An Industrial Point of View on Control

Teaching and Theory

Edgar H. Bristol
ABSTRACT: This paper presents some ideas
on the relation between the unified discipline
of control as taught in the university and the
differentiated practices of control as they are
developed in different industries.

Introduction
For the university, control is a single unified discipline, developed primarily in terms
of formal andacademic research interests
and subdivided into courses along theoretical
lines. As with most generalized academic
subjects, this has the pedagogic advantage of
building the broadest underlying background
in the most time-efficient way, but it is rnisleading with a broad impact subject like control. For an industrial organization, control
becomes an integration of many technologies
that carry out the organizations commercial
goals in terms of its own and its customers
technical needs and traditions. This totality
of technologies,needs,and
traditionsis
likely to occupy the attention of the control-application engineer far more than the
formal control material presently learned in
the university.
Because each industq is different, each
corresponding industrial view of control is
different. Within these commercial differences, each industry needs the simplest control designs compatible with its goals, its
competitive environment,and any salable
benefits. We know that industrial practice
differs from the available,teachabletheoretical technique in such obvious dimensions as nonlinearity, high-order dynamics,
and process uncertainty.Butmore
subtle
are the differences in the scale of industrial
designsoverwhatcanbetaught(the
100- 1000 loop design rather than the 2-3 loop
design) or the requirements for compatibility
with human operation and maintenance with
the multidisciplinary system design.
Thisis the crux of the theory-practice
gap: First,current theoretical perspectives
can only introduce the issues of real practice
and should not be expected to solve them.
Second, the university orientation to theory
Presented at the 1985 American Control Conference, Boston, MA, June 19-21. 1985. E. H.
Bristol is with The Foxboro Company, Foxboro,
MA 02035.

is justified by teaching efficiency as long

as its limitationsinaddressing
the needs
of practice are recognized. In particular, the
emphasis on a unified control theory tends
to minimize perspectives not fitting an application-independent or theoretical model.
But if a proper distinction is made between
the academic background, as an intellectual
training discipline, and the realities of prac:
tice, the teachers have done their main job.
Those engineering disciplines (chemical,
mechanical. or aeronauticalengineering),
whose university focus is on the design of
particular classes of physical systems (which
may then be incidentally controlled), have a
special advantage in putting control application in its proper perspective. Their primary
focus on a particular class of application
tends to provide a natural interest in the real
applicationneeds. In suchfields,control
courses can be used to unify the applicationrelated background, tending toward a practical view, and even, in some cases, to a
distinct pedagogy 111.
In this regard, electrical engineering students have a peculiar problem. Whereas electrical engineering departments teach most
control courses, the core curriculum of electrical engineering is not focused on design of
objects that can be expected to be the subject
of controls. The student electrical engineer
then naturally learns control as an end in
itself rather than an incidental means of accomplishinglargerandmoreinteresting
goals within the field. He is not given a sufficient grasp of the engineering and human
needs to be addressed by the control system.
For this reason. the control-oriented electrical engineering student has a great deal of
difficulty adjusting to practice and finding

No seriously advocated theoretical technique

has been properly taught unless demonstrated
experimentally, both in realistic success and
in realistic failure.
Experimentalmethodsshouldalsobe
presented as serious and rigorous alternatives
to theory, rather than as an incidental means
for proving theory or as a final fudge on a
fundamentallytheoreticaldesign.Inthe
practical world, most key results are arrived
at experimentally. Experiment supports the
design of systems too complex for current
theory. Experimentapplieswhenthe
theoretical abstractionsare too remotefrom
realityortootediousanddetailed;
[2]
presents a recently commercialized adaptive
controller whose formal basis of design was
entirely experimental,intuitivelyderived,
but without any significant theoretical component. Experiment and simulation are also
to be used when the alternative theoretical
development would resemble a simulation
in longhand.
Theillustration in thisarticle shows a
facetiousview of thepracticelteachingl
theory situation, plotting the usual disposition ofclever talent against the complexity of
the technical problems needing their attention. It divides the world into efforts easily
taught to anyone and, therefore, left to the
common man; efforts within the ability of
current formal analysis and, therefore, preempted by the cleverexperts; and efforts
beyond any systematic attack and, therefore,
again left to the common man.
From another point of view, noting the
direction of motion of the state of the art,
the commonmanhasyoubothcoming
and going. It is important that researchers
and teachers be aware and sensitive to the

