476 Ind. Eng. Chem. Res.

1993,32, 476-486

Dynamics and Control of Recycle Systems. 2. Comparison of

Alternative Process Designs
William L. Luyben
Department of Chemical Engineering, Iacocca Hall, Lehigh University 111, Bethlehem, Pennsylvania 18015

This paper is the second in a series of papers that explore the challenging problems associated with
the dynamics and control of recycle systems. In this paper we compare the steady-state economics
and the controllability of several alternative process designs. The designs are primarily limited to
a reactor/column configuration with recycle, but some consideration is given to several other process
configurations (several reactors in series and a tubular reactor). The reactor/stripper configuration
is shown to be an economically attractive process configuration even for a simple single irreversible
reaction system when fairly high conversions are desired. The optimum steady-state economic
design consists of a fairly large reactor and a moderately sized stripping column. As the reactor
is made larger, the energy consumption in the column decreases but the number of trays increases.
This translates into a taller stripping column (more trays), but one that is smaller in diameter, and
into smaller heat exchangers for the reboiler and the condenser. The dynamics of the column stay
about the same as its diameter changes, but the dynamics of the reactor slow down as the size of
the reactor increases. The gain around the recycle loop increases as the reactor size is decreased.
All these competing effects result in a process in which the dynamics of the various alternative
designs are quite similar. Temperature control in the reactor is shown to improve as the size of the
reactor increases, which is just the opposite of the normal situation when one scales up a reactor.
For the smaller reactor process, coupling a reactor, which is stable by itself, with a column, which
is stable by itself, is shown to sometimes lead to an unstable coupled system.

Introduction constant are calculated from the Francis weir formula. A

As discussed in part 1 (Luyben, 19931, a practical
procedure has yet to be developed for the design of an
effective control system for a plant that contains recycle
streams. The historical approach of installing large surge
tanks is no longer acceptable for a number of reasons:
capital investment, safety, and environmental concerns.
The literature on the subject of recycle dynamics is quite
limited and was reviewed in part 1.
This paper looks at the same simple recycle process
considered in part 1which consists of a continuousstirred-
tank reactor followed by a distillation column. But we
now explore the effects of changing the sizes of the
individual unita in the process. The effects of redesign on
both the steady-state economics and on the dynamics of
the individual unita and of the overall process are studied.
The Basic Process
-
An irreversible, first-order reaction A B occurs in a
single stirred-tank reactor. Some of the reactant A is
consumed. The reactor effluent is a mixture of A and B,
and it is fed into a distillation column which takes product
B out the bottom and component A out the top. The
distillate is recycled back to the reactor. Fresh feed is
introduced into the reactor at a rate FOand composition
20.
First-order kinetics and, for the moment, isothermal
operation are assumed in the reactor.
R = V,kz (1)
where R = rate of consumption of reactant A (lb-mol/h),
VR = holdup of reactor (lb-mol),k = specific reaction rate
(h-I), and z = concentration of reactant A in the reactor
(mol fraction A).
In the column we assume a constant relative volatility
between A and B of CY = 2, equimolal overflow, theoretical
trays, saturated liquid feed, total condenser, and partial
reboiler. Tray holdups and the liquid hydraulic time
0888-5885/93/2632-0476$04.oo/o 0 1993 American Chemical Society
1-in.weir height is used. Holdups in the column base and
reflux drum are set to provide 5 min of holdup based on
the total flow rate into the surge volume.
Alternative Reactor Designs
In this paper we will limit the choice of reactor types
to a single continuous stirred-tank (CSTR) reactor. This
is done because we want to concentrate on the plant-wide
effects of recycle and of changing the sizes of individual
units. We do not choose a single CSTR because it makes
the most sense from a kinetic point of view.
Just to illustrate this point, let us consider the process
in which the fresh feed is 239.5 lbmollhwith acomposition
of 0.9 mole fraction A. The desired product is to have a
concentration of 0.0105 mole fraction A. The reaction is
carried out at a temperature that gives a specific reaction
rate k = 0.34 h-l.
Instead of building a process with a CSTR and a column
with recycle, we could simply build a single CSTR and
make ita holdup large enough so that we convert 98.83 7%
of the A in the fresh feed.
20 - z 0.9 - 0.0105 = o.9883
conversion = -=
ZO 0.9
The equation describing the steady-state CSTR is
F$o F$ + V R k Z (3)
Solving for the size of the reactor gives V R = 59 700 lb-
mol. Thus a very large reactor would be required, but no
column would be needed and there would be no recycle
stream.
Now from a reactor design point of view, we know that
we could conduct this reaction more efficiently (use less
reactor volume) by using a number of CSTRs in series.
The relationship for a series of N reactors is
"=[
zo
Fo
Fo+ V,k
3 (4)
where ZN = concentration from Nth stage = 0.0105 mole
 Ind. Eng. Chem. Res., Vol. 32, No. 3, 1993 477
fraction A, LO = fresh feed concentration = 0.9 mole fraction
A, Vn = holdup in one stage (lb-mol), and k = specific
reaction rate = 0.34h-l. Solving for the individual reactor
holdups V, and for the total holdup in the process VtOtal
= VnN for several values of N gives
N=l V,, = 59670 lb-mol V,,l= 59670 lb-mol
N=2 V,, = 5820 lb-mol V,,l= 11640 lb-mol
N=3 V,, = 2400 lb-mol V,,l= 7200 lb-mol
N=4 V,, = 1440 lb-mol V,I = 5760 lb-mol

