Strategy for

Distillation-Column
Control

n chemical plants and petroleum refineries, there are, today, many distillation
columns that are working well. There are also many others that are not working
well, and at least a few that function very poorly, or not at all. Failure to obtain
performance specified by the column design engineer is due, in many cases, to
faulty or inadequate control system design. Troubleshooting of columns that
are already in operation is frequently necessary, but practical considerations
usually limit corrective measures to relatively minor items. Proper original
design is by far the best way to guarantee satisfactory operation and control.
Therefore, in this book we will approach the design of integrated distillationcolumn control systems as a systems problem in process design. The application
of feedforward, feedback, and protective controls wdl be coordinated with the
sizing and proper location of process holdups to achieve both automatic startup and shutdown and smooth, noninteracting control of column product
compositions.

1 .i

DISTILLATION CONTROL OBJECTIVES

The starring point of any design project is a definition of objectives. For

distillation there are many possible approaches, but the one chosen here is one
the authors have found broadly useful in virtually all kinds of processes. It
has three main facets: (1)material-balance control, (2) product quality control,
and (3) satisfaction of constraints. As applied specifically to distillation columns,
this philosophy suggests the following:
1. Material-balance control

* This term is sometimes used by others to mean a control system in which reflux is set
by reflux drum level control, and distillatelfeed ratio is set manually or by a composition (temperature)
controller. The authors of this book have been unable to find any special merit for this scheme
except for some high reflux ratio columns.
3

StrMeD J%- Dthdktim-Column Control

-The column control system must cause the average sum of the product
streams to be exactly equal to the average feed rate. Harbed4 has called
this requirement that of keeping the column in balance.
-The resulting adjustments in process flows must be smooth and gradual
to avoid upsetting either the column or downstream process equipment
fed by the column.
--Column holdup and overhead and bottoms inventories should be
maintained between maximum and minimum limits.
It is important to note that the material-balance controls on any
given column must be consistent with the material-balance controls on
adjacent process equipment. In most cases material balance will be controlled by so-called averaging liquid-level or pressure controls.
2. Product quality control
The control system for a binary distillation in most cases must:
-Maintain the concentration of one component in either the overhead
or bottoms at a specified value.
-Maintain the composition at the other end of the column as close as
possible to a desired composition.t
It is usually true that minimum operating cost is achieved when the
products are controlled at minimum acceptable purities. This is so because
the relationship between thermodynamic work of separation and purity
is nonlinear. For some columns compositions are allowed to vary at
one end, and sometimes both ends, to satisfjr certain economic constraints.
Both material-balance and composition controls must function satisfactorily in the face of possible disturbances in:
-Feed flow rate
-Feed composition
-Feed thermal condition
-Steam supply pressure
--Coohg-water supply temperature
--Cooling-water header pressure
-Ambient temperature, such as that caused by rainstorms

3. Satisfaction of constraints
For safe, satisfactory operation of the column, certain constraints
must be observed. For example:
-The column shall not flood.
--Column pressure drop should be high enough to maintain effective
column operation, that is, to prevent serious weeping or dumping.

t For multicomponent columns subjected to feed composition changes, it is not possible to

hold exactly constant the compositions at both ends of the column; the composition at one end
must change a little. If feed variations are in the nonkey components, composition may vary
somewhat at both ends of the column. With only two drawoffs, we can conml two keys, but
not the nonkey components.

1.I

Distillutk Control O&jemves

-The temperature difference in the reboiler should not exceed the

critical temperature difference.
--Column feed rate should not be so high as to overload reboiler or
condenser heat-transfer capacity.
-Boilup should not be so high that an increase will cause a decrease
in product purity at the top of the column.
--Column pressure should not exceed a maximum permissible value.
There are, in addition, several other facets of column control.

Startup and Shutdown

Column controls should facilitate startup and shutdown and, by implication,
should make it easy to achieve total reflux operation when desired.
Transitions

When it is desired to change product compositions, the column controls

should facilitate doing so. This is particularly important when feed stock composition varies widely and it is desired to optimize column or train operation,
as, for example, with a computer.
Heat Recovery

Increasingly there is an interest in recovering as much heat as possible. The

petroleum industry has frequently used the sensible heat in a column bottom
product to preheat column feed. Recently more ambitious schemes have been
employed in which the reboiler for one column is used as tha condenser for
another. Such schemes magnify control problems and sometimes limit process
turndown.
Testing

There should be enough instrumentation so that testing may be carried out

for tray efficiency, heat and material balances, flooding, and so on.
Miscellaneous

In addition to the above, column control should be designed with human

engineering in mind. For example:

-The operators work station, whether a cathode-ray-tubekeyboard console

or a panelboard with gages, switches, recorders, and so on, should be carefully
designed according to human engineering principles for easy use.
-The controls should be so designed as to require minimum maintenance.
The need for frequent or critical tuning should be avoided. The hardware
should be designed and arranged for convenient access and quick repair or
replacement.
-The control system design engineer should use the simplest possible
design procedures, not only to hold design costs down, but also so that the
work can be readily discussed with other design and plant personnel. This will
facilitate, at a later date, any necessary minor redesign at the plant site. Failure

Stratemfw Didhttim-Column Control

of design and plant personnel to achieve a mutual understanding of design

objectives, concepts, and methods is one of the frequent causes of unsatisfactory
plant operation.

