Professional Documents
Culture Documents
Advanced Chemical Process Control 1St Edition Morten Hovd Full Chapter
Advanced Chemical Process Control 1St Edition Morten Hovd Full Chapter
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
Contents
Preface xvii
Acknowledgments xxi
Acronyms xxiii
Introduction xxv
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
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],
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
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.
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.
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
Introduction
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.
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.
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
Production planning/
scheduling
Supervisory control
Regulatory control
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:
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:
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
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
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.
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.
yk = g(xk , uk , dk ) (1.4)
1.2 Models for Dynamical Systems 3
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.
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.
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.”
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.
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.
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.
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.
Something Wonderful.