The total volume of the reaction system can be reduced

by a factor of 10 in going from a single CSTR to four
I I R E R C T O R EFFLUENT

CSTRs in series. The capital costa and the safety and

environmental hazards of this smaller system would all be
greatly reduced. However, the control of this system may
PRODUCT
be more difficult.
Of course the smallest reactor volume would be achieved
Figure 1. Reactor/stripper process with recycle.
by using a plug-flow reactor. The equation describing
this system is Table I. Results of Steadu-State Designs
~~ ~

In(zL/zO)= -kL/v (5) V R (lb-mol) 800 1000 2500 5000 10,OOO

z (mole fraction) 0.7813 0.6250 0.250 0.125 0.0625
where ZL = concentration at the exit of the reactor = 0.0105 V (lb-mol/h) 2007 1079 407.2 304.1 241.5
mole fraction A, L = length of reactor (ft), and u = velocity F(1b-mol/h) 2247 1319 646.7 543.6 481.0
in reactor (ft/h). The velocity of flow in the reactor, u, is NT 14 14 17 16 13
YNT (mole 0.8732 0.7614 0.3909 0.2152 0.1141
the flow rate divided by the cross-sectionalarea AF. Since fraction)
the total volume VPFR is the length L times the cross- Dc (ft) 8.24 6.04 3.71 3.21 2.86
sectional area of flow AF, we can solve for the volume. DR(ft) 7.98 8.60 11.7 14.7 18.5
Q (106 Btu/h) 25.1 13.5 5.09 3.80 3.02
M B(lb-mol) 187 110 53.9 45.3 40.1
M D (lb-mol) 167 89.9 33.9 25.3 20.1
M,, (lb-mol) 9.40 4.68 1.65 1.22 0.97
B (8) 5.30 4.18 2.77 2.42 2.16
heat exchanger
areas (ft2)
reboiler 5020 2698 1018 760 604
condenser 8360 4496 1697 1267 1006
Using the numerical values for the case under study gives capital costa
VPFR = 3135 lb-mol. This is almost a factor of 20 times ($lOOO)
smaller than a single CSTR. reactor cost 162 186 329 507 780
We will not study these alternative reactor systems in column cost 304 218 152 124 93
tray cost 13 8 5 4 2
any more detail in this paper since we want to concentrate exchanger cost 749 500 266 220 189
on recycle systems, but some economic comparisons will energy cost 1099 591 223 167 132
be presented in a later section. total annual cost 1508 895 473 451 487
Parameters That Were Held Constant for All Designe
Reactor/Stripper Process
fresh feed 239.5 lb-mol/h bottoms 0.0105 mole
The process studied in part 1was a fixed process that fresh feed 0.90 mole composition fraction A
was taken from the literature. It had a reactor with 1250 composition fraction A relative volatility 2
lb-mol of holdup and a 20-tray distillation column. The specific 0.34 086 h-' weir height 1in.
reaction rate
steady-state reactor composition was 0.50, and the steady- bottoms 239.5 lb-mol/h
state distillate composition was0.95. The column operated flow rate
with a reflux ratio of 2.2, and the recycle flow rate was
260.5 lb-mol/h. The energy consumption in the column fresh feed flow rate Fo,the fresh feed composition LO, the
was equivalent to 833.2 lb-mollh of vapor boilup. specific reaction rate k, the relative volatility a,and the
Since we assume that there is no need to fix the distillate product purity XB.
composition X D in this system, it becomes a design 1. Select a reactor holdup VR.
optimization variable. This was studied by varying the 2. Calculate reactor concentration 2.
specified value of X D in the design of the column. As
expected, it was found that both energy consumption and
the number of trays decreased as XD was reduced. The (7)
lowest energy consumptionand the lowest achievable value
of XD occurred when no reflux was used, i.e., when the 3. Calculate the minimum vapor boilup Vmin from the
column was operated as a simple stripper. known values of z and X B (which would require an infinite
All cases considered in this paper use a stripper as number of trays in the stripper). Note that the liquid rate
sketched in Figure 1. The reactor effluent F is fed to the in the stripper is the reactor effluent F and the feed is
top tray (NT). The overhead vapor is condensed, passes assumed to be saturated liquid.
through a surge drum, and is recycled back to the reactor.
A. Design Procedure. The following steady-state
design procedure was used to size the stripper for a given
reactor holdup. Other variables that were fixed were the where yz = composition of vapor in equilibrium with liquid
478 Ind. Eng. Chem. Res., Vol. 32, No. 3, 1993
SYSTEM 1 CVR-23BW)

