Chapter 13

Control Strategies at the Process

Unit Level
CHAPTER CONTENTS

13.1 Degrees of Freedom Analysis for Process Control

13.1.1 Control Degrees of Freedom
13.1.2 Effect of Feedback Control
13.2 Selection of Controlled, Manipulated, and Measured Variables
13.2.1 Controlled Variables
13.2.2 Manipulated Variables
13.2.3 Measured Variables
13.3 Applications
13.3.1 Distillation Column
13.3.2 Fired-Tube Furnace
13.3.3 Catalytic Converters for Automobiles
13.3.4 Plasma Etching in Semiconductor Processing
Summary

Previous chapters have emphasized process control
problems with a single controlled variable and single
manipulated variable. In this chapter, we show that these
concepts and analysis methods are also applicable to
control problems at the process unit level that have mul-
tiple controlled variables (CVs) and multiple manipu-
lated variables (MVs). These types of control problems
are considered further in Chapters 18, 20, 25, and 26.
For control system design and analysis, it is conve-
nient to classify process variables as being either out-
put variables or input variables. The output variables
(or outputs) are dependent variables that typically are
associated with exit streams or conditions within a
process vessel (e.g., compositions, temperatures, lev-
els, and flow rates). Some outputs must be controlled
in order to operate a process in a satisfactory manner.
They are called controlled variables (CVs). Input
variables are process variables that affect one or more
output variables.
Input variables are classified as either manipulated
variables (MVs) or disturbance variables (DVs). Manip-
ulated variables are used to adjust the rates of material
and energy that enter or leave a process. The MVs are
often flow rates adjusted by control valves, variable-
speed pumps or compressors, or conveyor belts (for solid
materials). An energy input, such as the power to an
electrical heater, can also be an MV. If an MV is a flow
rate, there must be some place for the material to accu-
mulate. For example, it is not feasible to place two
control valves at different locations on the same pipe.
Manipulated variables are often inlet flow rates. How-
ever, an exit flow rate can also be an MV, for example,
when the liquid level in a tank is controlled by manipu-
lating an exit flow rate.

236
 13.1 Degrees of Freedom Analysis for Process Control
By definition, disturbance variables are input vari-
ables that cannot be manipulated. Common DVs include
ambient conditions and feed streams from upstream
process units.
13.1 DEGREES OF FREEDOM ANALYSIS
FOR PROCESS CONTROL
The important concept of degrees of freedom, NF, was
introduced in Section 2.3 in connection with process
modeling. It is the number of process variables that
must be specified in order to be able to determine the
remaining process variables. If a dynamic model is
available, NF can be determined from a relation in
Chapter 2,
NF = Nv- NE (13-1)
where Nv is the number of process variables and N E is
the number of independent equations.
For process control applications, it is very important
to determine the maximum number of process vari-
ables that can be independently controlled, that is, to
determine the control degrees of freedom, N Fe:
Definition. The control degrees of freedom, N FO is
the number of process variables that can be controlled
independently.
In order to make a clear distinction between N F and
N Fe, we refer to N F as the model degrees of freedom
and toNFe as the control degrees of freedom. They are
related by,
NF = NFe + Nv (13-2)
where Nv is the number of DVs.
13.1.1 Control Degrees of Freedom
The control degrees of freedom N Fe is closely related
to the number of independent MVs that are available:
General Rule. For most practical control problems,
the control degrees of freedom N Fe is equal to the
number of independent input variables that can be
manipulated.
It is important that the manipulated inputs be inde-
pendent. For example, if a process stream splits, or if
two process streams merge to form a third stream, it is
not possible to adjust all three flow rates indepen-
dently. These situations are shown in Fig. 13.1.
237
(a) Stream splits
(b) Streams merge
Figure 13.1 Two examples where all three streams cannot be
manipulated independently.
Two examples illustrate the General Rule.
EXAMPLE 13.1
Determine Npand Npc for the steam-heated, stirred-tank
system model in Eqs. 2-50 through 2-52 in Chapter 2.
Assume that only steam pressure P8 can be manipulated.
SOLUTION
To calculate Np from Eq. 13-1, we need to determine Nv
and N E· The dynamic model contains three equations
(NE = 3) and six process variables (Nv = 6): T8 , P8 , w, T;, T,
and Tw. Thus, Np = 6 - 3 = 3. If feed temperature T; and
mass flow rate w are considered to be disturbance vari-
ables, Nn = 2 and thus Npc = 1 from Eq. 13-2. This single
degree of freedom could be used to control temperature
T by manipulating steam pressure, P8 •
EXAMPLE 13.2
A conventional distillation column with a single feed
stream and two product streams is shown in Fig. 13.2. The
feed conditions are disturbance variables. Determine the
control degrees of freedom Npc and identify potential
MVsandCVs.
SOLUTION
For a typical distillation column, five input variables can
be manipulated: product flow rates, Band D, reflux flow
rate R, coolant flow rate qc, and heating medium flow rate
qh· Thus, according to the General Rule, Npc = 5. This
result can also be obtained from Eqs. 13-1 and 13-2, but
considerable effort is required to develop the required
dynamic model. Although five output variables could be
selected as CVs, xn, XB, hB, hn, and P, for many distillation
control problems, it is not necessary to control all five.
Also, if it not feasible to measure the product composi-
tions on-line, tray temperatures near the top and bottom
of the column are often controlled instead, as discussed in
the next section.
238 Chapter 13 Control Strategies at the Process Unit Level

