnostic Strategies Based on Quantitative Models 145 A.8 Artificial Neural Network Strategies 147 A.9 Knowledge-Based System Strategies 147 A.10 Methodology Choice Conclusions 148 References 149 Appendix B The FALCON Project 163 B.1 Introduction 163 B.2 Overview 164 B.3 The Diagnostic Philosophy Underlying the FALCON System 164 B.4 Target Process System 165 B.5 The FALCON System 167 B.5.1 The Inference Engine 168 B.5.2 The Human–Machine Inference 169 B.5.3 The Dynamic Simulation Model 169 B.5.4 The Diagnostic Knowledge Base 172 B.6 Derivation of the FALCON Diagnostic Knowledge Base 173 B.6.1 First Rapid Prototype of the FALCON System KBS 173 B.6.2 FALCON System Development 173 B.6.3 The FALCON System’s Performance Results 182 B.7 The Ideal FALCON System 183 B.8 Use of the Knowledge-Based System Paradigm in Problem Solving 184 References 185 Appendix C Process State Transition Logic Used by the Original FALCONEER KBS 187 C.1 Introduction 187 C.2 Possible Process Operating States 187 C.3 Significance of Process State Identification and Transition Detection 189 C.4 Methodology for Determining Process State Identification 189 C.4.1 Present-Value States of All Key Sensor Data 189 C.4.2 Predicted Next-Value States of All Key Sensor Data 190 C.5 Process State Identification and Transition Logic Pseudocode 191 C.5.1 Attributes of the Current Data Vector 191 C.5.2 Method Applied to Each Data Vector 192 C.6 Summary 196 Appendix D FALCONEERTM IV Real-Time Suite Process Performance Solutions Demos 197 D.1 FALCONEERTM IV Demos Overview 197 D.2 FALCONEERTM IV Demos 197 D.2.1 Wastewater Treatment Process Demo 197 D.2.2 Pulp and Paper Stock Chest Demo 199 Index 203