502 Book Reviews
Chemical Process Control: An Introduction to Theory and
Practice*
George Stephanopoulos
Reviewer: ODD A. ASBJORNSEN
Department of Chemical Engineering, University of Houston,
Texas, U.S.A.
CHEMICAL engineers are generally unfamiliar with the theory of
linear systems, networks of cause and effect relationships and
even more unfamiliar with electronics, instrumentation and on-
line computer hardware. On the other hand, control and
electrical engineers are usually unfamiliar with the physical and
chemical background and the complexity of process technology.
The reasons for this are found in the university education.
Furthermore, the chemical industry is conservative and has a
slower pace of development in comparison with other fields like
electronics. The reason for this is found primarily in the very large
plant investments, where only well-proven control technology
will be accepted at the design stage.
These are some of the explanations why traditional control
engineering based on the transfer function concepts, frequency
response analysis, complex algebra, Nyquist plots and similar
concepts from linear systems theory have found only limited
interests and applications in the process industry. This situation
has led some authorities in the field to criticize control
engineering as not being applicable to process control. This
situation has also led to great expectations whenever a new
textbook in process control is brought on the market.
George Stephanopoulos obviously had some of these general
thoughts in mind when he wrote his book. He tries to motivate
the reader to increased emphasis on process control problems
and application of modern systems engineering and mathe-
matical modeling techniques to process engineering. He also tries
to present the modeling concepts in a simplified way which would
attract control engineers to the subject. However, he does this in a
traditional way where some of the less applied design theories
and criteria for process control systems are given too much
weight. Examples again are the transfer function concepts, the
Nyquist stability measures for single loops and the frequency
response.
Chemical process control has some distinct features which
should be emphasized. The structure of the production plant
itself is that of a geographically distributed one. The structure of
the control functions to be executed is that of a multivariable,
decentralized and hierarchical system. Most control com-
putations and actions are executed in parallel. The economic
significance of a control function is increasing upwards in the
hierarchy, but so are the time constants. The nature of the
mathematical models for the process and plant dynamics is
extremely complex, but the general principles of conservation,
lumping (or integration) transport phenomena, kinetics and
equilibria, make it possible to generalize and simplify. The
control variables are bounded and the operating conditions
usually subject to inequality constraints, nonlinearities and
bottlenecks.
Traditional control engineering basics usually do not take the
real effects of bounded control, process constraints, and non-
linearities into account, apart from some applications of
*Chemical Process Control: An Introduction to Theory and
Practice, by G. Stephanopoulos. Published by Prentice-Hall,
Englewood Cliffs, New Jersey (1984). ISBN 0-13-128629-3.
696 pp., 184 illustrations. U.S. $47.20.
linearization. This statement indeed applies to the present book
which has the following structure.
It starts with an overview of process control, its objectives, the
process variables involved, the cause-and-effect relationships
needed and a general motivation of the reader to dig deeper into
the subject.
Then the reader is introduced to the technique of modeling
from the basic physical principles by conservation equations,
transport phenomena, kinetics and equilibrium. At this point
very little is said about alternative modeling techniques like time
series modeling or step or impulse response modeling.
The static and dynamic characteristics are analyzed by
linearization procedures and LaPlace transforms applied to
linear differential equations. The author includes a refresher of
some fundamentals of the LaPlace transform, where the useful
relationships between derivatives in the s-domain and moments
in the time domain are missing.
The introduction of the LaPlace transform naturally leads to
the transfer function concepts in a traditional way. This is used
extensively in the analysis of transient behavior of first-, second-
and higher-order systems. The traditional extension of the
transfer function concepts into frequency response and Nyquist
stability analysis is also followed.
At this point, about 55°o of the pages are covered in a
traditional way. The author then starts a series of chapters on the
analysis and design of advanced control systems where modern
concepts of adaptive and inferential control, multivariable
control and decoupling are included. This involves qualitative
analysis of control systems for complete plants.
Only about 20% of the last part of the book covers the most
significant breakthrough in the history of Chemical Process
Control, the digital computer. This is unfortunate, because this is
the area where most chemical engineers will meet their challenges
in their future career in process control.
The dynamic characteristics of the operation and control of
chemical plants have some distinct features which may qualify for
a special treatment. Therefore, one has great expectations
whenever a chemical engineer writes a textbook in process
control, to see if advantages have been taken of these special
features. Unfortunately, the author did not quite hit the target.
