| Theory of Grey-box Process Identification | |
| Prospects and Problems | p. 3 |
| Introduction | p. 3 |
| White, Black, and Grey Boxes | p. 4 |
| White-box Identification | p. 5 |
| Black-box Identification | p. 6 |
| Grey-box Identification | p. 10 |
| Basic Questions | p. 13 |
| Calibration | p. 14 |
| How to Specify a Model Set | p. 15 |
| ...and a Way to Get Answers | p. 17 |
| Tools for Grey-box Identification | p. 18 |
| Available Tools | p. 18 |
| Tools that Need to Be Developed | p. 21 |
| The MoCaVa Solution | p. 23 |
| The Model Set | p. 23 |
| Time Variables and Sampling | p. 24 |
| Process, Environment, and Data Interfaces | p. 25 |
| Multi-component Models | p. 27 |
| Expanding a Model Class | p. 29 |
| The Modelling Shell | p. 31 |
| Argument Relations and Attributes | p. 34 |
| Graphic Representations | p. 37 |
| Prior Knowledge | p. 41 |
| Hypotheses | p. 42 |
| Credibility Ranking | p. 43 |
| Model Classes with Inherent Conservation Law | p. 43 |
| Modelling 'Actuators' | p. 44 |
| Modelling 'Input Noise' | p. 46 |
| Standard I/O Interface Models | p. 49 |
| Fitting and Falsification | p. 51 |
| The Loss Function | p. 52 |
| Nesting and Fair Tests | p. 54 |
| Evaluating Loss and its Derivatives | p. 55 |
| Predictor | p. 56 |
| Equivalent Discrete-time Model | p. 56 |
| Performance Optimization | p. 57 |
| Controlling the Updating of Sensitivity Matrices | p. 58 |
| Exploiting the Sparsity of Sensitivity Matrices | p. 59 |
| Using Performance Optimization | p. 60 |
| Search Routine | p. 62 |
| Applicability | p. 65 |
| Applications | p. 65 |
| A Method for Grey-box Model Design | p. 67 |
| What is Expected from the User? | p. 68 |
| Limitations of MoCaVa | p. 69 |
| Diagnostic Tools | p. 69 |
| What Can Go Wrong? | p. 71 |
| Tutorial on MoCaVa | |
| Preparations | p. 77 |
| Getting Started | p. 77 |
| System Requirements | p. 77 |
| Downloading | p. 77 |
| Installation | p. 77 |
| Starting MoCaVa | p. 78 |
| The HTML User's Manual | p. 78 |
| The 'Raw' Data File | p. 78 |
| Making a Data File for MoCaVa | p. 78 |
| Calibration | p. 83 |
| Creating a New Project | p. 83 |
| The User's Guide and the Pilot Window | p. 85 |
| Specifying the Data Sample | p. 86 |
| The Time Range Window | p. 86 |
| Creating a Model Component | p. 88 |
| Handling the Component Library | p. 89 |
| Entering Component Statements | p. 90 |
| Classifying Arguments | p. 92 |
| Specifying I/O Interfaces | p. 95 |
| Specifying Argument Attributes | p. 98 |
| Specifying Implicit Attributes | p. 100 |
| Assigning Data | p. 100 |
| Specifying Model Class | p. 101 |
| Simulating | p. 103 |
| Setting the Origin of the Free Parameter Space | p. 103 |
| Selecting Variables to be Plotted | p. 104 |
| Appraising Model Class | p. 105 |
| Handling Data Input | p. 106 |
| Fitting a Tentative Model Structure | p. 107 |
| Search Parameters | p. 108 |
| Appraising the Search Result | p. 111 |
| Testing a Tentative Model Structure | p. 113 |
| Appraising a Tentative Model | p. 116 |
| Nesting | p. 118 |
| Interpreting the Test Results | p. 119 |
| Refining a Tentative Model Structure | p. 121 |
| Multiple Alternative Structures | p. 122 |
| Augmenting a Disturbance Model | p. 124 |
| Checking the Final Model | p. 132 |
| Terminals and 'Stubs' | p. 134 |
| Copying Components | p. 135 |
| Effects of Incorrect Disturbance Structure | p. 138 |
| Exporting/Importing Parameters | p. 140 |
| Suspending and Exiting | p. 141 |
| The Score Table | p. 142 |
| Resuming a Suspended Session | p. 143 |
| Checking Integration Accuracy | p. 143 |
| Some Modelling Support | p. 147 |
| Modelling Feedback | p. 147 |
| The Model Class | p. 148 |
| User's Functions and Library | p. 153 |
