1890 Ind. Eng. Chem. Res.

2003, 42, 1890-1904

Plantwide Control of Continuous Multiproduct Processes:

Two-Product Process
Kulchanat Kapilakarn† and William L. Luyben*
Process Modeling and Control Center, Department of Chemical Engineering, Lehigh University,
Bethlehem, Pennsylvania 18015

This paper considers the plantwide control of continuous processes that produce multiple
products. A typical example is a process in which there are two reversible reactions producing
two products M and D: A + B S M + C and A + M S D + C. The control structure must be able
to achieve different production rates of the two products. Several conventional control structures
are studied in which the flow rates of the fresh feed streams are fixed or manipulated by level
or composition controllers and the production rates of the two products are not directly set.
Several “on-demand” control structures are also developed in which both product streams are
flow-controlled. The control system must adjust the conditions in the plant and the fresh feed
streams to achieve the desired product flow rates. The most effective on-demand control structure
requires no reactor composition analyzer and no recycle of product streams.

1. Introduction
Many continuous chemical processes feature consecu-
tive or simultaneous reactions that result in the produc-
tion of multiple products. A typical example is a process
in which there are two reversible reactions producing
two products M and D.

A+BSM+C
A+MSD+C

The demand for the two products may vary, so the plant
must be designed so that different product ratios can
be attained. The optimum steady-state economic design
changes as the desired product ratio changes.
The control system for the process must also be able
to achieve the desired product ratio. This paper dis-
cusses the plantwide control of this type of process. The
flowsheet includes a reactor, three distillation columns,
and (in some cases) three recycle streams. Because the
reactions are reversible, chemical equilibrium limits
conversion, so the reactor effluent contains significant
amounts of unreacted A and B. These components must
be recycled back to the reactor from the downstream
separation section. If chemical equilibrium results in an
amount of one of the products leaving the reactor that
is in excess of demand, it may be necessary to also
recycle some of the product stream.
Several alternative plantwide control structures are
developed and compared. Some structures use the
conventional approach of setting the fresh feed streams.
Other control structures are based on an “on-demand”
control objective; i.e., the flow rates of the product
streams are flow-controlled, and the control structure
must adjust the fresh feed flow rates and conditions
throughout the process to satisfy the specified product
production rates.
* To whom correspondence hould be addressed. E-mail:
WLL0@Lehigh.edu. Phone: 610-758-4256.
†
Current address: Kulchanat Kapilakarn, Chemical Engi-
neering Department, Prince of Songkla University, Songkla,
Thailand, 90112.
10.1021/ie0205905 CCC: $25.00 © 2003 American Chemical Society
Published on Web 03/28/2003
Figure 1. Base case.
The design of on-demand plantwide control structures
has been the subject of only a few papers in the
literature. A general discussion of the inherent dis-
advantages of on-demand control was presented by
Luyben.1 Other on-demand structures for several spe-
cific processes have been presented: the Eastman
process,2 vinyl acetate process,3 and methylamines
process.4
2. Process Studied
Figure 1 shows the flowsheet of the process. Compo-
nents A and B are reactants, component C is a byprod-
uct, and components M and D are the desired products.
The desired products leave in the streams D2 and B3.
The isothermal continuous stirred-tank reactor (CSTR)
is fed by two fresh feed streams (F0A and F0B) and two
recycle streams (D1 and RD). The reactor effluent
contains all five chemical components and is fed to a
three-column distillation separation section.
The desired amount of each component can vary with
market conditions. The desired product split ratio is
defined as a ratio between the amount of product M and
the total product amount: M/(M + D). Different ratios
give different flowsheets and different control struc-
tures. For example, consider the process with relative
volatilities of each component assumed to be RA ) 16,
 Ind. Eng. Chem. Res., Vol. 42, No. 9, 2003 1891

Table 1. Optimum Design for Several M/(M + D)