For example, a general phenomenon in the conservation
equations is that they always give rise to bilinear terms from the
products of extensive and intensive thermodynamic properties. A
large number of interaction problems in the processes may be
explained by this basic phenomenon. In the book it is treated as
special cases for heat and species in a stirred tank balance.
Another example, the most fundamental and general principle
for chemical plant and process control, is the conservation
principle of energy, mass and species. Hence, the process control
system should be based on these principles in the first place. I n the
book, these principles are referred to only as examples in
feedforward control and in a total plant control. Process control
following the conservation principle will always lead to a
combined feedforward and feedback solution, where the main
disturbances are identified and measured and even parameters
estimated.
Yet another example, the application of the simple
conservation principle for control, solves most of the decoupling
problems caused by the bilinear nature of the extensive and
intensive properties. This is now standard practice in real
industrial systems, where the computer plays the central role. In
the book, the decoupling is solved as a general diagonalization of
linearized transfer function matrices which does not make use of
the special features of the conservation principle in the process
dynamics.
 Book Reviews
The feedback cascaded ratio controller is the industrial analog
solution to conservation principle control, and this has been in
operation for generations. However, apart from the ratio
controller and some computing elements, traditional analog
instrumentation and control did not practically allow for
dedicated control structures and models of the kind the process
industry really needed and wanted. Consequently, one may look
upon the digital computer at any position and level of the process
control hierarchy as a tremendous technological breakthrough
which has changed the attitude of the process industry to process
control dramatically. One should have liked to see the emphasis
on digital techniques and discrete time modeling and control as
80% of the book rather than 20%.
As emphasized in the book, the computer technology has
changed many of the process industries from being conservative
and reluctant to apply new control technology, to setting a much
higher pace in developing its own tailor-made applications of
models and control structures to solve their process control and
instrumentation problems. In this development process, some of
the ideas brought forward in the chemical engineering world are
mere transcriptions of what control engineers developed 20-40
years ago, but which they failed to market and apply to the
process industry. The technological barrier in hardware and its
costs, lack of reliability, and user-friendliness are among the
reasons why they failed. The improper understanding of the
process dynamics, both qualitatively and quantitatively, and the
mismatch between control and chemical engineers are other
reasons why a lot of the control engineering theory failed to be
applied. The book certainly contributes to improve this situation
very much, but it does not take the full step to do it on the process
engineering ground. It still has far too much of the traditional
control engineering material which has been proved to be
difficult to find practical excitement for. The author seems to have
limited experience in practical industrial process control.
As an example, consider the classical complaints from the
practitioner, that too much emphasis is placed on control design
methods which are practically never used in industry. Some of the
methods frequently mentioned are the Nyquist plots and the
Bode plots, left aside the whole area of transfer functions and
frequency response analysis. The reason is that data are difficult
to generate for the analysis at the design stage and that much
simpler tuning methods, manual or automatic, are available once
the plant is ready to start up. Hence, a lot of this traditional
material in the book could be sacrificed to the benefit of tuning
methods.
Industrial process control, in its widest possible context, has
grown out of the laboratory and the classroom. The single loop
control and its instrumentation is the least significant problem
and is usually mastered at the technicians level. Furthermore,
many plants have recommended control structures developed
and improved over the years as a well proven technology, and the
design is reduced to specifying equipment. The book would
benefit from a more practical input.
As an example from the analysis of cascaded systems in the
book, the fluid flow control loop is in most cases considered as an
extended valve positioner, where the feedback is taken from the
flow rather than the valve stem position. It is recommended
where precise flow manipulations are required and where valve
characteristics and pressure drops should be automatically
compensated for. But this implies that feedback from the valve
stem may be removed. In cascaded systems, the flow control loop
may be considered the final element, similar to the valve
positioner.
Another practical example, is the rule of using only one control
valve per fluid pipeline, unless the valve is a stable pressure
reducing valve. The author breaks this rule in his attempt to
illustrate conflicting loops on page 482. In industrial practice,
situations like those would hardly occur, or at least be eliminated
by an a priori analysis of variables and degrees of freedom, in fact
what the author starts with. In removing the loop conflicts, the
author fails to obey the conservation principle by forgetting the
cascaded control from liquid level in a separator to the setpoint of
the outlet flow controller, which happens to be an inlet to a
distillation column. He also introduces plant changes by an
unnecessary purge stream, disregarding the practical experience
that it is definitely possible to control two compressors working
on the same manifold provided one is the master of the manifold
503
pressure and the other is the master of its own suction pressure.