1.2 ARRANGEMENTS FOR MATERIAL-BALANCE CONTROL

As shown elsewhere, the size and location of tanks and the concept of
overall material-balancecontrol used can have a great influence on plant investment
and process control. If the design engineer uses the concept of control in the
directwn opposite t o w , tanks may be smaller and plant fned investment and
is used. The
working capital can be lower than if umtroZ in the directwn
meaning of these expressions is illustrated, for simple tanks with level controllers,
in Figures 1.1 and 1.2.
For the last tank in a series (final product storage), the demand flow is
always the shipments to a customer. As shown by Figures 1.3 and 1.4, the
choice in control strategy is between adjusting the flow into the last tank (control
in direction opposite to flow) or adjusting flow into the first process step
(control in direction of flow). In the first case, we can easily use simple automatic
controls. In the second case, it is more common to have an operator make the
adjustment.
When more than one tank is involved, other advantages of control in the
direction opposite to flow are (1) less difficulty with stability problems, and
(2) reduced internal turndown requirements. Turndown, as used here, is the
ratio of maximum required flow rare to minimum required flow rate. In this
instance the meaning is that, in response to a given change in demand flow,
the required change in the manipulated flows will be smaller in one case than
in the other.
Once the basic concept of material-balance control has been selected for a
process, one must apply the same concept to all process steps. It is for this
reason that the first step in designing column controls is to determine the
material-balance control arrangement. Control in the direction of flow is the
most commonly used concept (although the least desirable), and a frequently
encountered arrangement is shown on Figure 1.5. Here level in the condensate
receiver (also commonly called reflux drum or accumulator) sets the top product,
or distillate flow, while the level in the base of the column sets the bottom
product flow; in other columns base level sets steam or other heat-transfer
media to the reboiler, in which case the condensate receiver level sets top
product flow.
Generally speaking the direction of material-balance control is determined
by the demand stream. In recycle systems we may find some material-balance
controls in the direction of flow while others are in the direction opposite to

@e

Material-balance control, in the direction opposite to flow, can lead to many

interesting level-control and flow-ratio options. These are discussed in derail
in Chapter 6 .

1.2 A w a q m fm
~ M a t d - B a l a n c e Control

FIGURE 1.1
Material balance control in direction opposite to flow

FIGURE 1.2
Material balance control in direction of flow

Ly

er

VI

Strazcm fb.DistiuatiOn-column Control

-g
k=z
u

c
0
,

0
G
Q
c,

-s
a
a
0
c
0

a3

-E

U
C

--2

C
0

c,
0

-s3
--m

ki
@?I4

-W -E

a-

32

E8

1.2 Arrangements fm Material-Balance Control

C
0

Q)

3
C

3
-0

-8s
2
-8m
2
CI

-3
c
.e
-3

E8
8
-mc
d
-

-?$

z' E=e,o3
3

iz&%

U Q

10

Strategy for Dljtillatwn-Column Control

FIGURE 1.5
Distillation column with material balance control in direction of flow

1.3 Funahmentnls of Composition Control

11

1.3 FUNDAMENTALS OF COMPOSITION CONTROL

Let us consider briefly what must be done to a column to keep terminal
compositions constant on a steady-state basis when the column is subjected to
sustained changes in feed flow rate or feed composition. Methods of handling
other disturbances will be discussed later.
To simpl@ the analysis, let us limit our attention to an ideal, binary distillation.
This is somewhat restrictive, although the results will be applicable in a general
way to multicomponent systems, particularly those that may be treated as quasib&ary or pseudobinary systems.
As will be shown in Chapter 2, if feed composition and feed thermal
condition are constant, then we want the operating lines on a McCabeThiele diagram to remain constant when feed flow changes. The operating lines
(defined in the next chapter) will not change as long as the distillate-to-feed,
reflux-to feed, boilup-to-feed, and bottom-product-to-feed ratios are held constant.
Practically speaking one may hold all four ratios constant by fixing anv one of
the three pairs: (1) the reflux-to-feed ratio and the boilup-to-feed
ratio, (2) the reflux-to-feed ratio and the bottom-product-to-feed ratio, and
(3) the boilup-to-feed ratio and the distillate-to-feed ratio. In considering case
(1), for example, we see that if the rectification section vapor-to-feed ratio is
fixed (and it will be if the boilup-to-feed ratio is fixed) and the reflux-to-feed
ratio is fixed, then the distillate-to-feed ratio will be fixed, since the distillate
flow is the difference between the reflux flow and the vapor flow. Similarly,
the bottom-product-to-feed ratio will be fixed, since the bottom product flow
is the difference between the stripping section liquid and the boilup.
If feed rate and feed composition are not constant, Rippin and Lamb have
shown that, for small perturbations, one should change the boilup and reflux
according to the following equations:

AV

where

ALR

K ~ A z F+ KfzAF

= Kf3AzF

+ Kf4A.F

vapor flow from the reboiler

LR = internal reflux flow at the

top of the column
The As represent departures from average operating conditions. The constants
KO Kf4 may be calculated approximately by the column designer. Luyben6
has shown that it is necessary to be quite careful in designing feedforward
compensation for feed composition changes, particularly when the column is
not making a sharp separation.
Visualization of column operation, in terms of reflux-to-feed and boilupto-feed ratios, was suggested by Uitti7 and has since been proposed in varying
degrees bv many others. Throughout the rest of this book, it will be used as
the primar). basis for column composition control.

12

StrateBy fm DirtillatMn-Column Control

It should be noted, however, that this "feedforward" approach to column

control has a particular limitation: In general one cannot calculate the constants
Kfl--.Kf4 with great accuracy. For columns that are not operating too close
to either upper dr lower h i t s of capacity, small changes in feed rate, and
consequent changes in boilup and reflux, will not change tray efficiency appreciably.
The terms Kf2and Kf4 therefore will be constants. If the feed composition
changes are not too large (as will usually be the case), then Kfland Kf3may
also be treated as constants. To determine the control accuracy obtainable by
this approach, one should make the necessary calculations or tests for each
individual column. Where really close control is required, one must supplement
feedforward control with measurement of the column terminal compositions
and subsequent feedback control, at least at one end of the column. The usual
philosophy will be to use feedforward for fast, approximate control and feedback
for long-term, accurate control of composition.
It should be noted, too, that feedforward from feed composition may not
be needed if the feed comes from a process step with discharge composition
control. Feedfonvard compensation for other process variables, such as bottom
product or distillate demand flow, is discussed in Chapter 6.
As will be shown later (Chapters 5 and 20), a properly designed column
feed system can play a very important role in filtering out hsturbances in feed
rate, feed composition, and feed enthalpy, thereby making composition control
much easier.

1.4 COMPENSATION FOR VARIOUS DISTURBANCES

Feed Thermal Condition
Feed should enter the column with a constant enthalpy. When significant
changes are anticipated, a heat exchanger and feed-enthalpy control system
should be provided. This is discussed in more detail in Chapters 5 and 11.

Steam Supply Pressure

Changes here can cause changes in boiling rate. The best preventive is a
steam-to-feed ratio control system combined with temperature and pressure
compensation of steam flow. A high pressure drop across the steam valve favors
smooth control but velocity-limiting trim may be required to minimize noise
and plug and seat wear. A high pressure drop is also undesirable for energy
conservation. If the pressure drop is high enough, sonic flow through the valve
results, and reboiler steam-side pressure has no effect on flow rate. The design
should avoid having sonic flow corresponding to low feed rates and nonsonic
flow corresponding to high feed rates, since required controller gain changes
make tuning very difficult.

1.5 Startup and Shutdown

13

Cooling-Water Supply Temperature

Cmling-water temperature changes are usually seasonal, and will require

no specific correction. If, for some reason, they are large and rapid, then it
may be desirable to provide an enthalpy control system for the condenser. By
measuring the temperature rise of the cooling water across the condenser, and
multiplying it by the cooling-water flow rate, one has a measure of the heat
transferred, 4.. This calculated value of qc can serve as the measured variable
in an enthalpy control system. For column pressure control, the enthalpy control
system can serve as the secondary loop in a cascade system.

Cooling-Water Header Pressure

One of the best ways to make a flow system immune to pressure changes
is to provide a high system pressure drop. This, however, is costlv. Another
way, which should be satisfactory for some distillation columns, is to use a
cooling-water flow control system. It does, however, have limitations, which
will be discussed later.

Ambient Temperature
If the column, auxiliaries, and piping are properly insulated, and if the
column is properly controlled, ambient changes should cause little difficulty
unless the condenser is of the air-cooled type. In this event it may be desirable
to use an internal reflux computer, discussed in Chapter 11. If, as is often the
case, the vapor piping from the top of the column to the condenser is both
uninsulated and long, ambient temperature changes mav cause fluctuations in
pressure and the rate of condensation.

1.5 STARTUP AND SHUTDOWN

Startup and shutdown are often dismissed as relatively unimportant, since
they happen so seldom that it is not economical to spend much time and monev
on improvements. This may be true in a petroleum refinery where shutdowns
may occur at intervals of two or three years. In the chemical industry, however,
where process equipment is often plagued by severe corrosion or by plugging
process materials, startups and shutdowns are far more common- perhaps
monthly, weekly, or even daily. Further, elaborate heat and material recycle
schemes may require inmcate startup/shutdown procedures as part of the original
design. To put it another way, columns and their control systems may have
to be designed specifically to accommodate a particular startup/shutdown
procedure.
Columns are commonly started up in total reflux-no product is taken off
at top or bottom. Limited experience, however, suggests that faster and smoother

14

Stratgy for Dtddhtkfi-Cul~mn

Cuntrui

startups may be achieved by recycling top and bottom products back to feed
during part of the startup sequence.
Startup/shutdown will be discussed in more detail in Chapter 9.

