Advanced Chemical Process Control

1st Edition Morten Hovd

Visit to download the full and correct content document:
https://ebookmass.com/product/advanced-chemical-process-control-1st-edition-morte
n-hovd/
 WILEY END USER LICENSE AGREEMENT
Go to www.wiley.com/go/eula to access Wiley’s ebook EULA.
Advanced Chemical Process Control
Advanced Chemical Process Control

Putting Theory into Practice

Morten Hovd
Author All books published by WILEY-VCH are carefully
produced. Nevertheless, authors, editors, and
Prof. Morten Hovd publisher do not warrant the information
Norwegian University of Science and contained in these books, including this book,
Technology to be free of errors. Readers are advised to keep
7491 Trondheim in mind that statements, data, illustrations,
Norway procedural details or other items may
inadvertently be inaccurate.
Cover Image: © Voranee/Shutterstock
Library of Congress Card No.: applied for

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available
from the British Library.

Bibliographic information published by

the Deutsche Nationalbibliothek
The Deutsche Nationalbibliothek lists
this publication in the Deutsche
Nationalbibliografie; detailed bibliographic
data are available on the Internet at
<http://dnb.d-nb.de>.

© 2023 WILEY-VCH GmbH, Boschstr. 12,

69469 Weinheim, Germany

All rights reserved (including those of

translation into other languages). No part of
this book may be reproduced in any form – by
photoprinting, microfilm, or any other means –
nor transmitted or translated into a machine
language without written permission from the
publishers. Registered names, trademarks, etc.
used in this book, even when not specifically
marked as such, are not to be considered
unprotected by law.

Print ISBN: 978-3-527-35223-4

ePDF ISBN: 978-3-527-84247-6
ePub ISBN: 978-3-527-84248-3
oBook ISBN: 978-3-527-84249-0

Typesetting Straive, Chennai, India

To Ellen
 vii

Contents

Preface xvii
Acknowledgments xxi
Acronyms xxiii
Introduction xxv

1 Mathematical and Control Theory Background 1

1.1 Introduction 1
1.2 Models for Dynamical Systems 1
1.2.1 Dynamical Systems in Continuous Time 1
1.2.2 Dynamical Systems in Discrete Time 2
1.2.3 Linear Models and Linearization 3
1.2.3.1 Linearization at a Given Point 3
1.2.3.2 Linearizing Around a Trajectory 6
1.2.4 Converting Between Continuous- and Discrete-Time Models 6
1.2.4.1 Time Delay in the Manipulated Variables 7
1.2.4.2 Time Delay in the Measurements 9
1.2.5 Laplace Transform 9
1.2.6 The z Transform 10
1.2.7 Similarity Transformations 11
1.2.8 Minimal Representation 11
1.2.9 Scaling 14
1.3 Analyzing Linear Dynamical Systems 15
1.3.1 Transfer Functions of Composite Systems 15
1.3.1.1 Series Interconnection 15
1.3.1.2 Parallel Systems 16
1.3.1.3 Feedback Connection 16
1.3.1.4 Commonly Used Closed-Loop Transfer Functions 17
1.3.1.5 The Push-Through Rule 17
1.4 Poles and Zeros of Transfer Functions 18
1.4.1 Poles of Multivariable Systems 19
1.4.2 Pole Directions 19
1.4.3 Zeros of Multivariable Systems 20
1.4.4 Zero Directions 22
viii Contents

1.5 Stability 23
1.5.1 Poles and Zeros of Discrete-Time Transfer Functions 23
1.5.2 Frequency Analysis 24
1.5.2.1 Steady-State Phase Adjustment 26
1.5.3 Bode Diagrams 27
1.5.3.1 Bode Diagram Asymptotes 27
1.5.3.2 Minimum Phase Systems 29
1.5.3.3 Frequency Analysis for Discrete-Time Systems 30
1.5.4 Assessing Closed-Loop Stability Using the Open-Loop Frequency
Response 31
1.5.4.1 The Principle of the Argument and the Nyquist D-Contour 31
1.5.4.2 The Multivariable Nyquist Theorem 32
1.5.4.3 The Monovariable Nyquist Theorem 32
1.5.4.4 The Bode Stability Criterion 32
1.5.4.5 Some Remarks on Stability Analysis Using the Frequency Response 35
1.5.4.6 The Small Gain Theorem 36
1.5.5 Controllability 37
1.5.6 Observability 38
1.5.7 Some Comments on Controllability and Observability 39
1.5.8 Stabilizability 40
1.5.9 Detectability 40
1.5.10 Hidden Modes 41
1.5.11 Internal Stability 41
1.5.12 Coprime Factorizations 43
1.5.12.1 Inner–Outer Factorization 44
1.5.12.2 Normalized Coprime Factorization 44
1.5.13 Parametrization of All Stabilizing Controllers 44
1.5.13.1 Stable Plants 45
1.5.13.2 Unstable Plants 45
1.5.14 Hankel Norm and Hankel Singular Values 46
Problems 47
References 49

2 Control Configuration and Controller Tuning 51

2.1 Common Control Loop Structures for the Regulatory Control Layer 51
2.1.1 Simple Feedback Loop 51
2.1.2 Feedforward Control 51
2.1.3 Ratio Control 54
2.1.4 Cascade Control 54
2.1.5 Auctioneering Control 55
2.1.6 Split Range Control 56
2.1.7 Input Resetting Control 57
2.1.8 Selective Control 59
2.1.9 Combining Basic Single-Loop Control Structures 60
2.1.10 Decoupling 61
 Contents ix

2.2 Input and Output Selection 62

2.2.1 Using Physical Insights 63
2.2.2 Gramian-Based Input and Output Selection 64
2.2.3 Input/Output Selection for Stabilization 65
2.3 Control Configuration 66
2.3.1 The Relative Gain Array 66
2.3.2 The RGA as a General Analysis Tool 68
2.3.2.1 The RGA and Zeros in the Right Half-Plane 68
2.3.2.2 The RGA and the Optimally Scaled Condition Number 68
2.3.2.3 The RGA and Individual Element Uncertainty 69
2.3.2.4 RGA and Diagonal Input Uncertainty 69
2.3.2.5 The RGA as an Interaction Measure 70
2.3.3 The RGA and Stability 70
2.3.3.1 The RGA and Pairing of Controlled and Manipulated Variables 71
2.3.4 Summary of RGA-Based Input–Output Pairing 72
2.3.5 Partial Relative Gains 72
2.3.6 The Niederlinski Index 73
2.3.7 The Rijnsdorp Interaction Measure 73
2.3.8 Gramian-Based Input–Output Pairing 74
2.3.8.1 The Participation Matrix 75
2.3.8.2 The Hankel Interaction Index Array 75
2.3.8.3 Accounting for the Closed-Loop Bandwidth 76
2.4 Tuning of Decentralized Controllers 76
2.4.1 Introduction 76
2.4.2 Loop Shaping Basics 77
2.4.3 Tuning of Single-Loop Controllers 79
2.4.3.1 PID Controller Realizations and Common Modifications 79
2.4.3.2 Controller Tuning Using Frequency Analysis 81
2.4.3.3 Ziegler–Nichols Closed-Loop Tuning Method 86
2.4.3.4 Simple Fitting of a Step Response Model 86
2.4.3.5 Ziegler–Nichols Open-Loop Tuning 88
2.4.3.6 IMC-PID Tuning 88
2.4.3.7 Simple IMC Tuning 89
2.4.3.8 The Setpoint Overshoot Method 91
2.4.3.9 Autotuning 95
2.4.3.10 When Should Derivative Action Be Used? 95
2.4.3.11 Effects of Internal Controller Scaling 96
2.4.3.12 Reverse Acting Controllers 97
2.4.4 Gain Scheduling 97
2.4.5 Surge Attenuating Controllers 98
2.4.6 Multiloop Controller Tuning 99
2.4.6.1 Independent Design 100
2.4.6.2 Sequential Design 102
2.4.6.3 Simultaneous Design 103
2.4.7 Tools for Multivariable Loop-Shaping 103
x Contents

2.4.7.1 The Performance Relative Gain Array 103

2.4.7.2 The Closed-Loop Disturbance Gain 104
2.4.7.3 Illustrating the Use of CLDG’s for Controller Tuning 104
2.4.7.4 Unachievable Loop Gain Requirements 107
Problems 108
References 112

3 Control Structure Selection and Plantwide Control 115

3.1 General Approach and Problem Decomposition 115
3.1.1 Top-Down Analysis 115
3.1.1.1 Defining and Exploring Optimal Operation 115
3.1.1.2 Determining Where to Set the Throughput 116
3.1.2 Bottom-Up Design 116
3.2 Regulatory Control 117
3.2.1 Example: Regulatory Control of Liquid Level in a Deaeration Tower 118
3.3 Determining Degrees of Freedom 121
3.4 Selection of Controlled Variables 122
3.4.1 Problem Formulation 123
3.4.2 Selecting Controlled Variables by Direct Evaluation of Loss 124
3.4.3 Controlled Variable Selection Based on Local Analysis 125
3.4.3.1 The Minimum Singular Value Rule 127
3.4.3.2 Desirable Characteristics of the Controlled Variables 128
3.4.4 An Exact Local Method for Controlled Variable Selection 128
3.4.5 Measurement Combinations as Controlled Variables 130
3.4.5.1 The Nullspace Method for Selecting Controlled Variables 130
3.4.5.2 Extending the Nullspace Method to Account for Implementation
Error 130
3.4.6 The Validity of the Local Analysis for Controlled Variable Selection 131
3.5 Selection of Manipulated Variables 132
3.5.1 Verifying that the Proposed Manipulated Variables Make Acceptable
Control Possible 133
3.5.2 Reviewing the Characteristics of the Proposed Manipulated
Variables 134
3.6 Selection of Measurements 135
3.7 Mass Balance Control and Throughput Manipulation 136
3.7.1 Consistency of Inventory Control 138
Problems 140
References 141

4 Limitations on Achievable Performance 143

4.1 Performance Measures 143
4.1.1 Time-Domain Performance Measures 143
4.1.2 Frequency-Domain Performance Measures 145
4.1.2.1 Bandwidth Frequency 145
4.1.2.2 Peaks of Closed-Loop Transfer Functions 146
 Contents xi

4.1.2.3 Bounds on Weighted System Norms 146

4.1.2.4 Gain and Phase Margin 147
4.2 Algebraic Limitations 148
4.2.1 S + T = I 148
4.2.2 Interpolation Constraints 148
4.2.2.1 Monovariable Systems 148
4.2.2.2 Multivariable Systems 149
4.3 Control Performance in Different Frequency Ranges 149
4.3.1 Sensitivity Integrals and Right Half-Plane Zeros 149
4.3.1.1 Multivariable Systems 150
4.3.2 Sensitivity Integrals and Right Half-Plane Poles 150
4.3.3 Combined Effects of RHP Poles and Zeros 150
4.3.4 Implications of the Sensitivity Integral Results 150
4.4 Bounds on Closed-Loop Transfer Functions 151
4.4.1 The Maximum Modulus Principle 152
4.4.1.1 The Maximum Modulus Principle 152
4.4.2 Minimum Phase and Stable Versions of the Plant 152
4.4.3 Bounds on S and T 153
4.4.3.1 Monovariable Systems 153
4.4.3.2 Multivariable Systems 153
4.4.4 Bounds on KS and KSGd 154
4.5 ISE Optimal Control 156
4.6 Bandwidth and Crossover Frequency Limitations 156
4.6.1 Bounds from ISE Optimal Control 156
4.6.2 Bandwidth Bounds from Weighted Sensitivity Minimization 157
4.6.3 Bound from Negative Phase 158
4.7 Bounds on the Step Response 158
4.8 Bounds for Disturbance Rejection 160
4.8.1 Inputs for Perfect Control 161
4.8.2 Inputs for Acceptable Control 161
4.8.3 Disturbances and RHP Zeros 161
4.8.4 Disturbances and Stabilization 162
4.9 Limitations from Plant Uncertainty 164
4.9.1 Describing Uncertainty 165
4.9.2 Feedforward Control and the Effects of Uncertainty 166
4.9.3 Feedback and the Effects of Uncertainty 167
4.9.4 Bandwidth Limitations from Uncertainty 168
Problems 168
References 170