At the same time as the design, tuning and maintenance of the
single loops at the bottom level of the process control hierarchy
have moved down to the technicians level and responsibility, the
chemical engineers are faced with great challenges in plant
operation and higher level control system design. They would
have to think more in the systems engineering frameworks to be
able to understand multivariable process dynamics and control,
hierarchical control structures, and information systems,
geographically distributed and functionally decentralized and
decomposed systems, general aspects of process modeling and
optimization and even human factors and process-operator
interfaces. The chemical engineer is very poorly trained for this in
the present curriculum and this is probably the greatest challenge
the author faced in writing his book. It is possible that the
training in process control for chemical engineers should follow
along the lines, now under experimentation at some universities,
namely that commercial software systems like for example ACS
from IBM should be introduced in the classroom directly and
from the very beginning. But that would require another type of
textbook which is different from that of the present book.
Regardless of the future approach to the teaching of process
control, the author has focused on some very important factors
and deficiencies in the mathematical background of chemical
engineers. One would desire that mathematical modeling should
have a deep root in process engineering and mathematics which
would be common to and essential for many other fields, i.e.
design, unit operations, transport phenomena, material and
energy balances, etc. As it now stands, this seems to belong more
or less exclusively to process control which turns out to be an
unnecessary voluminous course covering up for the deficiency in
mathematical training. The present book with its 680 pages is a
good example.
This situation is very unfortunate because process control has
a tendency to be considered a stranger in the chemical
engineering curriculum, while it is really an integrated part with
common tools. Treatment of linear algebra, general systems of
algebraic and differential equations, and LaPlace transforms and
transfer functions should be common to a lot of chemical
engineering topics like stoichiometry, conservation and balance
of materials and energy, transport phenomena, reaction kinetics
plant start-up and shut-down and even economics. It should be
dealt with in the mathematics courses parallel to or prior to
process control. The author spends approximately 20 % of the
book on these mathematical topics, which may mislead the
student to think that the operational type of mathematics is
something special for process control.
In modern process control, the concepts of states and state
space, parameters, and parameter space are crucial both for
continuous and discrete systems. Similarly, the concepts of linear
algebra and vector formulation are essential for computer
simulations, plant design, and optimization. This requires a
similar mathematical background and a general application of
this in the chemical engineering curriculum to a greater extent
than what is the case today.
The technology push by the computers on the industrial
process control market has been recognized for several years;
now the market pull becomes visible as a desire to implement
special structures of process control like Dynamic Matrix
Control, Model Algorithmic Control, Self-Tuning Regulators,
Inferential Control, etc. These topics are covered in the last part
of the book in a professional way. However, one should have
liked to see this part expanded and the first part as a process and
systems engineering foundation leading up to the modern
computer control algorithm in the last part. The adaptive nature
of modern control systems plays an important role in this picture,
and the treatment of parameter estimation schemes in the book
represents the bare minimum of what one should expect a future
chemical engineer to know.
The consequence of this development should be recognized in
chemical engineering as a greater emphasis on the systematic and
phenomenological approach with the chemical engineering
specialities as examples. In the 1930s, the concepts of unit
operations were introduced as a concept for systematization of
chemical engineering. In the 1980s, one would expect systems
engineering concepts to play a similar role. The present book is a
step in this direction, although limited to the control aspects.
504 Book
It must be taken Ior granted that the chemical engineering
departments take up thesc challenges and that the future student
in undergraduate process control masters the nlost elementary
concepts of linear algebua, ordinary algebraic and differential
equations including LaPlace and Fourier transforins. Further-
more, it must be taken for granted thal the future student masters
the chemical engineering production process technology from
courses prior to process control. When this symbiosis of systems
and chemical engineering is achieved the present book may be
reduced to one-third and process control becomes what it should
be an integrated part of the process technology with emphasis
on thc real challeuges of process control to the chemical
engineers, namely the last one-third of the book. This covers the
Probability Theory and Mathematical Statistics*
V. S. Pugachev
Reciewer: A R U N A B H A BAGCHI
Department of Applied Mathematics, Twente University of
Technology, 7500 AE. The Netherlands
A ~L Xq--BOOKon probability and statistics for social or medical
science is published more often than one specifically designed for
engineers. The reason probably is twofold: engineers are
confronted more with stochastic processes in practice than with
purely statistical questions (explaining an overabundance of
books with the general theme of probability and stochastic
processes for engineers) and more importantly, engineers have
sutticient mathematical background to follow a regular course on
mathematical statistics. In fact, with the exception of the last two
chapters, Academician Pugachev's book is really a general
introductory text on probability theory and mathematical
statistics.