r-.;;:-"t-VCQ-- Coolant
qc

Disti Ilate
Feed------~ D
Reflux
R
xo

Heating
medium --1~c1-----,,r-,.,

qh
Figure 13.2 Schematic Bottoms
diagram of a distillation B
column. xs

Although the General Rule is simple and widely fraction, f Thus, it is not possible to compensate for sus-
applicable, there are exceptions where it is not valid. tained disturbances in x1 by adjusting f For this reason,
For example, N FC should be reduced by 1 when a the proposed control scheme is not feasible.
MV does not have a significant steady-state effect Because f does not appear in (13-4), the steady-
on any of the CVs, that is, when these steady-state state gain between x and f is zero. Thus the bypass
gains are very small. This situation is illustrated in flow rate can be adjusted, but it does not provide a
Example 13.3. control degree of freedom. However, if Wz could also
be adjusted, manipulating both f and w 2 could pro-
duce excellent control of the product composition.
EXAMPLE 13.3
The blending system in Fig. 13.3 has a bypass stream that
allows a fraction f of inlet stream Wz to bypass the stirred
tank. It is proposed that product composition x be
controlled by adjusting/via the control valve. Analyze the
feasibility of this control scheme by considering its steady-
state and dynamic characteristics. In your analysis, assume
that x1 is the principal disturbance variable and that xz, w1,
and w 2 are constant. Variations in liquid volume V can be
neglected because Wz << w1. h

SOLUTION
The dynamic characteristics of the proposed control
scheme are quite favorable because product composition x Xc X
responds rapidly to changes in the bypass flow rate. In L----~--~----L~~ w
w1 +fw2
order to evaluate the steady-state characteristics, consider
a component balance over the entire system: Figure 13.3 Blending system with bypass stream.