M/(M + D)
0.0 0.2 0.4 0.5 0.6 0.8 1.0
product M (lb‚mol/h) 0.0 50.0 100.0 125.0 150.0 200.0 250.0
VR (lb‚mol) 12000 8000 8000 5000 8000 5000 12000
F0A 508.65 456.89 406.12 380.74 355.36 304.59 253.83
F0B 255.10 254.84 254.59 254.46 254.33 254.07 253.81
RD (lb‚mol/h) 0 0 0 0 0 400 1000
D1 (lb‚mol/h) 1580 1147 696 883 1784 1448 10448
NT1 27 17 19 18 19 21 20
NT2 29 31 31 30 31 31 31
NT3 30 30 30 30 30 30
VS1 (lb‚mol/h) 6053 2274 1923 2356 4140 3729 3172
VS2 (lb‚mol/h) 1738 767 870 917 1102 1183 1463
VS3 (lb‚mol/h) 1780 1547 1429 1787 1556 1995
DC1 (ft) 17.00 10.64 10.42 10.61 15.36 13.35 12.31
DC2 (ft) 9.11 6.05 6.05 6.62 7.25 7.52 8.36
DC3 (ft) 9.23 9.22 8.26 9.24 8.62 9.76
RR1 2.83 1.11 1.76 1.67 1.77 1.56 2.04
RR2 2.41 14.19 7.61 6.26 6.26 4.85 4.79
RR3 2.87 2.79 2.73 4.00 4.08 6.82
reactor (106 $) 8.56 6.65 6.65 4.96 6.65 4.96 7.64
column 1 (106 $) 1.66 0.64 0.64 0.75 1.10 1.00 0.90
column 2 (106 $) 0.85 0.57 0.61 0.63 0.70 0.73 0.82
column 3 (106 $) 0.86 0.80 0.76 0.87 0.80 0.92
heat ex 1 (106 $) 5.20 4.09 3.66 2.82 4.56 3.79 3.42
heat ex 2 (106 $) 2.31 2.02 2.19 1.52 1.72 1.80 2.07
heat ex 3 (106 $) 3.49 3.18 2.03 2.35 2.15 2.53
energy (106 $/yr) 8.53 5.28 4.75 5.15 5.41 7.08 7.26
TAC (106 $/yr) 14.73 11.38 10.67 9.64 11.40 12.16 13.35

RB ) 8, RM ) 4, RC ) 2, and RD ) 1. If the ratio is zero,

we need only product D. The flowsheet will have one
reactor and two distillation columns. Any M in the
reactor effluent is recycled back to the reactor with A
and B in the distillate stream D1 from the first column.
If we need only product M, the flowsheet will consist of
one reactor and three distillation columns with two
recycle streams from the columns back to the reactor
(one recycle stream of components A and B and one
recycle with component D). Both steady-state economic
design and dynamic control of the process depend on
the product split ratio.
2.1. Optimum Steady-State Design. Table 1 gives
steady-state design conditions and economic results for
the optimum flowsheets over a range of product split
ratios from 0 to 1. The total production rate of both
products is held constant at 250 lb‚mol/h. The two
design degrees of freedom (reactor holdup VR and the
RD recycle flow rate) are varied for each product ratio
to minimize the total annual cost (TAC), which is the
sum of the annual energy cost (for providing heat in the Figure 2. Effect of RD and the product ratio on the recycle flow
column reboilers) and the annual capital cost (assuming rate D1 (a) and TAC (b).
a 3-year payback period). Equipment costs are esti-
mated from the correlations given in work by Douglas.5 at each product ratio and no RD. For low product ratios,
Product purities are set at 99 mol %, and the impurity a large amount of M is recycled back to the reactor. This
of component M in the D1 recycle from the first column implicit recycle is achieved by increasing the impurity
is assumed to be 10 mol % for most cases. The effect of of M in the D1 recycle stream. The optimum D1
the impurities will be discussed below. The numbers of composition is about 15 mol % M for product ratios
trays in the columns are set at twice the minimum and between 0.1 and 0.2. If high product ratios (between 0.32
the reflux ratios at 1.2 times the minimum in the and 0.5) are required, the impurity of M in D1 is about
calculation of the steady-state economics. 10 mol %, which is the value used in the base case.
Figure 2 shows that the economic steady-state design 2.2. Base Case for Dynamic Studies. The base case
indicates that there should be no RD recycle for product used in the dynamic control studies has a product split
ratios of less than about 0.6. For higher ratios (more ratio of 0.5 (equal amounts of the two products). The
production of M and less of D), it is economical to recycle optimum reactor holdup for this case is 5000 lb‚mol, and
some of the bottoms from the third column, which is the three columns have 18, 30, and 30 theoretical trays.
component D, back to the reactor. As shown in Table 2, the results of rating calculations
Figure 3 shows the effect of the impurity of component showed that the vapor boilup rates in some of the
M in D1 recycle on the vapor boilup rate VS1 and TAC columns had to increase quite significantly in order to
of the first column with the optimum number of trays achieve different product ratios using the 5000 lb‚mol
1892 Ind. Eng. Chem. Res., Vol. 42, No. 9, 2003