z0=0.9

DISTILLATIOH

SYSTEtl 2 C V R = l B 0 0 >

FRESH FEED

OlSTILLATIOH
F J - 5 2 9 . 2 cpn
TJO :78

AH * Y 6 Y . 8

c Z l l B X B - B ,0165
8.239 m 5
PRODUCT

Figure 2. Two cases studied in dynamic simulations.

of composition z. Using eq 8 and the relationship that
+
Vmin Fo = F, we can solve for F and Vmin.
4. Set the actual vapor boilup at 1.1times Vmin.
5. Starting at the bottom of the stripper at XB, calculate
up the column tray-to-tray until zn is equal to or greater
than z , the column feed composition. The total number
of NT trays is then equal to n.
6. Use a rating program to determine the precise vapor
boilup needed when an integer number of trays (NT) exist.
This is an iterative calculation in which a value of vapor
boilup Vis guessed (a good initial guess is 1.1times Vmin).
Knowing V, the feed rate to the stripper can be calculated
from a total mass balance around the reactor.
F=V+F, (9)
From a component balance around the reactor, we can
calculate the composition of the vapor leaving the top of
the stripper YNT.
YNT = (Fz+ V& - F$,)/V (10)
With the guessed value of V, tray-to-tray calculations up
the column NT trays give a second value for the composition
of the vapor leaving tray NT. If this value is not the same
as that calculated from the reactor balance (eq 101, a new
value of Vis guessed. A simple interval-halving algorithm
was used to converge this loop.
7. The diameter of the column was calculated by
assuming an F-factor of 1.
F-factor = uc(pv)1’2 (11)
 Ind. Eng. Chem. Res., Vol. 32, No. 3, 1993 479
Table 11. Tuning Parameters for Two Cams: Bottoms System 1 Step Change m 10, soltd=O 8. duhed=l

Composition and Reactor Temperature Control I0581 I

CompositionControl in Strippef
VR 2600 lo00
p u (h) 0.24 0.15
K" 6.62 8.22
KZN 3.01 3.74
~ Z (h)
N 0.20 0.125
TemperatureControl of Reach@
VR (Ib-mol) 2500 lo00
from ATV test
Ku 6.60 4.29
~u (h) 0.42 0.43
KZN 3.0 1.95
TZN 0.344 0.35
from linear model 1 042' I
K,. 9.04 6.15 0 1 2 3 4 5 6 7 8 9 10
Kmin 0.763 1.32 Time (houn)
an -0.0857 +3.809
open-loop eigenvalues
-0.0393 -1.131 System 1 Step Change m 10. solrd=O S, dashed4
+1.802 +5.027
-31.82 -56.58
empirical settings
KC 3.0 1.95
71 1.38 2.8
a Controller gains are dimensionless, uaing flow transmitterspans
of twice the steady-state value and compoeition transmitter spans
of 10 mol 5%. Gains are dimensionleaa: temperature transmitter
span = 100 O F and valve span is 4 times the steady-state flow rate
Of Coolant (FJO).
System 1 Smppcr Alone: Sop Chan8e in 2: solid=+ OS. dyhed=-.05
1.3

380L ' i
0 1 2 3 4 5 6 7 8 9 10
Tune (hours)
Figure 4. Response of coupled system 1.

ft3/(lb-mol OR), T = temperature = 300 + 460 = 760 OR,

and uc = vapor velocity in column (ft/s). Using them
numerical values gives eq 12 for the diameter of the column
Dc .

0'85A O:? 0:4 0.6

,
0.8
.
1
.
1.2
.
1.1 1:6 1:s
Dc = 0.1838V'/2 (12)
Time (hours) where V = vapor boilup (lb-mol/h).
8. The height of the column was determined assuming
System 1 Smpper Alone. Step Change in L. soLd=t 05, duhcd=. 05 a 2-ft tray spacing and allowing 20 9% more height for base
500, I level volume.
I I
4s0t < A 1 L, = 2.m, (13)
9. Tray liquid holdup (M, lb-mol) was calculated from
the height of liquid over the weir how(from the Francis
weir formula) and the weir height (1 in.).

how)'"/ M
F/60 = 200pLwL( (14)
where F = liquid flow rate (lb-mollh), WL = weir length
(in feet) = 0.8Dc, and p~ = liquid density (501b/ft3).Using
.......... the numerical values gives eq 15 for the liquid height over
..,'. the weir.
02
,
, I