As mentioned in the Introduction, the author avoided measure
theory altogether. Statistical decision theory was also barely
touched upon. In this sense, the material covered is entirely
classical. Within its stated objective, however, this is a
comprehensive book containing all aspects of probability and
statistics. This is a formidable undertaking in which the author
succeeded admirably well. Given the constraint of including such
a vast material in one book of managable size, many topics are
discussed somewhat hurriedly. Some of them may be omitted in
teaching a course based on this book and for the rest, additional
material should be provided for the beginning students.
The first five chapters deal with probability theory. Chapter 1
is concerned with probability of events. Along with the frequency
interpretation of probability, standard statistical measures of
describing random phenomena are introduced. Calculations of
probabilities of events are treated in detail, leading to the axioms
of probability theory. Conditional probabilities, repeated trials
(binomial distribution) and Poisson distribution are also covered
in the first chapter. Chapter 2 deals with random variables. Both
discrete and continuous random variables are studied, along with
their distribution and density functions. The additional feature is
the discussion of entropy of a distribution. Entropy is a central
notion in information theory and the author did well to focus on
this concept throughout the book. Chapter 3 is on various
numerical measures of random variables. The key idea is, of
course, that of expectation. This leads to moments and central
moments of various orders. Quantiles are also introduced in this
chapter. Finally, all these characteristics are analyzed in detail for
the case of one-dimensional normal distribution. An interesting
topic in this chapter is canonical expansions of random vectors.
* Probability Theol 3' and Mathematical Statistical, by V. S.
Pugachev. Published by Pergamon Press, Oxford (1900). U.S.
$90.
Reviews
concepts of muhivariablc control, material and energy balance,
ratio and feedforward control, model algorithmic control, and
inferential control together with parameter estimation, analysis
and filtering of time-series. AR MA-models and general computer
control.
As it now stands, the book is a step towards this symbiosis but
hardly the final text. It should certainly be read and discussed as
another contribution to a proper understanding of the complex
nature of chemical plants and their equally complicated control
structure. It should also be regarded as a step in the righl
direction, a ~mified treatment of process control on the premises
of process technology.
This and corresponding results for stochastic process are very
important in the area of signal processing. Chapter 4 discusses
random vectors. The material, although standard, is presented in
an original manner. Thus, projection of random vectors are used
to define marginal distributions and densities and thereby, study
conditional distributions and densities as well. This leads to
conditional expectation, the central motion in filtering and
smoothing problems arising in control and communication
systems. Characteristic functions of random variables are
exhaustively treated in this chapter. Chapter 5 considers
functions of random variables. The results are used to derive,
a m o n g others, / : , t and F-distributions. This sets the stage for
studying mathematical statistics, h also includes a brief
discussion on limit theorems in probability theory.
The next three chapters are concerned with mathematical
statistics. Chapter 6 discusses estimation of parameters of
distribution. Both point estimation and determination of
confidence intervals are studied, as also the elementary concepts
related to testing hypotheses about parameters of distribution. In
fact, this chapter contains many useful statistical techniques one
needs in practice. Chapter 7, by contrast, is more of theoretical
nature. It contains detailed discussions about asymptotic
properties of estimators, with special attention paid to the
method of m a x i m u m likelihood. Recursive estimations, and more
specifically, stochastic approximation schemes, are also included.
Chapter 8 deals with estimation of distribution and density
functions. Various methods of testing hypotheses about
distributions are considered and the chapter ends with a short
discussion about statistical simulation methods.
The last two chapters are concerned with some statistical
models which occur often in engineering. Chapter 9 covers
regression models and contains an introduction to analysis of
variance and experimental design. These topics are handled too
abruptly and could be deleted altogether. Chapter 10 deals
primarily with autoregressive models and these are becoming
increasingly important in engineering and in other disciplines as
well. The last three sections deal with factor models, recognition
models and decision-making models. The choice is somewhat
arbitrary and reflect the personal taste and research interest of
the author. It is not clear why Neyman Pearson theory is
introduced only in the last chapter.
The book contains numerous examples which are instructive
and form an integral part of the text. For direct application of the
proposed statistical techniques to practical problems, the author
has given reference throughout to corresponding standard
programs given in IBM Programmer's Manual. The reviewer,
however, is of the opinion that other statistical packages are more
commonly used nowadays, rendering the references to IBM PM
to limited usefulness. The book includes a comprehensive list of
statistical tables at the end. In summary, the book under review is
very well written and appears free of typing errors. It is a good
introductory text in probability and statistics.