proportional-only with gains of 2 for the columns and

10 for the reactor.
The reflux flow rates in each column are ratioed to
the column feed flow rate. The column feed is measured,
this flow signal is multiplied by the desired ratio, and
the output of the ratio is the setpoint of the reflux flow
controller. The desired ratio is set at the steady-state
value.
2.2.2. Reactor. Knowing the conditions in the reactor
for various product ratios provides important insight
about how the control system must change conditions
in the process to achieve different production rates.
Figure 5 shows how the compositions of A and B in the
reactor must change as the product ratio changes. When
more M is desired, the concentration of A in the reactor
(zA) must decrease and the concentration of B in the
reactor (zB) must increase for the same total production
rate. This results in a lower production rate of D from
the second reaction because the second reaction depends
on the concentration of both M (zM) and B (zB).
This information is useful in understanding how
alternative control structures can be set up to adjust
Figure 3. Effect of xD1,M on the vapor boilup rate VS1 (A) and process conditions to achieve the desired product pro-
TAC (B). duction rates.
Table 2. Comparison of the Column Vapor Rates for Two
Reactor Sizes 3. Control Structure Development
M/(M + D) We consider only conventional PI diagonal control
VR ) 5000 lb‚mol VR ) 8000 lb‚mol structures in this paper and use the plantwide control
0.4 0.5 0.6 0.4 0.5 0.6 procedure discussed by Luyben et al.3 to develop alter-
native basic regulatory control structures. Different
F 1652 1637 1803 1448 1480 1590
control objectives yield different control structures.
zA 0.345 0.255 0.176 0.283 0.207 0.138
zB 0.191 0.270 0.388 0.180 0.258 0.377 Process understanding and insight are employed to
VS1 1430 1250 1420 1000 958 1200 generate alternative control schemes, whose dynamic
VS2 730 770 790 730 770 780 performances are evaluated by rigorous simulations.
VS3 1745 2230 970 1745 2230 920 As is the case with all complex multiunit processes
D1 990 1000 1190 788 845 980 with recycles, there are a very large number of alterna-
xD1,A 0.575 0.417 0.267 0.520 0.362 0.223
xD1,B 0.318 0.440 0.586 0.331 0.451 0.611
tive control structures. As the “First Law of Plantwide
RR1 0.44 0.24 0.19 0.28 0.13 0.23 Control” states, it is easy to find a plantwide control
RR2 6.21 4.86 4.17 6.21 4.87 4.15 structure that does not work! We present several
RR3 3.25 4.88 1.71 3.25 4.88 1.59 workable control structures in this paper, but there is
no claim that any of these are “the best” from any
reactor. It is unrealistic to assume that column vapor perspective. We only claim that they provide stable
rates can be changed by more than 50%. operation. If optimum steady-state economic operation
To achieve more flexibility, the base case reactor is desired, the setpoints of the basic regulatory control-
holdup was increased to 8000 lb‚mol. The number of lers in these structures can be adjusted by an online
trays and the diameters of the columns were fixed at optimizer.
the optimum design values. As shown in Table 2, the The following control loops are use in all structures:
required changes in column vapor rates to achieve 1. The reactor temperature is controlled by manipu-
different product ratios are reduced. lating the cooling water flow rate.
2.2.1. Columns. The initial conditions for the col- 2. Column temperatures are controlled by changing
umns in the dynamic simulations were found by using the reboiler heat inputs.
a rigorous steady-state simulation of the columns (the 3. Column reflux flows are ratioed to column feeds.
Wang-Henke method). Column base and reflux drum 4. Column pressures are controlled by condenser heat
holdups were set at 10 min (at 50% level). removal.
Inferential control of product compositions in all 5. The reflux drum level in the first column is
columns is achieved by selecting an appropriate control controlled by manipulating the distillate stream D1,
tray. Figure 4a shows the temperature profiles in the which is recycled back to the reactor.
three columns. Figure 4b shows the sensitivity of the 6. The reflux drum level in the third column is
tray temperatures in column 2 to small changes in heat controlled by manipulating the distillate stream D3,
input and reflux ratio. Using the temperature profiles which is the byproduct component C.
and the sensitivity results, we selected tray 6 in column The control of the reflux drum level in the second
1, tray 23 in column 2, and tray 6 in column 3. Two column is different in the various control structures.
first-order temperature measurement lags of 1 min are Similarly, the variables used to control the column base
used in all temperature control loops. Relay-feedback levels vary with the control structure.
tests are used to obtain the ultimate gain and frequency. 3.1. Conventional Control Structures. In the
Then the temperature controllers are tuned using the conventional control structure, the production rates of
Tyreus-Luyben tuning rules. All level controllers are the two products are not directly set. The fresh feed