1.6 CONTROL SYSTEM DESIGN PHILOSOPHY

Current Design Practices
In the preface we noted that we would try, in this book, to present a
multivariable control approach to distillation column control. Before discussing
this, however, let US look at typical controls in existing plants.
There is a traditional pattern of what is called instrumentation in the
chemical and petroleum industries based on single-loop control (sometimes
called SISO-single input, single output). Each process operation has a number
of independent or single loops for feedback control of temperatures, pressures,
flows, liquid levels, and sometimes compositions. The term single loop means
there is one measurement, one controller, and one final control element, usually
a valve. Many, but not all, of these loops are represented in the central control
room (CCR) by control stations.
The controllers usually have neither antireset windup nor automatic tracking,
and there is little or no logic circuitry to tie the many loops together. This
statement is true for both analog and some digital hardware. As a consequence
the operators must perform startup operations with the control stations switched
to manual, and must implement process logic by switching in and out of
ccautomatic. Since the original design procedures are usually qualitative and
intuitive, with heavy emphasis on conformity to past practices, it is not surprising
that some loops never work in automatic. Others, although stable in automatic,
are so sluggish that they are ineffective in dealing with typical disturbances.
For newer plants with higher throughputs, operation closer to hard constraints,
smaller and fewer holdups, energy-recovery systems, and elaborate materialrecycle systems, the traditional instrumentation approach to control is often
seriously inadequate.
For some years now, progress in single-loop design has been essentially at
a standstill. As recently as 1950, process control was hardware limited. Since
then primary measuring elements, controllers, computing devices, control stations,
and control valves have improved greatly in reliability, sensitivity, and speed
of response. Consequently these characteristics less and less frequently pose
limitations to the single-loop designer. Further, the quality of control achieved
by single-loop systems is not greatly affected by type of hardware; it makes
little difference, for a given algorithm, whether one uses analog pneumatic or
electronic gear, microprocessor controls, or a digital computer.* In addition,

* Digital computers and microprocessor controls, however, usually offer a wider range of
controller parameter adjustments and facilitate the design of control systems more sophisticated
than most of those discussed in this book.

1.6 Control System Des@ Philosop@

15

optimized tuning procedures for unaided feedback controllers have limited

practical value for continuous processes; they yield results that are far inferior
to those obtainable with well-damped feedback controls with simple feedforward
and override control.*
For plants being designed today (late 1983), it is increasingly common to
use microprocessor controls instead of analog (see discussion under Hardware
Conventions and Considerations).

Multivariable Control
To avoid the limitations of single-loop design and to provide a more flexible
and sophisticated process operating logic than can be implemented by human
operators, we use an approach we call multivariable control. Many definitions
of this term may be found in the literature, but most of them are expressed
in mathematical terms rather than in terms of process hctions. For our purposes
we define a multivariable control svstem as one that has the built-in intelligence
to look simultaneously at two or more process variables and to choose, in a
given situation, the best of several preprogrammed strategies (algorithms) for
manipulating one or more control valves (or other final control elements).
For example, the steam valve for a distillation column reboiler, depending
on circumstances, may respond to controllers for:
Steam flow rate
Column AP
Column pressure
Base temperature
Column feed rate
Column base level
Column bottom-product rate
The seven variables listed may also exert control on five or six other valves.
To provide automatic control of this sort, we make extensive use of variable
configuration controls that are usually implemented by overrides. If, for instance,
base composition is normally controlled by steam flow that can be taken over
or overriden by high column AP, this is a variable configuration. If base level
control is normally achieved bv a PI controller that can be overridden by high
or low base level proportion&-only controls, we call that variable structure.
Multivariable control may involve both variable configuration and variable structure
controls. Hardware permits us to automate this kind of control with a speed,
precision, and reliabilinr that are completely beyond the capabilities of human
operators.
It is common to think of process control fimctions stacked one above another
in a pyramid or hierarchic arrangement as with traditional business or military
organization structures, particularly when computers are involved. Multivariable
control structures, however, with their extensive lateral and diagonal crossovers,
are functionally more like the modern matrix concept of management. From

16

Strategy fm Distillation-Column Control

a control engineering standpoint, the shorter lines of communication and decentralized control functions permit more rapid and stable control, and more
reliable, troublefree operation.
By now it is probably apparent that we are striving for control system
designs whose performance and design parameters are specified in advance of
plant startup. In practice we furnish calibration data for controller parameters
and computational devices for the majority of control loops prior to startup.
We calculate these from simulations or simple linear models. For microprocessor
computer controls, we calculate scaling parameters for computation blocks
(either in software or hardware). Our design procedures are accurate enough
that only a modest amount of empirical controller tuning is required at startup.