5 Model-Based Predictive Control 173

5.1 Introduction 173
5.2 Formulation of a QP Problem for MPC 175
5.2.1 Future States as Optimization Variables 179
5.2.2 Using the Model Equation to Substitute for the Plant States 180
xii Contents

5.2.3 Optimizing Deviations from Linear State Feedback 181

5.2.4 Constraints Beyond the End of the Prediction Horizon 182
5.2.5 Finding the Terminal Constraint Set 183
5.2.6 Feasible Region and Prediction Horizon 184
5.3 Step-Response Models 185
5.4 Updating the Process Model 186
5.4.1 Bias Update 186
5.4.2 Kalman Filter and Extended Kalman Filters 187
5.4.2.1 Augmenting a Disturbance Description 188
5.4.2.2 The Extended Kalman Filter 189
5.4.2.3 The Iterated Extended Kalman Filter 189
5.4.3 Unscented Kalman Filter 190
5.4.4 Receding Horizon Estimation 193
5.4.4.1 The Arrival Cost 195
5.4.4.2 The Filtering Formulation of RHE 196
5.4.4.3 The Smoothing Formulation of RHE 196
5.4.5 Concluding Comments on State Estimation 198
5.5 Disturbance Handling and Offset-Free Control 199
5.5.1 Feedforward from Measured Disturbances 199
5.5.2 Requirements for Offset-Free Control 199
5.5.3 Disturbance Estimation and Offset-Free Control 200
5.5.4 Augmenting the Model with Integrators at the Plant Input 203
5.5.5 Augmenting the Model with Integrators at the Plant Output 205
5.5.6 MPC and Integrator Resetting 208
5.6 Feasibility and Constraint Handling 210
5.7 Closed-Loop Stability with MPC Controllers 212
5.8 Target Calculation 213
5.9 Speeding up MPC Calculations 217
5.9.1 Warm-Starting the Optimization 218
5.9.2 Input Blocking 219
5.9.3 Enlarging the Terminal Region 220
5.10 Robustness of MPC Controllers 222
5.11 Using Rigorous Process Models in MPC 225
5.12 Misconceptions, Clarifications, and Challenges 226
5.12.1 Misconceptions 226
5.12.1.1 MPC Is Not Good for Performance 226
5.12.1.2 MPC Requires Very Accurate Models 227
5.12.1.3 MPC Cannot Prioritize Input Usage or Constraint Violations 227
5.12.2 Challenges 227
5.12.2.1 Obtaining a Plant Model 228
5.12.2.2 Maintenance 228
5.12.2.3 Capturing the Desired Behavior in the MPC Design 228
Problems 228
References 231
 Contents xiii

6 Some Practical Issues in Controller Implementation 233

6.1 Discrete-Time Implementation 233
6.1.1 Aliasing 233
6.1.2 Sampling Interval 233
6.1.3 Execution Order 235
6.2 Pure Integrators in Parallel 235
6.3 Anti-Windup 236
6.3.1 Simple PI Control Anti-Windup 237
6.3.2 Velocity Form of PI Controllers 237
6.3.3 Anti-Windup in Cascaded Control Systems 238
6.3.4 A General Anti-Windup Formulation 239
6.3.5 Hanus’ Self-Conditioned Form 240
6.3.6 Anti-Windup in Observer-Based Controllers 241
6.3.7 Decoupling and Input Constraints 243
6.3.8 Anti-Windup for “Normally Closed” Controllers 244
6.4 Bumpless Transfer 245
6.4.1 Switching Between Manual and Automatic Operation 245
6.4.2 Changing Controller Parameters 246
Problems 246
References 247

7 Controller Performance Monitoring and Diagnosis 249

7.1 Introduction 249
7.2 Detection of Oscillating Control Loops 251
7.2.1 The Autocorrelation Function 251
7.2.2 The Power Spectrum 252
7.2.3 The Method of Miao and Seborg 252
7.2.4 The Method of Hägglund 253
7.2.5 The Regularity Index 254
7.2.6 The Method of Forsman and Stattin 255
7.2.7 Prefiltering Data 255
7.3 Oscillation Diagnosis 256
7.3.1 Manual Oscillation Diagnosis 256
7.3.2 Detecting and Diagnosing Valve Stiction 257
7.3.2.1 Using the Cross-Correlation Function to Detect Valve Stiction 257
7.3.2.2 Histograms for Detecting Valve Stiction 258
7.3.2.3 Stiction Detection Using an OP–PV Plot 260
7.3.3 Stiction Compensation 262
7.3.4 Detection of Backlash 263
7.3.5 Backlash Compensation 264
7.3.6 Simultaneous Stiction and Backlash Detection 265
7.3.7 Discriminating Between External and Internally Generated
Oscillations 266
7.3.8 Detecting and Diagnosing Other Nonlinearities 266
xiv Contents

7.4 Plantwide Oscillations 269

7.4.1 Grouping Oscillating Variables 269
7.4.1.1 Spectral Principal Component Analysis 269
7.4.1.2 Visual Inspection Using High-Density Plots 269
7.4.1.3 Power Spectral Correlation Maps 270
7.4.1.4 The Spectral Envelope Method 271
7.4.1.5 Methods Based on Adaptive Data Analysis 272
7.4.2 Locating the Cause for Distributed Oscillations 273
7.4.2.1 Using Nonlinearity for Root Cause Location 273
7.4.2.2 The Oscillation Contribution Index 273
7.4.2.3 Estimating the Propagation Path for Disturbances 274
7.5 Control Loop Performance Monitoring 278
7.5.1 The Harris Index 278
7.5.2 Obtaining the Impulse Response Model 279
7.5.3 Calculating the Harris Index 280
7.5.4 Estimating the Deadtime 281
7.5.5 Modifications to the Harris Index 282
7.5.6 Assessing Feedforward Control 283
7.5.7 Comments on the Use of the Harris Index 285
7.5.8 Performance Monitoring for PI Controllers 286
7.6 Multivariable Control Performance Monitoring 287
7.6.1 Assessing Feedforward Control in Multivariable Control 287
7.6.2 Performance Monitoring for MPC Controllers 288
7.7 Some Issues in the Implementation of Control Performance
Monitoring 290
7.8 Discussion 290
Problems 291
References 291

8 Economic Control Benefit Assessment 297

8.1 Optimal Operation and Operational Constraints 297
8.2 Economic Performance Functions 298
8.3 Expected Economic Benefit 299
8.4 Estimating Achievable Variance Reduction 300
8.5 Worst-Case Backoff Calculation 300
References 301

A Fourier–Motzkin Elimination 303

B Removal of Redundant Constraints 307

Reference 308

C The Singular Value Decomposition 309

 Contents xv

D Factorization of Transfer Functions into Minimum Phase Stable and

All-Pass Parts 311
D.1 Input Factorization of RHP Zeros 312
D.2 Output Factorization of RHP Zeros 312
D.3 Output Factorization of RHP Poles 313
D.4 Input Factorization of RHP Poles 313
D.5 SISO Systems 314
D.6 Factoring Out Both RHP Poles and RHP Zeros 314
Reference 314

E Models Used in Examples 315

E.1 Binary Distillation Column Model 315
E.2 Fluid Catalytic Cracker Model 318
References 320

Index 321
 xvii

Preface

Half a century ago, Alan Foss [1] wrote his influential paper about the gap between
chemical process control theory and industrial application. Foss clearly put the
responsibility on the chemical process control theorists to close this gap. Since then,
several advances in control theory, some originating within academia, while others
originated in industry and was later adopted and further developed by academia,
have contributed to addressing the shortcomings of chemical process control theory,
as addressed by Foss:
● The extension of the relative gain array (RGA) to nonzero frequencies and

Graminan-based control structure selection tools have extended the toolkit for
designing control structures.1 Self-optimal control [4] provides a systematic
approach to identifying controlled variables for a chemical plant.
● Model predictive control (MPC) has great industrial success [3].

● Robustness to model error received significant research focus from the 1980s

onward.
● Control Performance Monitoring has, since the seminal paper by Harris [2],

resulted in tools both for monitoring and diagnosing the performance of

individual loops as well as for identifying the cause of plantwide disturbances.
The aforementioned list notwithstanding, many would agree to the claim that
there is still a wide gap between control theory and common industrial practice in
the chemical process industries. This book is the author’s attempt to contributing to
the reduction of that gap. This book has two ambitious objectives:
1. To make more advanced control accessible to chemical engineers, many of whom
will only have background from a single course in control. While this book is
unlikely to effortlessly turn a plant engineer into a control expert, it does contain
tools that many plant engineers should find useful. It is also hoped that it will give
the plant engineer more weight when discussing with control consultants – either
from within the company or external consultants.

1 This author is aware that many colleagues in academia are of the opinion that the
frequency-dependent RGA is not “rigorous.” This book will instead take the stance that the
dynamic RGA has proved useful and therefore should be disseminated. It is also noted that the
“counterexamples” where the steady-state RGA is claimed to propose a wrong control structure are
themselves flawed.
xviii Preface

2. To increase the understanding among control engineers (students or graduates

moving into the chemical process control area) of how to apply their theoretical
knowledge to practical problems in chemical processes.
The third approach to reducing the gap, i.e. to develop and present tools to sim-
plify the application of control, is not a focus of this book – although some colleagues
would surely claim that this is what proportional integral derivative (PI(D)) tuning
rules are doing.
The reader should note that this book does not start “from scratch,” and some
prior knowledge of control is expected. The introduction to the Laplace transform
is rudimentary at best, and much more detail could be included in the presenta-
tion of frequency analysis. Some knowledge of finite-dimensional linear algebra is
expected. Readers who have never seen a linear state-space model will face a hur-
dle. Still, the book should be accessible to readers with background from a course in
process control.
Readers with a more extensive knowledge of control theory may find the book
lacks rigor. Frequently, results are presented and discussed, without presenting
formal proofs. Readers interested in mathematical proofs will have to consult the
references. This is in line with this author’s intent to keep the focus on issues of
importance for industrial applications (without claiming to “know it all”).