(13-3)
Solving for the controlled variable gives,
w1x 1 +w2x 2
x=---- (13-4) 13.1.2 Effect of Feedback Control
w
The addition of a feedback controller can change the
Thus x depends on disturbance variable x1 and four con-
control degrees of freedom, NFc· In general, adding a
stants (w 1 , w2 , x 2 and w), but it does not depend on bypass
feedback controller utilizes a control degree of freedom,
 13.2 Selection of Controlled, Manipulated, and Measured Variables
because an MV is now adjusted by the controller. How-
ever, if the controller set-point is adjusted by a higher-
level (or supervisory) control system, neither NF nor
NFc changes. The reason is as follows. Adding a con-
troller introduces a new equation, the control law, and
a new variable, the set point. Thus both N E and Nv in-
crease by one. But Eqs. 13-1 and 13-2 indicate that NF
and N FC do not change.
13.2 SELECTION OF CONTROLLED,
MANIPULATED, AND MEASURED
VARIABLES
A general representation of a control problem is
shown in Fig. 13.4. In general, it is desirable to have at
least as many MVs as CVs. But this is not always possi-
ble, and special types of control systems sometimes
need to be utilized (see Chapter 16). It may not be fea-
sible to control all of the output variables for several
reasons:
1. It may not be possible or economical to measure all
of the outputs, especially chemical compositions.
2. There may not be enough MVs.
3. Potential control loops may be impractical because
of slow dynamics, low sensitivity to the MVs, or
interactions with other control loops.
In general, CVs are measured on-line, and the
measurements are used for feedback control. But
sometimes it is possible to control an unmeasured CV
by using a process model (a soft sensor) to estimate it
from measurements of other process variables. This
strategy is referred to as inferential control (see
Chapter 16).
13.2.1 Controlled Variables
Consideration of plant and control objectives has pro-
duced guidelines for the selection of CV s from the
available output variables (Newell and Lee, 1989).
Guideline I All variables that are not self-regulating
must be controlled. In Chapter 5, a non-self-regulating
Disturbance
variables
Manipulated Controlled
variables variables
Figure 13.4 Process with multiple inputs and multiple
outputs.
239
variable was defined to be an output variable that ex-
hibits an unbounded response after a sustained input
change such as a step change. A common example is
liquid level in a tank that has a pump on an exit line
(see Chapter 11). Non-self-regulating variables must
be controlled in order for the controlled process to be
stable.
Guideline 2 Choose output variables that must be kept
within equipment and operating constraints (e.g., tem-
peratures, pressures, and compositions). The con-
straints are due to safety, environmental, and
operational requirements.
Guideline 3 Select output variables that are a direct
measure of product quality (e.g., composition, refrac-
tive index) or that strongly affect it (e.g., temperature
or pressure).
Guideline 4 Choose output variables that seriously in-
teract with other controlled variables. The pressure in
a steam header that supplies steam to downstream
units is a good example. If this supply pressure is not
well regulated, it will act as a significant disturbance
to downstream units.
Guideline 5 Choose output variables that have favor-
able dynamic and static characteristics. Output vari-
ables that have large measurement time delays, large
time constants, or are insensitive to the MVs are
poor choices.
Except for Guideline 1, these guidelines are not strict
rules. For specific situations, the guidelines may be in-
consistent or conflicting. For example, suppose that one
output variable must be kept within specified limits for
safety reasons (Guideline 2), whereas a second inter-
acts strongly with other output variables (Guideline 4).
Guideline 2 would prevail because of safety considera-
tions. Thus, the first output variable should be con-
trolled if there is only a single MV.
13.2.2 Manipnlated Variables
Based on the process and control objectives, a number
of guidelines have been proposed for the selection of
MVs from among the input variables (Newell and Lee,
1989). Inlet or exit flow rates can be manipulated in
order to adjust mass balances and thus control CVs
such as liquid level and pressure. Temperatures and
vapor pressures are controlled by adjusting the energy
balance.
Guideline 6 Select inputs that have large effects on
controlled variables. Ideally, an MY should have a
significant, rapid effect on only one controlled vari-
able. In other words, the corresponding steady-state
gain should be large. Furthermore, it is desirable that
the effects of this MV on the other CVs should be
240 Chapter 13 Control Strategies at the Process Unit Level
negligible (that is, the other steady-state gains should
be small or zero). It is also important that each ma-
nipulated variable be able to accommodate a wide
range of conditions. For example, if a distillation col-
umn has a reflux ratio of 5, it will be much more
effective to control the reflux drum level by
manipulating the large reflux flow rate rather than
the small distillate flow rate, because larger distur-
bances in the vapor flow rate could be handled.
However, the effect of this choice on the control of
product compositions must also be considered in
making the final decision.
Guideline 7 Choose inputs that rapidly affect the con-
trolled variables. For multiloop control, it is desirable
that each manipulated variable have a rapid effect on
its corresponding controlled variable.
Guideline 8 The manipulated variables should affect
the controlled variables directly, rather than indi-
rectly. Compliance with this guideline usually re-
sults in a control loop with favorable static and
dynamic characteristics. For example, consider the
problem of controlling the exit temperature of a
process stream that is heated by steam in a shell and
tube heat exchanger. It is preferable to throttle the
steam flow to the heat exchanger rather than the
condensate flow from the shell, because the steam
flow rate has a more direct effect on the steam pres-
sure and on the rate of heat transfer.
Guideline 9 A void recycling of disturbances. As Newell
and Lee (1989) have noted, it is preferable not to
manipulate an inlet stream or a recycle stream, be-
cause disturbances tend to be propagated forward,
or recycled back, to the process. This problem can
be avoided by manipulating a utility stream to ab-
sorb disturbances or an exit stream that allows the
disturbances to be passed downstream, provided
that the exit stream changes do not unduly upset
downstream process units.
Note that these guidelines for MVs may be in con-
flict. For example, a comparison of the effects of two
inputs on a single controlled variable could indicate
that one has a larger steady-state gain (Guideline 6)
but slower dynamics (Guideline 7). In this situation, a
trade-off between static and dynamic considerations
must be made in selecting the appropriate manipulated
variable from the two candidates.
13.2.3 Measured Variables
Safe, efficient operation of processing plants requires
on-line measurement of key process variables. Clearly,
the CVs should be measured. Other output variables
can be measured to provide additional information