Stability, Speed of Response, and Interactions

Most existing literature on automatic control is concerned with the stability
and speed of response of single loops. The traditional objectives of feedback
control system design are:

1. To get the fastest possible response to set point changes.

2. To compensate for or to attenuate disturbances as much and as quickly
as possible. These must be accomplished with a reasonable degree of closedloop stability.
In process control the objectives are often quite different. The objectives
of averaging level control, for example, clearly are different from those just
mentioned. A typical chemical plant or refinery has hundreds of single loops
with many interactions among them. It is usually far more important to design
for a dynamic balance among these loops with a minimum of interaction than
to strive for maximum speed of response. Further, it is usually undesirable to
make rapid changes in manipulated variables since these may upset the process.
For example, distillation column reflux flow and boilup should not be changed
too rapidly since these might cause transient flooding or weeping in part of
the column. Our preferred philosophy of controller tuning is discussed in a
book and a paper.
There are five simple methods by which to make a system noninteracting:

1. Design material-balance control loops to be at least a factor of 10 slower

than related composition control loops. Similarly, in cascade systems, make the
secondary or slave loop at least a factor of 10 faster than the master loop.
2. Avoid designs that are intrinsically interacting, such as pressure control
and flow control at the same point in a pipeline. One of the two controllers
must be detuned.
3. Select process designs that eliminate or minimize interaction. For example,
* Some processes are open-loop unstable, which means that controls must be added and
must be kept in automatic for stable operation.

1.6 Control System Des@ Philosqphy

17

if a tubular reactor is fed at several points along its length from a common
header, flow-rate interactions may be reduced to an arbitrary level by providing
a very high pressure drop across each feed valve in comparison with the reactor
pressure drop.
4. Use override circuits. Although not specifically intended for this purpose,
override circuits, by permitting only one controller at a time to control a given
valve, eliminate interactions.
5. Use interaction compensators (decouplers). If two control loops, such
as top and bottom temperature controls on a distillation column, interact, we
can eliminate the interactions by installing two compensators. One compensates
for the action of the top temperature controller on the bottom loop while the
other compensates for the action of the bottom temperature control loop on
the top one. This is discussed in Chapter 20.

In addition, there are some very sophisticated mathematical methods for

dealing with interaction^.^^,^^'^^ Some are intended for noninteracting design
while others seek a design that provides an optimum amount of interaction.

Hardware Conventions and Considerations*

For control loops represented in the CCR (central control room), it is
normal practice (as mentioned earlier) to furnish control stations. These may
be analog pneumatic, analog electronic, or microprocessor based. In the last
case, the station may be physically distinct, like an analog station, or may be
represented on a CRT display as a faceplate. Each provides an indication of
the process variable (flows, level, temperature, etc.), the desired value (set point),
and the valve loading signal (controller output, really). There is also a manualautomatic switch, which some vendors label hand-automatic. In the manual
mode, the feedback controller is disconnected and there is a knob that enables
the operator remotely to set the valve position. This may or may not be subject
to restrictions imposed bv feedforward compensation, overrides, and so on,
depending on the design philosophy for a particular project.
If cascade control is involved, as, for example, liquid level control cascaded
to flow control, the secondary station not only has manual-automatic switching,
but also another fUnction-remote-locaI. In the remote position, the secondary
controller set point comes from the output of the primary controller. In the
cloca position, valve position may be set manually or the controller set point
may be set by the operator (clocal-auto).Although cascade functions are
sometimes combined into one station (or CRT faceplate) for space and
money-saving reasons, we recommend dual stations. Most single-station designs
with which we are familiar are very inflexible and complicated; they do not
permit ready implementation of feedforward, overrides, and so forth.
* Much of this section may seem superfluous to instrument and plant personnel. Our experience,
however, with process design engineers, column designers, and so for& is that this discussion
may be very helpfid.

18

Strategy fm Dirtillation-Column Control

As of this writing (late 1983), the two biggest hardware needs appear to
be:

1. Better measurements, especially of compositions.

2. Better control valves. With regard to these, more progress has been made
in the design of valve bodies and trim than in the design of actuators. Valve
positioners should always be used (it is assumed in this book that they will
be).* Piston actuators with double-acting, two-stage positioners are recommended.