The structure of the Book

This book has grown out of more than three decades of learning, teaching, and dis-
cussing the control of chemical processes. What has become clear is that process
control engineers are faced with a wide variety of tasks and problems. The chapters
of this book therefore address a range of different topics – most of which have been
the subject of entire books. The selection of material to include is therefore not trivial
nor obvious.
● Chapter 1 presents some mathematical and control theory background. Readers
with some knowledge of control may choose to skip this chapter and only return
to it to look up unfamiliar (or forgotten) concepts that appear in the rest of the
book. This chapter definitely has a more theoretical and less practical flavor that
much of the rest of the book.
● Chapter 2 addresses controller tuning for PI(D) controllers, as well as control
configuration. The term control configuration here covers both the control func-
tionality often encountered in the regulatory control layer (feedback, feedforward,
ratio control,...) and determining which input should be used to control which
output in a multi-loop control system.
● Chapter 3 focuses on determining what variables to use for control. Typically,
there are more variables that can be measured than can be manipulated, so the
most focus is given to the choice of variables to control.
● Chapter 4 presents limitations to achievable control performance. Clearly, if it
is not possible to achieve acceptable performance, it makes little sense trying to
design a controller. Understanding the limitations of achievable performance is
 Preface xix

also very useful when designing controllers using loop shaping, as presented in
Chapter 2.
● Chapter 5 is about MPC, which is the most popular advanced control type in
the chemical process industries.2 In addition to presenting MPC problem formu-
lations per se, important issues such as model updating, offset-free control, and
target calculation are also discussed.
● Chapter 6 presents some practical issues in controller implementation. This list
is far from complete, some of the issues included are well known and may be
considered trivial – but all are important.
● Chapter 7 addresses control performance monitoring (CPM). The number
of controllers in even a modestly sized chemical process is too large for plant
personnel to frequently monitor each of them, and automated tools are therefore
needed for the maintenance of the control. The chapter also includes some tools
for finding the root cause of distributed oscillations – oscillations originating at
one location in the plant will tend to propagate and disturb other parts of the
plant, and hence it is often nontrivial to find where the oscillation originates and
what remedial action to take.
● Chapter 8 addresses control benefit analysis, i.e. tools to a priori assess the eco-
nomic benefit that can be obtained from improved control. This author admits
that the chapter is disappointingly short. Developing generic tools to estimate such
economic benefit is indeed difficult. On the other hand, the inability to estimate
economic benefit with some certainty is also one of the major obstacles to more
pervasive application of advanced control in the chemical process industries.

What’s Not in the Book

Selecting what to cover in a textbook invariably requires leaving out interesting top-
ics. Some topics that are relevant in a more general setting, but which this book does
not make any attempt to cover, include
● Nonlinear control. Real-life systems are nonlinear. Nevertheless, this book almost
exclusively addresses linear control – with the main exception being the handling
of constraints in MPC.3 Linearization and linear control design suffices for most
control problems in the chemical process industries, and in other cases static non-
linear transforms of inputs or outputs can make more strongly nonlinear systems
closer to linear. The book also describes briefly approaches to nonlinear MPC
(with nonlinearity in the model, not only from constraints). Such complications
are needed mainly for control of batch processes (where continuous changes in the
operating point are unavoidable), or for processes frequently switching between
different operating regimes (such as wastewater treatment plants with anaerobic
and aerobic stages). Although linear control often suffices, it is clearly prudent to
verify a control design with nonlinear simulation and/or investigate the control at
different linearization points (if a nonlinear model is available).

2 And, indeed, the chemical process industries is where MPC was first developed.
3 Constraints represent a strong nonlinearity.
xx Preface

● Modeling and system identification. The availability of good system models are
of great importance to control design and verification. This book only briefly
describes how to fit a particularly simple monovariable model – more complete
coverage of these topics is beyond the scope of this book.
● Adaptive and learning control. While classical adaptive control seems to have been
out of favor in the chemical process industries for several decades, there is cur-
rently a lot of research on integrating machine learning with advanced control
such as MPC. This author definitely accepts the relevance of research on learn-
ing control but is of the opinion that at this stage a research monograph would be
more appropriate than a textbook for covering these developments.
In leaving out many of the more theoretically complex areas of control theory,
readers from a control engineering background may find the book title somewhat
puzzling – especially the word Advanced. Although some of the topics covered
by this book are relatively standard also within the domain of chemical process
control, this author would claim that much of the book covers topics that are
indeed advanced compared to common application practice in the chemical process
industries. The Introduction will hopefully enlighten readers from outside chemical
process control about the unique challenges faced by this application domain and
contribute to explaining the prevalence of linear control techniques.

August, 2022 M. Hovd

Trondheim, Norway

References

1 Foss, A.S. (1973). Critique of chemical process control theory. AIChE Journal
19: 209–214.
2 Harris, T.J. (1989). Assessment of control loop performance. The Canadian Journal
of Chemical Engineering 67: 856–861.
3 Qin, S.J. and Badgwell, T.A. (2003). A survey of industrial model predictive con-
trol technology. Control Engineering Practice 11 (7): 733–764.
4 Skogestad, S. (2000). Plantwide control: the search for the self-optimizing control
structure. Journal of Process Control 10: 487–507.
 xxi

Acknowledgments

It is noted in the preface that this book is a result of more than three decades of
learning, teaching, discussing, and practicing chemical process control. Therefore,
a large number of people have directly or indirectly influenced the contents of this
book, and it is impossible to mention them all. The person with the strongest such
influence is Professor Sigurd Skogestad at NTNU, whom I had the fortune to have
as my PhD supervisor. At that time, I had an extended research stay in the group
of Professor Manfred Morari (then at Caltech). Discussions these outstanding pro-
fessors as well as with my contemporaries as a PhD student, both in Skogestad’s
group and in Morari’s group, greatly enhanced my understanding of process con-
trol. I would particularly like to mention Elling W. Jacobsen, Petter Lundström, John
Morud, Richard D. Braatz, Jay H. Lee, and Frank Doyle.
In later years, I have benefited from discussions with a number of people, includ-
ing Kjetil Havre, Vinay Kariwala, Ivar Halvorsen, Krister Forsman, Alf Isaksson,
Tor A. Johansen, Lars Imsland, and Bjarne A. Foss. My own PhD candidates and
Postdocs have also enriched my understanding of the area, in particular, Kristin
Hestetun, Francesco Scibilia, Giancarlo Marafioti, Florin Stoican, Selvanathan
Sivalingam, Muhammad Faisal Aftab, and Jonatan Klemets.
Special thanks are due to my hosts during sabbaticals, when I have learned a
lot: Bob Bitmead (UCSD), Jan Maciejowski (University of Cambridge), Sorin Olaru
(CentraleSupelec), and Andreas Kugi (TU Vienna).

M. H.
 xxiii

Acronyms

AKF augmented Kalman filter

AR auto-regressive (model)
BLT biggest log modulus (tuning rule)
CCM convergent cross mapping
CLDG closed-loop disturbance gain
CPM control performance monitoring
DNL degree of nonlinearity
EKF extended Kalman filter
EnKF ensemble Kalman filter
FCC fluid catalytic cracking
FOPDT first-order plus deadtime (model)
GM gain margin
IAE integral (of) absolute error
IEKF iterated extended Kalman filter
IMC internal model control
ISE integral (of) squared error
KISS keep it simple, stupid
LHP left half-plane (of the complex plane)
LP linear programming
LQ linear quadratic
LQG linear quadratic Gaussian
MHE moving horizon estimation
MIMO multiple input multiple output
MPC model predictive control
MV manipulated variable
MVC minimum variance controller
MWB moving window blocking
NGI non-Gaussianity index
NLI nonlinearity index
OCI oscillation contribution index
OP controller output
PCA principal component analysis
PID proportional integral derivative
xxiv Acronyms

PSCI power spectral correlation index

P&ID piping and instrument diagram
PM phase margin
PRGA performance relative gain array
PV process variable
QP quadratic programming
RGA relative gain array
RHE receding horizon estimation
RHP right half-plane (of the complex plane)
RHPZ right half-plane zero
RSS residual sum of squares
RTO real time optimization
SIMC simple (or Skogestad) internal model control (tuning rules)
SISO single input single output
SP setpoint (reference value)
SOPDT second-order plus deadtime (model)
SVD singular value decomposition
UKF unscented Kalman filter
 xxv

Introduction

I.1 Why Is Chemical Process Control Needed?

Many texts on control implicitly assume that it is obvious when and why control is
needed. It seems obvious that even a moderately complex process plant will be very
difficult to operate without the aid of process control. Nevertheless, it can be worth-
while to spend a few minutes thought on why control is needed. In the following, a
short and probably incomplete list of reasons for the need of control is provided, but
the list should illustrate the importance of control in a chemical process plant.

1. Stabilizing the process. Many processes have integrating or unstable modes. These
have to be stabilized by feedback control, otherwise the plant will (sooner or later)
drift into unacceptable operating conditions. In the vast majority of cases, this
stabilization is provided by automatic feedback control.1 Note that in practice,
“feedback stabilization” of some process variable may be necessary even though
the variable in question is asymptotically stable according to the control engineer-
ing definition of stability. This happens whenever disturbances have sufficiently
large effect on a process variable to cause unacceptably large variations in the pro-
cess variable value. Plant operators therefore often use the term “stability” in a
much less exact way than how the term is defined in control engineering. A con-
trol engineer may very well be told that, e.g. “this temperature is not sufficiently
stable,” even though the temperature in question is asymptotically stable.
2. Regularity. Even if a process is stable, control is needed to avoid shutdowns due to
unacceptable operating conditions. Such shutdowns may be initiated automati-
cally by a shutdown system but may also be caused by outright equipment failure.
3. Minimizing effects on the environment. In addition to maintaining safe and stable
production, the control system should also ensure that any harmful effects on
the environment are minimized. This is done by optimizing the conversion of
raw materials,2 and by maintaining conditions which minimize the production
of any harmful byproducts.

1 However, some industries still use very large buffer tanks between different sections in the
process. For such tanks, it may be sufficient with infrequent operator intervention to stop the
buffer tank from overfilling or emptying.
2 Optimizing the conversion of raw materials usually means maximizing the conversion, unless
this causes unacceptably high production of undesired byproducts or requires large energy inputs.
xxvi Introduction

4. Obtaining the right product quality. Control is often needed both for achieving the
right product quality and for reducing quality variations.
5. Achieving the right production rate. Control is used for achieving the right produc-
tion rate in a plant. Ideally, it should be possible to adjust the production rate at
one point in the process, and the control system should automatically adjust the
throughput of up- or downstream units accordingly.
6. Optimize process operation. When a process achieves safe and stable operation,
with little downtime and produces the right quality of product at the desired
production rate, the next task is to optimize the production. The objective of
the optimization is normally to achieve the most cost-effective production. This
involves identifying, tracking, and maintaining the optimal operating conditions
in the face of disturbances in production rate, raw material composition, and
ambient conditions(e.g. atmospheric temperature). Process optimization often
involves close coordination of several process units and operation close to process
constraints.
The list aforementioned should illustrate that process control is vital for the oper-
ation of chemical process plants. Even plants of quite moderate complexity would
be virtually impossible to operate without control. Even where totally manual oper-
ation is physically feasible, it is unlikely to be economically feasible due to product
quality variations and high personnel costs, since a high number of operators will be
required to perform the many (often tedious) tasks that the process control system
normally handles.
Usually many more variables are controlled than what is directly implied by the
list above, and there are often control loops for variables which have no specifica-
tion associated with them. There are often good reasons for such control loops – two
possible reasons are:
1. To stop disturbances from propagating downstream. Even when there are no direct
specification on a process variable, variations in the process variable may cause
variations in more important variables downstream. In such cases, it makes sense
to remove the disturbance at its source.
2. Local removal of uncertainty. By measuring and controlling a process variable,
it may be possible to reduce the effect of uncertainty with respect to equipment
behavior or disturbances. Examples of such control loops are valve positioners
used to minimize the uncertainty in the valve opening, or local flow control
loops which may be used to counteract the effects of pressure disturbances up- or
downstream of a valve, changes in fluid properties, or inaccuracies in the valve
characteristics.