commercially appropriate role models. There

differencesbetweenthose

is a Catch 22 here; the electrical engineer

may be in the best position to generalize control design. but only after he understands the
applications being generalized.
The usual method of introducing practice
involves experimental work. (In this country,
experimental work is often sadly neglected
for a variety of reasons.) But experimental
pedagogical methods can be extended: control engineers should not only see real and
simulatedsystems thatverify classroom
work, but, more cynically, systems that show
the failure of overstated theoretical positions.

practice thatrepresenttherecedingpast
from those that need to be seen by their
student engineersas essential elements of
the future and of relating to a human world.
Successfulindustrialpracticedependson
uniquely proprietary capabilities not accessible to simple teaching.
One view is that the university can accomplish no more than it does; practice has to be
taught by practice. Industrial knowledge is
certainly more diverse, disjoint, and informal
than can be easily packaged for pedagogy. It
more logically constitutes the irregular sub-

0272-1708/86/0200-0024 $01 .OO

aspects of the

1986 IEEE
IEEE Control Systems Magazine

24

~
~

Basic Building Block

Functions such as
theApplication
PID.Controller

>8

Practices
such as the PID Application
Practices and Idioms

cc
0

b!

41.

:FEATURING

dP*

m e mBwnmy

Technical Complexity
The Lower Domain of the

COMMON
FORMALLY
MAN

The DckaiA of the

The Upper Domain of the

CLEVER

COMMON
MAN

The disposition of clever people in a well-run (control) technical world (OR W H Y WE A L L NEED AN EDGE!).
ject of some expert system. And industrial
practice is more logically presented, divided
along industrial grounds ratherthan technical
grounds. But there is a less extreme position.
Minimally, at some point, an academic experience might introduce a sensitivity to the
overwhelming impact of economics, market,
and the technology underlying the system to
be controlled on the practical methods.
Beyond this, practice is driven not only
by random practical considerations but also
progressively, by the same forces that drive
the university and evolve unique, but teachable, bodies of coherent practice. Whether or
not the university teaches these,it should
sensitize the young engineer to them, honor
thembecause they definethelivelihood
of most engineers, and recognize them as a
base on which truly useful creative research
will grow.
This paper first appearedas the introduction to a session chaired by the author
at the 1985 American Control Conference
(ACC) on The Practical Teaching of Control.Severalleadingteachers
and practitionerspresentedcurrentthinkingand
methods of teaching the practical view [3].
These papers focused on experimental work.
This paper will develop an industrial view of
other perspective differences of practice from

Februory 1986

the usual teachings in a form that could be the

basis of some leavening pedagogy. The discussion will f i t examine the general range
of industrial economic and operational concerns and then illustrate them with a more
specific example: process control. Even more
than the physical nature of the controlled system, these concernsshape practice. The
physical system is, after all, a derived response to more fundamental human needs.

Industrial Control in General

It is axiomatic that reality is more complex, nonlinear, and of a higher order than
theacademicmodel.Perhapsitistautological also- teaching must illustrate the
complex with the simple. More centrally, industry operates on a competitive edge. A
practical issue, once solved by theory, belongs to everyoneand to no one. A new edge
must then be sought, relative to theissue, by
refining details, replacing it by a more basic
issue, or minimizing its role with a broader
issue. The university most consistently leads
t e c ~ o l o g yin thein;cepbon of a field.-Thereafter, it ignores the needs and insighs of
practice to its peril. Control is an oldfield.
The industrial view of control differs from
the normal university abstraction in at least

these ways:
6 Control is a broader activity in industry,

muchmore so thanimplied by linear

theory -it programs and integrates the
function of the entire controlled system.
Industrial controls are applied to nonlinear
processes operatingunder large disturbances and oftenderived unconventionally from standard analytic formulations
in the engineering field of the system to
be controlled.
High-order dynamics must be directly
faced even when not sensibly modeled.
Practical models are likely tobe sound
only if they are standard parts of the relevant industrial practice. Few engineers are
good modelers.
Industry is much more concerned about
the economic value of time (to design
- and
operate).
The
of failure are
serious and
varied.
A successful engineer contributes to his
companys commercialdifferentiation
ratherthansomegeneralizedconcept
of control.
Industrial controls are designed and used
by ordinary, ordinarily motivated people