04 06 08 1

Time (hours)
12 14 1'6 1's 1
Figure 3. System 1 (VR= 2600) response of stripper by iteelf.

where pv = vapor density in the column = MPIRT = 0.2741

lb/ft3,M = molecular weight = 50 lb/lb-mol, P = pressure
= 44.7 psia, R = perfect gas law constant = 15451144 psia where M,, = holdup on tray (lb-mol). The liquid hydraulic
480 Ind. Eng. Chem. Res., Vol. 32,No. 3, 1993
Syrrem 2 Step Change u1 20 sohd-0 8 duhed=l
Syriem 1 Smpper Alone. Step Change YI z. roLd=+ M.duhed=-.05
I
' 0 6 j 5

--..
....___

0 95

103
09 0 1 2 3 4 5 6 i 8 9 IO
0 02 04 06 OS 1 12 14 16 18 2
Time (hours)
Time (hours)

Sysrcm 2 Step Chvlgc in 10. rohd.0 8, dahed.1

Svsrem 2 Smpper Alone, Step Change u1 z.soLd=t 05, dashed=- 05
1400, I
1400

1
loo0

.__.-
...._. ___ ... . ..___. . -
___.....
900 -

s co I
800b 02 0.1 06 08 1 12 11 16 18 i 0 1 2 3 4 5

Time (hours)
6 i 8 9 10

Time (hours)
Figure 6. Response of coupled system 2.
Figure 5. System 2 ( V R = 1OOO) response of stripper by itself.

constant 8 (in minutes) was calculated from eq 17.

trays = 36.26(0c)1.56NT (19)

10. The reactor was assumed to have a diameter DRto
height LR ratio of 0.5.
11. Column base and overhead surge drum was sized
for 5 min of holdup.
12. Reboiler and condenser heat-transfer rates were
determined from V using a heat of vaporization of 250
Btu/lb and a molecular weight of 50 lb/lb-mol. Reboiler
area ARwas calculated assuming an overall heat-transfer
coefficient U = 100 Btu/(hr O F ft2) and a temperature
differential of 50 OF. Condenser area Ac was calculated
assuming an overall heat-transfer coefficient U = 150Btu/
(hr O F ft2) and a temperature differential of 20 O F .
Once all the process flow rates had been determined
and the equipment had been sized, the cost of operation
(energy) and the capital costs of each plant could be
determined. The annual energy cost was calculated
directly from the vapor boilup V, using a value of $5/106
Btu.
The capital costs of the reactor, the column, and the
heat exchangers (reboiler and condenser) were estimated
using the correlations given in Douglas (1988).
column = 1917(Dc)'~oss(Lc)0~82 (20)
heat exchangers = 1557(Ac + AR)0.66 (21)
The annual capital cost was assumed to be one-third of
the capital cost (3-year payback). Materials of construc-
tion were stainless steel. Design pressures were 300 psig.
An M&S index of 800 was used.
Results of Steady-State Designs
Table I gives a complete set of results of the steady-
state design calculations for the reactorlstripper process
for several values of reactor holdup VR. There are a
number of interesting observations that can be made about
these results.
1. Reactor composition decreases as reactor holdup is
increased. Since the fresh feed flow rate, fresh feed
composition, product composition, and product flow rate
are the same in all designs, the rate of consumption of
reactant A has to be the same in all designs. The specific
reaction rate k is the same in all designs since the reactors
runat the same temperature. Therefore, the product k VRZ