I.2 What Knowledge Does a chemical Process Control

Engineer need?
The aforementioned list of reasons for why process control is needed also indicates
what kind of knowledge is required for a process control engineer. The process
 Introduction xxvii

control engineer needs to have a thorough understanding of the process. Most

stabilizing control loops involve only one process unit (e.g. a tank or a reactor),
and most equipment limitations are also determined by the individual units.
Process understanding on the scale of the individual units is therefore required.
Understanding what phenomena affect product quality also require an under-
standing of the individual process units. On the other hand, ensuring that the
specified production rate propagates throughout the plant, understanding how the
effects of disturbances propagate, and optimizing the process operation require an
understanding of how the different process units interact, i.e. an understanding of
the process on a larger scale.
Most basic control functions are performed by single loops, i.e. control loops with
one controlled variable and one manipulated variable. Thus, when it is understood
why a particular process variable needs to be controlled, and what manipulated vari-
able should be used to control it, 3 the controller design itself can be performed using
traditional single-loop control theory (if any theoretical considerations are made at
all). Often a standard type of controller, such as a proportional integral derivative
(PID) controller, is tuned online, and there is little need for a process model. Other
control tasks are multivariable in nature, either because it is necessary to resolve
interactions between different control loops or because the control task requires
coordination between different process units. Process models are often very useful
for these types of control problem. The models may either be linear models obtained
from experiments on the plant or possibly nonlinear models derived from physical
and chemical principles. Some understanding of mathematical modeling and system
identification techniques are then required. Nonlinear system identification from
plant experiments are not in standard use in the process industries.
Optimizing process operation requires some understanding of plant economics,
involving the costs of raw materials and utilities, the effect of product quality on
product price, the cost of reprocessing off-spec product, etc. Although it is rare
that economics is optimized by feedback controllers, 4 an understanding of plant
economics will help understanding where efforts to improve control should be
focused and will help when discussing the need for improved control with plant
management.
A process control engineer must thus have knowledge of both process and con-
trol engineering. However, it is not reasonable to expect the same level of expertise
in either of these disciplines from the process control engineer as for “specialist”
process or control engineers. There appears to be a “cultural gap” between process
and control engineers, and the process control engineer should attempt to bridge

3 Determining what variables are to be controlled, what manipulated variables should be used for
control, and the structure of interconnections between manipulated and controlled variables are
quite critical tasks in the design of a process control system. This part of the controller design is
often not described in textbooks on “pure” control engineering but will be covered in some detail
in later sections.
4 It is more common that economic criteria are used in the problem formulation for so-called real
time optimization (RTO) problems, or for plant production planning and scheduling, as shown in
Figure I.1.
xxviii Introduction

this gap. This means that the process control engineer should be able to commu-
nicate meaningfully with both process and control engineers and thereby also be
able to obtain any missing knowledge by discussing with the “specialists.” However,
at a production plant, there will seldom be specialists in control theory, but there
will always be process engineers. At best, large companies may have control the-
ory specialists at some central research or engineering division. This indicates that a
process control engineer should have a fairly comprehensive background in control
engineering, while the process engineering background should at least be sufficient
to communicate effectively with the process engineers.
In the same way as for other branches of engineering, success at work will not
come from technological competence alone. A successful engineer will need the
ability to work effectively in multidisciplinary project teams, as well as skills in com-
municating with management and operators. Such nontechnical issues will not be
discussed further here.

I.3 What Makes Chemical Process Control Unique?

Control engineering provides an extensive toolbox that can be applied to a very wide
range of application domains. Still, process control is characterized by a few features,
the combination of which make process control uniquely challenging:

● Physical scale. Although small-scale industrial plants do exist, more common are
large-scale industrial production plants, with hundreds of meters or even kilo-
meters from one end to the other, and the need for some degree of coordination
between different sections of the plant.
● One-of-a-kind plants. Most plants are unique. Although many of the individual
components may be standard issue, their assembly into the overall plant is typi-
cally unique to the plant in question. Even in the rather rare cases when plants are
built to be nominally identical, differences in external disturbances, raw materials,
and operating and maintenance practices will mean that differences accumulate
over time. The sheer scale of industrial plants also mean that the number of ways
in which plant behavior may differ is very large. This differs from, say, automo-
biles or airplanes, where the aim is to produce a large number of identical units.
From a control perspective, the main consequence for chemical process control is
that the cost of modeling and model maintenance must be borne by each individual
plant. Model-based control design or plant operation will therefore often have to
be based on rather simple and inaccurate models.
● Plant experimentation. Another aspect of the point above is that in order to learn
plant behavior or verify control designs, experiments often have to be performed
on the plant itself. Experiments disrupt plant operation and can also involve the
risk of damage or accident. Again this differs from control applications in mass
produced items, where damage to a few items during testing is often expected.
● Lots of data, little information. In many chemical process plants, fast and reliable
measurement of key variables are not available. In some processes, measurements
 Introduction xxix

of temperatures, pressures, and flows are readily available, while composition

measurements require either costly online analyzers with long time delays or
require manual sampling and laboratory analysis. In other cases, harsh process
conditions means that not even pressures and temperatures can be measured,
and control must be based on measurements obtained externally to the process
itself (such as currents and voltages in electrochemical plants). Again, this is
in contrast to other application areas of control, such as motion control where
positions and velocities often can be measured fairly accurately.

I.4 The Structure of Control Systems in the Chemical

Process Industries
When studying control systems in the chemical process industries, one may observe
that they often share a common structure. This structure is illustrated in Figure I.1.
The lower level in the control system is the regulatory control layer. The structure
of the individual controllers in the regulatory control layer is normally very simple.
Standard single-loop controllers, typically of proportional integral (PI)/PID type, are
the most common, but other simple control functions like feedforward control, ratio
control, or cascaded control loops may also be found. Truly multivariable controllers
are rare at this level. The regulatory control system typically controls basic process
variables such as temperatures, pressures, flowrates, speeds, or concentrations, but
in some cases the controlled variable may be calculated based on several measure-
ments, e.g. a component flowrate based on measurements of both concentration and
overall flowrate or a ratio of two flowrates. Usually, a controller in the regulatory
control layer manipulates a process variable directly (e.g. a valve opening), but in
some cases the manipulated variable may be a setpoint of a cascaded control loop.
Most control functions that are essential to the stability and integrity of the process
are executed in this layer, such as stabilizing the process and maintaining acceptable
equipment operating conditions.
The supervisory control layer coordinates the control of a process unit or a few
closely connected process units. It coordinates the action of several control loops
and tries to maintain the process conditions close to the optimal while ensuring that
operating constraints are not violated. The variables that are controlled by super-
visory controllers may be process measurements, variables calculated or estimated
from process measurements, or the output from a regulatory controller. The manip-
ulated variables are often setpoints to regulatory controllers, but process variables
may also be manipulated directly. Whereas regulatory controllers are often designed
and implemented without ever formulating any process model explicitly, supervi-
sory controllers usually contain an explicitly formulated process model. The model
is dynamic and often linear and obtained from experiments on the plant. Typically,
supervisory controllers use some variant of model predictive control (MPC).
The optimal conditions that the supervisory controllers try to maintain may be
calculated by a RTO control layer. The RTO layer identifies the optimal conditions
by solving an optimization problem involving models of the production cost, value of
xxx Introduction

Production planning/
scheduling

Real time optimization

Supervisory control

Regulatory control

To manipulated variables From measurements

Process

Figure I.1 Typical structure of the control system for a large plant in the process industries.

product (possibly dependent on quality), and the process itself. The process model is
often nonlinear and derived from fundamental physical and chemical relationships,
but they are usually static.
The higher control layer shown in Figure I.1 is the production planning and
scheduling layer. This layer determines what products should be produced and
when they should be produced. This layer requires information from the sales
department about the quantities of the different products that should be produced,
 Introduction xxxi

the deadlines for delivery, and possibly product prices. From the purchasing depart-
ment, information about the availability and price of raw materials are obtained.
Information from the plant describes what products can be made in the different
operating modes and what production rates can be achieved.
In addition to the layers in Figure I.1, there should also be a separate safety system
that will shut the process down in a safe and controlled manner when potentially
dangerous conditions occur. There are also higher levels of decision-making which
are not shown, such as sales and purchasing, and construction of new plants. These
levels are considered to be of little relevance to process control and will not be dis-
cussed further.
Note that there is a difference in timescale of execution for the different layers.
The regulatory control system typically have sampling intervals on the scale of 1
second (or faster for some types of equipment), supervisory controllers usually
operate on the timescale of minutes, the RTO layer on a scale of hours, and the
planning/scheduling layer on a scale of days (or weeks). The control bandwidths
achieved by the different layers differ in the same way as sampling intervals differ.
This difference in control bandwidths can simplify the required modeling in the
higher levels; if a variable is controlled by the regulatory control layer, and the
bandwidth for the control loop is well beyond what is achieved in the supervisory
control layer, a static model for this variable (usually the model would simply be
variable value = setpoint) will often suffice for the supervisory control.
It is not meaningful to say that one layer is more important than another, since
they are interdependent. The objective of the lower layers are not well defined with-
out information from the higher layers (e.g. the regulatory control layer needs to
know the setpoints that are determined by the supervisory control layer), whereas
the higher layers need the lower layers to implement the control actions. However,
in many plants, human operators perform the tasks of some the layers as shown
in Figure I.1, and it is only the regulatory control layer that is present (and highly
automated) in virtually all industrial plants.
Why has this multilayered structure for industrial control systems evolved? It is clear
that this structure imposes limitations in achievable control performance compared
to a hypothetical optimal centralized controller which perfectly coordinates all avail-
able manipulated variables in order to achieve the control objectives. In the past,
the lack of computing power would have made such a centralized controller vir-
tually impossible to implement, but the continued increase in available computing
power could make such a controller feasible in the not too distant future. Is this the
direction industrial control systems are heading? This appears not to be the case.
In the last two decades, development has instead moved in the opposite direction,
as increased availability of computing power has made the supervisory control and
RTO layers much more common. Some reasons for using such a multilayered struc-
ture are:

● Economics. Optimal control performance – defined in normal control engineering

terms (using, e.g. the H2 - or H∞ norm) – does not necessarily imply optimal
economic performance. To be more specific, an optimal controller synthesis
xxxii Introduction

problem does not take into account the cost of developing and maintaining the
required process (or possibly plant economic) models. An optimal centralized
controller would require a dynamic model of most aspects of the process behavior.
The required model would therefore be quite complex and difficult to develop
and maintain. In contrast, the higher layers in a structured control system can
take advantage of the model simplifications made possible by the presence of
the lower layers. The regulatory control level needs little model information to
operate, since it derives most process information from feedback from process
measurements.5
● Redesign and retuning. The behavior of a process plant changes with time, for a
number of reasons such as equipment wear, changes in raw materials, changes
in operating conditions in order to change product qualities or what products are
produced, and plant modifications. Due to the sheer complexity of a centralized
controller, it would be difficult and time-consuming to update the controller to
account for all such changes. With a structured control system, it is easier to see
what modifications need to be made, and the modifications themselves will nor-
mally be less involved.
● Start-up and shutdown. Common operating practice during start-up is that many
of the controls are put in manual. Parts of the regulatory control layer may be in
automatic, but rarely will any higher layer controls be in operation. The loops of
the regulatory control layer that are initially in manual are put in automatic when
the equipment that they control are approaching normal operating conditions.
When the regulatory control layer for a process section is in service, the supervi-
sory control system may be put in operation, and so on. Shutdown is performed
in the reverse sequence. Thus, there may be scope for significant improvement
of the start-up and shutdown procedures of a plant, as quicker start-up and shut-
down can reduce plant downtime. However, a model, which in addition to normal
operating conditions, is able to describe start-up and shutdown and is necessarily
much more complex than a model which covers only the range of conditions that
are encountered in normal operation. Building such a model would be difficult
and costly. Start-up and shutdown of a plant with an optimal centralized control
system which does not cover start-up and shutdown may well be more difficult
than with a traditional control system, because it may not be difficult to put an
optimal control system gradually into or out of service.
● Operator acceptance and understanding. Control systems that are not accepted by
the operators are likely to be taken out of service. An optimal centralized control
system will often be complex and difficult to understand. Operator understanding
obviously makes acceptance easier, and a traditional control system, being easier
to understand, often has an advantage in this respect. Plant shutdowns may be
caused by operators with insufficient understanding of the control system. Such
shutdowns should actually be blamed on the control system (or the people who
designed and installed the control system), since operators are an integral part of

