8.

10
Chemical Reactors: Control and Optimization
F. G. SHINSKEY
(2005)
INTRODUCTION It should be noted that the coverage of this section and of the pre
vious one (8.9) are similar. If the reader is interested in a detailed and compl
ete treatment of the subject of chemical reactor control, it is recommended to a
lso read that section. On the other hand, if you have little time and you are an
experienced process control engineer who is familiar with the basics of chemica
l reactor control and want only to quickly refresh your memory about the most im
portant aspects of this area of process control and optimization, you should rea
d this section.
produce heat, much like the combustion of fuel. That heat must be removed as it
is produced, to maintain a constant temperature. If it is not, the temperature w
ill rise, which increases the rate of reaction, creating the possibility of a th
ermal runaway—in the extreme case resulting in a re or explosion. For the protectio
n of life, equipment, and the environment, such an outcome must be prevented. Bu
t less extreme conditions can damage product, deactivate catalyst, and foul heat
transfer surfaces—these also must be avoided. Maximizing Yield Operating a reactor
safely is not enough. It must also operate pro tably. This means maximizing yield
and minimizing operating costs. Yield is the fractional or percentage salable p
roduct that can be made from a given feedstock. Ideally, 100% of the ethylene fe
d to an oxidation reactor should be converted to ethylene oxide, but some is inv
ariably converted to carbon dioxide and other by-products of the reaction. If th
e reaction that produced some of the by-products is reversible, then they may be
recycled to reduce their further production and do not represent a loss in yiel
d. But some, like carbon dioxide, will have to be purged from the system and the
refore lost. To maximize yield, it may be necessary to limit conversion, which r
epresents the fraction of the feed that is converted to product on one pass thro
ugh the reactor. The unconverted feed must be separated from the product and rec
ycled. Operating costs increase with recycling, as it increases the loading on p
umps, compressors, and separation equipment. So, there must be a balance struck
between yield and conversion, re ecting the cost of lost yield and the operation o
f the equipment carrying the recycle, along with their constraints. The Catalyst
Most reactions are catalyzed, as well. A catalyst is a substance that does not
change during the reaction, but increases the reaction rate, allowing a higher y
ield, or higher conversion, or production to take place under less severe condit
ions of pressure and temperature. Reactions take place on the catalyst surface,
so that a large, active surface is essential. Most catalysts are porous, and por
e size can affect their 1697
OBJECTIVES The optimization of batch and continuous chemical reactors has many p
otential bene ts, including increase in productivity and improvement in safety, pr
oduct quality, and batch-tobatch uniformity. Such results are the consequences o
f many individual control loops and strategies. These loops will maintain materi
al balances, hold optimum temperature and pressure, and control concentrations,
while providing safe operating conditions. All elements of the overall chemical
reactor control system are discussed in this chapter. The reactor is the heart o
f most chemical production facilities. Other operations such as heat transfer, s
eparation, recycle, puri cation, and drying all are dependent on the reactor’s produ
ction and demands for services. Most reactors are “base-loaded,” i.e., their feed ra
te is xed, as determined by market demand and constraints. All other operations a
re then dependent on the reactor’s production rate. Therefore, it is essential tha
t the reactor operation be stable. If the reactor is run ef ciently, with steady p
roduct of uniform quality, all other operations may proceed smoothly and the pla
nt should be pro table. But if temperature control is erratic, for example, produc
t quality will suffer and all downstream plant operations will be upset. Safety
Several objectives must be met for successful reactor operation. The rst is alway
s safety. Most chemical reactions in production facilities are exothermic, which
means that they
© 2006 by Béla Lipták
1698
Control and Optimization of Unit Operations
selectivity for particular reactions. Many are inert ceramics coated with active
metals like platinum. The catalyst may be xed, such as shaped material packed or
spaced in a bed. Some catalysts are very ne and uidized by the upward ow of the re
action mass, as in a uid cat-cracker, which breaks heavy oil into smaller gasolin
e molecules. Some catalysts travel with the reaction mass as a slurry, or may ev
en be dissolved. Polymerization reactions usually require an initiator, which is
used to create an active site on a molecule, attracting other molecules into a
chain reaction. Catalysts are costly, many having active surfaces of noble metal
s. While they are not changed by the reaction, they can be degraded, and even po
isoned. Degradation consists of clogging the pores or coating the active surface
with coke or tar, for example. The uid cat-cracker contains an auxiliary unit fo
r burning the coke continuously off the catalyst, thereby regenerating its surfa
ce and recirculating it to the reactor. A xed catalyst bed must be regenerated of
f-line, which takes time and interferes with production. Some catalysts are pois
oned by heavy metals and sulfur, which may require their eventual replacement. C
lose control of operating conditions can minimize the cost of catalyst loss and
regeneration.
treated as rst-order. The reason for this is that favorable conditions for most r
eactions have one reactant in dominant concentration and the other in a much sma
ller concentration, the smaller thereby controlling the reaction. At equilibrium
, forward and reverse reaction rates are equal, the equilibrium constant being t
heir ratio. Batch Reactions In a rst-order reaction, the rate of conversion of re
actant of concentration x varies linearly with that concentration and rate coef ci
ent k, expressed in units of inverse time: − dx = kx dt 8.10(2)
In a batch reactor, that concentration will change with time t, as the rate equa
tion is integrated between initial and nal conditions:
∫
dx = x0 x
x
∫ −kdt
t 0
where x0 is the initial concentration and x is its value at time t. The solution
of the above is x = x0 e− kt where e is 2.718, the base of natural logarithms. Th
e fractional conversion of feed to product in a batch reactor at time t is y= x0
− x = 1 − e− kt x0 8.10(3)
CHEMICAL REACTION KINETICS The conduct of chemical reactions follows an exact sc
ience called chemical kinetics, which describes the relationships that produce r
eactions. The rst relationship is one of equilibrium. Most chemical reactions are
reversible and terminate in a condition of equilibrium, where forward and rever
se rates are equal. As an example, consider the partial oxidation of ethylene to
ethylene oxide: C2H 4 +
1 2
O 2 ⇔ C2H4O + heat
Continuous Plug Flow Reactors Continuous reactors operate with a uniform ow of fe
ed in and product out. The average time spent by a molecule in a continuous reac
tor is called its residence time and is simply reactor volume V divided by total
 volumetric ow F, i.e., V/F. In an ideal plug ow reactor, ingredients ow from inlet
to outlet without any longitudinal mixing. In this way, the ideal plug ow reacto
r resembles a batch reactor, in that the fractional distance a molecule travels
from inlet to outlet is also its fractional residence time in the reactor. As a