Time (hours)
Svsrem I (Rencror Alone): Sreo Chvlee in TO: rolid.0.8. d x h e d = l
Time (hours)
Figure 7. Temperature control of reactor by itaeif (system 1, V R = 2500). (a, left) Half Ziegler-Nichols gain and twice Zieglel-Nichols reset;
(b, right) Ziegler-Nichols gain and 2 and 4 times Ziegler-Nichols reset.
must be the same in all designs. This is why the
concentration of reactant in the reactor z must decrease
inversely with reactor holdup VR.
2. The vapor boilup in the column (and the energy
consumption,column diameter,and condenser and reboiler
areas) increases drastically as reactor holdup is reduced
below about 1500 lb-mol.
3. The composition of the vapor leaving the stripper
decreases as reactor holdup is increased. This is due to
the decrease in z.
4. The minimum-cost design is a reactor with 5000 lb-
mol of holdup and a 16-traystripping column. The reactor
is a vessel 14.7 ft in diameter. The column diameter is
3.21 ft. Energy consumption is 3.80 X 106 Btu/hr. Total
annual cost of energy is $167 OOO/yr. Total capital cost is
$855 000, which gives an annual capital cost of $285 OOO/
year, using a 3-yearpayback. Total annual cost is $451 000/
year.
Designs with smaller reactor sizes have energy costa that
increase more rapidly than capital costa decrease. Designs
with larger reactor sizes have capital costa that increase
more rapidly than energy costa decrease.
The alternative processes discussed earlier in this paper
(a series of CSTRs) would have no column and no recycle,
just one or more reactors. These systems would have zero
energy cost. For one huge reactor (33.6 f t in diameter),
the capital cost would be $2 374 000, giving an annual cost
of $791 300/year, which is much higher than the reactor/
stripper system ($451 000/year). The process with two
CSTRs in series would have two 15.47-ft-diameter reac-
Ind. Eng. Chem. Res., Vol. 32, No.3, 1993 481
System 1 (Reactor Alone) rolidr2'ZN Reset. duhed=?*W Reset
25
20
V J
-15'
0 1 2 3 4 5 6 7 8 9 IO
Time (hours)
System 1 (Reactor Alone) solid=Z*ZN Reset, duhed=?'M Reset
' 1
T i e (hours)
tors, each costing $557 000. Therefore the total annual
cost would be $371 400, which is somewhat less expensive
than the reactor/column process. The process with three
CSTRs in series would have three 11.5-ft-diameter
reactors, each costing $321 000, with a total annual cost
of $321 000.
A plug-flow tubular reactor (6 f t in diameter and 111f t
long) would cost $565000, giving an annual cost of
$188 300/year. However, achieving plug-flow conditions
in this huge reactor would be quite difficult since the
velocity would be only 0.00235 ft/s. Temperature control
problems may also be severe.
The reactor/stripper configuration appears to be an
attractive process configuration even for this simple single
irreversible reaction system when fairly high conversions
are desired.
Dynamics and Control
The discussion up to this point has considered only the
steady-state design of the reactor/stripper system. Now
we wish to look at the dynamics of these systems.
It is natural to expect that the dynamics of the system
will be dominated by the reactor since ita holdup is quite
a bit larger than the holdup in the column. For example,
the optimum economic design has a reactor holdup of 5000
lb-mol. The total holdup in the column is 16 trays at 1.22
lb-mol/tray, column base at 45.3 lb-mol, and overhead surge
drum at 25.3 lb-mol, giving a total of 90.1 lb-mol. Thus
the ratio of holdup in the reactor to holdup in the column
is over 50.
482 Ind. Eng. Chem. Res., Vol. 32,No. 3, 1993
System 1 Step Change m 20. d d = O 8. dashed=: System 1: Slev C h m s in 10;soM=O.8.duhed=l

1056

Time (hours) Time (hours)

System 1 Step Change in 20, solid=0.8, dashed-1 System 1 S a p Change m 20. soLd=O 8. dashedol
440 1 I 330

320

310

300

290

280

270

260

250

240

230
0 1 2 3 4 5 6 7 8 9 10

Time (hours) Time (hours)

Figure 8. Reactor temperature and bottoms composition control: coupled system 1.

In this process we have assumed that the only compo-
sition that needs to be controlled is the base of the
distillation column, the final product purity. Since the
dynamics of the column are quite fast compared to the
reactor, the slow changes in composition z coming from
the reactor will seem to the column almost like ramp
disturbances.
Dynamicsimulation studies of two different designswere
conducted to explore these dynamic effects. Figure 2 shows
the ateady-state conditions and parameter values for the
two caaes. One case has a reactor holdup of lo00 lb-mol
and a large-diameter column (6.04 ft). The other case
features a reactor holdup of 2500 lb-mol and a smaller
3.71-ft-diameter column.
The XB-vloop in the base of each column was tuned by
first determiningthe ultimate gain and ultimate frequency
from a relay-feedback test of the stripper by itself. A
relay height of A 5 % of scale was used. A 3-min dead time
was assumed in the xg compositionmeasurement. Results
for the two cases are given in Table 11.
Figure 3 shows the response of the stripper by itself,
isolated from the reactor, in system 1 when the Ziegler-
Nichols settings were used. The disturbances are f0.05
mole fraction step changes in stripper feed composition.
Control is quite tight with the maximum error about 0.2
mol %, and the system settles out in about 1 h. Figure
4 shows what happens to the entire coupled system when
f0.1 mole fraction step changes are made in fresh feed
cornpositionto the reactor. The maximum error in product
purity is only 0.006 mol %, but the change in reactor
composition takes almost 10h to occur, so the column sees
a disturbance that lasts for a very long time and product
purity is away from its setpoint for many hours.
The same procedure was followed with system 2 with
the smaller reactor. Results are shown in Figures 5 and
6. The response of the stripper by itself is given in Figure
5. The closed-loopdynamics of the stripper in this system
are about the same as that in the other system, despite the
fact that this stripper is much bigger. Comparing Figures
4 and 6 reveals that the dynamics of the overall coupled
systems are quite similar,despite the fact that the reactors
and columns are much different in size.
The reason for this similarity is due to the difference
in the recycle loop gains. The reactor gains were calculated
from a linearized model of the reactor.
for system 1 (V, = 2500): ["I
YNT reactor
= 0.2717
for system 2 (V, = 1o00): [1'
YNT reactor
= 0.6502
The reactor gain decreases as the reactor size increases.
The stripper gains were calculated numerically, using
the steady-state rating program, by making f l % changes
in feed compoeition z and calculatingthe resulting changes
in overhead composition YNT.