5 A good process model may be of good use when designing control structuresfor regulatory
control. However, after the regulatory controllers are implemented, they normally do not make any
explicit use of a process model.
 Introduction xxxiii

the plant operation, and their understanding of the control system must therefore
be ensured.
● Failure of computer hardware and software. In traditional control systems, the
operators retain the help of the regulatory control system in keeping the process
in operation if a hardware or software failure occurs in higher levels of the con-
trol system. A hardware backup for the regulatory control system is much cheaper
than for the higher levels in the control system, as the regulatory control system
can be decomposed into simple control tasks (mainly single loops). In contrast,
an optimal centralized controller would require a powerful computer, and it is
therefore more costly to provide a backup system. However, with the continued
decrease in computer cost, this argument may weaken.
● Robustness .The complexity of an optimal centralized control system will make it
difficult to analyze whether the system is robust with respect to model uncertainty
and numerical inaccuracies. Analyzing robustness need not be trivial even for tra-
ditional control systems. The ultimate test of robustness will be in the operation of
the plant. A traditional control system may be applied gradually, first the regula-
tory control system, then section by section of the supervisory control system, etc.
When problem arise, it will therefore be easier to analyze the cause of the problem
with a traditional control system than with a centralized control system.
● Local removal of uncertainty. It has been noted earlier that one effect of the lower
layer control functions is to remove model uncertainty as seen from the higher lay-
ers. Thus, the existence of the lower layers allow for simpler models in the higher
layers and make the models more accurate. The more complex computations in
the higher layers are therefore performed by simpler, yet more accurate models. A
centralized control system will not have this advantage.
● Existing traditional control systems . Where existing control systems perform rea-
sonably well, it makes sense to put effort into improving the existing system rather
than to take the risky decision to design a new control system. This argument
applies also to many new plants, as many chemical processes are not well under-
stood. For such processes, it will therefore be necessary to carry out model identi-
fication and validation on the actual process. During this period, some minimum
amount of control will be needed. The regulatory control layer of a traditional
control system requires little information about the process and can therefore be
in operation in this period.

It should be clear from the above that this author believes that control systems in
the future will continue to have a number of distinct layers. Two prerequisites appear
to be necessary for a traditional control system to be replaced with a centralized one:

1. The traditional control system must give unacceptable performance.

2. The process must be sufficiently well understood to be able to develop a process
model which describes all relevant process behavior.

Since it is quite rare that a traditional control system is unable to control a chemi-
cal process for which detailed process understanding is available (provided sufficient
effort and expertise have been put into the design of the control system), it should
xxxiv Introduction

follow that majority of control systems will continue to be of the traditional struc-
tured type.
In short, the layered control system is consistent with the common approach
of breaking down big problems into smaller, more manageable parts, and as such
agrees with the keep it simple, stupid (KISS) principle.

I.5 Notation

x Vector of system states.

ẋ The time derivative of the state vector (for continuous-time systems).
u The vector of manipulated variables (the variables manipulated by the con-
trol system to control the plant), sometimes also referred to as inputs. In some
literature, the vector u is also called the control variables.
y The controlled variables (the variables that the control system attempts to con-
trol). Often, the vector y is also identical to the vector of measured variables.
d The vector of disturbance variables.
vk The vector v at timestep k (for discrete-time systems).
vi , vj Elements i and j of the vector v.
M Capital letters are used for matrices.
Mij Element (i, j) of the matrix M.
For the linear(ized) system in continuous time:
ẋ = Ax + Bu + Ed
y = Cx + Du + Fd
A, B, C, D, E, F are matrices of appropriate dimension, and
G(s) = C(sI − A)−1 B + D
Gd (s) = C(sI − A)−1 E + F
are the corresponding plant and disturbance transfer function matrices, respectively.
An alternative notation which is often used for complex state-space expressions, is
[ ]
A B
G(s) =
C D
That is, matrices in square brackets with a vertical and a horizontal line contain
expressions for the state-space representation of some transfer function matrix.
Matrices A, B, C, D, E, F are used also to define dynamical linear(ized) models in
discrete time:
xk+1 = Axk + Buk + Edk
yk = Cxk + Duk + Fdk
 Introduction xxxv

where the subscript (k or k + 1) defines the discrete sampling instant in question.

For simplicity of notation, the same notation is used often for continuous- and
discrete-time models, and it should be clear from context whether continuous or
discrete time is used. Note, however, that the model matrices will be different for
discrete and continuous time, i.e. converting from continuous to discrete time (or
vice versa) will change the matrices A, B, E.6

6 Whereas the matrices C, D, F describe instantaneous effects (not affected by the passing of time)
and will be the same for continuous- and discrete-time models.
 1

Mathematical and Control Theory Background

1.1 Introduction
This chapter will review some mathematical and control theory background, some
of which is actually assumed covered by previous control courses. Both the coverage
of topics and their presentation will therefore lack some detail, as the presentation
is aiming
● to provide sufficient background knowledge for readers with little exposure to con-
trol theory,
● to correct what is this author’s impression of what are the most common miscon-
ceptions
● to establish some basic concepts and introduce some notation.

1.2 Models for Dynamical Systems

Many different model representations are used for dynamical systems, and a few of
the more common ones will be introduced here.

1.2.1 Dynamical Systems in Continuous Time

A rather general way of representing a dynamical system in continuous time is via a
set of ordinary differential equations:
ẋ = f (x, u, d) (1.1)
where the variables x are termed as the system states and ẋ = dxdt
is the time derivative
of the state. The variables u and d are both external variables that affect the system. In
the context of control, it is common to distinguish between the manipulated variables
or (control) inputs u that can be manipulated by a controller, and the disturbances
d that are external variables that affect the system but which cannot be set by the
controller.
The system states x are generally only a set of variables that are used to describe
the system’s behavior over time. Whether the individual components of the state

Advanced Chemical Process Control: Putting Theory into Practice, First Edition. Morten Hovd.
© 2023 WILEY-VCH GmbH. Published 2023 by WILEY-VCH GmbH.
2 1 Mathematical and Control Theory Background

vector can be assigned any particular physical interpretation will depend on how
the model is derived. For models derived from fundamental physical and chemical
relationships (often termed as “rigorous models”), the states will often be quanti-
ties like temperatures, concentrations, and velocities. If, in contrast, the model is an
empirical model identified from observed data, it will often not be possible to assign
any particular interpretation to the states.
Along with the state equation (1.1), one typically also needs a measurement
equation such as:
y = g(x, u, d) (1.2)
where the vector y is a vector of system outputs, which often correspond to available
physical measurements from the systems. Control design is usually at its most simple
when all states can be measured, i.e. when y = x.
Disturbances need not be included in all control problems. If no disturbances
are included in the problem formulation, Eqs. (1.1) and (1.2) trivially simplify to
ẋ = f (x, u) and y = g(x, u), respectively.
Since we are dealing with dynamical systems, it is hopefully obvious that the vari-
ables x, y, u, d may all vary with time t. In this section, time is considered as a con-
tinuous variable, in accordance with our usual notion of time.
Together, Eqs. (1.1) and (1.2) define a system model in continuous time. This
type of model is rather general and can deal with any system where it suffices to
consider system properties at specific points in space, or where it is acceptable to
average/lump system properties over space. Such models where properties are aver-
aged over space are often called lumped models.
For some applications, it may be necessary to consider also spatial distribution
of properties. Rigorous modeling of such systems typically result with a set of
partial differential equations (instead of the ordinary differential equations of (1.1)).
In addition to derivatives with respect to time, such models also contain derivatives
with respect to one or more spatial dimensions. Models described by partial differ-
ential equations will not be considered any further here. Although control design
based on partial differential equations is an active research area, the more common
industrial practice is to convert the set of partial differential equations to a (larger)
set of ordinary differential equations through some sort of spatial discretization.

1.2.2 Dynamical Systems in Discrete Time

Although time in the “real world” as we know it is a continuous variable, control sys-
tems are typically implemented in computer systems, which cyclically execute a set
of instructions. Measurements and control actions are therefore executed at discrete
points in time, and to describe system progression from one time instant to subse-
quent instants we will need a discrete-time model. Such models may be represented
as:
xk+1 = f (xk , uk , dk ) (1.3)

yk = g(xk , uk , dk ) (1.4)
 1.2 Models for Dynamical Systems 3

where xk , yk , uk , and dk are the discrete-time counterparts to the system states,

outputs, inputs, and disturbances introduced above for continuous-time systems,
and the subscript (k) identify the timestep (or sampling instant). Thus, xk is the state
x at timestep k, while xk+1 is the state at the subsequent timestep. Note that although
the same letter f is used to represent the system dynamics for both continuous-
and discrete-time systems, these functions will be different for the two different
model types. The measurement equation, however, will often be identical for the
two model types.

1.2.3 Linear Models and Linearization

Many control design methods are based on linear models. It is therefore necessary
to be able to convert from a nonlinear model to a linear model which is (hopefully) a
close approximation to the nonlinear model. This is called linearization of the non-
linear model.
A systems is linear if the functions f and g (in (1.1) and (1.2) for the case of
continuous-time models, or in (1.3) and (1.4) for the case of discrete-time models)
are linear in all the variables x, u, and d. Thus, a linear continuous-time model may
be expressed as:
ẋ = Ax + Bu + Ed (1.5)

y = Cx + Du + Fd (1.6)
where A, B, C, D, E, F are matrices of appropriate dimensions, and the matrix ele-
ments are independent of the values of x, u, d. Linear models for discrete-time sys-
tems follow similarly.
Linearization is based on the Taylor series expansion of a function. Consider a
function h(a). We want to approximate the value of h(a) in the vicinity of a = a∗ .
The Taylor series expansion then provides the approximation:
𝜕h || 1 𝜕2 h |
h(a) = h(a∗ + 𝛿a) ≈ h(a∗ ) + | 𝛿a + 𝛿aT 2 || 𝛿a + · · · (1.7)
𝜕a |a∗ 2 𝜕a |a=a∗
where the notation |a=a∗ indicates that the value a = a∗ is used when evaluating the
derivatives.