result, fractional conversion at the exit of a plug ow reactor is y = 1 − e − kV / F
8.10(4)
This equilibrium is de ned by an equilibrium constant K, de ned as K= [C2H 4O] [C2H
4 ][O 2 ]1/2 8.10(1)
where the bracketed terms are molar concentrations of the components. To shift t
he equilibrium to the right, i.e., to make more product, the concentration of ei
ther ethylene or oxygen may be increased, or the concentration of ethylene oxide
reduced. The reactant concentrations can be increased by increasing the operati
ng pressure, as they are gases, and the product concentration can be reduced by
condensing it to a liquid. This forward reaction is of 1.5 order, in that it dep
ends on the rst power of ethylene concentration and 0.5 power of oxygen concentra
tion; the reverse reaction is rst order. Many reactions are second order, but mos
t ultimately can be
At kV/F = 1, fractional conversion is 0.632. Actual plug ow reactors may have some
longitudinal mixing, but the intent of the design is to avoid it, as back mixin
g reduces conversion.
© 2006 by Béla Lipták
8.10 Chemical Reactors: Control and Optimization
1699
Continuous Back Mixed Reactors If a vessel is thoroughly mixed, reactant concent
ration is the same everywhere, including in the product stream withdrawn. The ra
te of reaction is therefore determined by the outlet concentration, which is low
er than that at any point within the plug ow reactor. Production rate in a perfec
tly backmixed reactor is its volume times the rate of loss of reactant concentra
tion, which is also its feed rate multiplied by the loss in reactant concentrati
on from inlet to outlet. − V dx = kVx = F ( x 0 − x ) dt
4
3 Rate coe . k/k70
2
1
Solving for exit concentration gives x= x0 1 + kV /F
0 50
60
70 80 Temperature, °C
90
100
FIG. 8.10b Reaction increases exponentially with temperature.
And fractional conversion is y = 1− 1 1 + kV /F 8.10(5) where a and E are constant
s particular to the reaction, R is the universal gas constant, and Tr is the abs
olute temperature of the reaction mass. The nature of the relationship is that r
eaction rates tend to double with each increase in temperature of about 10°C (18°F).
The rate coef cient for a typical reaction normalized to what it would be at 70°C i
s plotted vs. temperature in Figure 8.10b. This effect is so dominant that preci
se control of temperature is the most important aspect of operating reactors of
any type. For all reactors, then, conversion is a function of both temperature a
nd time. Because conversion in plug ow and back mixed reactors differs as a funct
ion of residence time, they can also be expected to differ as a function of temp
erature. This is shown in Figure 8.10c, where both reactors are operated at kV/
F = 1 at 70°C. Other values of V/ F produce
For the same residence time and rate coef cient, conversion will always be less in
the back mixed than the plug ow reactor, as shown in Figure 8.10a. At kV/F = 1,
fractional conversion is 0.5. However, a back mixed reactor is lagdominant and i
s, therefore, easier to control than the plug ow reactor, which is dominated by de
ad time. Effect of Temperature Temperature has a profound effect on reaction rat
e, through an exponential relationship to the rate coef cient: k = ae − E/RTr 8.10(6
)
1.00 Plug ow 0.80 Fractional conversion Fractional conversion
1.00 Plug ow 0.80 Backmixed
0.60 Backmixed 0.40
0.60
0.40
0.20
0.20
0.00
0
1
2 kV/F
3
4
5
0.00 50
60
70
80
90
100
Temperature, °C
FIG. 8.10a Plug ow reactors are used for their higher conversion.
FIG. 8.10c Conversion in a plug ow reactor is more temperature sensitive.
© 2006 by Béla Lipták
1700
Control and Optimization of Unit Operations
parallel curves. Because Figure 8.10b shows a doubling of k for about 9°C temperat
ure rise, parallel curves for doubling or halving of V/F will be spaced 9°C apart.
For example, the plug ow reactor will produce 63.2% conversion at 61°C and half th
e feed rate, and also at 79°C with twice the feed rate.
In Figure 8.10d, the conversion vs. temperature plot of the plug ow reactor is eq
uated to heat release by multiplying by the reactant feed rate F and the heat of
reaction Hr: Qr = yx 0 FHr 8.10(7)
Here, it is normalized to 100% conversion at nominal feed conditions. On the sam
e plot is a line representing heat transfer from the reactor to the coolant: EXO
THERMIC REACTOR STABILITY Qt = UA∆Tm Endothermic reactions are inherently self reg
ulating, because heat must be continuously applied for them to proceed—remove the
heat and they die. Ordinary heat transfer is also self regulating. Increasing th
e temperature of a hot uid entering a heat exchanger will increase the rate of he
at transferred to the coolant without any change in its ow. Exothermic reactors,
by contrast, can be steady state unstable—depending on their heat transfer capabil
ity—and even uncontrollable. Moreover, some have been observed to change between s
table and uncontrollable states. 8.10(8)
Steady State Stability Consider an exothermic reactor operating at steadystate—con
stant temperature—in the open loop. A small upset causes its temperature to rise,
which increases the rate of reaction, thereby releasing more heat. The increase
in heat release then raises the temperature further—positive feedback. At the same
time, however, the rising temperature increases the rate of heat transfer to th
e coolant—negative feedback. Which has the stronger in uence will determine the stea
dystate stability of the reactor. Figure 8.10d compares the heatrelease and heat
transfer capabilities of a stable exothermic reactor.
where U and A are the heat transfer coef cient and area respectively, and ∆Tm is the
mean temperature difference between the reactor and the coolant. (If the reacto
r were a well mixed vessel and the coolant rapidly circulated past its heat tran
sfer surface, ∆Tm could be a simple difference between reactor and coolant tempera
tures. In practice, this is not likely to be the case, but a mean temperature di
fference formula can be tted to whatever heat transfer con guration exists.) Heat ow
in and out of the reactor is set equal at 70°C, the control point. If the slope o
f the heat transfer line exceeds that of the heat release curve at the operating
point, as is the case in Figure 8.10d, the reactor is stable. A rise in tempera
ture transfers more heat than is released, resulting in a stabilizing of tempera
ture. The key to achieving this stability lies therefore in designing the heat t
ransfer system to have suf cient area for the expected heat load. Area cannot be t
raded for colder coolant—they are not equivalent as a guarantor of reactor stabili
ty. Note that ∆Tm for the stable reactor in this example is only 16°C. The slope of
the heat release curve can be determined by differentiating the conversion curve
. For the plug ow reactor, the slope is dy E =− (1 − y) ln(1 − y) dT RT 2 8.10(9)
1.00
Heat transfer, Qt UA
And for the back mixed reactor, it is dy E (1 − y) y = dT RT 2
Stable intersection
0.80 Fractional heat ow
8.10(10)
0.60
0.40
Heat release, Qr
0.20 ∆Tm
40 60 Temperature, °C 80 100
0.00 20
FIG. 8.10d Steady state stability is provided by suf cient heat transfer surface.
The maximum slopes of both curves occur where kV/F = 1. At that point, the slope
of the plug ow curve is exactly twice that of the back mixed curve. The plug ow r
eactor is therefore twice as demanding of the heat transfer surface to remain st
able. For pilot scale reactors, jacket cooling may be suf cient for stability, esp
ecially with stirred tanks. However, there is a scale up problem: Production inc
reases with volume, which increases with the cube of diameter, whereas jacket ar
ea increases only with its square. At some scale, jacket cooling is insuf cient, a
nd additional heat transfer surface is required. This can be provided by interna