or for use in model-based control schemes such as
inferential control. It is also desirable to measure MVs
because they provide useful information for tuning
controllers and troubleshooting control loops (see
Chapter 12). Measurements of DVs provide the basis
for feedforward control (see Chapter 15).
In choosing sensor locations, both static and dynamic
considerations are important, as discussed in Chapter 9.
Guideline 10 Reliable, accurate measurements are
essential for good control. Inadequate measurements
are a key factor in poor process control performance.
Hughart and Kominek (1977) cite common measure-
ment problems that they observed in distillation-
column control applications: orifice runs without
enough straight piping, analyzer sample lines with
large time delays, temperature probes located in
insensitive regions, and flow rate measurement of
liquids that are at, or near, their boiling points,
which can lead to liquid flashing at the orifice
plate. They note that these types of measurement
problems can be readily resolved during the
process design stage, but changing a measurement
location after the process is operating can be both
difficult and costly.
Guideline 11 Select measurement points that have
an adequate degree of sensitivity. As an example,
consider product composition control in a tray-
distillation column. If the product composition cannot
be measured on-line, it is often controlled indirectly
by regulating a tray temperature near that end of the
column. But for high-purity separations, the location
of the temperature measurement point can be quite
important. If a tray near an end of the column is se-
lected, the tray temperature tends to be insensitive,
because the tray composition can vary significantly,
even though the tray temperature changes very little.
For example, suppose that an impurity in the vapor
leaving the top tray has a nominal value of 20 ppm.
A feed composition change could cause the impurity
level to change significantly (for example, from 20 to
40 ppm) but produce only a negligible change in the
tray temperature. By contrast, suppose that the tem-
perature measurement point were moved to a tray
that is closer to the feed tray. Then the temperature
sensitivity is improved because the impurity level is
higher, but disturbances entering the column at ei-
ther end (e.g., from the condenser or the reboiler)
would not be detected as quickly.
Guideline 12 Select measurement points that mini-
mize time delays and time constants. Reducing time
delays and time constants associated with mea-
surements improves closed-loop stability and re-
sponse characteristics. Hughart and Kominek (1977)
have observed distillation columns with the sample
connection for the bottom analyzer located 200 ft
 13.2 Selection of Controlled, Manipulated, and Measured Variables
downstream from the column. This large distance in-
troduced a significant time delay and made the col-
umn difficult to control, particularly because the
time delay varied with the bottom flow rate.
An evaporator control problem will now be used to
illustrate the Guidelines.
EXAMPLE 13.4
The evaporator shown in Fig. 13.5 is used to concentrate a
dilute solution of a single, nonvolatile solute in a volatile
solvent by evaporating solvent using heat supplied by a
steam coil. Three process variables can be manipulated:
steam pressure, Ps, product flow rate, B, and vapor flow
rate of solvent, D. The chief DVs are feed composition,
xF and feed flow rate, F. The compositions are expressed
as' mole fractions of solute, and the flow rates are in
molar units.
Propose multiloop control strategies for two situations:
(a) The product composition xs can be measured on-line
(b) xs cannot be measured on-line
SOLUTION
Case (a): Product Composition x8 Is Measured On-Line
First, we select the CVs. Because the chief objective for
an evaporator is to produce a product stream with a
specified composition, mole fraction xs is the primary
CV (Guideline 3). Liquid level h must be controlled be-
cause of operating constraints and safety considerations
(Guideline 2). If the level is too high, liquid could be en-
trained in the solvent stream; if the level is too low, the
tubes of the steam chest would be exposed to vapor,
rather than liquid, a potentially dangerous situation. In
this latter situation, the heat transfer rate from the steam
to the evaporator liquid would be significantly lower,
and thus overheating and damage to the steam chest
could result. Pressure P should also be controlled, be-
cause it has a major influence on the evaporator opera-
tion (Guideline 2). Large pressure variations affect the
Solvent
.-----~D. xn = 0, T
P, T,p
Ps
Steam _.....:;..-+-'7
h
Product
B,x8 , T
Figure 13.5 Schematic diagram of an evaporator.
241
temperature T and could shift the boiling regime from
film boiling to nucleate boiling, or vice versa. This type
of regime shift could produce a major process upset. For
these reasons, three CVs are selected: xs, h, and P.
Next we select the MVs. Because the feed conditions
cannot be adjusted, the obvious MVs are B, D, and Ps.
Product flow rate B has a significant effect on h, but
relatively small effects on P and xs. Therefore, it is
reasonable to control h by manipulating B (Guideline 6)
unless B is only a small fraction of F (for example, less
than 10% ). In this latter case, it would be desirable to
have Fbecome an MV. Vapor flow rateD has a direct and
rapid effect on P but has less direct effects on h and xs.
Thus, P should be paired with D (Guideline 6). This leaves
the Ps-XB pairing for the third control loop. This pairing is
physically reasonable, because the most direct way of
regulating xs is by adjusting the evaporation of solvent via
the steam pressure (Guideline 8).
Finally, we consider which process variables to measure.
Clearly, the three CVs should be measured. It is also
desirable to measure the three MVs because this information
is useful for controller tuning and troubleshooting. If large
and frequent feed disturbances occur, measurements of
disturbance variables F and xF could be used in a
feedforward control strategy that would complement the
feedback control scheme. It is not necessary to measure
TF because sensible heat changes in the feed stream
ar~ typically small compared to the heat fluxes in the
evaporator.
A schematic diagram of the controlled evaporator for
Case (a) is shown in Figure 13.6.
Case (b): Product Composition Cannot Be Measured
On-Line
The CVs are the same as in Case (a), but, because the
third controlled variable xs cannot be measured, standard
feedback control is not possible. A simple feedforward
control strategy can be developed based on a steady-state
component balance for the solute,
o= FxF- Bxs (13-5)
where the bar denotes the nominal steady-state value.
Rearranging gives,
- -XF
B=F- (13-6)
xs
Equation 13-6 provides the basis for the feedforward control
law. Replacing B and F by the actual flow rates, B(t) and
F(t), and replacing xs by the set-point value, Xssp• gives
B(t) = F(t) XF (13-7)
XBsp
Thus B is adjusted based on the measured value of F,
the set point XBsp• and the nominal value of the feed
composition, XF.
The MVs are the same as for Case (a): D, B, and Ps.
Bottom flow rate B has already been used in the
feedforward control strategy of (13-7). Clearly, the P-D
242 Chapter 13 Control Strategies at the Process Unit Level