System '2 (Rcrctor Alone): Step Chrnpe in 20: rolld=0.8 duhed.1
200
System 2 (Reactor Alone): Siep Chmge in 10:solid.0.8; duhed.1
Time (hours)
Figure 9. Temperature control of reactor by itself (system 2, V R = 1OOO). (a, left) Ziegler-Nichols gain and 8 times Ziegler-Nichols reset;
(b, right) Ziegler-Nichols gain and 16 times Ziegler-Nichols reset.
for system 1 (VR = 2500):
[y]
for system 2 (vR= 1000):
[F]
The total recycle loop gains are 0.397 for system 1 and
0.724 for system 2. The closer this recycle loop gain is to
1, the slower the dynamics of the process will be. The
larger the volume of the reactor, the slower its dynamics
will be. Since the larger reactor system has a smaller
recycle loop gain, these two effectstend to compensate for
each other. Thus the two systems have about the same
dynamic response, despite the fact that one has a much
bigger reactor.
Temperature Control of the Reactor
Up to this point, we have assumed isothermal operation
of the reactor. As a final part of this paper, let us consider
the temperature control problems that occur in the two
different reactors associated with the two different pro-
cesses.
The equations describing a jacketed, constant-holdup,
nonisothermal CSTR are (Luyben, 1990)
where FO = fresh feed flow rate to reactor (lb-mol/h) =
Ind. Eng. Chem. Res., Vol. 32, No. 3, 1993 483
System 2 (Reactor Alone): Step Change in 20; roLid=0.8: d u h e d = l
I
.lSO[ I
0 1 2 3 4 5 6 7 8 9 10
Time (hours)
System 2 (Rca.ctor Alone): Step Change in LO; rolid=0.8 dashed=]
rn)
239.5, zo = fresh feed composition (mole fraction A) =
column
= 1.462 0.90, D = flow rate from overhead surge tank to reactor
(lb-mol/h) = V, XD = composition in surge tank (mole
fraction A) = YNT only at steady state, F = flow rate from
coumn
= 1.114 +
reactor (lb-mol/h) = FO D,ko = preexponential factor
(h-l) = 2.7829 X lolo, E = activation energy (Btu/lb-mol)
= 30000, R = 1.99 Btu/(lb-mol OR), and TR = reactor
temperature (OF) = 140 OF at steady state.
where TO= temperature of fresh feed (OF) = 140, TD =
temperature in overhead surge drum (OF) = 140, h = heat
of reaction (Btu/lb-mol of A) = -30 000, A4 = molecular
weight (lb/lb-mol) = 50, U = overall heat-transfer coef-
ficient (Btu/(h O F ft2)) = 300, AH = heat-transfer area
through reactor wall (ft2)= TDRLR,DR = reactor diameter
(ft), LR = reactor height (ft) = ~ D Rcp, = heat capacity of
process material (Btu/(lb OF)) = 0.5, and TJ= temperature
in cooling jacket (OF).
dt = ($)(Tj0 - Tj) + -(TR
UAH
dT, CJPJVJ
- Tj) (25)
where FJ = coolant flow rate (ft3/h), VJ = jacket volume
(ft3) = AH (0.25) since a 3-in. jacket was assumed, Tjo =
inlet coolant temperature (OF) = 70, CJ = heat capacity of
coolant (Btu/(lb OF)) = 1.0, and PJ = density of coolant
484 Ind. Eng. Chem. Res., Vol. 32, No. 3,1993
System 12 Step Change in 20. whd=0.8, duhedd.95 System 1 Step Change in 20, roLdd.8. dvhed=O 95
109
II ,,

0 1 2 3 4 5 6 7 8 9 10

Time (hours) Time (hours)

System 2 Srep Change n z o , solid-0 8, duhed=0.95 System 2 Step Change m zo. sohd=O 8, dvhed=O 95

2M)0 1 2 3 4 5 6 7 8 9 10

Time (hours) Time Oloun)

Figure 10. Reactor temperature and bottoms composition control: coupled system 2.