l coils, by a re ux condenser, or by an external exchanger through which reactor c
ontents
© 2006 by Béla Lipták
8.10 Chemical Reactors: Control and Optimization
1701
are circulated. Most full scale plug ow reactors have external exchangers. Unstab
le but Controllable The heat transfer surface in Figure 8.10e has been reduced b
y half from that in Figure 8.10d—increasing ∆Tm to 32°C— and now the reactor is steady s
tate unstable. Any rise in temperature above the control point will release more
heat than is transferred, augmenting the rise until the upper stable intersecti
on of the line and the curve is reached. Conversely, any drop in temperature wil
l cause heat release to fall more than heat transfer until the lower stable inte
rsection is reached. In the open loop, the temperature will seek one of the two
stable intersections, running away from the desired control point. A simple mode
l of the unstable reactor is the inverted pendulum—a stick balanced vertically in
the hand, for example. Its balanced position is an unstable steady state, in tha
t the slightest disturbance will cause it to accelerate away from that state. Ho
wever, it is possible for a person, or even a trained seal, to balance a stick f
or a time. To be successful in doing so reveals the control effort required. As
soon as the stick begins to deviate from a true vertical position, the hand must
move farther in the same direction to restore balance. In other words, the gain
of the (human) controller must exceed unity to succeed in controlling a steady
state unstable process. That is, the process is controllable if the gain of the
controller can be set high enough. Reference 1 identi es the limit of controllabil
ity of the steady state unstable process as when the dead time in the loop appro
aches its time constant. In other words, a lag dominant unstable reactor is cont
rollable, whereas a dead time dominant one is not.
A stirred tank is more likely to be stable than a plug ow reactor, because the ma
ximum slope of its heat release curve is only half as great. But even if unstabl
e, it is controllable because it is dominated by the large lag of its well mixed
contents. To be sure, its control loops need to be properly structured, a subje
ct to be covered. But the proportional gain of a typical temperature controller
on a back mixed reactor is in the order of 3:5, which is entirely capable of con
trolling it even if it is unstable. Uncontrollable Reactors Plug ow reactors are
dominated by dead time and, if unstable, are therefore also uncontrollable. Duri
ng start up, as temperature is raised, they may behave well because reaction rat
e is low at the lower temperatures. But as the desired operating temperature is
approached, there is a tendency for the reactor to speed right past it. Even a c
arefully structured control system will not be able to stop it in time. As the t
emperature rises further, the slope of the heat release curve moderates, and the
manipulated cooling begins to bring the temperature down. But it then goes past
the control point in a downward direction, until the control system arrests the
fall. The result is a limit cycling of the temperature—at a given set of conditio
ns, the cycle will be of uniform amplitude and period and nonsinusoidal, possibl
y sawtooth in shape. Its amplitude will depend on the difference between the slo
pes of the heat release curve and the heat transfer line at the control point. O
ne of the characteristics of a limit cycle is its resistance to change. Variatio
ns in the tuning of the temperature controller(s) tend to have little effect—its p
eriod and amplitude may change slightly, but the cycle persists. The only way to
stop the temperature from cycling is to lower the production rate, which reduce
s the slope of the heat release curve, or to improve the heat transfer condition
. A reactor that was once controllable and is no longer, may be suffering from a
fouled heat transfer surface—cleaning has been observed to restore 2 stability.
1.00
Heat release, Qr
0.80 Fractional heat ow Unstable intersection Heat transfer, Qt
TEMPERATURE CONTROL
UA
0.60
0.40
Unless reactor temperature can be controlled precisely, product quality will var
y. This is especially true of polymerization reactions, where molecular weight,
density, and viscosity are all functions of reactor temperature. Variations as s
mall as 1°C may produce unsatisfactory product. Measuring Temperature
0.20 ∆Tm 40 60 Temperature, °C 80 100
0.00 20
FIG. 8.10e Insuf cient heat transfer surface makes a reactor unstable.
Resistance bulb sensors are universally used to measure reactor temperature, as
they are more accurate than thermocouples. In a well stirred tank, the measureme
nt may be made anywhere within the mixed zone, velocity being necessary for dyna
mic response and to avoid fouling. The sensor must
© 2006 by Béla Lipták
1702
Control and Optimization of Unit Operations
be protected against corrosion and erosion, requiring installation in a well in
most cases—then close thermal contact of the sensor with the well is necessary to
maximize dynamic response. The temperature will vary along the length of a plug o
w reactor, due to the variable rate of reaction within. In an adiabatic reactor
(no heat removal), temperature will rise throughout. For one that is cooled by a
jacket, a peak temperature will usually be reached along the rst third of travel
from the inlet. In a packed bed, that hot spot may move downstream gradually as
catalyst decays. There may also be a radial gradient. Therefore, most plug ow re
actors have multiple sensors, one of which may be manually selected as the contr
ol point. Selection may even be automatic, as the highest of several measurement
s. An average of two or more temperatures may also be used for control, or the m
edian of three or ve measurements, as being more representative of the reaction m
ass. However, if the combined sensors are spaced too far apart, a difference in
their dynamic responses may require the controller settings to be compromised. R
eactor pressure, or rather its change, has been used to predict reactor temperat
ure change. The rationale is that pressure changes directly with temperature, bu
t faster, because it lacks the lags in the bulb and thermowell. However, the ste
ady state value of pressure is misleading, because it is too easily in uenced by v
ariations in composition, particularly the presence of noncondensible gas. There
fore, the predicted temperature is best calculated as the measured temperature p
lus a gain factor times the rate of change of pressure. This method was tried by
the author on an uncontrollable reactor, but unsuccessfully. The predicted temp
erature led the measured temperature when falling, but not when rising, and so w
as ineffective in stabilizing the loop. It was only a half predictor. Manipulati
ng Coolant Flow It is possible to control the temperature of a continuous reacto
r by manipulating feed rate at a constant coolant ow—but it is also very dangerous!
If the reaction mass is cold, reaction rate and heat release will be low, and a
dding more feed will not raise the temperature. Under these conditions, it is po
ssible to overfeed a reactor, so that when reaction temperature is reached, ther
e will not be enough cooling available to remove the heat released, resulting in
a thermal runaway. But even beginning at a steady state, manipulating feed rate
is poor practice. A change in feed rate must rst change reactant concentration t
hrough the concentration lag, and then change temperature through the thermal la
g. These are two large lags in 3 series. The concentration time constant is
and the thermal time constant is
τT =
MC UA
8.10(12)