As far as CCR hardware is concerned, we have a decided preference for

microprocessor controls. They are technically more versatile and are less expensive
(some versions) than analog. As of late 1983, many are featuring satisfactory
antireset windup and override capabilities. In adchion, they provide more
advanced logic capability, dead-time simulation, and adaptive tuning. Some of
the last named achieve self-tuning via stochastic techniques or by pattern recognition. Others have gain scheduling, where reset time and proportional gain
are functions of some process variable or the controller error signal.
Microprocessor controls usually have a sampling time of a fraction of a
second. Although slightly slower than analog controls, their performance can
generdy be approximated by analog control algorithms.
Other advantages include freedom from drift, and the fact that they can be
calibrated more precisely, can be reconfigured or restructured without wiring
changes, have a larger range of tuning parameters, and contain more control
algorithms.
For most projects today, it is possible to find worthwhile applications for
a supervisory digital computer with a good data historian, regardless of the
type of basic controls selected (pneumatic, electronic, or microprocessor). Digital
readouts for important variables are worthwhile because they permit seeing
their magnitudes with sensitivity approaching that of the original analog measurements. Most typical analog measurements have a sensitivity ranging from
one part per 1000 to one part per 10,000. Most analog readout devices,
however, are limited to 0.5-2 percent.
For maximum advantage a supervisory computer should be programmed
to have the control algorithms discussed in Chapter 12. These are position
rather than velocity algorithms. It is our opinion that using such a computer
to imitate unenhanced two- and three-mode analog controls is poor practice.
Some worthwhile applications for computers will be discussed later.
Computer consoles were originally provided in the CCR for the convenience
of the operators. Sometimes a consolidated console is also provided for production
supervision. Engineers consoles, perhaps at another location, facilitate technical
studies. Separate consoles for maintenance personnel (the computer can be a
powerful maintenance tool) are highly desirable.
* Most control valves today are either single-seated global types or rotary types. For both
there are substantial, coercive stem forces when the valves are in flowing streams. Positioners
compensate for this and maintain the valves inherent flow characteristic expressed as a function
of controller output signal. It should be noted, however, that some users prefer not to use
positioners.

1.8 Column Des@ Philosophy and Control System Des@

19

1.7 PROCEDURE FOR OVERALL CONTROL SYSTEM DESIGN

We now have enough information to suggest a sequence of design steps
to follow that will lead to a quantitative definition of column and hardware
performance.

1. After careful discussion with the process engineer and column designer,
and after careful review of the overall process flow sheet, prepare a simplified
flow sheet that defines the control concepts:
a. Select the overall material-balance control scheme first, preferably
proceeding from final product inventory to raw material inventory. All
individual equipment-piece material-balance controls must be consistent
with this scheme.
b. Select composition control schemes.
c. Add feedforward and interaction compensators as required.
d. Add protective controls and automatic startup/shutdown circuits.
e. Add miscellaneous temperature and pressure controls.
2. Prepare material-balance and composition-control signal flow diagrams.
3. Determine holdup volumes required for smooth material-balance control
and for liquid-level override controls at each end of the column.
4. With holdups determined calculate column-composition transfer functions.
5. Select measurement spans and calculate control valve sizes.
6 . Calculate feedforward and interaction compensators.
7. Calculate all other overrides.
8. Calculate feedback-controller gain and reset settings and control-loop
natural frequencies. Check feed-tank material balance and mixing time constants
for adequacy.
9 . Use simulation for some columns, particularly those in critical service
or with a new untested control system or process configuration. Simulation of
the column and its control system will be usefd in confirming control concepts
and controller tuning parameters. It may also save startup time.
Hardware vendors may now be selected and the measurement and control
equipment may be ordered.
Part I1 of this book deals with the qualitative and heuristic aspects of steps
1 and 3. Quantitative information for the other steps is presented in Part 111.

1.8 COLUMN DESIGN PHILOSOPHY

AND CONTROL SYSTEM DESIGN
Experience on many projects shows that even small, simple columns benefit
from a modest application of feedforward compensation and overrides. This is
due to the trend toward increasinglv tight column design. The number of trays
is held down because smaller allowan& are made for uncertainty in tray efficiency.
Column diameter and tray spacing are now kept to smaller values, with the

20

Strategy fm Dljtillatwn-Column Control

result that columns typically operate closer to flooding. The combined effect
of these design policies is to make columns much touchier and harder to control.
Experience indicates that typical incremental instrument investment over
that required for unenhanced feedback controls is 5-10 percent (large projects
tend toward the lower figure). This incremental investment not only provides
better normal control, but it also helps to avoid inadvertent shutdowns. It is,
therefore, our opinion that these controls should be used to some extent on
almost every column.
Suppose, however, that the customer insists on minimum application of
feedback controls with no feedfonvard compensation or overrides. What column
design philosophy should be followed?Having had considerable adverse experience
with columns with primitive controls, particularly sidestream drawoff columns,
and columns with heat recovery schemes, we suggest the following:

1. Design for normal operation at 60 percent of the rate for flooding.*

2. Provide five extra trays or increase the number of trays by 10 percent,
whichever is larger.
3. Provide increased tray spacing.
4. Provide larger condensers. If water cooled, use tempered water.
5. Provide larger reboilers (lower heat flux).
6. Control column Al by boilup, or flow control steam.
7. Provide 100 percent reserve capacity in heat-recovery schemes or avoid
them altogether.
8. Avoid sidestream drawoff designs.
9. Provide surge tanks between columns with at least 30-60 minutes of
holdup each.