Figure 4. (a) Temperature profiles of three columns. (b) Temperature differences of column 2 with 5% ∆VS and 5% ∆RR.
Figure 5. Compositions of A and B in the reactor for different
product ratios.
streams are set directly or are manipulated by some
other controller (level or composition). The three alter-
native conventional control structures explored in this
work are shown in Figures 6 and 8-10.
3.1.1. Structure CS1. Figure 6 gives the most
obvious conventional control structure. The two fresh
feed streams (F0A and F0B) are flow-controlled. Setting
these two flows fixes the amounts of the two products
M and D that will be produced in the plant. Every 1
mol of M produced consumes 1 mol of A and 1 mol of B.
Every 1 mol of D produced consumes 2 mol of A and 1
mol of B. If we neglect for the moment the small amount
of component B that is lost as an impurity in the D
product (stream D2), we can relate the production rates
of the two products M and D to the flow rates of the
Ind. Eng. Chem. Res., Vol. 42, No. 9, 2003 1893
pure fresh feeds F0A and F0B. The stoichiometry of the
two reactions requires that
M + 2D ) F0A
M + D ) F0B
Solving for M and D gives
D ) F0A - F0B
M ) 2F0B - F0A
From the equations above, a limitation on the range of
possible values of the ratio between F0A and F0B can be
deduced. Physically, the ratio can vary between 1 and
2. If the ratio is 1, the process produces only product D.
Figure 7 shows the relationship between the F0A/F0B
ratio and the recycle D1 back to the reactor from the
first column. In this figure we assume that the first
column removes all of components A and B from the
top and all of components M, C, and D from the bottom.
As we approach either of the limiting cases (all M or
all D), the concentration of one of the product compo-
nents must become very small. This means that the
concentrations of the reactants A and B must become
very large, which produces a large D1 recycle.
The conditions in the reactor (concentrations and
flows) and in the columns must adjust themselves to
generate exactly these amounts of products in streams
D2 and B3. Of course, equipment limitation may limit
the achievable range in which the process can oper-
ate.
1894 Ind. Eng. Chem. Res., Vol. 42, No. 9, 2003
Figure 6. CS1 conventional control structure.
Figure 7. Relationship between F0A/F0B and D1.
The key loops in the CS1 structure are as follows:
1. The reactor level is controlled by manipulating the
flow rate of the reactor effluent F.
2. Reflux drum levels are controlled by distillate flows.
3. Base levels are controlled by bottoms flows.
Although this control structure appears to be an
obvious and straightforward one, the results presented
later show that its performance is unsatisfactory. We
should not be too surprised with these results. This
structure has the potential to exhibit the “snowball”
effect because all of the levels in the recycle loop from
the first column are on level control.
3.1.2. Structure CS2. Figure 8 gives a control
structure that provides good stable regulatory control
of the process. The key loops in the structure are as
follows:
1. The reactor level is controlled by manipulating the
fresh feed F0B.
2. The reactor composition zA is controlled by changing
the fresh feed F0A.
3. The reactor effluent F is flow-controlled.
4. Reflux drum levels are controlled by distillate flows.
5. Base levels are controlled by bottoms flows.
Two modifications of this control structure are also
studied. In the first, the composition of A in the distillate
of the first column xD1,A is controlled instead of the
reactor composition. In the second, the temperature of
the top tray in the first column is used to infer the
composition of A. These alternatives are shown in
Figure 9. The control structure works well, as is shown
in the next section. The flow controller in the reactor
effluent stream prevents the snowball effect. However,
this structure has the disadvantage that the production
rates are not directly set. If different flow rates of the
product streams D2 and B3 are desired, the two available
handles are the setpoints of the composition (or tem-
perature) controller and the reactor effluent flow con-
troller. These would have to be adjusted in an open-
loop fashion to influence the production rates of the two
products. Historical plant data or a model of the process
would typically be used to estimate the values of zA and
F required.
3.1.3. Structure CS3. Figure 10 gives an alternative
control structure that eliminates the need for the
composition analyzer. The features of the scheme are
as follows:
1. The reactor level is controlled by manipulating the
fresh feed F0B.
2. The reactor effluent F is flow-controlled.
3. The fresh feed F0A is ratioed to the reactor effluent
flow.
4. Reflux drum levels are controlled by distillate flows.
5. Base levels are controlled by bottoms flows.
The key feature is item 3. Fresh feed F0A is changed
as the reactor effluent flow is changed. The two handles
to adjust the product production rates are the setpoint
of the reactor effluent flow controller and the ratio
signal. This structure provides stable regulatory control
of the process without requiring a composition analyzer.

Figure 8. CS2A conventional control structure.
Figure 9. Modified CS2B/C conventional control structure.
However, it still has the problem of not directly setting
product production rates.
3.2. On-Demand Control Structures. In on-de-
mand control structures, the flow rates of the product
streams are set. We have looked at several alternatives,
as discussed below. It is important to remember that
reactor compositions must change to achieve different
product ratios. This implies that any on-demand control
structure that attempts to hold the composition in the
reactor constant will not work.
Ind. Eng. Chem. Res., Vol. 42, No. 9, 2003 1895
Three different on-demand control structures are
discussed below. In all of these schemes, there are flow
controllers on the two product streams D2 and B3.
3.2.1. Structure CS4. Figure 11 shows an on-demand
control structure in which the flow rates of the two
product streams (D2 and B3) are flow-controlled. Recycle
streams of both products (streams RM and RD) are used
to control the reflux drum level in column 2 and the
base level in column 3.
The reactor composition is not controlled. Fresh feed
1896 Ind. Eng. Chem. Res., Vol. 42, No. 9, 2003
Figure 10. CS3 conventional control structure.
Figure 11. CS4 on-demand control structure.
F0B comes in on reactor level control. Fresh feed F0A is
manipulated to maintain a constant ratio of the two
recycle flows. The ratio of the flow rates of the two
recycle streams is calculated and fed into a ratio
controller RC as the “PV” signal. The output signal of
the ratio controller is the setpoint of the flow controller
on the fresh feed F0A.
This scheme is workable, but it has the disadvantage
that the use of product recycles increases the energy
consumption and increases the load on the columns.
3.2.2. Structure CS5. Figure 12 gives an alternative
on-demand control structure in which recycles of prod-
ucts are not used. The basic idea is to develop a
relationship that shows how reactor composition zA