where M and C are he reac ion mass and speci
c hea , respec ively. There is
also

he problem

of

inverse
response:
Increasing he
ow of
cold
feed will cause he e

mpera

ure of
he reac

ion mass
o
fall before he effec of he increased
concen
ra
ion of he
reac an causes i o rise. This reversal could las for several

minu es,
resul
ing in very
unfavorable
dynamic response.


This scheme is
no reco
mmended o con rol an
exo hermic

reac

or. Manipula
ing he
ow of coolan

is
neces
sary
for successful

empera

ure con rol.

If
i can


be injec ed direc ly in o
he
reac

ion mass, hen
he lags

associa

ed

wi h hea
ransfer
are avoided.

If no ,
hen a
boiling coolan is he bes hea sink, as

i is iso
hermal, bu such a ui
d may no
be available

or

safe or economical in
he empera ure range desired.
B
oiling par
of he reac ion mass
and condensing i
in a re

ux condenser
is ano her
self-regula

ing
means of hea

removal. However,



he mos common coolan used fo
r exo hermic reac ors is wa er owing pas a hea - ransfer surface. Tempering Loop







s Manipula

ing coolan ow in a single
pass by
he
hea - ransfer

surface
is no go
od prac
ice. Consider, for example, he jacke ed ba ch reac

or shown in

he prev
ious
sec ion in Figure
8.9f, whose

cooling

load
varies
wi h ime. When
li le co

olan is required,
i s low ow
hrough he jacke
crea es a

long dead ime—jacke

vo
lume

divided
by he low
ow. A
higher demands,

his dead
ime decreases. The emp
era ure con roller canno be op imally uned for all of he possible owing
0.50


0.40 Frac ional ow Linear valve
0.30
0.20
e = 1.333
1.125




0.10 Eq.-pc . valve 0.40 0.60 Frac ional hea load
0.00 0.00
0.20
0.80
1.00
τx =
1 k + F/V
8.10(11)






FIG. 8.10f The nonlinear

rela ionship of coolan ow o hea load is compensa ed b
y an equal-percen age valve.