1.9 EXISTING COLUMNS-TYPICAL

PRACTICES
AND TROUBLESHOOTING

Although this book deals primarily with &stillation control in design projects,
it is pertinent to consider briefly the controls of typical, existing columns, the
opportunities for their improvement, and how to troubleshoot them when
necessary. Frequently encountered problems include unstable or ineffective controls, off-specification product or products, and f l d g or dumping. In addition,
it is fairly common practice to use excessive boilup and reflux to make sure of
meeting or exceeding product specifications. This not only wastes energy; it
also reduces column capacity. To provide a perspective on energy savings, one
may note that 100 Ibm/hr of steam is worth $3200 per year (basis $4.00/1000
pounds, 8000 hours per year). To save this amount of steam would probably
be only a modest accomplishment for most columns.
If composition control of each product stream is desired (and this is usually

* In some companies columns are designed by the probabilistic methods recommended by

Fractionation Research, Incorporated.

1 .IO Cunventwns Followed in thlj Book

21

the case), the most obvious deficiency of most exisring columns is the lack of
appropriate composition measurements. Most commonly temperature in the
upper or lower section of the column (or both) is used in lieu of true composition
measurements. Frequently composition control is attempted at only one end
of the column, and sometimes at neither end.
Another shortcoming frequently observed is the use of fixed flow controls
for steam, reflux, or product drawoff. Any such unaided flow control should
be regarded with suspicion. With rare exceptions flow-control set points must
be changed to accomplish either composition control or material balance control.
In view of the preceding comments about problem areas and likely opportunities for improvement of composition control and reduction in energy
consumption, the following guidelines are suggested:

1. Make sure that column material-balance controls are properly designed

and tuned, and that hardware, especially level and flow transmitters and control
valves, is in good working condition. If PI level controllers areused, follow
the tuning procedures of Chapter 16; auto overrides or nonlinear controllers
should be used. It is usually desirable to cascade level control to flow control,
in which case flow measurement should be linear.
2. Provide averaging level control of column feed. Column feed rate should
also be held between maximum and minimum limits.
3. Provide a linear flow measurement for steam flow control. Also provide
temperature and pressure compensation.
4. If condenser controls are a problem, review the control schemes in
Chapter 3.
5. At this point with flows established, smoothed, and in some cases limited,
it will probably be possible to see some improvement in composition control,
at least for part of the time. For further improvement provide steam-to-feed
ratio control, internal reflux-to-feed or distillate-to-feed ratio control, high A P
override on steam to protect against flooding, and a minimum steam flow
ltmiter to protect against dumping.
6. If composition is measured and is cascade controlled via reflux or boilup,
or both, ratio controls should be replaced by impulse feedforward compensation
.
.
(see Chapter 12) if feed flow turndown is greater than 2: 1.
7. For pressurized or vacuum columns, make sure an adequate scheme is
provided for maintaining an inert balance.
8. If temperature by itself is not an adequate measure of composition,
consider one of the schemes in Chapter 10 for using temperature, pressure,
and, in some cases, flow measurements to deduce compositions.

1.I 0

CONVENTIONS FOLLOWED IN THIS BOOK

Throughout the remainder of this book, we will frequently illustrate control

schemes with pneumatic components. This is done for convenience. Pneumatics,

22

Strategy fm Dirtillation-ColumnControl

with few exceptions, features a standard signal span of 3- 15 psig. ALL pneumatic
controllers, as far as we know, can achieve antireset windup by the same external
reset feedback method. This appears to be the most universally usell method
(we use it extensively), and has been adopted by one manufacturer of analog
electronic instruments and by several vendors of microprocessor-based distributed
controls. Other vendors of electronic analog and digital controls feature a wide
variety of techniques; some of these work fairly well, while some are quite
inflexible. A brief discussion is presented in Chapter 12.
Units used in this book are those commonly employed in chemical engineering:
pounds, feet, degrees Celsius, mols, and so on. To facilitate the calculation of
control engineers time constants, we have mostly used time units of minutes
or seconds [e.g., Ibm/sec, Ibm/min, (pcu/sec>/C fi]. For projects that make
partial use of metric or SI units, we found it convenient to convert them to
the above units. Generally speaking, we have found no advantage in writing
equations with SI units. Instead, in programs for computers or programmable
calculators, we mostly write the equations in the older units and add subroutines
for going back and forth to metric or SI units.
The most common type of controller used in the chemical and petroleum
industries was once called proportional plus automatic reset, later shortened
to proportional reset. Today it is more common to use PI, which stands
for proportional-integral. It is also becoming common today to speak of iontroller
gain rather than proportional band (PB = 100/Kc).We will also use reset
time, usually in minutes, rather than its older reciprocal, repeats per minute.
For drawings in which we are trying to present a perspective of control
concepts (configurations), we use very simplified symbols. In drawings where
we are trying to illustrate concepts of structure (overrides, feedfonvard, etc.),
we use a more detailed symbolism that we have found useful in our design
work.
In pneumatics it is c o m o n to refer to most signal-conditioning devices
other than controllers as relays. Included are adders, subtractors, multipliers,
and so on. For electronic analog and digital controls, it is more common to
use terms such as signal scalers and mult@lien.
A table of nomenclature and symbols will be found at the end of this book.