Figure 12. CS5 on-demand control structure.
Figure 13. Composition of A in the reactor for different D2 and
B3.
must change to achieve different product flow rates D2
and B3. Figure 13 shows this relationship as developed
from the steady-state model of the process. The setpoint
of the reactor composition is changed by the signal from
the “product ratio” function block, given the actual flow
rates of the two products D2 and B3.
The CS5 control structure is similar to CS2A in the
reactor section. However, in the separation section there
are important differences.
1. The base levels in all three columns are controlled
by adjusting the flow rates of the feed streams to the
columns. Thus, the material balance control structure
is in the opposite direction to flows through the system;
i.e., flows into a unit are manipulated to hold an
inventory in the unit.
2. The level in the reflux drum of the second column
is controlled by manipulating the setpoint of the tem-
perature controller in the first column. The idea is to
adjust the amount of component M that is recycled back
Ind. Eng. Chem. Res., Vol. 42, No. 9, 2003 1897
to the reactor in the D1 stream to feed into the second
column just the amount of M that is called for by the
flow controller on stream D2. For example, if the reflux
drum level is rising in the second column, there is too
much component M being fed into the column. So, the
level controller raises the setpoint of the column 2
temperature controller, which drives more M overhead.
This control structure requires a composition ana-
lyzer. It also depends on the accuracy of the correlation
(Figure 13) between product flow rates and reactor
composition. Its effectiveness could also be affected by
inaccuracies in flow measurements. The impact of these
errors is explored later in this paper.
3.2.3. Structure CS6. Figure 14 gives the final on-
demand control structure studied. It is similar to CS5,
but it has two important differences:
1. The reactor composition analyzer is eliminated.
2. The flow rate of the fresh feed F0A is simply ratioed
to the flow rate of the recycle stream D1 from the first
column.
This structure evolved from looking at how several
variables and ratios of variables had to change to
achieve different product flow rates at steady state. As
Figure 15 shows, the ratio of F0A to F is fairly constant
over a wide range of values of D2 and B3 product flow
rates. This process insight led us to try CS6.
Table 3 summarizes the important loops and require-
ments of the various control structures.
4. Results and Discussion
Each control structure was subjected to several
disturbances, as discussed below.
4.1. Control Structure CS1. Figure 16 gives the
results when the conventional CS1 control structure is
1898 Ind. Eng. Chem. Res., Vol. 42, No. 9, 2003

Figure 14. CS6 on-demand control structure.

Table 3. Summary of Control Structures

CS1 CS2 CS3 CS4 CS5 CS6
mode conv conv conv on-demand on-demand on-demand
flow-controlled streams F0A and F0B F F F, D2, and B3 D2 and B3 D2 and B3
reactor composition analyzer no yes no no yes no
ratio F0A/F RM/RD F0A/D1
product recycles no no no yes no no
setpoint of TC1 adjusted no no no no yes yes

structure can handle a very small (2%) decrease in F0A,

Figure 15. Ratios between F0A and D1 for different D2 and B3.
used. The disturbances are small changes in the fresh
feed flow rate of component A (F0A).
One set of curves in Figure 16 shows that the control
structure can handle a +10% change in F0A, but it takes
more than 10 days for the final changes in the produc-
tion rates to occur. This very slow response is due to
the presence of the recycle between the first column and
the reactor. This very sluggish responsiveness would
probably be unacceptable with the present industrial
requirements for “agile” processes. Note also the snow-
ball effect: a 10% change in F0A causes a 30% change
in the flow rate of the D1 recycle.
An even more severe problem is shown in the other
two sets of curves in Figure 16. The CS1 control
but it cannot handle a modest 5% decrease. The D1
recycle increases to the point that the vapor boilup in
the first column reaches its limit, which is set at 3 times
the steady-state design value.
Thus, this intuitively straightforward control struc-
ture does not provide effective control.
4.2. Control Structure CS2. Figure 17 gives the
results when the conventional CS2A control structure
is used. The disturbances are positive and negative
changes in the setpoint of the flow controller on the
reactor effluent stream F, which occur at a time equal
to 3 h. The control structure is very robust, handling
very large changes ((50%).
Changing F has a drastic effect on the recycle D1 but
has less effect on the flow rates of the fresh feed streams
and the product streams. Deceasing F changes the
production rates more significantly than increasing F.
Reactor composition zA is held constant, and the produc-
tion rates of both products change by about the same
amount. Using the distillate composition or the top tray
temperature to control F0A gives the same response as
using the reactor composition analyzer for the distur-
bances in F.
Figure 17B shows how the impurities in the two
product streams (D2 and B3) are affected by these large
disturbances. Although there are momentary large
deviations, the column temperature controllers do a
good job in returning product purities to close to the