1.2.3.1 Linearization at a Given Point

When linearizing a dynamical system model, we terminate the Taylor series expan-
sion after the first-order term. The underlying nonlinear system is therefore natu-
rally assumed to be continuous and have continuous first-order derivatives. Assume
that the linearization is performed at the point:
⎡ x ⎤ ⎡ x∗ ⎤
a = ⎢ u ⎥ = ⎢ u∗ ⎥ = a ∗ (1.8)
⎢ ⎥ ⎢ ∗⎥
⎣d⎦ ⎣d ⎦
The terminated Taylor series expansion of (1.1) then becomes
dx d𝛿x 𝜕f || 𝜕f || 𝜕f ||
= ≈ f (a∗ ) + | 𝛿x + | 𝛿u + 𝛿d (1.9)
dt dt 𝜕x |a=a ∗ 𝜕u |a=a ∗ 𝜕d ||a=a∗
4 1 Mathematical and Control Theory Background

Similarly, we get for (1.2)

𝜕g || 𝜕g || 𝜕g |
y = y∗ + 𝛿y ≈ g(a∗ ) + | 𝛿x + | 𝛿u || 𝛿d (1.10)
𝜕x |a=a∗ 𝜕u |a=a∗ 𝜕d |a=a∗
where it is understood that y∗ = g(a∗ ).
𝜕f | 𝜕f | 𝜕f | 𝜕g | 𝜕g |
Next, define A = 𝜕x | ∗ , B = 𝜕u | ∗ , E = 𝜕d | ∗ , C = 𝜕x | ∗ , D = | ,
|a=a |a=a |a=a |a=a 𝜕u |a=a∗
|
F = 𝜕d |
𝜕x |a=a∗

Linearizing at an Equilibrium Point The point a∗ used in the linearization is usually

an equilibrium point. This means that
f (a∗ ) = 0 (1.11)

g(a∗ ) = y∗ (1.12)
Thus, we get
dx
= A𝛿x + B𝛿u + E𝛿d (1.13)
dt
𝛿y = C𝛿x + D𝛿u + F𝛿d (1.14)
Linearizing a discrete-time model is done in the same way as for continuous-time
models. The only slight difference to keep in mind is that for a discrete-time model
at steady state xk+1 = xk , and therefore f (a∗ ) = xk when linearizing at a steady
state.

Deviation Variables It is common to express the system variables (x, u, d, and y) in

terms of their deviation from the linearization point a∗ . When doing so the 𝛿’s are
typically suppressed for ease of notation – as will be done in the remainder of this book.
It is, however, important to beware that when converting from deviation variables
to “real” variables, the linearization point has to be accounted for.
To illustrate: A model for a chemical reactor is linearized at steady-state conditions
corresponding to a reactor temperature of 435 K. If the linearized model, expressed
in deviation variables, indicates a temperature of −1, the corresponding “real” tem-
perature would be 434 K.

Linear Controllers Are Not Linear! It appears that many students, even after intro-
ductory control courses, do not appreciate that our so-called “linear” controllers are
only linear when expressed in deviation variables. In “natural” variables, the typi-
cal “linear” controller is in fact affine, i.e. they have a constant term in addition to
the linear term. This can lead to many frustrations, until the misunderstanding has
been clarified – which might actually take some time, because the importance of this
issue will depend on both controller structure and controller type. Consider a simple
feedback loop, with a (linear) controller K controlling a system G, as illustrated in
Figure 1.1.
 1.2 Models for Dynamical Systems 5

u*

r e u + y
K G
–

Figure 1.1 A simple feedback loop with a one-degree-of-freedom controller and possible
“output bias.”

This type of controller is called a “one-degree-of-freedom controller,” since

it has only one input, the control offset e = r − y. We can make the following
observations:
● Clearly, it does not matter whether the reference r and measurement y are
expressed in “physical” variables or deviation variables, as long as the same scale
is used for both. This is because the controller input is the difference between
these two variables.
● Consider the case when the controller K is a pure proportional controller,
i.e. u = K(r − y) with K constant. It is then necessary to add u∗ as an “output
bias”1 to the controller output, as indicated by the dashed arrow in the figure.
● Consider next the case when the controller K contains integral action. In this
case, the “output bias” is not strictly necessary, since the value of the integrating
state will adjust for this when the system reaches steady state. However, an out-
put bias may improve transient response significantly when putting the controller
into operation.2
Consider next a loop where the controller has separate entry port for the reference
and the measurement, as shown in Figure 1.2. This type of controller is used

r u*

y* – u y
+
K G
–

Figure 1.2 A simple feedback loop with a two-degree-of-freedom controller and possible
“bias” on both controller inputs and controller output.

1 Some system vendors may use different terminology.

2 See also Section 6.4 on Bumpless Transfer.
6 1 Mathematical and Control Theory Background

when one wants to treat the measurement and reference signals differently in the
controller. We note that
● In this case, we need to subtract the value of the measurement at the linearization
point, y∗ , from both the reference and the measurement.
● Whether to add u∗ to the controller output is determined by the same considera-
tions as for the one-degree-of-freedom controller.
1.2.3.2 Linearizing Around a Trajectory
It was noted above that it is most common to linearize around a steady state.
However, in some cases, one may want to linearize around a trajectory, i.e. around
a series of consistent future values of x, u, and d. This most commonly occurs in
nonlinear model predictive control (MPC). Each time an MPC controller executes,
it solves an optimization problem that optimizes system behavior over a “predic-
tion horizon.” However, for some strongly nonlinear problems, using the same
linearized model for the entire prediction horizon may not give sufficient accuracy.
In such cases, one may choose to linearize around a trajectory instead.
Given the present state, a prediction of the future manipulated variables (typically
obtained from the previous execution of the MPC), and predicted values for future
disturbances, the nonlinear model can be used to simulate the system in the future.
This gives predicted future states that are consistent with the present state and the
predicted future manipulated variables and disturbances.
For each timestep in the future, the linearization is performed around the pre-
dicted state, manipulated variable, and disturbance values. This will give different
matrices A, B, CD, E, F for each timestep. In this way, a nonlinear system is approxi-
mated by a linear, time-varying model.
Linearizing around a trajectory clearly complicates the model. In addition to the
added complexity of having to ensure that the right model matrices are used at the
right timestep in the future, one also has to remember that the linearization point
varies from timestep to timestep (resulting from f (a∗ ) ≠ xk in the discrete-time equiv-
alent of (1.9)). This adds additional complexity when converting between physical
variables and deviation variables.

1.2.4 Converting Between Continuous- and Discrete-Time Models

It will often be necessary to convert from continuous- to discrete-time models (and
less frequently necessary to convert the other way). Process models based on first
principles modeling will typically result in continuous-time models. Often, control
design is performed with a continuous-time model. The continuous-time controller
is thereafter converted to a discrete-time controller for implementation in a com-
puter. There are also controller types that are more conveniently designed using
discrete-time models. The most notable example of such controllers are the so-called
MPC controllers which will be described in some detail later in the book.
To convert from continuous to discrete time, we need to
● choose a numerical integration method for the system dynamics, and
● determine (assume) how the external variables (u and d) change between the time
instants for the discrete-time model.
Another random document with
no related content on Scribd:
dangerous situations. The tide rose very rapidly, and all the
temporary embarrassments of our situation vanished with our
footprints in the sand. The mounting sun soon burned up the fog,
which in dispersing produced its usual singular and fantastic effects
upon the rugged and precipitous shores that lay on each side; and
retaining the services of our old friend as pilot, we ran through the
river, which is about four miles long, and connected with the harbor
of Gloucester by a short canal, through which we passed, and spent
another pleasant day in that town previous to starting for Boston;
which place we had left just three weeks before. We arrived there
the next day, meeting with nothing worthy of particular notice in the
course of it.

Our vessel ashore on Squam Bar.

“Such is a brief outline of our excursion, from which we returned
much invigorated in mind and body. A thousand little incidents
occurred, serving to enhance the pleasure of the trip, which it would
be impossible to condense into so small a space as is here allotted
us. We had finer opportunities of obtaining picturesque sketches of
our New England coast scenery, than could be obtained by any other
method. One of our company made a sketch of our mischance upon
the bar, and an engraving of it is presented to the reader. We had a
good opportunity of observing the peculiar traits that characterize the
hardy race that inhabit our rough and rock-bound coast, and always
found them a freehearted, hospitable people, ever ready to yield any
assistance we might need. We were obliged to submit to many little
inconveniences, it is true, which, had they not been voluntary, or had
they come under other than the then existing circumstances, would
have been deemed hardships; but there was so much excitement, so
much novelty, such an endless variety of new objects from day to
day to attract and interest us, that we were a thousand times repaid
for all our petty privations.”

Proverb.—A person who is suspicious, ought to be suspected.

Travels, Adventures, and Experiences of Thomas
Trotter.

CHAPTER V.
Departure from Malta.—​Arrival at Sicily.—​Syracuse Ruins.—​Ear of
Dionysius.

Our vessel landed her cargo at Malta, and then took in ballast
and sailed for Palermo, in Sicily, to load with fruit. I preferred to cross
immediately over to Syracuse, and take Mount Ætna in my way,
being very desirous not to lose a sight of this celebrated volcano. I
found a Sicilian vessel about to sail, and took passage in her. She
was a polacre, having the masts of single sticks from top to bottom,
instead of three or four pieces joined together, like the masts of
English and American vessels. I could not help laughing at the
oddities of the crew: there were fifteen of them, although the vessel
was not above seventy tons burthen. They were the queerest ship’s
company I ever saw; all captains and mates, and no common
sailors. Whatever was to be done was everybody’s business: there
was no discipline, no order, no concert; all was hurly-burly, and
scampering here and there, and tumbling head over heels.
Which was the commander, nobody could tell, for every one was
giving orders. The slightest manœuvre caused a clatter and bawling
that made me think the masts were going overboard. If there was a
rope as big as a tom-cod-line to be pulled, the whole crew would
string themselves along it, yo! heave ho! tug it an inch and a half,
puff and blow, thump and clamor, as if it were a case of life and
death. Every man must have a finger in what was going on, even to
cuffing the cabin-boy. The men squatted down upon deck to their
meals all in a group, and fell to cracking jokes and cutting capers
together. The helmsman sat in a chair to steer, and moved his seat
as often as he luffed or bore away. A little hop-off-my-thumb fellow,
with a comically dirty face and ragged breeches, sat upon a bucket
to watch the hour-glass in the binnacle. We had only seventy or
eighty miles to sail from Malta to Sicily, with a fair wind and a smooth
sea, but the fuss and clatter during the navigation of this short space
were prodigious. All hands were running fore and aft, looking out
ahead and astern, bustling around the man at the helm, peeping at
the compass, and jabbering and gesticulating as if they were in the
most imminent danger.
At daylight the next morning, we found ourselves close under the
Sicilian shore, with Mount Ætna in the north, towering up majestically
to the heavens, like a huge pyramid of snow with a black spot at the
top. It was more than seventy miles off. About ten in the forenoon we
arrived at Syracuse, a city which was once ten times as big as
Boston, but is now almost entirely depopulated. It has a noble
harbor, but we found only a few fishing-boats there; and when we
landed at the quay, hardly a living being was to be seen: everything
looked solitary, ruinous, and forlorn. I walked through the streets, but
saw no signs of trade, commerce, or industry. A few people were
sitting lazily before their doors, sunning themselves; and numbers of
beggars dogged my heels wherever I went. Now and then I met a
donkey with a pannier of greens, but no such thing as a wagon or
chaise.
When I got to the market-place, I saw groups of people sitting in
the sun or lounging idly about, but no business doing. I could not
help smiling to see a constable, who was strutting up and down to
keep the peace among this pack of lazy fellows. He wore a great,
long, tattered cloak, a huge cocked hat, a sword, and he had a most
flaming, fiery visage, with a nose like a blood-beet. I never saw such
a swaggering figure in my life, before. He happened to spy a little
urchin pilfering a bunch of greens, on which he caught him by the
nape of the neck with one hand, and drawing his sword with the
other, gave him a lusty thwacking with the flat of the blade. The little
rogue kicked and squalled, and made a most prodigious uproar,
which afforded great amusement to the crowd: they seemed to be
quite familiar with such adventures.
I walked out into the country, and was struck with astonishment at
the sight of the ruins scattered all round the neighborhood. They
extend for miles in every direction. Walls, arches, columns, remains
of temples, theatres and palaces met the eye at every step. Here
and there were little gardens among the ruins, where artichokes
were growing, but hardly a human being was to be seen. I came at
length to the remains of a large theatre, consisting of a semicircle of
stone steps, and found a mill stream tumbling down the middle of it.
A ragged peasant was lying lazily in the sun among the ruins. I
asked him what building it was, but he was totally ignorant of the
matter, and could only reply that it was “cosa antica”—something
ancient. Presently I discovered an enormous excavation in the solid
rock, as big as a house, which excited my curiosity very strongly. I
could not imagine the use of it, till I luckily met an old Capuchin friar,
plodding along in his coarse woollen gown; and learnt from him that
this was the famous “Ear of Dionysius,” where that tyrannical king
used to confine such persons as fell under his suspicion. It is a most
curious place, hollowed out in the shape of the human ear, and
forming a vast cavern: in the top is a little nook or chamber, where
the tyrant used to sit and hear what the prisoners said. The lowest
whisper was heard distinctly in this spot; so that the prisoners were
sure to betray themselves if they held any conversation together.
While I stood wondering at this strange perversion of human
ingenuity, I was startled by the appearance of a grim-looking fellow,
who pulled out a pistol as he approached me. My first impulse was to
grasp my trusty cudgel, and flourish it at him with a fierce air of
defiance, for I took him to be a robber, of course. To my surprise he
burst out a laughing, and told me he had come on purpose to show
me the wonderful effect of sound in the Ear. He bade me go into the
further end of the cavern, while he fired the pistol at the entrance. I
did so, and the effect was like the roaring of thunder: I was glad to
clap my hands to my ears and run out as fast as I could. I gave the
fellow a few cents for his trouble, and told him I had never before got
so much noise for so little money.
 I continued to ramble about among the ruins, which seemed to
have no end. The almond trees were in full bloom, and the orange
trees were bowing down under loads of ripe fruit. Flocks of magpies
were flitting about, but everything was silent and deserted. Now and
then I met a countryman jogging lazily along upon a donkey, or an
old woman driving her beast with a load of vine-stalks, which are
used in the city to heat ovens. I could not help wondering to see so
fine a territory lie utterly neglected; but the indolence of the
inhabitants is the cause of all. A very little labor will earn a loaf of
bread, and most of them are satisfied with this. The climate is so
mild, that ragged clothes occasion no discomfort, and hardly
anybody minds going in rags. The soil is so rich as scarcely to
require art or industry in the cultivation. The oranges and the grapes
grow with hardly any care, and the husbandman lives a lazy life, with
but little to do except to pick the fruit and make the wine.
 Sketches of the Manners, Customs, and History
of the Indians of America.