1 .I I

LITERATURE

For anyone seriously interested in distillation control, two books are highly
recommended. The first is an easy-to-read, nontheoretic (as far as control is
concerned) work by F. G. Shinskey. The treatment of energy conservation
alone is worth the price of the book. The second book is by Rademaker,
Rijnsdorp, and Maarleveld.12It relies heavily on conventional, single-loop control
theory, and explores painstalungly a large number of possible control systems.
It also contains an extensive bibliography.

Refevences

23

For basic reference books on distillation, we have made much use of those
Treybal,l6 and Hengstebeck
by Van Winkle13 and King.14Others by
also have been useful.
For basic books on control, we recommend two written by the
one by Harriott, and one bv Murri11.28 For more advanced treatments, we
suggest texts by Koppel,20 Douglas,21 and Gould.22 The last contains some
perceptive comments about the difficulties of applving advanced control theory
developed by electrical and mechanical engineers to chemical processes. Recent
books bv Rav3 and by S t e p h a n o p o l ~ u discuss
s ~ ~ applications of modern control
theor). in chemical processes. M c A v o ~has
~ addressed the specific subject of
interaction analysis.
Nonchemical engineers with no background in distillation may find an
introductonr text bi7 Nisenfeld and Seemann useful.24It is clearly written and
easv to read.

REFERENCES
A. Maarleveld, DynamicJ and Control
1. Buckley, P. S., Techniques of Process
of Continuous Distillation Units, ElControl, Wiley, New York, 1964.
sevier, New York, 1975.
2. Harbert, W. D., Pet. R$27(4): 10613. Van Winkle, M., DG-tilhim, McGraw109 (1948).
Hill. New York, 1967.
3. Harbert, W. D., Pet. Ref 29(10): 11714. King, C. J., Separation Processes,
122 (1950).
McGraw-Hdl, New York, 1971.
4. Harbert, W. D., Pet. Ref 35(11): 15115. Holland, C. D., Fundamentals and
159 (1956).
Modelling of Separation Processes,
5. R~ppin,D. W. T., and D. E. Lamb,
Prentice-Hall, Englewood Cliffs,
A Theoretical Study of the DyN.J., 1975.
namics and Control of Binan Distillation Columns, Bulletin from 15A. Holland, C. D., and A. I. Liapis,
Computer Methodr fm Soltin8 Dythe Universitv of Delaware, 1960.
m u Separmim P r o b h , McGraw6. Luyben, W. L., Chem. End. P q .
Hill, New York, 1983.
61(8): 74-78 (1965).
7. Uitti, K. D., ISA Paper no. 49-9-2. 16. Trevbal, R. E., Mms-Tran$er Operatwns, McGraw-Hill, New York,
8. Buckle?, P. S., Override Controls on
1968.
a Chemical Reactor, in Proceedings
of Texas A&M Instrumentation 17. Hengstebeck, R. J., Distillation, Reinhold, New York, 1961.
Symposium, Jan. 1970.
9. Buckley, P. S., MultivariableControl 18. Luyben, W. L., ProcessMod.ellin8, Simulation, and Control f i Chemical
in the Process Industries, presented
at Purdue University, Apr. 1975.
Enginem, McGraw-Hdl, New York,
1973.
10. Buckley, P. S., A Modem Perspective
on Controller Tuning, presented 19. Harriott, P., Process Control, McGrawHill, New York, 1964.
at Texas A&M University, Jan. 17,
20. Koppel, L. B., Introduction to Control
1973.
T b e o q . . . , Prentice-Hd, Engle11. Shinskey, F. G., DljtiLLatwn Control,
wood Cliffs, N.J., 1968.
McGraw-Hill, New York, 1977.
12. Rademaker, O., J. E. Rijnsdorp, and 21. Douglas, J. M., Process Dynamics and

24

22.
23.
24.
25.

Stratgy fm Dirtillation-Column Control

in Recycle Systems, INTECH, May
Control ( 2vols.), Prentice-Hall,En1974, pp 29-34.
glewood Cliffs, N.J., 1972.
Gould, L. A., Chemical Process Control, 26. McAvoy, T., Interactwn Analysir, Instrument Society of America, 1983.
Addson-Wesley, Reading, Mass.,
27. Stephanopolous, G., Chemical Process
1969.
Control, Prentice-Hall, Englewood
Ray, W. H., Advanced Process Control,
Cliffs, N.J., 1983.
McGraw-Hd, New York, 1981.
Nisenfeld, E. I., and R. C. Seemann, 28. Murrill, P. W., Fundumntals $Process
Control Theory, Instrument Society
Dictdlation Columns, ISA, Research
of America, 1981.
Triangle Park, N.C., 1981.
Buckley, P. S., Materral Balance Control