(lb/ft3) = 62.3. Two 1-min first-order lags were assumed

to exist in the temperature measurement loop.
Coolant flow rate FJ was the manipulated variable.
Reactor temperature TRwas the controlled variable. The
ultimate gain and frequencywere obtained for each reactor
by the relay-feedback test. Table I1 gives these results
and the corresponding Ziegler-Nichols settings. Also
shown are the maximum and minimum gain found from
a linearized mathematical model of the reactor.
The linear model was used to calculate the roots of the
characteristicequation of the open-loopsystem (the open- b3 = -
TJO - TJ
loop eigenvalues) as well as the maximum and minimum VJ
gains. This model had the form
The total open-loop transfer function used to calculate
& = Ax + B m + DL (26) the maximum and minimum gains was (Deviaand Luyben,
1978)
where x = vector of state variables = [ z TR T J ] ~ ,m =
manipulated variable = FJ, and L = vector of load 1 a&,(s - '11)
disturbances = IZO TOFO XD VIT. (30)
dz
G ~ ( s ) +
= (TMs 1)' 8 3 B 9 2 B,s + + + Bo)
- = U112 + U 1 2 T R + ... (27) where T M = temperature measurement lag (h) = 1/60.
dt
--
dTR- 021% + U ~ ~ T+ UR ~ ~ +
dt
T ...
J
(28) B2 = -all - u~~- u~~
dTJ = 0 3 2 T +
- ~ U +
~ ~ T b$J
J (29) = '11'22 + '11'33 + '22'33 - '12'21 - '23'32
dt
where the other terms in eqs 27 and 28 involve disturbance BO = '12'!21'33 - '11'22'33 + '11'23'32 (31)
inputs.
-2EK Table I1 gives some interesting results.
1. The ultimate gains found from the relay-feedback
'I1 =-K- test are somewhat lower than the maximum gains found