| Rescaling | p. 154 |
| Importing External Models | p. 159 |
| Using Dymola2+TM as Modelling Tool for MoCaVa | p. 160 |
| Detecting Over-parametrization | p. 166 |
| Assigning Variable Input to Imported Models | p. 170 |
| Selective Connection of Arguments to Dymola2+TM Models | p. 173 |
| Case Studies | |
| Case 1: Rinsing of the Steel Strip in a Rolling Mill | p. 185 |
| Background | p. 185 |
| Step 1: A Phenomenological Description | p. 185 |
| The Process Proper | p. 185 |
| The Measurement Gauges | p. 188 |
| The Input | p. 189 |
| Step 2: Variables and Causality | p. 189 |
| The variables | p. 189 |
| Cause and effect | p. 190 |
| Data Preparation | p. 191 |
| Relations to Measured Variables | p. 192 |
| Step 3: Modelling | p. 194 |
| Basic Mass Balances | p. 194 |
| Strip Input | p. 201 |
| Step 4: Calibration | p. 203 |
| Refining the Model Class | p. 206 |
| The Squeezer Rolls | p. 206 |
| The Entry Rolls | p. 211 |
| Continuing Calibration | p. 213 |
| Refining the Model Class Again | p. 215 |
| Ventilation | p. 215 |
| More Hypothetical Improvements | p. 217 |
| Effective Mixing Volumes | p. 217 |
| Avoiding the pitfall of 'Data Description' | p. 219 |
| Modelling Disturbances | p. 222 |
| Pickling | p. 222 |
| State Noise | p. 223 |
| Determining the Simplest Environment Model | p. 225 |
| Variable Input Acid Concentration | p. 225 |
| Unexplained Variation in Residual Acid Concentration | p. 225 |
| Checking for Possible Over-fitting | p. 229 |
| Appraising Roller Conditions | p. 233 |
| Conclusions from the Calibration Session | p. 233 |
| Case 2: Quality Prediction in a Cardboard Making Process | p. 235 |
| Background | p. 235 |
| Step 1: A Phenomenological Description | p. 235 |
| Data Preparation | p. 237 |
| Step 2: Variables and Causality | p. 244 |
| Relations to Measured Variables | p. 247 |
| Step 3: Modelling | p. 248 |
| The Bending Stiffness | p. 248 |
| The Paper Machine | p. 253 |
| The Pulp Feed | p. 260 |
| Control Input | p. 262 |
| The Pulp Mixing | p. 265 |
| Pulp Input | p. 267 |
| The Pulp Constituents | p. 269 |
| Step 4: Calibration | p. 271 |
| Expanding the Tentative Model Class | p. 279 |
| The Pulp Refining | p. 279 |
| The Mixing-tank Dynamics | p. 284 |
| The Machine Chests | p. 287 |
| Filtering the "Kappa" Input | p. 289 |
| Checking for Over-fitting: The SBE Rule | p. 290 |
| Ending a Calibration Session | p. 293 |
| 'Black-box' vs 'White-box' Extensions | p. 293 |
| Determination vs Randomness | p. 294 |
| Modelling Disturbances | p. 295 |
| Calibrating Models with Stochastic Input | p. 296 |
| Determination vs Randomness Revisited | p. 299 |
| A Local Minimum | p. 304 |
| Conclusions from the Calibration Session | p. 306 |
| Mathematics and Algorithms | p. 313 |
| The Model Classes | p. 313 |
| The Loss Derivatives | p. 316 |
| The ODE Solver | p. 317 |
| The Reference Trajectory | p. 317 |
| The State Deviation | p. 318 |
| The Equivalent Discrete-time Sensitivity Matrices | p. 318 |
| The Predictor | p. 321 |
| The Equivalent Discrete-time Model | p. 322 |
| Mixed Algebraic and Differential Equations | p. 322 |
| Performance Optimization | p. 326 |
| The Sensitivity Update Control Function | p. 327 |
| Memoization | p. 330 |
| The Search Routine | p. 330 |
| Library Routines | p. 331 |
| Output Conversion | p. 331 |
| Input Interpolators | p. 331 |
| Input Filters | p. 334 |
| Disturbance Models | p. 335 |
| The Advanced Specification Window | p. 337 |
| Optimization for Speed | p. 337 |
| User's Checkpoints | p. 338 |
| Internal Integration Interval | p. 338 |
| Debugging | p. 339 |
| Glossary | p. 341 |
| References | p. 345 |
| Index | p. 349 |
| Table of Contents provided by Ingram. All Rights Reserved. |