(Continued from page 119.)

CHAPTER II.
The West Indies continued.—​Discovery of Hayti.—​Generosity of the
Cacique.—​Testimony of Columbus in favor with the Indians.—​
Character of the natives.—​Columbus erects a cross.—​Indian
belief.—​Effect of the Spanish invasion.—​The Cacique.

Columbus entered a harbor at the western end of the island of

Hayti, on the evening of the 6th of December. He gave to the harbor
the name of St. Nicholas, which it bears to this day. The inhabitants
were frightened at the approach of the ships, and they all fled to the
mountains. It was some time before any of the natives could be
found. At last three sailors succeeded in overtaking a young and
beautiful female, whom they carried to the ships.
She was treated with the greatest kindness, and dismissed finely
clothed, and loaded with presents of beads, hawk’s bells, and other
pretty baubles. Columbus hoped by this conduct to conciliate the
Indians; and he succeeded. The next day, when the Spaniards
landed, the natives permitted them to enter their houses, and set
before them bread, fish, roots and fruits of various kinds, in the most
kind and hospitable manner.
Columbus sailed along the coast, continuing his intercourse with
the natives, some of whom had ornaments of gold, which they
readily exchanged for the merest trifle of European manufacture.
These poor, simple people little thought that to obtain gold these
Christians would destroy all the Indians in the islands. No—they
believed the Spaniards were more than mortal, and the country from
which they came must exist somewhere in the skies.
The generous and kind feelings of the natives were shown to
great advantage when Columbus was distressed by the loss of his
ship. He was sailing to visit a grand cacique or chieftain named
Guacanagari, who resided on the coast to the eastward, when his
ship ran aground, and the breakers beating against her, she was
entirely wrecked. He immediately sent messengers to inform
Guacanagari of this misfortune.
When the cacique heard of the distress of his guest, he was so
much afflicted as to shed tears; and never in civilized country were
the vaunted rites of hospitality more scrupulously observed than by
this uncultivated savage. He assembled his people and sent off all
his canoes to the assistance of Columbus, assuring him, at the same
time, that everything he possessed was at his service. The effects
were landed from the wreck and deposited near the dwelling of the
cacique, and a guard set over them, until houses could be prepared,
in which they could be stored.
There seemed, however, no disposition among the natives to take
advantage of the misfortune of the strangers, or to plunder the
treasures thus cast upon their shores, though they must have been
inestimable in their eyes. On the contrary, they manifested as deep a
concern at the disaster of the Spaniards as if it had happened to
themselves, and their only study was, how they could administer
relief and consolation.
Columbus was greatly affected by this unexpected goodness.
“These people,” said he in his journal, “love their neighbors as
themselves; their discourse is ever sweet and gentle, and
accompanied by a smile. There is not in the world a better nation or
a better land.”
When the cacique first met with Columbus, the latter appeared
dejected, and the good Indian, much moved, again offered
Columbus everything he possessed that could be of service to him.
He invited him on shore, where a banquet was prepared for his
entertainment, consisting of various kinds of fish and fruit. After the
feast, Columbus was conducted to the beautiful groves which
surrounded the dwelling of the cacique, where upwards of a
thousand of the natives were assembled, all perfectly naked, who
performed several of their national games and dances.
Thus did this generous Indian try, by every means in his power, to
cheer the melancholy of his guest, showing a warmth of sympathy, a
delicacy of attention, and an innate dignity and refinement, which
could not have been expected from one in his savage state.
He was treated with great deference by his subjects, and
conducted himself towards them with a gracious and prince-like
majesty.
Three houses were given to the shipwrecked crew for their
residence. Here, living on shore, and mingling freely with the natives,
they became fascinated by their easy and idle mode of life. They
were governed by the caciques with an absolute, but patriarchal and
easy rule, and existed in that state of primitive and savage simplicity
which some philosophers have fondly pictured as the most enviable
on earth.
The following is the opinion of old Peter Martyr: “It is certain that
the land among these people (the Indians) is as common as the sun
and water, and that ‘mine and thine,’ the seeds of all mischief, have
no place with them. They are content with so little, that, in so large a
country, they have rather superfluity than scarceness; so that they
seem to live in a golden world, without toil, in open gardens, neither
entrenched nor shut up by walls or hedges. They deal truly with one
another, without laws, or books, or judges.”
In fact, these Indians seemed to be perfectly contented; their few
fields, cultivated almost without labor, furnished roots and
vegetables; their groves were laden with delicious fruit; and the coast
and rivers abounded with fish. Softened by the indulgence of nature,
a great part of the day was passed by them in indolent repose. In the
evening they danced in their fragrant groves to their national songs,
or the rude sound of their silver drums.
Such was the character of the natives of many of the West Indian
Islands, when first discovered. Simple and ignorant they were, and
indolent also, but then they were kind-hearted, generous, and happy.
And their sense of justice, and of the obligations of man to do right,
are beautifully set forth in the following story.

Columbus erecting a Cross.

It was a custom with Columbus to erect crosses in all remarkable
places, to denote the discovery of the country, and its subjugation to
the Catholic faith. He once performed this ceremony on the banks of
a river in Cuba. It was on a Sunday morning. The cacique attended,
and also a favorite of his, a venerable Indian, fourscore years of age.
While mass was performed in a stately grove, the natives looked
on with awe and reverence. When it was ended, the old man made a
speech to Columbus in the Indian manner. “I am told,” said he, “that
thou hast lately come to these lands with a mighty force, and hast
subdued many countries, spreading great fear among the people;
but be not vain-glorious.
“According to our belief, the souls of men have two journeys to
perform after they have departed from the body: one to a place
dismal, foul, and covered with darkness, prepared for such men as
have been unjust and cruel to their fellow-men; the other full of
delight for such as have promoted peace on earth. If then thou art
mortal, and dost expect to die, beware that thou hurt no man
wrongfully, neither do harm to those who have done no harm to
thee.”
When this speech was explained to Columbus by his interpreter,
he was greatly moved, and rejoiced to hear this doctrine of the future
state of the soul, having supposed that no belief of the kind existed
among the inhabitants of these countries. He assured the old man
that he had been sent by his sovereigns, to teach them the true
religion, to protect them from harm, and to subdue their enemies the
Caribs.
Alas! for the simple Indians who believed such professions.
Columbus, no doubt, was sincere, but the adventurers who
accompanied him, and the tyrants who followed him, cared only for
riches for themselves. They ground down the poor, harmless red
men beneath a harsh system of labor, obliging them to furnish,
month by month, so much gold. This gold was found in fine grains,
and it was a severe task to search the mountain pebbles and the
sands of the plains for the shining dust.
Then the islands, after they were seized upon by the Christians,
were parcelled out among the leaders, and the Indians were
compelled to be their slaves. No wonder “deep despair fell upon the
natives. Weak and indolent by nature, and brought up in the
untasked idleness of their soft climate and their fruitful groves, death
itself seemed preferable to a life of toil and anxiety.
“The pleasant life of the island was at an end; the dream in the
shade by day; the slumber during the noontide heat by the fountain,
or under the spreading palm; and the song, and the dance, and the
game in the mellow evening, when summoned to their simple
amusements by the rude Indian drum.
“They spoke of the times that were past, before the white men
had introduced sorrow, and slavery, and weary labor among them;
and their songs were mournful, and their dances slow.
 “They had flattered themselves, for a time, that the visit of the
strangers would be but temporary, and that, spreading their ample
sails, their ships would waft them back to their home in the sky. In
their simplicity, they had frequently inquired of the Spaniards when
they intended to return to Turey, or the heavens. But when all such
hope was at an end, they became desperate, and resorted to a
forlorn and terrible alternative.”
They knew the Spaniards depended chiefly on the supplies raised
in the islands for a subsistence; and these poor Indians endeavored
to produce a famine. For this purpose they destroyed their fields of
maize, stripped the trees of their fruit, pulled up the yuca and other
roots, and then fled to the mountains.
The Spaniards were reduced to much distress, but were partially
relieved by supplies from Spain. To revenge themselves on the
Indians, they pursued them to their mountain retreats, hunted them
from one dreary fastness to another, like wild beasts, until thousands
perished in dens and caverns, of famine and sickness, and the
survivors, yielding themselves up in despair, submitted to the yoke of
slavery. But they did not long bear the burden of life under their
civilized masters. In 1504, only twelve years after the discovery of
Hayti, when Columbus visited it, (under the administration of
Ovando,) he thus wrote to his sovereigns: “Since I left the island, six
parts out of seven of the natives are dead, all through ill-treatment
and inhumanity; some by the sword, others by blows and cruel
usage, or by hunger.”
No wonder these oppressed Indians considered the Christians the
incarnation of all evil. Their feelings were often expressed in a
manner that must have touched the heart of a real Christian, if there
was such a one among their oppressors.
When Velasquez set out to conquer Cuba, he had only three
hundred men; and these were thought sufficient to subdue an island
above seven hundred miles in length, and filled with inhabitants.
From this circumstance we may understand how naturally mild and
unwarlike was the character of the Indians. Indeed, they offered no
opposition to the Spaniards, except in one district. Hatney, a cacique
who had fled from Hayti, had taken possession of the eastern
extremity of Cuba. He stood upon the defensive, and endeavored to
drive the Spaniards back to their ships. He was soon defeated and
taken prisoner.
Velasquez considered him as a slave who had taken arms against
his master, and condemned him to the flames. When Hatney was
tied to the stake, a friar came forward, and told him that if he would
embrace the Christian faith, he should be immediately, on his death,
admitted into heaven.
“Are there any Spaniards,” says Hatney, after some pause, “in
that region of bliss you describe?”
“Yes,” replied the monk, “but only such as are worthy and good.”
“The best of them,” returned the indignant Indian, “have neither
worth nor goodness; I will not go to a place where I may meet with
one of that cruel race.”
(To be continued.)