© 2006 by Béla Lip ák





8.10 Chemical Reac ors: Con rol and Op imiza ion
1703







condi ions,
and mus favor
he
wors case
of low ow,
giving less-effec
ive con ro
l during

he
major por ion
of he reac

ion. This
dead ime varying
wi h
ow also c
rea es
a limi
cycling
condi ion in he
empera ure loop. A rising
empera
ure ca
uses

he con roller

o

increase coolan
ow, which
decreases

jacke
dead ime
and
hereby

drops empera
ure quickly. Bu a falling
empera ure causes
he
con roll
er o decrease


coolan ow, which increases




dead ime and delays empera

ure respo
nse.

A empera
ure cycle
is crea ed
ha
hen has sharp peaks and a valleys, and


uning may

no be able

o remove i . As
he cooling

load increases,
higher ow ra
es end

o decrease
he

period and ampli ude of he cycle,
and i may
even disa
ppear a
he highes ra
es. This
behavior can be
avoided en irely
by he use of
a circula ing pump
on he
coolan
, as
shown in
he previous

sec
ion
in
Figure
8.
9g. Adding

coolan
lowers he empera


ure of

he jacke


con en

s
wi hou affec i
ng

in ernal

veloci
ies. The coolan



en ers
he jacke a cons an ow and variable

empera

ure, ins


ead of cons
an empera
ure and
variable ow. Valve Selec ion Th
e
ra e of
hea
ransfer is
no linear wi h coolan
ow. Consider,

for
example, a b
a ch reac or
con
rolled a
70°C
by cooling
wa er supplied a 20°C.

If here
is no co
oling

load,

hen
he coolan

in he circula
ing loop will be

a he same

empera
ure as
he
reac
or, wi h
no fresh

coolan added. Opening he coolan


valve will
add

wa er
a he supply empera
ure of

20°C while



displacing
wa er a

70°C—each incre
men of
wa er carries
away
hea propor

ional


o ha

50°C of empera ure rise. How
ever, a full hea load,
he coolan exi empera
ure will be much
lower,
possib

ly
as low
as 30°C. Now he same change in wa er added
removes hea

propor

ional


o
a empera ure rise
of only 10°C. The gain in conver ing
coolan ow
o hea
ransfe
r varies by 5:1 be
ween zero
and
full
load.
When a logari

hmic-mean

empera ure

difference


is used
o
rela e hea ransfer
o coolan
empera ures
in a jacke ed
s irred ank, he ra
io of fresh
coolan mass ow WC o circula ing ow W is derive
d as a hyperbolic func ion of hea load: WC 1 = W WCC (Tr − TC )/Qt − 1/(ε − 1) whr CC
is th spci c hat of th coolant, and
givn stady stat; if U, A, and W ar constant,  is also constant. This rlati
onship is plottd for th ractor dscribd abov as th numbrd curvs in Figu
r 8.10f, for a rang of coolant ow up to half th circulatd ow, and two valus o
f ε . ε = 1.125 corrsponds to jackt inlt and outlt tmpraturs of 25 and 30°C, an
d 1.333 corrsponds to jackt inlt and outlt tmpraturs of 30 and 40°C at full
hat load. Th lattr cas rprsnts a valu of UA/WCC that is 23% highr. Lin
ar and qual prcntag valv charactristics ar plottd for comparison, with
th rangability of th qual prcntag valv bing 50:1. Th lattr matchs th
 rquird charactristic for  = 1.125 almost prfctly. Matching th procss c
urv by th valv mans that hat ow will b linar with valv strok, and thrf
or so should tmpratur. Howvr, th curv for ε = 1.333 shows that such a clos
 match of valv charactristic to th hat transfr procss is not always attai
nd. In fact, Rfrnc 2 dscribs a gasphas plug ow ractor coold by rcircul
ation of gas through a countr ow coolr, whr th procss curv is closr to tha
t of ε = 1.333. Th qual prcntag valv is still th prfrrd choic, bcaus
its variabl gain compnsats th loop for th ffct of changs in coolant supp
ly tmpratur and ractor tmpratur st point. Equation 8.10(13) shows that t
h rlationship btwn coolant ow WC and hat transfr Qt is affctd by th tm
pratur diffrnc (Tr − TC) as a gain factor. A lowr tmpratur diffrnc dc
rass th procss gain dQt /dWC and rquirs a highr coolant ow for th sam l
vl of hat transfr. But th gain of an qual prcntag valv incrass dirct
ly with ow dlivrd, thrby compnsating for th procss gain chang. Cascad C
ontrol Cascad control is rcommndd for critical variabls that ar dif cult to
control and subjct to svral disturbancs, and th tmpratur of an xothrmi
c ractor crtainly ts this cas. Cascad control is possibl whnvr th procs
s can b dividd into sparat dynamic pics by th masurmnt of an appropria
t scondary variabl. In th cas of an xothrmic ractor, th two pics ar
th ractor itslf and its cooling systm, and th scondary variabl is th tm
pratur of th coolant as it ntrs or xits th hat transfr surfac. As show
n in Figur 8.9l in th prvious sction, th scondary or slav loop is closd
by a tmpratur controllr whos st point is adjustd by th primary or mastr
tmpratur controllr. Th primary controllr ss th scondary closd loop a
nd th ractor in sris. It is imprativ that th scondary loop b fastr tha
n th primary, and yt it should contain signi cant dynamics and disturbanc varia
bls to b ffctiv. Th priods of th two loops should diffr by a factor of
4 or mor to avoid rsonanc. For th stirrd tank, controlling
8.10(13)
ε = UA/WCc =
Tr − TC1 . Tr − TC 2
Tmpraturs TC1 and TC2 ar thos of th circulating coolant ntring and lavi
ng th jackt, rspctivly, at any
© 2006 by Béla Lipták
1704
Control and Optimization of Unit Oprations
th jackt outlt tmpratur closs th loop around th jackt, thus spding t
h rspons of th primary loop and prvnting cooling systm disturbancs from
upstting th ractor. But plug ow ractors ar fastr, and controlling coolant i
nlt tmpratur as th scondary variabl givs th 2 two loops bttr dynamic
sparation. Batch xothrmic ractors rquir hating as wll as cooling, both f
or startup purposs and to cur th product and driv off solvnt at th nd of
th raction. Th scondary controllr thrfor opns cooling and hating valv
s in split rang. Th split ranging is usually accomplishd with valv position
rs as shown in th prvious sction in Figur 8.9m, calibratd to opn th hati
ng valv from 50 to 100% of th controllr output, and th cooling valv from 50
to 0% (th cooling valv fails opn). Bcaus of th dif culty of controlling xo
thrmic ractors and of th critical rol playd by valvs, smart digital positi
on2 rs ar rcommndd. Thy liminat stiction and dad band, and maximiz rs
pons spd. Each valv positionr closs a loop around its valv and thrfor
constituts a scond lvl of cascad control, as was shown in Figur 8.9m. With
th provision of hating, additional prcautions ar advisd. A high limit shou
ld b st on coolant tmpratur as shown, and for glass lind ractors, th tm
pratur diffrnc btwn th ractor and jackt should b limitd to avoid x
cssiv thrmal strss. Som polymrization ractors ar also ttd with intrlock
s that prvnt hating aftr th initiator has bn introducd into th ractor.
In som polyvinyl chlorid (PVC) procsss, th raction mass is hatd to abou
t 70°C without any initiator, and allowd to stabiliz at that tmpratur without
any raction taking plac. Thn initiator is injctd, and th hating valv lo
ckd out for safty. A rising tmpratur thn signals th bginning of th rac
tion. Th tmpratur controllrs rspond by opning th cooling valv, and aft
r a pak dviation of 2–3°C is rachd, th tmpratur rturns toward st point, an
d in fact, ovrshoots. At this point, th scondary controllr output riss abov
 50%, but th hating valv is lockd out, so that th scondary loop is opn.
Th tmpratur slowly bgins to ris again, but during th opn loop condition,
th primary controllr has bn winding up—it has raisd th st point of th sc
ondary controllr, which was unabl to rspond bcaus of th intrlock. As a r
sult, th scondary st point is too high, causing th primary tmpratur to ri
s past its st point again. Th scnario thn rpats, sttling into a limit cy
cl of 1°C or thrabouts, which dos not rspond to tuning changs and advrsly a
ffcts product quality. Th limit cycling is avoidd through th us of xtrnal
rst (ER) from th scondary tmpratur masurmnt to th intgral fdback
port of th primary controllr, as was shown in Figur 8.9m. As long as th sco
ndary tmpratur is at its st point, th ER signal and th primary controllr
output ar idntical, allowing th primary controllr to intgrat normally. But
whn th scondary is prvntd from controlling (by th lockd out valv, or a
ny othr obstacl
such as a st point limit), primary intgration stops. Th scondary controllr
must hav intgral action for this stratgy to b succssful, bcaus any scond
ary loop offst will caus a proportional primary loop offst. This fatur succ
ssfully liminatd limit cycling in th PVC ractors with th lockd out hatin
g valv. It was also usd on batch ractors for latx paint and found to b vry
robust. In trials, th primary tim constant of th ractor was varid ovr a r
ang of 4:1 by changing batch siz and hattransfr ara, ach by a factor of 2:
1—yt th control systm did not rquir any tuning adjustmnts. Th rat of tmp
ratur chang was obviously affctd by ths procss conditions, but th dynami
c rspons of th scondary tmpratur altrd th rat of intgration of th p
rimary controllr suf cintly to accommodat thm. Th scondary controllr in a c
ascad systm must b tund bfor th primary, as it is a part of th primary l
oop. It should b tund for optimum load rspons—not st point rspons—rgardlss
of having to rspond to its st point. Th st point changs that it ss ar gr
adual, not stpwis, and thrfor affct th loop dynamically much th sam as
load changs do. For th sam rason, ltring should nvr b applid to th sco
ndary st point, and th st point should rciv full proportional action (som
controllrs hav th option of liminating proportional action to th st point
). Othrwis, th scondary control action will ffctivly lag th primary outp
ut, ngating th dynamic bn t of cascad control. Batch Ractor Start Up Th pro
blm of intgral windup is particularly troublsom during start up of a batch r
actor. Th goal is to rais ractor tmpratur from its initial valu (possibl
y ambint) to an lvatd st point (whr th raction procds) as quickly as
possibl, but without ovrshoot. In this situation, a convntional PID controll
r will invariably ovrshoot st point. Following a larg st point chang, its o
utput will saturat, and during th ris tim, th intgral mod will also satur
at. Th rsult is that th output rmains at its high limit until th st point
is crossd. If th nw stady stat rquirs no hating or cooling in absnc o
f a raction, th nw stadystat valu of th controllr output should b 50%;
if an xothrmic raction has bgun, th nw output should b vn lowr. As a r
sult, thr will b an unaccptabl tmpratur ovrshoot, such as that shown i
n th top curv in Figur 8.10g. Intgral windup in th start up of a batch proc
ss can b ovrcom by th circuit, which was shown in th prvious sction in F
igur 8.9x. Th output of th controllr is compard with a high limit stting,
and thir diffrnc snt to a high gain ampli r, whos function is to control th
 controllr output at that limit by manipulating th intgral fdback trm. If
th controllr output is blow th high limit (say 100%), that output is slct
d by th low signal slctor (<)