25

living ordinary lives. Outstanding industrial design work is ultimately based

on the convenience of the end-user, not of
the designer.
9 Expertise in industry is a matter of adequate grasp of the large and the whole, not
excellence with the small and the partial.
But industries differ from each other:
0

Eachindustryhas
a uniquecontrolapplication technology, differing in state
and detail of development. Here, the process designer has an edge over the control
specialist in matters that may radically affect the best control-design approach.
Differing industries use control in different ways:
- as part of a volume product hidden
from the end-user (which can then be designed and specialized carefully for optimal realperformanceandproduction
cost).
- as a product to be used directly to control some other engineers system by some
knowledgeable user.
- as the means of controlling some proprietary production facility.
The kinds of control and operational objectives may vary: different economic,
safety, or optimality objectives may prevail, stated in formal or informal, steadystate or dynamic, terms, as goals or constraints.
The normal dynamics or nonlinearity of
the processes may build a particular specialized understanding; for example, the
processes may be dominantly resonant or
dominantly dead time.
Who designs and maintains the controls
for the process (or the process itself): the
finaluser, a consultant,or themanufacturer? Where is it most economical or
efficient to position control expertise and
of what form?
Does the industry support many different
or nonstandard controlled systems, or few,

or standard ones?
How long has the industry been practicing control?
How varied is the experience of industry
control designers?
These considerations generate many different forms of control practice.

Process Control as an Example

We can take fluid process control as an
example of a coherent body of practice about
a single body of related industries to be mas26

tered by students of industrial control practice. (It really contains many industries: the
processing industries, the controls suppliers,
sensor suppliers, and all kinds of consulting
and systems-oriented combinations.) The
processing industries include stages of petroleum,chemicals, power, paper and films,
metals, textiles, plastics, food,drugs, and
cement processing. All of these, in some
sense, prpduce homogeneous material products or services that are easily manipulated
(e.g., by valves) and measured. As a consequence. some of these industries have been
automated since 1920, but many date back to
the dawn of history. Thus, the field has stabilized unique expertise and practices.
Fluid processing divides itself along many
lines. One of the basic distinctions is between the use of batch and continuous processes. Batch processes produce products in
batches, with the processingsteps taking
place in succession in a single vessel. In continuous processes, the different processing
steps occur in different, specially designed
vessels. The former are the fluid processing
equivalent of the lowvolume job-shop production. Control of these processes is more
like computer programming. The latter are
the equivalent of the high-volume production
linesanduseconventionalfeedback/
feedforwardcontrol to maintain the state
constancy of the separate units as materials
flow through them. This discussion will be
restricted to the latter case.
In general, the processes are practical to
model for design, but not for control (lowfrequency versus high-frequency behavior).
They are stable and nonresonant. But within
the broad field, there are subdivisions that
support differing degrees of predictability
in their processes. The use of comparatively
pure feedstocks in the petroleum and chemical industries makes these processes predictableandsubjectsthemtotheirown
sophisticated analytical and chemical engineering-derived control design techniques.
On the other hand, the paper, food, and cement industriesusehighlyvariablefeedstocks and take a correspondingly more ad
hoc approach to control. The range of products in the chemical andfoodindustries
changes rapidly with the development of new
productsand new productionfacilities,
putting a premium on the ability to design
and operate varied control systems. However, industries such as power, paper, and
cement are more constant. In the case of the
power industry, the result has been a highly
refined and complex control system.
While the preceding in no sense defines
process control, it does allow one to derive
some logical consequences about the necessary character of process control practice:

Processes that must be well controlled and

can be only partially modeled lead to control designs that must be tuned.Thisis
a critical area of academicignorance.
Except for the occasional references to
Ziegler-Nichols, teaching has not faced
thepossibilityoftreatingmanually
carried-out tuning and system debugging
as a systematic effort, as subject to analysis as any other control issue. And yet,
the experienced engineer is quite capable
of tuninganycontrolstructure
to any
process. Tuning is one way of building
real intuitive control understanding. And
if more engineers understood tuning of
real processes and were not limited by
the theoretical fables necessary in current
theory of statistically derived models, then
practical adaptive control would be further developed.
Processes requiring tuned controls must be
somehow matched or cataloged to suggest
the most likely control structure and anticipate the most likely tuning behavior. The
theoretically or academicallydisturbing
success of the RGA [4], [5] as an orderly
cataloging tool for multivariable processes
comes from this need.
If processes are unique and varied, there
is not the usual chancetoevolvethe
complete plantcontrol design through
progressive commercial installations. But,
effective controlrequires evolution of
standard control or process system modules, and combinations for which standard
human interfaces and tuning and operations practices can independently be developed. Only if the design of the control
elements and standard structures is decoupled from the application in a given
process can the evolutionary limitations of
these unique control designs be overcome.
In process controls, these elements and
practices, which I callidioms 161, [7],
have been evolving for 40 years to a teachably refiied state.
With idiomaticcontrolcomponents in
place, the process of designing complicated controls becomes a logical process,
like amplifier design or computer circuit
design. Dynamics are built in a priori as
experiencesuggests, or as retrospective
tuning and modification (debugging).
This approach speeds the control design
process. And its success creates and depends on practice based on comparative
experience.Ourmajordesigners
may
design or consult on a hundred or more
unrelated,structurallydifferentcontrol
systems during a year and are allowed to
accumulate this experience without being
promoted out of control practice. These

IF Control

Systems Magazine

are not simplyPID-baseddesigns,

although the PID occupies a central
place in
this practice the way the AND gate occupies a central position in logic design.

Concluding Thoughts
Control is currently taught as a unifiedsubject, split along theoretical lines and slightly
colored by differencesintroduced by the
different base engineering fields. Practice is
introduced by laboratoryexperiences of
varying rigor. Depending on theoverall academiccontext, this can be effective.But
experimental methods may play a more fundamental role in control practice, addressing
problems not accessible to theory. And real
practice is divided along different lines, depending on the industry and its needs.When
such anapplication area stabilizes, unique
practicesandadaptationsofthetheory
evolve. Thesecan be taught eitheras specializations based on the engineering field dealing with the system to be controlled or as a
broadening of the normal teaching of unified
control. Wider appreciation of the different
shades of practice must have many benefits
for teaching and research.
Industries with uniquepracticescannot
expect entry-level engineersto know of these

practices. The university challengeis to provide students with a basic design capability,
an awareness of the diversity of points of
view, an abstract cynicism or awareness of
the limitations of all analysis, and a limited
experimentalexperiencethatvalidatesthe
cynicism but c o n f i that real design problems can be solved simply.

[6] E.H. Bristol, Strategic Design: A Practical

ChapterinaTextbookonControl,

1980

Joint Automatic Control Conference, San

Francisco, CA, Aug. 1980.

J. McAvoy,and E. H.
Bristol, A Method for the Analysis of Complex Control Schemes,1982 American Control Conference, Arlington, VA, June 1980.

[7] A.Prassinos,T.

References
G . Stephanopolis, Chemical Process Control-An Introductwn to Theory and Practice, EnglewccdCliffs,NJ:Prentice-Hall,

Inc.,1984.
E.H. Bristol, The EXACT Pattern RecognitionAdaptiveController,aCommercial
Success, in English, 4th Yale Workshop on
ARplicationsofAdaptiveSystemTheory,
Center for Computer Science, Yale University, New Haven, CT, May 29-31, 1985.
K.J.AstromandA.Ostberg,AModern
TeachingLaboratoryforProcessControl,
1985 American Control Conference, Boston,
MA, June19-21,1985.
E.H. Bristol, A New Measure of Interaction
for Multivariable Control,IEEE, PPTGAC,
V O ~ .AC-11, no. 1, Jan. 1966.
T. J. McAvoy, Interactwn Analysis, Principles, and Applications, Instrument Society
of America, 1983.