Something Wonderful.

The thing to which we refer is a seed. How wonderful that an

acorn should contain within it a little plant, capable of growing up into
an enormous oak, which will produce other acorns, capable of
growing into other oaks, and so on forever! and yet there are seeds
not one hundredth part as big as an acorn, which produce trees
almost, if not quite, as large as an oak.
Or think of a grain of wheat. It is just as useful for food as if it
contained nothing but a little flour mixed with a little bran. In fact,
when it is ground there is nothing else to be seen; but beside these it
contains a little plant, too small to be made out by common sight.
When one of these grains or seeds is put into moist earth, it
begins to suck in water, which softens it and makes it swell. The little
plant inside begins to grow, and in a few days a small, delicate root
peeps out from one end of the seed. The seed may be lying on its
side, or with the root end uppermost; but the little root, whether it
comes out at the top or bottom of the seed, immediately turns
downward, and grows in that direction.
Soon after, a little white shoot comes out at the other end, which
turns upwards, and becomes green as soon as it gets into air and
light; and thus we have a little plant.
In the mean time, the seed itself spoils and decays; or, as St. Paul
calls it, dies. The flour changes into a kind of gummy sugar, which is
sucked up by the young plant as its first nourishment; the husk
shrivels and rots, and the plant grows up until it becomes a thousand
times as large as the seed. At last it produces many other seeds, just
as wonderful as that from which it grew.
In all the works of man, there is nothing like this. A watch is a
remarkable invention, and a man would be set down as mad who
should think it should be made by chance. But how much more
wonderful would a watch be, if it could make other watches like itself!
Yet a seed does this; and every cornfield in harvest-time contains
millions of seeds, each of which is far more wonderful than the best
watch.
The reason is, that men make watches, but God makes seeds. It
is true that the skill by which men make watches comes from God,
and should be acknowledged as his gift; but the more wonderful
power by which a seed is made, he keeps in his own hands, that we
may know that we have a Maker and a Master in heaven, and may
serve him with reverence and godly fear.
 Fanny Gossip and Susan Lazy;
a dialogue.

Susan. Well, Fanny, I was on my way to your house. I thought I

never should see your face again. Did you ever know such a long,
stupid storm? nothing but rain, rain, rain for three everlasting days!
Fanny. And in vacation-time too! it did seem too bad. If our house
had not been on the street, so that I could see something stirring, I
believe I should have had the blues.
Susan. And I did have the blues outright. I never was so dull in my
life, moping about the house. Mother won’t let me touch such books
as I like to read, and the boys went to school all day, so I had nothing
on earth to do but look at the drops of rain racing down the windows,
and watch the clouds to see if it was going to clear up. I assure you I
fretted from morning till night, and mother got out of all patience with
me, and said I was a perfect nuisance in the house; but I am sure it
was not my fault.
Fanny. Well! I was a little better off. I sat half the time making fun
of all the shabby cloaks and umbrellas that turned out in the rain.
There was Mr. Skimmer went by every day with a cotton umbrella;
and Mr. Saveals with an old faded silk one, three of the whalebones
started out on one side, as if he wanted to poke people’s eyes out,
and a great slit to let the rain through:—both of them misers, I know!
And there was Miss Goodbody! she goes to see sick poor folks in all
weathers, and won’t take a carriage, though she can afford it,
because she says that would be ridiculous. I wish you had seen her
come paddling through the wet! such shoes, and such stockings! I
do think it is unladylike. Then, when everything else failed to amuse
me, there were our neighbors opposite to be speculated upon.
 Susan. Ah! Laura Busy lives just across the street, I believe?
Fanny. Yes, and there she sat at the window, on purpose to be
seen, stitching away, and reading, and setting herself up as a pattern
to the whole neighborhood.
Susan. I would not have such a strict mother as she has for all the
world. I don’t believe she enjoys her vacation at all.
Fanny. I dare say it is her mother that keeps her at it so close. I
should think she was bringing her up to be a seamstress; and yet,
considering that everybody knows Mr. Busy is not rich, they dress
Laura extravagantly. Did you see that beautiful French calico she
wore on examination day?
Susan. Yes, I saw it across the room, and thought I would go over
and look at it, but couldn’t take the trouble.
Fanny. Why, how you do gape, Susan!
Susan. I know it; mother says I have a terrible trick of gaping. But
I do get so tired.
Fanny. Tired of what?
Susan. I don’t know; I am tired of the vacation, I believe: and
before the term was over I was wishing so for it! I was tired to death
of school, and dare say I shall be so again in a fortnight.
Fanny. Here comes Laura, glad enough to get away from
mamma’s workbasket. Just see how fast she walks;—ah ha! she is
going to the circulating library; look at that novel under her arm.
Susan. I shall tell my mother of that; she thinks everything right
that the Busy family do.
(Enter Laura.)
Fanny. Well, Laura, poor thing! you are so glad to get out of the
house that I suppose you are running away from it as fast as you
can.
Laura. I am not quite running, I believe, but you know I always
walk fast.
 Susan. I can’t think why, I am sure.
Laura. It saves a deal of time, and the exercise does me more
good than if I were to go sauntering along.
Susan. Saves time? and in the vacation too? why, of what
consequence is time now, when you have no school-hours to mind?
Laura. Because if I don’t take care I shall not get through what I
have planned. Only think how fast the vacation is going! Next
Monday school begins.
Fanny. So the studious Miss Laura Busy is sorry the vacation is
almost over. I thought you told the master, when school broke up,
that you wished there was no vacation.
Laura. I did wish so then, for I thought vacation would be a dull
time.
Susan. I am sure it has been horrid dull to me, and I should think
it must have been worse yet for you.
Laura. Why?
Susan. Because your mother keeps you at work all the time.
Laura. Indeed she does not. She sent me out to walk this very
afternoon, and she always makes me put my work away at just such
hours, for fear I should sit too close at my needle.
Susan. Mercy! do you love to sew? oh, I suppose you are learning
fancy work: well, I don’t know but I might like that for a little while.
Laura. No, mother says I must not learn fancy work till I can do
plain sewing extremely well. I was thinking how I should manage to
pass the vacation, and I took it into my head that I would try to make
a shirt by a particular time, and that is Saturday, my birthday. I shall
be twelve years old next Saturday, and then I shall present my father
with a shirt of my own making.
Susan. Did you do all the fine stitching yourself?
Laura. To be sure.
Susan. I am sure I would not make myself such a slave.
 Laura. There is no slavery about it; it was my own pleasure; and
you cannot think how fast it has made the time go. I set myself a task
every day, and then, you see, trying to get just so much done by
twelve o’clock, made me feel so interested!
Fanny. And the rest of the time you have been reading novels, I
see.
Laura. No, indeed; I never read one in my life. Did you think this
library-book was a novel?
Fanny. Let me see it; “Astoria;” is not that the name of some
heroine? let me look at it a little. (Turning over the leaves.)
Laura. You can’t think how interesting it is. It gives an account of a
place away on the western coast of North America; and of all that the
people suffered to get there; and about the very wildest Indians, and
the trappers, and the Rocky Mountains; and here is a map, you see,
Susan.
Susan. Oh, well! it is a sort of geography-book, I suppose.
Laura. Such books will make your geography pleasanter than
ever, I am sure; do read it.
Susan. Not I; I have hardly touched a book or a needle this
vacation, and I have no idea of it. These long summer days are
tedious enough without that.
Laura. But I do believe they would be pleasanter if you were only
occupied about something or other.
Fanny. And so, Laura, you have really spent this whole vacation
without a bit of amusement? I must say I think there is a little
affectation in that.
Laura. Oh no, indeed! I do not like to sit still from morning till night
any better than you do; and mother would not let me if I did. I have
taken a long, brisk walk every day.
Fanny. What, alone? I hate walking alone.
Laura. Not alone, very often; sister Helen sometimes walks over
the bridge into the country with me, and we get wild flowers, and she
explains all about them; that we call going botanizing, and it makes
the walks much more pleasant. It really made me stare when she
pulled a common head of clover to pieces and showed me how
curiously it is made up of ever so many florets, as she calls them;
and even the dandelion is very queer.
Susan. And did you go botanizing in the rain too?
Laura. No; of course we could not stir out then.
Susan. Then I rather think you found the last three days as dull as
any of us.
Fanny. Not she, Susan. No doubt it was very pleasant to sit
perched up at the window all day, for the passers to admire her
industry.
Laura. O, Fanny, how can you be so uncharitable! if you had not
been at the window so much yourself you would not have seen me.
Fanny. But I was not making a display of myself, with a book or a
needle forever in my hand.
Laura. No, Fanny; if you had been occupied, however, you would
not have been making such unkind remarks about your neighbors,
would you? Did you not observe that my mother sat at the window
with me? The reason was, we cannot see to work in any other part of
the room when it is cloudy. You know our little breakfast-room has
only one window.
Susan. So for the last three days you have been reading and
poking your needle in and out from morning till night? Well! it would
be the death of me. (Gaping.)
Laura. Why no; I tell you I do not like sitting still forever, any more
than you do; I like to use my feet every day as well as my hands, and
I presume they expect it. Too much stitching gives me a stitch in my
side; so when rainy weather came I played battledoor and
shuttlecock with father when he came home to dinner, and one day
we kept it up to five hundred and two. Then before tea I used to skip
rope along the upper entry sometimes; and then there was
something else—but I suppose Fanny will tell all the girls in school
and make them laugh at me; but I really enjoyed it best of anything.
Fanny. What was it? tell us, do. I hate secrets.
Laura. You like to find them out, I am sure; but it is no mighty
secret, after all; and I don’t know why I need be ashamed to tell, for
my father and mother made no objection. I went up into the nursery
every evening before the little ones went to bed, and played blind
man’s buff with them.
Fanny. And could you take any pleasure in it?
Laura. To be sure.
Fanny. Then I must say I had no idea you were such a baby. Mr.
Teachall’s best scholar playing romping games with little children! I
am six months younger than you, Laura, but I hold myself rather too
much of a woman for blind man’s buff! I gave that up three years
ago!
Laura. Well! it seemed to make the children enjoy their fun all the
better, and I am sure it did me a deal of good, and did nobody any
harm; so I am content to be called a baby.
Susan. I don’t see how you could take the trouble; it tires me just
to think of going racing about the room at that rate. I should as soon
think of sitting down to study French for amusement.
Fanny. I wonder you did not do that too, Miss Busy. I declare she
looks as if she had! Who would have thought of that?
Laura. I see no harm. You know how terrible hard those last
lessons were before the term ended, and I was afraid I should forget
them; so I have been reviewing the last thirty pages with sister
Helen, to keep what I had got, as she says, and make the next come
easier.
Susan. A pretty vacation, to be sure! How upon earth did you find
time for it all?
Laura. Why, I don’t know. There are no more hours in my day than
there are in yours, Susan. But good-by, girls; I am going to see if