~---
1

,...--- ---------------~
1
I
I
I
I
I
I

$
I

~
h

I
I I
I
I
Feed AT
F, xf, Tp

Figure 13.6 Evaporator control strategy

'---------1/'<J-----'----'~ Product
for Case (a). B,xB

pairing is still desirable for the reasons given for Case (a).
This leaves h to be controlled by adjusting the rate of
evaporation via Ps. A schematic diagram of the controlled
evaporator is shown in Fig. 13.7.
This control strategy has two disadvantages. First, it is
based on the assumption that the unmeasured feed
composition is constant at a known steady-state value.
Second, because the feedforward control technique was
based on a steady-state analysis, it may not perform well
during transient conditions. Nevertheless, this scheme
provides a simple, indirect method for controlling a
product composition when it cannot be measured.

~---
1

-----------~

I
I
I
I
I

--~
h

---------~

Feed forward
controller

Figure 13.7 Evaporator control strategy for Feed

Case (b). F, xr,Tp

13.3 APPLICATIONS
In this section, we describe four representative exam-
ples of control problems at the process unit level,
rather than at the individual control loop level, in order
to provide an introduction to more complex control
problems. For each of these case studies, key aspects of
their control system design are considered:
(a) Process objectives. For control system design, it
is essential to know the process objectives. For
simple processes, the process objectives are
fairly obvious. But for others, they may not be.
For example, a chemical reactor can be oper-
ated to maximize the yield, selectivity, or
throughput subject to satisfying process con-
straints (e.g., safety and the environmental con-
straints.) Similarly, a distillation column can be
operated to maximize throughput or minimize
energy consumption, while satisfying product
specifications and other constraints.
(b) Control objectives. The control objectives should
be carefully formulated based on a number of
considerations that include process objectives,
process constraints, and economic data. Even
for simple control problems, there can be alter-
native control objectives. For example, consider
a very common control application, liquid-level
control. In some applications, the control objec-
tive is to achieve tight level control at a specified
set point. This situation might occur for a con-
tinuous bioreactor, when maintaining a constant
residence time is important. On the other hand,
many process vessels are used as intermediate
storage tanks for surge control. Here, the
process objective is to reduce the effect of up-
stream disturbances on downstream units, by
having the exit stream from an intermediate
storage tank change gradually in response to
large, rapid changes in its inlet steams. In this
situation, tight level control would be undesir-
able, because inlet flow rate disturbances
would be propagated to the outlet stream.
The more appropriate control objective
would be averaging control, in which the liq-
uid level is allowed to vary between specified
upper and lower limits, thus providing surge
capacity and more gradual changes in the exit
flow rate.
(c) Choice of control configuration. A key decision in
control system design is to decide whether a con-
ventional multiloop control strategy, consisting of
individual feedback control loops, will provide
satisfactory control. If not, an advanced process
control strategy is required. This decision should
be based primarily on the control objectives and
13.3 Applications 243
knowledge of the static and dynamic behavior of
the process. Advanced control strategies are con-
sidered in Chapters 16, 18, and 20.
(d) Pairing of MVs and CVs. If a multiloop control
strategy is selected, the next step is to deter-
mine how the CVs and MVs should be paired.
A systematic method for making these deci-
sions, the Relative Gain Method, is considered
in Chapter 18.
13.3.1 Distillation Column
For continuous distillation, the primary process ob-
jective is to separate a feed mixture into two (or
more) product streams with specified compositions.
Thus the product compositions are the most impor-
tant CVs. For some distillation columns, one product
composition is much more important than the other.
For example, consider a series of columns where the
bottom stream of a column serves as the feed stream
to the next column. The distillate composition for
each column is more important because it is a product
stream (Guideline 3). By contrast, the bottom stream
for each column (except the last column in the series)
undergoes further separation, so the bottom composi-
tions are of less concern. Consequently, in these situa-
tions, the bottom compositions may not have to be
controlled.
In addition to one or more product compositions,
other process variables need to be controlled. Con-
sider the separation of a binary mixture and the con-
ventional tray-distillation column shown in Fig. 13.2.
Assume that the chief control objective is to control
both product compositions, xv and xB. However, the
liquid levels in the reflux drum, hv and the column
base (or sump), hB, must be kept b~tween upper and
lower limits (Guideline 2). The column pressure, P,
must also be controlled in order to avoid weeping or
flooding in the column and to control the vapor in-
ventory in the column. Thus, this column has a total
of five CVs.
The MVs are selected from among six input vari-
ables: feed flow rate F, product flow rates, D and
B, reflux flow rateR, and the heat duties for the con-
denser and reboiler, qv and qB. If the feed stream
comes from an upstream process, instead of from a
storage tank, it cannot be manipulated, and thus F
is considered to be a DV. In this situation, there are
five MVs.
Distillation column control can be difficult for the
following reasons.
1. There can be significant interaction between
process variables. One important example is that
changing a single MV can have significant effects
244 Chapter 13 Control Strategies at the Process Unit Level

on many CVs. For example, increasing heat duty 13.3.2 Fired-Tube Furnace
q8 by increasing the steam flow rate causes more
Fired-tube furnaces (or heaters) are widely used in the
liquid to be boiled and thus increases the vapor
process industries to heat process streams and to
flow in the column. Consequently, the q 8 increase
"crack" high-molecular-weight hydrocarbon feeds, in
causes sump level h8 , pressure P, and bottom
order to produce more valuable lower-molecular-
composition x 8 to change rather quickly (Guide-
line 7). However, the increase in q 8 also affects weight compounds. In this case study, we consider a
fired-tube furnace used to heat a liquid hydrocarbon
reflux drum level hv and distillate composition xv
feed steam that passes through the furnace in a set of
more slowly, after the increased vapor flow
tubes. A simplified schematic diagram is shown in
reaches the top of the column (Guideline 8).
Fig. 13.8. The combustion of the fuel gas (FG) generates
Similarly, changes in R or D affect hv and xv
heat, which is transferred to the hydrocarbon (HC). The
rather quickly and h8 and x 8 more slowly. Other
major gaseous combustion reactions in the furnace are
interactions arise when the hot bottom stream
from the column is used to heat the cold feed 3
stream in a bottom-feed heat exchanger, in order CH4 + 2 0 2 ~ CO + 2H20
to reduce energy consumption. 1
2. The column behavior can be very nonlinear, espe- C0+20z~COz
cially for high-purity separations. For example,
the amount of effort required to reduce an impu- A fired-tube furnace is one of the case studies in the
rity level in a product composition from 5% to 4% Process Control Modules (PCM) in Appendix E. The
is typically much less than the effort required to PCM furnace model is a nonlinear state-space model
reduce it from 1.5% to 0.5%. that consists of 26 nonlinear ordinary differential equa-
tions based on conservation equations and reaction
3. Distillation colunms often have very slow dynamics. rate expressions for combustion (Doyle et al., 1998).
The dominant time constants can be several The key process variables for the furnace model are
hours, or even longer, and long time delays are listed in Table 13.1.
also common. Because slow dynamics result in Important dynamic characteristics of the furnace
long response times with feedback control, the model include the different time scales associated with
addition of feedforward control can be very mass and energy transfer, the nonlinear behavior of
advantageous. the model, time delays, and the process interactions
4. Process constraints are important. The most prof- between the input and output variables. The term
itable operating conditions typically occur when process interaction means that changes in an input vari-
some MVs and CVs are at upper or lower limits. able affect more than one output variable. For exam-
For example, maximum separation, or maximum ple, the step responses in Fig. 13.9 illustrate that a step
recovery of a valuable feed component, often change greater than 20% in the inlet air temperature
occurs for maximum reboiler heating or condenser affects three of the four output variables, and their cor-
cooling. responding response times are quite different.
5. Product compositions are often not measured.
Although product compositions are the primary
Flue gas
CVs, their on-line measurement is often difficult,
expensive, and costly to maintain. Consequently,
tray or product stream temperatures are com-
monly measured and controlled as proxies for
product compositions. This strategy is easier to
implement but makes tight composition control
more difficult.
Another major complication is that there are many
different column configurations, especially for reboilers
and condensers, and many alternative process and con-
trol objectives. Consequently, each column control Air
application tends to be different and to require individual
analysis. Fortunately, there is an extensive literature
available on both the practical (Shinskey, 1984; Luyben.,
Fuel gas
1992) and theoretical (Skogestad, 1997) aspects of dis-
tillation column control. Figure 13.8 Schematic diagram of a tube-fired furnace.
 13.3 Applications 245