Figure 16. Sensitivity of CS1 to changes in F0A fresh feed flow rate.
Figure 17. (50% change in F for CS2.
Figure 18. (25% change in zA for CS2.
desired levels. Of course, this performance could be
improved by using steam-to-feed feedforward control to
assist the temperature controllers. For simplicity in
comparing the alternative structures, no feedforward
control is included in the simulation results.
Figure 18 illustrates the effect of changing the set-
point of the reactor composition controller. The original
steady-state value is 20 mol % A. The setpoint is
Ind. Eng. Chem. Res., Vol. 42, No. 9, 2003 1899
increased to 25 mol % or decreased to 15 mol %. This
change has a strong effect on the recycle D1 and on the
flow rates of the product streams but has less of an effect
on the flow rates of the fresh feed streams. However,
the product distribution is altered. Higher zA values
produce more of component D (stream B3 increases).
Figure 19 compares the responses of CS2A, CS2B, and
CS2C when the setpoint of the first column tray
1900 Ind. Eng. Chem. Res., Vol. 42, No. 9, 2003
Figure 19. 10% TC1 setpoint change for CS2A, CS2B, and CS2C.
temperature controller (tray 6) is change. Note that in
this column there are two temperature controllers for
CS2C: one is at the top tray, which controls the
composition A in D1, and another is at tray 6, which is
used to infer the impurity of B in the bottoms. All
structures are workable. Using the composition control-
ler in either the reactor or the column gives the same
response. The response when using temperature control
is different from that when using composition control;
i.e., increasing the temperature setpoint of the first
column drives more M up the column, and then the top
tray temperature controller increases the flow rate of
fresh feed F0A (because if the temperature is high, that
means there is less component A in the distillate flow).
This causes a higher production of component D.
4.3. Control Structure CS3. Figure 20 gives the
results when the conventional CS3 control structure is
used. The disturbances are (50% changes in the reactor
effluent flow rate F, which occur at a time equal 3 h.
The effect of the flow rate F on the production rate is
larger than that for CS2. Because the ratio of F0A/F is
fixed, a 50% change in F produces an immediate 50%
change in the F0A fresh feed. The total production of
Figure 20. (50% change in F for CS3.
products (M plus D) changes more significantly. The D
product (stream B3) is more affected than the M product
(stream D2), as expected, because a higher concentration
of A in the reactor favors the production of D via the
second reaction.
Notice that the reactor composition and the product
flow rates are still slowing changing after 60 h.
4.4. Comparison of Conventional Control Struc-
tures. Figure 21 gives a direct comparison between
CS2C and CS3 for the +50% increase in F. The larger
impact on the production rate of CS3 is clearly shown.
The third curves in this figure show the case where the
F0A/F ratio is not closed, so there is no change in F0A.
Only F has changed, and there is no composition
controller. Note that the product distribution changes.
For a +50% increase in F, the system ends up producing
less M (stream D2 decreases) and more D (stream B3
increases).
Figure 21B gives results for a -50% change in F.
Notice that the system shuts down when the F0A/F ratio
is not closed. The reactor level controller cuts back on
the fresh feed F0B, and the system fills up with compo-
nent A.
Figure 22 compares the responses of CS1, CS2C, and
CS3 structures when the process has 10 mol % impurity
of component B and 90 mol % A in fresh feed F0A. The
response is similar to the results of changing the fresh
feed F0A flow rate in the sense of the production rate.
CS1 fails to handle this change, but CS2C and CS3 are
workable for this disturbance. CS2C brings the process
to the new steady state faster than CS3. The advantage
of CS2C to CS3 is that the production rates of M and D
do not change significantly when the process has some
impurities in the fresh feeds because CS2C controls the
composition A indirectly by the temperature control of
the top tray in the first column.
4.5. Control Structure CS4. Figure 23 gives the
results for the on-demand control structure in which
recycle streams of both products are used. The reflux
drum level controller in column 2 manipulates recycle
RM and the base level controller in column 3 manipu-
lates recycle RD as the flow rates of the two product
streams are varied.
We show in these figures the maximum and minimum
values of the change in the product flow rate for which
the control structure is able to stabilize the system. Only
small increases in D2 or decreases in B3 can be handled,