© 2006 by Béla Lipták

8.10 Chmical Ractors: Control and Optimization
1705
90
No batch 70 Tmpratur, °C St point Prload = 0
Prload = load
50
30
10
0
10
20
30 Tim, min
40
50
60
FIG. 8.10g A prloadd batch PID controllr can avoid ovrshoot.
as intgral fdback, allowing th controllr to intgrat normally. If th outp
ut movs abov th limit stting, th ampli r will manipulat th intgral fdba
ck trm as ndd to hold it at th limit. Howvr, th fdback trm can only b
 rducd to th point whr th prload stting is slctd by th high signal
slctor (>); prload is adjustabl to match th xpctd load whn th nw sta
dy stat is rachd. Ovrshoot is causd by th intgral trm staying at th hig
h output limit until st point is rachd. Rducing that trm to match th xpc
td load—th controllr output rquird to hold tmpratur at st point in th n
w stady stat—should liminat th ovrshoot, and it dos. Without prloading th
controllr, howvr, undrshoot will rsult, as shown in th lowst curv in Fi
gur 8.9g. Ths curvs wr producd by an intracting PID controllr having d
rivativ action on th masurmnt (drivativ was not shown in Figur 8.9x), wh
ich also assists in limiting ovrshoot. If drivativ is not usd, thn th prl
oad stting will hav to b rducd somwhat blow th xpctd load, bcaus in
tgration bgins as soon as th controllr output lavs its limit, which is wl
l bfor th st point is crossd. Anothr mthod of avoiding ovrshoot whil mi
nimizing start up tim has bn usd succssfully on polystyrn ractors—th dual
mod control systm. It consists of an on/off controllr for start up and a PID
controllr to tak ovr at th nw st point, as was schmatically dpictd in
th prvious sction in Figur 8.9z. Th on/off controllr applis full hat unt
il th tmpratur has almost rachd st point—it thn switchs to full cooling.
Aftr a st tim dlay, control is transfrrd to a
prloadd PID. Switching from full hat to full cooling taks plac within 2–3°C of
st point, and th tim dlay is long nough to rplac th hot watr in th jac
kt with cold. Th squnc of opration is as follows: 1. Initial conditions ha
v both switchs in th A position. Full hat (100% output) is applid by th on
/off controllr until th dviation falls to th st point of Em. 2. Th on/off
controllr output thn drops to zro, producing full cooling. It also starts two
tim dlay rlays: TD 1 and TD 2. 3. Whn TD 1 tims out, th priod of full co
oling trminats, and th systm output is switchd by SS 1 to th prloadd PID
controllr, without intgration. 4. Whn TD 2 tims out, th tmpratur should
b sttld at st point. It thn transfrs switch SS 2 to th B position, nabl
ing intgration. Whil this systm is capabl of minimizing start up tim, it is
not foolproof. It has thr sttings mor than th batch controllr: Em, TD 1,
and TD 2. Furthrmor, ths sttings nd to b changd with procss conditions
such as batch siz and ractivity, tmpratur st point, and hat transfr co
f cint, th last typically dcrasing with th numbr of batchs producd, as sur
facs foul. Efforts to adapt ths paramtrs to changing conditions, or to prog
ram thm with ach rcip, hav mt with limitd succss. Th prloadd batch PI
D controllr with xtrnal rst fdback from scondary tmpratur is much sim
plr, mor univrsally applicabl, and mor robust undr changing conditions.
© 2006 by Béla Lipták
1706
Control and Optimization of Unit Oprations
MATERIAL BALANCE CONTROL
FY
X
AC St FY SO2 X2
Most chmical ractions rquir th prcis matching of two ractants to produc
a product. On of th challngs of procss dsign is to nsur that all of ac
h ractant is convrtd to product, rgardlss of th fractional convrsion in t
h ractor. A simpl xampl is th nutralization of acid and bas—thir individu
al concntrations in solution must b typically matchd within 0.01% for th solu
tion to fall within th pH rang of 6–8. A pH snsor is snsitiv nough to dtct
such a minut imbalanc, but on lin analyzrs ar not always availabl and v
n whn thy ar, nd point control can b dif cult, both in a batch and in a conti
nuous plant. Evn whn thr is but a singl ractant, as a monomr bing polym
rizd, nd point control is ncssary to trminat th raction at th dsird c
hain lngth or molcular wight or dnsity. Stoichiomtry Stoichiomtry is th t
rm usd to dscrib th balancing of a chmical raction so that all of th ra
ctants ar convrtd to th dsird product. A typical xampl is th oxidation
of hydrogn sul d to lmntal sulfur, which is condnsd: H 2S +
1
FC
Air Gas
FT Oxidizr FT Sulfur
AT
H2S Tail gas
FIG. 8.10h Minimum sulfur is lost whn th ratio of H2S to SO2 in th tail gas i
s controlld prcisly at 2:1.
Controlling Invntory Whn on of th ractants diffrs in phas from th produc
t(s) and othr ractants, its invntory is masurabl indpndntly from th oth
rs, and it can b fd at th rat at which it is consumd. Two xampls wr sh
own in Figur 8.9ii. A gasous ractant forming a gasous product in a liquidpha
s ractor will consum th liquid ractant at th rat of raction. Loss of liq
uid invntory is snsd by th lvl, which can b controlld to kp invntory
constant. Th manipulatd ow of liquid fd thn dirctly indicats raction rat
. Bcaus no liquid product lavs th ractor, howvr, liquid or solid contami
nants could accumulat at low points. Thr should b a liquid purg, ithr con
tinuous or priodic—just lik boilr blowdown—to liminat th problm. Most plants
hav facilitis for rcovring fdstocks and by products from purg strams—som
ar only opratd priodically, as wast invntory builds. Similarly, if th pro
duct(s) lav in th liquid phas, consumption of a gasous ractant lowrs prs
sur, indicating a loss in its invntory. A prssur controllr can add gasous
ractant as it is consumd. In this cas, a small gas purg is rquird to limi
nat inrts such as nitrogn or carbon dioxid. Molcular sivs ar oftn usd
to rcovr valuabl gass from purg strams. Whn fds and products ar in th
sam phas, invntory control is mor dif cult. For xampl, isobutan racts wit
h butylns in a 1:1 ratio to produc high octan gasolin calld alkylat, but
thy ar fd to th ractor in a 10:1 ratio to minimiz formation of butyln po
lymrs. Th procss thrfor contains a larg invntory of liquid isobutan, wh
ich must b controlld as it is consumd. Th ractor f unt, containing isobutan
, alkylat, and inrts such as normal butan, must b sparatd by distillation
, with th isobutan bing rcycld. Th rcycl storag tank provids an out ow t
hat is 10 tims th rat of consumption and mak up of isobutan, and so its tim
 constant is larg compard to that of typical liquid lvl loops. Proportional
only lvl control is rcommndd to allow th tank to absorb variations in pro
duction rat without th oscillations that intgral action brings to slow lvl
loops.
2
O 2 → H 2O + S ↓
A simpl nough raction, but if too much oxygn is fd, it will oxidiz sulfur
to sulfur dioxid in a conscutiv raction: S + O 2 → SO 2 Simply mtring th ga
s and air is not accurat nough, and th H2S contnt of th gas is also usually
variabl. Th ky to controlling this raction is stablishing an quilibrium b
twn H2S and SO2 in th tail gas laving th ractor: H 2S +
1
2
SO 2 ⇔ 112 S + H 2O
At quilibrium, whr th raction stops, H2S has twic th concntration as SO2
in th gas phas, and Rfrnc 4 dmonstrats that this ratio rsults in th m
inimum sulfur loss. So an analyzr on th tail gas rports both concntrations,
and a ratio controllr manipulats th air to gas ow ratio to hold th H2S/SO2 ra
tio in th tail gas at a valu of 2.0. Th systm is shown in Figur 8.10h. Thr
 is no xd or st valu for ithr H2S or SO2; thy oat with production rat and
othr conditions, but ar always at a minimum as long as thir ratio is maintain
d at 2:1. (Not that control of th ratio is achivd without th us of a divi
dr—placing a dividr in a closd loop is inadvisabl, bcaus its gain varis wit
h th valu of th dnominator.)
© 2006 by Béla Lipták
8.10 Chmical Ractors: Control and Optimization
1707
A similar r ning procss is th hydrocracking of havy oil into gasolin fraction
s. Hydrogn and oil ar hatd and fd into a sris of packd bd ractors, wh
r partial convrsion taks plac. Lightr products in th f unt ar thn rmov
d by distillation, with th unractd oil rcycld undr ow control from th colu
mn bas to th ractors. Th oil is vntually all consumd, but at a rat that
varis with tmpraturs and so on. Its invntory is snsd as th liquid lvl
in th bas of th distillation column, and that lvl controllr is usd to add
mor oil. It has also bn usd to st ractor tmpratur to convrt a xd ow of
oil, but this is mor dif cult dynamically, as it placs a tmpratur loop withi
n a lvl loop. Ractor Dynamics Closing th composition loop on a ractor can b
 dif cult, dpnding on th ractor dynamics. Idally, a plug ow ractor is pur d
ad tim In fact, static mixr lmnts ar oftn usd in a plug ow ractor to ma
ximiz radial mixing and minimiz longitudinal mixing. Actually, howvr, th st
p rspons will appar mor lik a sris of nonintracting lags, btwn 20 an
d 50 in numbr, as shown in Figur 8.10i. This would b quivalnt to 20–50 idal
back mixd ractors in sris. Th rspons dos approach dad tim, with 50% co
mpltion rachd whn th lapsd tim is qual to th rsidnc tim, V/F. It a
lso should b notd that th ffctiv dad tim is variabl with ow, although la
rg variations ar not xpctd in production ractors. Idally, a back mixd r
actor is a pur rst ordr lag. But prfct mixing is unattainabl, and thrfor
th actual stp rspons rvals a small amount of dad tim, dominatd by a rst
ordr lag. Th sum of th dad tim and tim constant is th rsidnc tim, V/F—a
t that point, th rspons curv crosss 63.2% complt, just as dos a singl l
ag. Th back mixd ractor curv in Figur 8.10i was simulatd by 20 intracting
lags, although incrasing th numbr
abov 20 dos not chang th shap of th curv. It is ssntially a distributd
lag that is an accurat rprsntation of a hat xchangr, whr hat capacity
and hat transfr surfacs ar distributd throughout th xchangr. Ractor co
mposition loops ar ordinarily slf rgulating, thir stp rspons approaching
a nal stady stat as th gur shows. Howvr, many ractors ar nclosd in rcyc
l loops that bring positiv fdback into th procss and liminat thir inhr
nt slf rgulation. Considr, for xampl, an acid bas raction carrid out in
an organic solvnt. At th ractor xit, gasous product is withdrawn, but any
xcss acid or bas rmains in th solvnt and is ultimatly rcycld to th ra
ctor. So if z concntration of xcss bas rmains in th solvnt, and th acid
bas ratio is not adjustd to allow for its prsnc in th rcycl stram, th
concntration will incras to 2z th nxt tim around, and continu to accumula
t. Th intgrating tim of this rcycl loop is th rsidnc tim of th solv
nt in th systm—th tim rquird to circulat th ntir solvnt invntory throu
gh th ractor. Th signi canc of this loss of slf rgulation cannot b undrsta
td. End point control must b applid, or th composition of th ractor f unt
will nvr rach a stady stat. And intgral only control cannot b usd to clo
s th loop, or th composition will cycl ndlssly—proportional action is ssnt
ial for stability. Batch End Point Control Thr ar two basic typs of batch r
actors: pur batch and fd batch. A pur batch ractor has all of its ingrdint
s chargd at onc, and th raction is allowd to procd ovr tim until compl
t. In a fd batch ractor, on or mor of th ingrdints is lft out of th ch
arg and thn fd at a controlld rat ovr tim until th raction is complt.
In both cass, som dtrmination must b mad of th raction’s approach to comp
l tion, to avoid wasting both chmicals and tim. Polymrization convrts light