Table 13.1 key process variables for the PCM furnace module
Measured Output Variables Disturbance Variables (DVs) Manipulated Variables {MVs)
HC outlet temperature HC inlet temperature Air flow rate
Furnace temperature HCflow rate FG flow rate
Flue gas (or exhaust gas) flow rate Inlet air temperature
0 2 exit concentration FG temperature
FG purity {CH4 concentration)

The control objectives for the furnace are the
following:
1. To heat the hydrocarbon stream to a desired exit
temperature
2. To avoid unsafe conditions resulting from the
interruption of fuel gas or hydrocarbon feed
3. To operate the furnace economically by maintain-
ing an optimum air-fuel ratio
Because the furnace model has two MVs, two CVs
should be specified. Of the four measured outputs in
Table 13.1, the most important are the primary CV, HC
outlet temperature, and the 0 2 exit concentration in
the flue gas. The latter provides a good indication of
the combustion efficiency of the furnace. A very low 0 2
measurement indicates that the FG combustion is
incomplete; a very high measurement indicates excess
air and thus low furnace efficiency. A high furnace
efficiency strongly depends on maintaining the optimum
air-fuel ratio. For these reasons, the HC outlet temper-
ature and the 0 2 exit concentration in the flue gas are
selected to be the CVs (Guideline 3).
The chief DV is the FG composition, which can
vary significantly depending on the source of the fuel
gas. Large composition changes affect the FG heating
value and thus upset the combustion process and heat
generation.
Conventional furnace control strategies involve both
feedforward and feedback aspects (Liptak, 2003,
Shinskey, 1996). The HC exit temperature can be
controlled by adjusting either the FG flow rate or FG
pressure. The 0 2 exit concentration is controlled by
adjusting the furnace draft, i.e., the difference between
air inlet and outlet pressures, by changing either the
inlet air flow rate or the damper in the furnace stack.
The air-fuel ratio can be controlled using a special type
of feedforward control referred to as ratio control (see
Chapter 15). The measured HC inlet flow rate can also
be used as a measured disturbance for feedforward con-
trol. The PCM include advanced model-based control
strategies for the furnace, including decoupling (Chap-
ter 18) and model predictive control (Chapter 20).
Safety considerations are a primary concern for fur-
nace operation because of the large amount of com-
bustible material that is present. In particular, it is
important to ensure that unreacted FG is not allowed
to accumulate. This unsafe condition can result if the
air-fuel ratio is too low or if burner flameout occurs
because of a FG interruption. Safety interlocks (see
Chapter 10) are used to shut off the fuel gas in these

626~
618

610 o 10 20 30 40 50 60 70 80 90 100

1480~
1460

1440 o 10 20 30 40 50 60 70 80 90 100

o~:~j
(.)
c ~
O"'
u E
~:::,

~ N~
E
0.86L-----~-------L------~------L-----~-------L------~------L-----~------~
0
0 10 20 30 40 50 60 70 80 90 100
Time (min)

Figure 13.9 Step response to a + 20% step change in inlet air temperature at t = 30 min.
246 Chapter 13 Control Strategies at the Process Unit Level

situations. Interruption of the feed stream also poses a

serious hazard, because it can result in overheating and
possible rupture of tubes in the furnace (Liptak, 2003).
Thus, an interlock shuts off the fuel gas if a low flow
alarm occurs for the feed flow rate.
For additional information on furnace and heater
control problems, see the books by Shinskey (1996)
and the Liptak (2003).