Figure 21. (50% change in F for CS2C, CS3, and CS3 (fixed F0A).
Figure 22. 10% impurity of B in F0A for CS1, CS2C, and CS3.
while significantly larger changes can be handled in the
opposite direction.
4.6. Control Structure CS5. Figure 24 gives the
results for the on-demand control structure in which the
“product ratio” function block predicts the reactor
composition required. This structure can handle very
large changes in the individual product production
rates.
Figure 25 indicates that CS5 is workable even when
there are some errors in flow measurement or errors in
prediction. The disturbance is a 20% decrease in the flow
rate of B3. In Figure 25A, the predicted value of the
reactor composition is in error by 10% (both positive and
negative), so the setpoint of the reactor composition
controller is not the correct. In Figure 25B, the flow-
rate signal sent to the predictor is in error by (10% (the
actual B3 flow rate is not what the predictor uses to
calculate the reactor composition setpoint). Despite
these errors, the system rides through the disturbance.
However, there are some large changes in D1 that may
adversely affect the operation of the first column.
Similar results were obtained for changes in D2 with
the same kinds of errors in prediction or flow measure-
ments.
4.7. Control Structure CS6. Figures 26 and 27 give
the results for the on-demand control structure in which
Ind. Eng. Chem. Res., Vol. 42, No. 9, 2003 1901
Table 4. Limits of Setpoint Changes of On-Demand
Control Structures
structure % D2 change % B3 change
CS4 -40, 25 -16, 12
CS5 -55, 40 -20, 40
CS6 -55, 40 -20, 40
the F0A/D1 ratio is maintained and there is no reactor
composition control. The fact that this control structure
does not require a composition measurement is a
significant advantage.
A variety of disturbances were made to test the
robustness of this control structure. Figure 26 shows
the dynamic results when increases and decreases in
the individual product flow rates are made. The system
can handle larger increases in either product flow rate
than decreases.
Figure 27 gives the results when simultaneous 20%
changes are made in both product flow rates.
1. Figure 27A: Both D2 and B3 are increased 20%.
2. Figure 27B: Both D2 and B3 are decreased 20%.
3. Figure 27C: D2 is increased 20%, and B3 is
decreased 20%.
4. Figure 27D: D2 is decreased 20%, and B3 is
increased 20%.
All of the fairly large disturbances are effectively
handled by the CS6 control structure.
4.8. Comparison of On-Demand Control Struc-
tures. Figure 28 compares the results for all on-demand
structures when the fresh feed F0A composition changes
from 100 mol % A to 90 mol % A and 10 mol % B. All
structures are workable.
CS4 with the recycle streams of two products takes a
longer time than the others to get to the new steady
state because the process must detect the change in
columns 2 and 3 and adjust the recycle flow rates, RM
and RT. The reflux drum level in column 2 is controlled
by the temperature setpoint of column 1, and the base
levels are controlled by the column feeds for CS5 and
CS6. That means level controller signals are opposite
to the direction of flows. This is faster than using the
recycle product stream to adjust the levels. To set fresh
feed F0A, CS5 uses the predictor and reactor composition
controller. The controller detects the change in composi-
tion A in the reactor and adjusts the fresh feed.
Therefore, the response of CS5 is the fastest. In any
event, the response of CS6 is faster than CS4. Table 4
1902 Ind. Eng. Chem. Res., Vol. 42, No. 9, 2003
Figure 23. Changes in D2 and B3 for CS4.
Figure 24. Changes in D2 and B3 for CS5.
Figure 25. CS5 with errors in prediction or flow measurement.
gives the limits of feasible changes in the product flow
rates for each of the three on-demand control structures.
Changes that are larger than these cannot be handled.
5. Operating Degrees of Freedom
This two-product process basically has three operating
degrees of freedom (the setpoints of three controllers can
be changed), so different flow rates of the two products
can be achieved, provided they lie within the feasible
region of the process and its control structure. In the
conventional control structures (CS1, CS2, and CS3),
besides the control tray temperature in the first column,
other operating degrees of freedom are as follows.
1. CS1: flow rates of two fresh feeds F0A and F0B.