hydrocarbons into highmolcular wight products having low vapor prssur. Thy
ar oftn conductd as pur batch procsss, with th initial charg undr prss
ur. At raction tmpratur, initially th prssur is clos to th vapor prss
ur of th monomr. As th last of th monomr is consumd, th prssur starts
falling, indicating compltion of th raction. Anothr indication is th dclin
 in hat volution as th last of th ractant is consumd. Still anothr is th
 total quantity of hat transfrrd to th coolant, which must b dtrmind by
calculation and intgration. Othrs masur or infr dnsity or viscosity of th
 product to dtrmin compltion. Fd batch procsss rquir trmination of f
d at th nd point, to avoid loss of ractant. This can b don simply by stopp
ing fd whn a dsird total is dlivrd, but this may not b accurat nough
in som cass. Th us of ndpoint analyzrs is rcommndd, whr availabl. A
cas in point is th control of pH in a batch of wastwatr from a
100 80 Rspons, prcnt 60 40 20 0 Back mixd, 20 intracting lags
Plug ow, 20–50 non int. lags
0
1
2 Tim, t/(V/F)
3
4
FIG. 8.10i Th plug ow ractor is simulatd by 20–50 nonintracting lags, th back
mixd ractor by 20 intracting lags.
© 2006 by Béla Lipták
1708
Control and Optimization of Unit Oprations
small procssing plant—it must mt nvironmntal spci cations bfor discharg is
allowd. This is not a typical composition loop rquiring a PID controllr, bca
us a batch procss has no ow laving it—it is ssntially a no load procss. Intg
ral action is ndd to liminat offst whnvr a controllr must mov its st
adystat output to match th currnt plant load. But whn th load is zro and s
tays thr, intgral action is unncssary and, in fact, causs irrcovrabl ov
rshoot. Considr a tank lld with impoundd industrial wastwatr at pH 3—its pH m
ust b raisd to 6 prior to discharg, and so 6 is th st point of th pH contr
ollr. Th larg initial dviation of 3 pH will driv th caustic valv fully op
n, and it will rmain opn as long as ncssary to nutraliz th wast. During
this tim intrval, if th intgral mod is activ it will wind up to th contr
ollr output limit, and th output will not fall blow that limit until th pH c
rosss st point. That is clarly too lat, and by th tim th caustic valv cl
oss, th pH will hav risn wll byond st point—prhaps to 9 or 10. In a contin
uous plant, ovrshoot is tmporary, bcaus its out ow will vntually purg away
xcss ragnt. But not in a batch plant—th ovrdos is prmannt. Th batch will
rquir r trating with acid or mor wastwatr to lowr th pH. Bcaus th l
oad is zro, th controllr output should also b zro whn th controlld varia
bl is at st point. This is achivabl by limination of intgral action, rpla
cing it with an output bias of zro—th controllr must b proportionalonly or pro
portional plus drivativ. This is not simply a mattr of shutting off intgrati
on, as this lavs th constant of intgration in plac of an output bias, and i
t oats, changing vry tim th controllr is transfrrd from manual to automati
c. Th bias must b a hard, prmannt zro, and som controllrs—of th incrmnta
l or vlocity typ—do not hav this capability. If a pur proportional controllr
is unavailabl, build on out of a calculation block, as follows: m = Kc (r − c) 8
.10(14)
fact, th dynamic gain of th procss varis invrsly with th volum of th ll.
To ffctivly control th composition of th variabl volum batch, th gain o
f th composition controllr nds to b changd in dirct proportion to th liq
uid lvl in th ractor.
REACTOR OPTIMIZATION Aftr satisfactorily controlling ractor tmpratur, matr
ial balanc, and nd point, at th ncssary production rat, optimization may b
 attmptd. Optimization gnrally mans maximizing conomic rturn on th plan
t capital and oprating costs, but thr ar many aspcts of this, bcaus thr
ar many cost factors. In an unlimitd markt, maximizing production maximizs
r turn on both capital and labor. Whn th markt is limitd, mting production