13.3.3 Catalytic Converters for Automobiles

In many urbanized areas of the world, automobiles are
the single greatest producer of harmful vehicle exhaust
emissions. For over 30 years, catalytic converters have
been used to significantly reduce harmful exhaust
emissions from internal combustion engines, especially
automobiles. In North America, automobiles with
standard gasoline engines manufactured since 2004 are
required to have three-way catalytic (TWC) converters.
This section presents an overview of control issues and
control strategies associated with TWC. More detailed
information is available elsewhere (Fiengo et al., 2005;
Balenovic et al., 2006; Guzzella, 2008).
Three-way catalytic converters (TWC) are designed
to reduce three types of harmful automobile emissions:
carbon monoxide (CO), unburned hydrocarbons in
the fuel (HC), and nitrogen oxides (NOx)· The term
nitrogen oxides refers to both N0 2 and NO. The TWC
accomplishes these tasks using precious metal catalysts
(platinum, palladium, and rhodium) for chemical reac-
tions that take place at high temperatures (e.g.,
1000-1600°F) and short residence times (~0.05 s).
The desired TWC oxidation reactions are (Schmidt,
2005),
Figure 13.10 TWC efficiency as a function of air-to-fuel ratio
(Guzzella, 2008).
and the desired reduction reaction is:
2NOx-xOz + Nz
The catalytic converter is most effective when the au-
tomobile engine operates with an air-to-fuel ratio (A/F)
that is slightly above the stoichiometric ratio of 1/14.7 for
gasoline. A normalized air-to-fuel ratio A is defined as:
Wair 1
A=-- (13-8)
Wfuel 14.7
where Wair is the mass flow rate of air, and Wfuel is the
mass flow rate of fuel. The pollutant removal efficiencies
for the three pollutants vary strongly with air-to-fuel
ratio, as shown in Fig. 13.10. For A ~ 1, the three pollu-
tants are essentially eliminated, with over 98% re-
moval. When A> 1, there is excess 0 2 , and the engine
is said to be running lean. For these conditions, the oxi-
dation reactions are favored, and excessive amounts of
NOx are emitted. Conversely, when A< 1, the engine is
said to be running rich. Here, the reduction reactions

CnH2n+2+ ( -3n2-+ 1) Oz-nCOz + (n+1)Hz0

are favored, and large amounts of hydrocarbons and
CO are emitted. The TWC has the capacity to store
oxygen for the oxidation reactions, when the engine is
1 running lean. From Fig. 13.10 and these considerations,
CO +20 2 -C0 2
it is evident that the desired value of A/F is slightly

Figure 13.11 Block diagram for the three way catalytic converter control system.