Figure 26. Changes in D2 and B3 for CS6.
Figure 27. Simultaneous changes in D2 and B3 for CS6.
2. CS2: reactor composition zA or the top-tray tem-
perature of the first column and reactor effluent flow
rate F.
3. CS3: reactor effluent flow rate F and the ratio of
F0A fresh feed to F.
There are no recycles of products back to the reactor
in these conventional control structures.
Ind. Eng. Chem. Res., Vol. 42, No. 9, 2003 1903
The three on-demand control structures use different
degrees of freedom and have different ways of recycling
excess component M. The 2 degrees of freedom are the
specified product flow rates.
The CS4 structure explicitly uses two recycle streams
(RM from the top of the second column and RD from the
bottom of the third column), and these provide an
1904 Ind. Eng. Chem. Res., Vol. 42, No. 9, 2003
Figure 28. 10% impurity of B in F0A for CS4, CS5, and CS6.
additional 2 degrees of freedom compared to the process
with no recycle streams. These additional degrees of
freedom in CS4 are the reactor effluent flow rate F and
the ratio of product M recycle (RM) to product D recycle
(RD).
Both the CS5 and CS6 structures have no explicit
recycle streams. However, recycle of excess M is achieved
by changing the temperature controller setpoint in the
first column so more or less M goes overhead in D1 and
is recycled back to the reactor. So, the relaxation of
holding a fixed tray temperature in the first column
provides an additional degree of freedom in CS5 and
CS6, beyond the two product flow rates. In CS5 the
reactor composition is the additional degree of freedom.
In CS6 the ratio of flow rate of fresh feed F0A to the
flow rate of D1 recycle is the additional degree of
freedom.
It is interesting to note that the different control
structures will not operate at the identical steady-state
conditions for the same product flow rates. This occurs
because other combinations of variables can produce the
same net production rates of the two products. For
example, the reactor effluent flow rate F, the recycle
D1 from the first column, the control tray temperature
in the first column and the reactor composition can
change from structure to structure. However, as dis-
cussed earlier in this paper, the two fresh feed flow rates
will be identical in all control structures for the same
product flow rates because of the stoichiometry.
6. Conclusion and Future Work
This paper has compared the dynamic performance
of several alternative control structures, both conven-
tional and on-demand. The most obvious conventional
control structure, which simply fixes the flow rates of
the two fresh feeds, does not provide effective control.
It exhibits the snowball effect and can handle only very
small disturbances.
The most promising on-demand control structure
(CS6) requires no reactor composition analyzer and can
handle quite large changes (about 20%) in the specified
flow rates of either or both products.
Future work includes extensions to a process in which
there are three consecutive reactions and three products
are produced. This case presents a much more chal-
lenging situation in which explicit recycles of some of
the products may be necessary because of a lack of
degrees of freedom. There is also no unique relationship
between the flow rates of the two fresh feeds and the
flow rates of the three products; i.e., setting the two
fresh feeds does not set the production rates of each of
the three individual products because the A and B
reactants can be consumed to make different amounts
of the three products.
Nomenclature
Bi ) bottom flow rate from the ith column (lb‚mol/h)
CC ) composition controller
CS ) control structure
DCi ) column diameter (ft)
Di ) distillate flow rate from the ith column (lb‚mol/h)
F ) reactor effluent flow rate (lb‚mol/h)
FC ) flow controller
FT ) flow transmitter
F0A ) fresh feed rate of the A component (lb‚mol/h)
F0B ) fresh feed rate of the B component (lb‚mol/h)
FLi ) column feed rate to the ith column (lb‚mol/h)
LC ) level controller
NFi ) feed tray of the ith column
NTi ) total number of trays of the ith column
PC ) pressure controller
ratio ) mulitplier
Ri ) reflux flow rate of the ith column (lb‚mol/h)
RD ) product D recycle flow rate (lb‚mol/h)
RM ) product M recycle flow rate (lb‚mol/h)
RR ) reflux ratio
SP ) controller setpoint signal
TC ) temperature controller
TR ) reactor temperature (°F)
VR ) reactor holdup (lb‚mol)
VSi ) vapor boilup in the ith column (lb‚mol/h)
xBi,j ) bottoms composition of the j component from the ith
column (mole fraction)
xDi,j ) distillate composition of the j component from the
ith column (mole fraction)
zi ) reactor composition of the i component (mole fraction)
Greek Symbols
Ri ) relative volatility of the i component
∆RR ) change in the reflux ratio
∆F ) change in the reactor effluent flow rate (lb‚mol/h)
∆F0A ) change in the fresh feed rate of the A component
(lb‚mol/h)
∆VS ) change in the vapor boilup (lb‚mol/h)
∆zA ) change in the reactor composition (mole fraction of
A)
Literature Cited
(1) Luyben, W. L. Inherent Dynamic Problems with On-
Demand Control Structures. Ind. Eng. Chem. Res. 1999, 38, 2315-
2329.
(2) Luyben, W. L. Simple Regulatory Control of the Eastman
Process. Ind. Eng. Chem. Res. 1996, 35, 3280-3289.
(3) Luyben, W. L.; Tyreus, B. D.; Luyben, M. L. Plantwide
Process Control; McGraw-Hill: New York, 1999.
(4) Luyben, W. L. Plantwide Dynamic Simulators in Chemical
Processing and Control; Marcel Dekker: New York, 2002.
(5) Douglas, J. M. Conceptual Design of Chemical Processes;
McGraw-Hill: New York, 1988.
Received for review August 4, 2002
Revised manuscript received December 30, 2002
Accepted February 13, 2003
IE0205905