goals with a minimum cost of utilitis is ssntial. Maintnanc is also costly
, and oprating a plant in a way that minimizs claning and war has its saving
s. Maximizing Production Production rat from a ractor is dtrmind by th rat
 cof cint and rsidnc tim. To incras ow rats through a ractor of xd volum
 rducs rsidnc tim and convrsion pr pass, and thrfor dos not rsult
in a proportional incras in production. Th production rat of a rst ordr rac
tion is P = x 0 Fy Incrasing production by raising th fd concntration x0 of
th controlling ractant is not oftn an option, in that it tnds to rduc yi
ld, through sid ractions and conscutiv ractions. In th cas of isobutan a
lkylation, for xampl, incrasing th ratio of butylns to isobutan abov 1:1
0 incrass th formation of polymrs having undsirabl proprtis as motor fu
l, and consums mor catalyst. Incrasing production by raising fd rat is via
bl, but inf cint. To dtrmin th ffct of fd rat on production it is nc
ssary to diffrntiat Fy as a function of F. For th plug ow ractor, d ( Fy) =
y + (1 − y) ln(1 − y) dF And for th back mixd ractor, d ( Fy) = y2 dF 8.10(16) 8.
10(15)
whr m is th controllr output, Kc its proportional gain, r th st point, and
c th controlld variabl. If drivativ action is dsird—usually hlpful in a c
omposition loop—a ladlag function can b applid to th controlld variabl bfor
 it ntrs th controllr. It is possibl to fd two ractants simultanously
into a batch ractor with no out ow. Th control problm is lik trying to ll a bat
htub in minimum tim. To rach th right tmpratur at th dsird volum, it r
ally nds to b controlld throughout th lling procss. This rquirs nding th
ow ratio that producs approximatly th right tmpratur at th start of th ll
ing procss, and adjusting it as ncssary until th nal volum is rachd. Initi
ally, changing th ratio producs a larg tmpratur chang, but as th tub lls,
this gain gradually dcrass. In
For th plug ow ractor oprating at kV/F = 1, d(Fy)/ dF = 0.264, and for th bac
k mixd ractor it is 0.250. Thrfor, incrasing fd rat by 1% incrass pro
duction only
© 2006 by Béla Lipták
8.10 Chmical Ractors: Control and Optimization
1709
0.25%, whil incrasing th load on th sparation and rcycl systms by 0.75%—no
t a favorabl rsult. To incras production without this pnalty rquirs incr
asing raction tmpratur along with fd rat, following th raction rat co
f cint curv. Only small changs ar rquird, givn th rat cof cint doubling a
pproximatly vry 10°C. In th rgion around 1.0 in Figur 8.10b, th normalizd
rat cof cint incrass about 10% for ach 1°C. Maximizing production in a back mi
xd xothrmic ractor amounts to oprating at th cooling constraints. Figur 8
.9dd dscribd how this is don whn manipulating fd rat(s) to a continuous r
actor. Th position of th coolant valv is snt to an optimizing controllr (O
IC), which adjusts th fd rat(s) to driv th valv to about 90% opn. In oth
r applications, such a valv position controllr would hav only th intgral m
od, but for an xothrmic ractor, all thr PID mods ar ncssary for stabl
opration. In addition, if a larg diffrnc dvlops btwn th positions of
th fd and coolant valvs—or mor accuratly btwn fd and hat transfr rat
s—this is an indication that th raction is not procding normally and should b
 shut down bfor xcssiv amounts of ractant(s) hav bn fd. Th OIC loop
must b slowr than th tmpratur loop that is insid it, and grat car must
b takn in closing it. Th OIC loop aros out of oprators’ practic of obsrving
th stady stat position of th coolant valv and gradually incrasing th f
d to a fd batch ractor to driv th valv almost fully opn.2 Th sam oprato
rs rspctd th conditions undr which this could b don safly, namly that t
h ractor had bn proprly chargd and initiatd, and was alrady at th contr
olld tmpratur. Thr is no fd rat to manipulat in a pur batch ractor.
Furthrmor, th raction rat tnds to vary as ractants ar consumd. To minim
iz run tim thn rquirs th OIC to adjust th tmpratur st point to driv
th coolant valv to 90% opn. This is a dif cult loop, howvr, xhibiting invrs
 rspons. To opn th coolant valv furthr rquirs incrasing tmpratur s
t point. Howvr, whn its st point is incrasd, th tmpratur controllr r
sponds by closing th coolant valv rst. As th tmpratur thn bgins to ris,
th controllr will vntually opn th coolant valv mor than bfor, but this
taks tim. Again, th OIC rquirs full PID control, and must b much slowr t
han th ractor tmpratur controllr. Maximizing production basd on th posit
ion of th coolant valv is not always applicabl on plug ow ractors. If th coo
lant valv is obviously limiting production, as can happn during pak summr w
athr, thn it may work. Howvr, Rfrnc 2 dscribs an uncontrollabl racto
r limitcycling in wintr, whn th qual prcntag coolant valv was only 10–20%
opn. Production rat was limitd by th hat transfr surfac of th coolr and
not by th coolant valv or supply tmpratur. Th maximum controllabl produc
tion rat of a plug ow ractor is dtrmind by th slop of th hat transfr lin
 in Figur 8.10d. At th limit of stability, th slops of

th hat rlas and hat transfr curvs ar th sam. Th slop of th hat r
las curv is dQr dy = x 0 FHr dTr dTr
8.10(17)
and that of th hat transfr curv is UA. Thrfor, th maximum controllabl f
d rat is Fmax = UA x 0 Hr (dy/dTr )

8.10(18)
To monitor UA rquirs th calculation of th logarithmicman tmpratur diffr
nc across th coolr: UA = Qt ∆Tm

8.10(19)
where for counter current heat transfer, ∆Tm = (TH 1 − TH 2 ) − (TC1 − TC 2 ) T −T ln H 1
C 2 TH 2 − TC1
8.10(20)
Qt may be calculated either from fresh or circulated coolant ow: Qt = WCCC (TC 2 −
TC ) = WCC (TC 2 − TC1 ) 8.10(21)
TH1 and TH2 are temperatures of the reaction mass entering and leaving the heat
transfer surface, respectively. For a plug ow reactor prone to fouling, UA or ∆Tm s
hould be monitored regularly and compared to production schedules. In this way,
cleaning can be scheduled in advance, instead of waiting until a limit cycling c
ondition develops that forces a maintenance shut down at an inconvenient time. S
upplemental Cooling Some reactors have two sources of cooling, at different cost
s. For example, cooling water can be used up to the point where its valve is ful
ly open, and then a supplemental medium such as chilled water may be used for ad
ditional cooling. If the two media can be mixed, then accommodating the addition
al source is as simple as sequencing the two valves. For example, the cooling wa
ter valve could open from 50 to 25% of controller output, and the chilled water
valve from 25 to 0. However, the media are likely to be separated from each othe
r to avoid cross contamination, with auxiliary cooling by chilled water passing
through an internal tube bundle, and cooling water circulated through the jacket
. This complicates
© 2006 by Béla Lipták
1710
Control and Optimization of Unit Operations
a control system that features cascade control of jacket outlet temperature. The
cascade system should be retained for its effectiveness, and the supplemental c
ooling manipulated to keep the jacket temperature operating in its control range
. This is done by the use of a valve position controller (VPC) connected to the
secondary output, as was shown in Figure 8.9o in the previous section. Using int
egral only control, it manipulates the chilled water ow to keep the cold water va
lve about 90% open, which corresponds to a controller output signal of 5% (its r
ange being 50–0%). As long as the heat load stays in the range of the cold water v
alve, the chilled water valve would remain closed. But when the VPC set point is
passed, the chilled water valve will be opened as necessary to match the load.
This loop is not fast, but it need not be any faster than the expected load chan
ges, because the jacket temperature loop is always in control. Whether the addit
ional cost of the chilled water can be justi ed depends on the demand for product.
In most cases where supplemental cooling is used, it is only needed during peak
load, which only represents part of a typical batch operation.
References
1. 2. 3. 4. Shinskey, F. G., Process Control Systems, 4th edition, New York: McG
raw Hill, 1996, p. 317. Shinskey, F. G., “Exothermic Reactors: The Stable, the Uns
table, the Uncontrollable,” Chemical Engineering, March 2004. Shinskey, F. G., Pro
cess Control Systems, p. 312. Ibid., pp. 324–325.
Bibliography
Corripio, A. B., Design and Application of Process Control Systems, Research Tri
angle Park, NC: ISA, 1998. Juba, M. R., and Hamer, J. W., “Progress and Challenges
in Batch Process Control,” Chemical Process Control CPC III, Amsterdam: Elsevier,
1986. Lipták, B. G., Optimization of Unit Operations, Radnor, PA: Chilton, 1987.
Luyben, W. L., “Batch Reactor Control,” InTech, August 1975. Luyben, W. L., Tyreus,
B.D., and Luyben, M. L., Plantwide Process Control, New York: McGraw Hill, 1998.
Shinskey, F. G., Controlling Multivariable Processes, Research Triangle Park, N
C: ISA, 1971.
© 2006 by Béla Lipták