greater than stoichiometric, or, equivalently, A. should
be slightly greater than 1.
A general representation of a TWC control strategy
is shown in the block diagram of Fig. 13.11. The A/F is
measured both upstream, (A/F)u, and downstream
(A/F)ct, of the TWC, using 0 2 sensors (also called
lambda sensors). Based on the A/F measurements and
the desired value, (A/F)sp' the feedback controller cal-
culates an appropriate output signal u that adjusts the
fuel injection system for the engine (Guidelines 6-8).
Sometimes, the two A/F measurements are used in a
cascade control configuration, a topic that is considered
in Chapter 16. The static and dynamic behavior of the
TWC varies with the operating conditions (e.g., engine
load) and aging of the components, including the 0 2
sensors. Thus automatic adjustment of the controller
settings (adaptive control) is a promising approach
(Guzzella, 2008).
Many TWC control systems are operated so that A/F
rapidly alternates between being slightly rich and
slightly lean (A. = 1 ± 0.05) to ensure that the reduction
catalyst (rhodium) does not become overloaded and
that the oxidation catalysts (platinum and palladium)
do not become oxygen-starved. The switching time be-
tween the two modes is very small, less than a second.
The switching strategy can be implemented by cycling
the set-point, (A/F)sp·
The performance of a TWC is strongly affected by its
temperature, as well as the A/F value. The TWC does
not begin to operate properly until it heats up to ap-
proximately 550°F; efficient operation does not occur
until the temperature reaches about 750°F. Conse-
quently, a significant amount of emissions occur during
cold starts of the engine. This problem can be alleviated
by the addition of an electrical heater that can heat the
TWC prior to cold starts. The TWC can operate properly
up to sustained temperatures of 1500°F.
13.3 Applications 247
If the A/F ratio is rich, unburned fuel from the engine
undergoes combustion in the TWC, which can raise the
TWC exit temperature to several hundred degrees above
the inlet temperature. Consequently, temperature sen-
sors located before and after the TWC can provide useful
diagnostic information (Guideline 11). If the difference
in temperature measurements is unusually large, it indi-
cates that rich conditions occur. On the other hand, if the
difference is essentially zero, the TWC has stopped func-
tioning (e.g., as a result of catalyst poisoning).
As emission standards for automobiles become tighter,
improved closed-loop TWC control strategies become
even more critical. The development of advanced TWC
control strategies includes custom model-based methods
(Fiengo et al., 2005; Balenovic et al., 2006; Guzzella,
2008).
13.3.4 Plasma Etching in Semiconductor
Processing
Solid-state devices are manufactured on circular disks
of semiconducting material called wafers (Edgar et al.,
2000). These devices are three-dimensional structures
made up of stacked layers. Each layer is typically man-
ufactured in batch operations, such as deposition and
etching. The purpose of deposition is to grow a thin
layer of a specific material on the wafer surface. In
etching, part of the layer is removed chemically, using
gases such as CF4 and HF. Etching can remove silicon
dioxide, silicon nitride, polysilicon, aluminum, photore-
sist, and other thin film materials. It creates the final
layer definition by transforming a single layer of semi-
conductor material into the patterns, features, lines,
and interconnects that make up an integrated circuit.
The polysilicon (poly) gate etch process is shown
schematically in Fig. 13.12. Photoresist (PR) etching and
polysilicon etching are the most critical batch steps for
Figure 13.12 Inputs and
outputs for polysilicon gate
etch process in semicon-
ductor manufacturing. The
measured inputs (CDin
and ein) in the incoming
wafer can be used in
feedforward control, while
the measured outputs
(CDout and eout) are used in
feedback control. BARC is
bottom anti-reflective
coating.
248 Chapter 13 Control Strategies at the Process Unit Level
creating the profile of polysilicon (side views are shown
in Fig. 13.12). Photoresist etching entails isotropic etch-
ing of the top layer of photoresist, which determines the
critical dimension (CD), or width, of polysilicon. This
step is followed by polysilicon etching, which is
anisotropic (etches in a single downward direction); the
final profile of polysilicon is determined in this step.
The etching process can be used to illustrate the appli-
cation of Guidelines 3, 6, 8, and 10 from Section 13.2. For
Guideline 3, the key CVs in plasma etching, CD and e
(sidewall angle), are shown in Fig. 13.12. The CD affects
transistor speed, which is the most important electrical
property of a logic chip (a product quality variable). Ide-
ally, e should be goo, but a target of 87° represents a
trade-off between 8 and CD because of the interactions
between the variables. Attempting to control e closer to
goo causes the CD to move further from the target. The
uniformity of the CD over the wafer is a third CV af-
fected by the inputs. Excessive nonuniformity makes the
wafer lower quality because of chip inconsistency.
A plasma etcher has a number of MVs that can be
adjusted in order to achieve the desired chip geometry.
By applying Guidelines 6 and 8, several input variables
can be selected from the four possible MVs: etch time,
SUMMARY
This chapter has considered two important issues in
control system design. The first issue was that the con-
trol system design is strongly influenced by the control
degrees of freedom that are available, N FC· In most sit-
uations, Npc is simply the number of process variables
that can be manipulated. In general, N FC < N p, where
Np is the model degrees of freedom that was intro-
duced in Chapter 2. The second issue concerned the
REFERENCES
Balenovic, M., J. Edwards, and T. Backx, Vehicle Application of
Model-Based Catalyst Control, Control Eng. Practice, 14, 223
(2006).
Doyle, F. J. III, E. P. Gatzke, and R. S. Parker, Practical Case Studies
for Undergraduate Process Dynamics and Control Using Process
Control Modules, Computer Applications in Engineering Educ. 6,
181-191 (1998).
Edgar, T. F., S. W. Butler, W. J. Campbell, C. Pfeiffer, C. Bode, S. B.
Hwang, K. S. Balakrishnan, and J. Hahn, Automatic Control in
Microelectronics Manufacturing: Practices, Challenges and Possi-
bilities, Automatica, 36, 1567 (2000).
Fiengo, G., J. W. Grizzle, J. A. Cook, and A. Y. Karnik, Dual-UEGO
Active Catalyst Control for Emissions Reduction: Design and Ex-
perimental Verification, IEEE Trans. Control Systems Techno/.,
13,722 (2005).
Guzzella, L., Automotive System-An Automatic Control Bonanza,
Plenary Talk, 2008 American Control Conf, Seattle, WA (June,
2008).
Hughart, C. L. and K. W. Kominek, Designing Distillation Units for
Controllability, Instrum. Techno/. 24 (5), 71 (1977).
pressure, plasma power, and flow rates of gases such as
N 2 , 0 2 , and Cl2 . Steady-state nonlinear models can be
obtained from experimental test wafers by specifying
the inputs and measuring each CV; the data can then
be fitted using polynomial models as the input-output
relationships. These models also allow the process gain
to be calculated for each input-output pair.
A controller determines the set of input variables
(known as the etch recipe) that keep CD, e, and unifor-
mity as close as possible to their targets while satisfying
MV constraints. Consistent with Guideline 10, inte-
grated metrology (IM), shown in Fig. 13.12, uses opti-
cal techniques such as ellipsometry or scatterometry to
measure the incoming wafer profile ( CDin and ein) at
multiple sites. IM then sends the measurements to a
computer that calculates the values of the MVs for the
batch. At the end of the etch process, the output CVs
for the batch are measured using IM. The errors for the
CVs are calculated and used to adjust the control strat-
egy (Parkinson et al., 2010).
Advanced control strategies for microelectronics ap-
plications such as including plasma etching are discussed
elsewhere (Edgar et al., 2000; Moyne et al., 2001). Batch
process control is discussed in Chapter 22.
selection of the controlled, manipulated, and measured
variables, a key step in the control system design.
These choices should be based on the guidelines pre-
sented in Section 13.2.
The chapter concluded with four case studies that il-
lustrated control problems at the process-unit level,
rather than at the individual control-loop level.
Liptak, B. (Ed.), Instrument Engineers' Handbook, 4th ed., Section 8.19,
Furnace and Reformer Controls, Chilton Pub. Co., Radnor, PA, 2003.
Luyben, W. L. (ed.), Practical Distillation Control, Van Nostrand,
NY, 1992.
Moyne, J., E. del Castillo, and A. M. Hurwitz (Eds.), Run to Run
Control in Semiconductor Manufacturing, CRC Press, Boca Raton,
FL, 2001.
Newell, R. B. and P. L. Lee, Applied Process Control, Prentice-Hall
of Australia, Brookvale, NSW, Australia, 1989.
Parkinson, B., H. Lee, M. Funk, D. Prager, A Yamashita, R. Sundarajan,
and T. Edgar, Addressing Dynamic Process Changes in High Volume
Plasma Etch Manufacturing by using Multivariable Process Control,
IEEE Trans. Semiconductor Manufacturing, in press, 2010.
Schmidt, L. D., The Engineering of Chemical Reactions, 2nd ed.,
pages 293-295, Oxford University Press, NY, 2005.
Shinskey, F. G., Distillation Control, 2nd ed., McGraw-Hill, NY, 1984.
Shinskey, F. G., Process Control Systems, 4th ed., Chapter 9, Mc-
Graw-Hill, NY, 1996.
Skogestad, S., Dynamics and Control of Distillation Columns-A
Tutorial Introduction, Trans !ChernE, 15, Part A, 539-562 (1997).