from the linear model, but the agreement is fairly good.
The maximum gain decreasesas the reactor size decreases.
2. The minimum gains increase as reactor size decreases.
This gives a much smaller range of stable gains for the
smaller reactor, indicating that the smaller reactor will be
more difficult to control.
3. Both reactors are open-loopunstable (one of the open-
loop eigenvalues is positive), but the positive pole of the
smaller reactor lies farther to the right in thes-plane. Note
that the a22 element of the smaller reactor is positive, but
it is negative for the larger reactor.
These indications of the smaller reactor being more
difficult to control are born out by simulations of the
nonlinear process. First the reactor was simulated by itself
and different temperature controller settings were tried.
The Ziegler-Nichols settings settings were found to give
unstable response.
For system 1with the large reactor ( VR = 2500),Figure
7ashows that halving the gain and doubling the reset gave
stable responses, but still oscillatory. Empirical tuning
showed that the reset time had to be increased to 4 times
the Ziegler-Nichols value to get reasonable responses (see
Figure 7b where the disturbance is a decrease in fresh feed
composition from 0.9 to 0.8).
Then the entire process was simulated: reactor tem-
perature is controlled by jacket cooling and stripper
bottoms composition is controlled by vapor boilup. The
controller settings used for the individual units were used
for the coupled system. Figure 8 gives the responses for
f0.1 changes in fresh feed composition. The coupled
system is stable, and temperature and product composition
control is fairly good.
Figures 9 and 10 give results for system 2 (VR= 1OOO).
The tuning of the reactor temperature controller was found
to be much more difficult, as predicted by the linear model.
The temperature controller reset time had to be made
very large to get stable, nonoscillatory responses. Figure
9a shows that increasing the Ziegler-Nichols reset time
by a factor of 8still gives oscillatory responses for increases
in fresh feed composition. The reset had to be increased
to 16 times the Ziegler-Nichols value to get reasonable
responses (see Figure 9b).
Comparing Figure 7b for the large reactor with Figure
9b for the small reactor, we can see that the maximum
error in reactor temperature for the same disturbance in
fresh feed composition is smaller for the bigger reactor
(0.23OF) than for the smaller reactor (2.3O F ) . This is due
to the smaller gain (3.0 versus 1.95) and the larger reset
time (1.38 versus 2.8 h). Notice also that the disturbance
affects the larger reactor (system 1, Figure 7b) somewhat
more quickly than it does the smaller reactor (system 2,
Figure Qa). This is contrary to what we would expect.
However, remember that the steady-state compositions
in the reactors are different: 0.25for the large reactor and
0.625 for the small reactor. Therefore the difference
between the fresh feed composition (20 = 0.9 at steady
state) and the reactor composition is larger for the larger
reactor. This may cause changes in feed composition to
have more of an effect on reactor temperature (i.e., 0.1
change in zo is a larger fraction of 0.25 than it is of 0.625).
Clearly the performance of the reactor temperature
controller could have been improved by the use of some
derivative action (Cheung and Luyben, 1979). We limited
ourselves to using only proportional-integral controllers
in this study to get a simpler comparison between the two
processes.
Finally the entire coupled process of system 2 was
simulated. As shown in Figure 10, this process is stable
Ind. Eng. Chem. Res., Vol. 32,No. 3, 1993 485
for decreases in fresh feed composition, but it goes unstable
for even a fairly small (+0.05) increase in fresh feed
composition.
Thus system 2 with the smaller reactor and bigger
column is less robust and less controllable than the
configuration with the larger reactor.
Conclusion
The steady-state economics and the controllability of
several alternative process designs for a reactor/stripper
process have been compared. Each design has identical
feed and product streams, but the sizes of the reactors and
columns are different.
The optimum steady-state economic design consists of
a fairly large reactor and a moderately sized stripping
column. This design is competitive with designs which
have only CSTRs in series with no distillation column
and no recycle.
The dynamics of the column stay about the same as its
diameter changes, but the dynamics of the reactor slow
down as the size of the reactor increases. The gain around
the recycle loop increases as the reactor size is decreased.
All these competing effects result in a process in which
the dynamics of the various alternative designs are quite
similar.
Temperature control in the reactor is shown to improve
as the size of the reactor increases, which isjust the opposite
of the normal situation when one scales up a reactor. Thus
the larger reactor system is not only economically more
attractive, but it is also more controllable.
Nomenclature
A = matrix in state variable representation
Ac = area of condenser (ft2)
AF = cross-sectional area of flow (ft2)
aij = element of A matrix
AH = heat-transfer area in jacketed reactor (ft2)
B = bottoms flow rate (lb-mol/h)
B, = parameters in open-loop transfer function
CJ = heat capacity of jacket coolant (Btu/(lb OF))
cp = heat capacity of process liquids (Btu/(lb OF))
D = distillate flow rate (lb-mol/h)
DC = diameter of column (ft)
D R = diameter of reactor (ft)
F = column feed flow rate = reactor effluent (lb-mol/h)
Fo = fresh feed rate to reactor (lb-mol/h)
h, = weir height (ft)
how= height of liquid over weir (ft)
k = specific reaction rate (hr')
ko = preexponential factor in Arrhenius equation (h-1)
K, = feedback controller gain
L = length of tubular reactor (ft)
LC = length of column (ft)
LR = length of reactor (ft)
M = molecular weight (lb/lb-mol)
M B = holdup in column base (lb-mol)
M D = holdup in overhead surge tank (lb-mol)
N = number of CSTR's
NT = total number of trays in column
R = perfect gas law constant (appropriate units)
R = reflux flow rate (lb-mol/h)
R = reaction rate (lb-mol of A/(h ft3))
s = Laplace transform variable
TJ= jacket temperature (OF)
TJO = inlet coolant temperature (OF)
TO= fresh feed temperature (OF)
TR= reactor temperature (OF)
U = overall heat-transfer coefficient (Btu/(h ft2 OF))
u = velocity (ft/h)
486 Ind. Eng. Chem. Res., Vol. 32, No.3, 1993
VC = velocity in column (ft/s)
V = vapor boilup (lb-mol/h)
VJ = jacket volume (ft3)
V,,= holdup in nth CSTR (lb-mol)
VPFR= holdup of plug-flow reactor
VR = reactor holdup (lb-mol)
V,,, = total holdup of N CSTRs in series
WL= weir length (ft)
XB = bottoms composition (mole fraction A)
XD = distillate composition (mole fraction A)
z = column feed composition = reactor composition (mole
fraction A)
ZL = composition leaving tubular reactor (mole fraction A)
ZN = composition leaving last stage of series of CSTRs (mole
fraction A)
zo = fresh feed composition to reactor (mole fraction A)
0 = liquid hydraulic time constant (min)
X = heat of reaction (Btu/lb-mol)
TI = reset time of controller
p = density of reaction liquids (lb/ft3)
p~ = density of coolant (lb/ft3)
pv = vapor density (lb/ft3)
Literature Cited
Cheung,T. F.;Luyben, W. L. PD Control ImprovesReactor Stability.
Hydrocarbon Process. 1979, 215-218.
Devia, N.; Luyben, W.L. Reactors: Size versus Stability. Hydro-
carbon Process. 1978,119-122.
Douglas, J. M. Conceptual Design of Chemical Proceeses;
McGraw-Hill: New York, 1988; appendix.
Luyben, W. L. Process Modeling, SimuZation and Control For
Chemical Engineers, 2nd ed.; McGraw-Hill: New York, 1990; 129.
Luyben, W. L. Dynamiceand Control of Recycle Systems. 1. Simple
Open-Loop and Closed-loop Systems. Znd. Eng. Chem.Res. 1993,
preceding paper in this issue.
Received for review January 23, 1992
Revised manuscript received May 5, 1992
Accepted November 28,1992