Edgar H. Bristol is an

ElectricalEngjneer (MIT
B.Sc.,1958)with
a
degree in mathematics
(BeloitB.A.,1957).He
has beenemployed by
The FoxboroCompany
since 1959. He is active in
AIChE,IEEE,ISA,and
ACM.Hisprofessional
activities have centered in
thefieldsofcontroland
computer science, notably
adaptive and multivariable control, and application
language design. In the past, he has taught in the
ChemicalEngineeringDepartmentatLouisiana
State University and currently holds an Honorary
Lectureship at MIT. Mr. Bristol is a Member of
the JEEE.

Conference Calendar
1986 IEEE International Conferenceon
Robotics and Automation, Apr. 7-10, 1986,
San Francisco,California.Contact:
Rajan
Suri,University of Wisconsin-Madison,
Lkpt. of Industrial Engineering, 1513 University Ave., Madison, WI 53706, USA.

The 18th SoutheasternSymposium on

System Theory, April 7-8, 1986,Hilton
Hotel, Knoxville, Tennessee. GeneralChairman, Dr. E W. Symonds,Electrical Engineering Department, Ferris Hall, University
of Tennessee, Knoxville, TN 37996-2100,
phone: (615) 974-5475.

Sixth Power Plant Dynamics, Control

and Testing Symposium, Apr. 1616,1986,
Knoxville,Tennessee.Contact:B.R.
Upadhyaya,Department of Nuclear Engineering, The University of Tennessee, Knoxville. TN 37996-2300,USA,phone:
(615) 974-5048.

Optimization Days 1986, April 30-May

2,1986. Montreal, Canada. Send contributed
papers before January 31, 1986, to Alain
Haurie,DirectorofGERAD,Ecoledes
H.E.C.,5255avenueDecelles,Montreal
(Quebec) H3T1V6,Canada, phone: (514)
3406042.
Febnrary 1986

1986 IEEE International Symposiumon

Circuits and Systems, May 5-7, 1986,

2nd WAC Workshop on Adaptive Systems in ControlandSignal Processing,

LeBaronHotel,San Jose, California. Program Char

iman: Prof.W.K. Chen,Dept.
of ElectricalEngineeringandComputer
Science, University of Illinois, Box 4348,
Chicago, IL 60680, phone: (312) 996-2462.

July 1-3, 1986, Lund,Sweden.Contact:

K. J . A s t r h , Department of Automatic
Control, Lund Institute of Technology, Box
118, S-221 00 Lund, Sweden.

1986 American Control Conference

June 18-20,1986
Seattle, Washington
General Chairman:
Dr. Edwin B. Stear
The Washington Technology Center
376 Loew Hall, FH-10
The University of Washington
Seattle, WA 98195
Phone: (206) 545-1920

Fourth IFAC Symposiumon Control

of Distributed Parameter Systems,
June 30 to July 3, 1986, California Institute of Technology, Pasadena, California.
Contact: G.Rodriguez,JetPropulsion
Laboratory, Mail Station 198-326, 4800
Oak Grove Drive, Pasadena, CA 91109,
USA, phone: (818) 354-4057.

Japan-USA Symposiumon Flexible Automation, July 14-16, 1986, Osaka, Japan.

Contact:Prof.M.Tomizuka,Dept.
of
Mechanical Engineering, Universityof California, Berkeley, CA 94720, USA, phone:
(415) 642-0870.

SIAM Conference on Linear Algebra

in Signals, Systems, and Control, August
12-14, 1986, Boston,Massachusetts.Send
abstracts of contributed papers by March 14,
1986, to SIAM, LASSC Conference, 117
South 17th St., Suite 1400, Philadelphia, PA
19103-5052, phone: (215) 564-2929.

IFAC Aquaculture 86- Automation

and Data Processing in Aquaculture, Aug.
18-21, 1986, Trondheim, Norway. Contact:
Prof. J. G. Balchen, Divisionof Engineering
Cybernetics,TheNorwegianInstitute
of
Technology, University of Trondheim, 7034
Trondheim -NTH, Norway.
Continued on p. 52
27