ACKNOWLEDGMENT
proportional controller
The authors are indebted to Professor
Hans Kramers for several valuable sug-
gestions made in the final preparation of
these papers on control while he was Minne-
sota Mining and Manufacturing-Visiting
Professor of Chemical Engineering a t the
University of Minnesota.
NOTATION
A = a chemical compound or its
concentration (mole/liter)
B = same
C = specific heat of the reacting
mixture, calories/g.
C = concentration, mole/liter
e = volumetric C.S.T.R. (continu-
ously stirred tank reactor) to
batch efficiency ratio
F = flow rate, liters/hr.
h = reactor heat input, "K./hr.
H = reactor heat input, OK./hr.
A H R = heat of reaction, calories/mole
of A
I = control quality
i = pure imaginary, d 7
k = reaction-rate constant, differ-
ent units.
11. Frequency-response Relations for Reactor Chains, Tubular
Reactors, and Unit Reactor Control Loops
REACTOR CHAINS
The results which were obtained in
Part I for a single reactor can be used to
calculate step by step the response of a
chain of reactors. Interest in these calcu-
lations is twofold. First the safe operation
of the whole chain may put upon a
reactor in the chain more stringent control
requirements than would individual oper-
ation, especially in the case of reactions
highly sensitive to temperature. Second,
these calculations can be used to deter-
mine simpler modes of control for the
whole chain. In some cases, for instance,
the feed Row is controlled by inlet
sampling and the whole heating system
by a sampling of the final product, the
individual heat-control loop on each
reactor thus being eliminated. The calcu-
lations arc straightforward and the theory
is easily devcloped with the introduction
of matrices. It will be developed here
on the basis of the example used pre-
viously, of a reaction involving com-
ponents A and B, with a rate R(x, T).
First Case: Variations in Concentration Only
Equations (17) and (18) of Part I are
Page 256
K = proportionality constant for a
reactor, hr .
M = a matrix
n = mole fraction, or number of
reactors in a chain or reaction
order
Q = modified heat of reaction, (OK.)
(mole)/liter of A
R = general reaction rate
t = time, hr.
T = temperature, OK., or controller
time constant
V = reactor volume, liters or varia-
tion vector
2 = concentration of feed com-
ponent, mole/liter
X = variation in concentration of
feed component, mole/liter
Y = concentration of product com-
ponent, mole/liter
Y = variation in concentration of
product component, mole/liter
2 = variation in reactor tempera-
ture, OK.
a,p, y = stoichiometric coefficients
A = fractional control quality
e = C.S.T.R. holding time, hr.
P = density of reacting mixture,
g./liter
used with the feed concentration varia-
tions X o and Y o
(1+ aOR,' + jcoO)X Xo =
-OR,'X + (1 + jUe)y= yo
These may be written in matrix form, the
variation vector V i representing the
variations in the composition of the feed
to reactor i + 1:
v.= X;
iyil
and the matrix M , of the variation CO-
efficients being introduced.
I n matrix notation Equations (1) and
(2) may be written
and for the reactor i in a chain
1 + aOiRzi' + j d , ,
- OiR,i',
A.1.Ch.E. Journal
7 = reaction time for a tubular
= signal frequency, radians/hr.
6J
Subscripts
0, 1 = C.S.T.R. feed and product
streams
D = derivative controller
I = integral controller
n = reactor number in a chain
P = proportional controller
S = steady state
LITERATURE CITED
1. Acton, F. S., and Leon Lapidus, Znd.
Eng. Chem., 47, 706 (1955).
2. Bilous, Olegh, and E. L. Piret, Chem.
Eng. Progr. (to be published).
3. Devyat,ov, B. N., and G. N. Bogashev,
Zhuy. Priklad. Khirn., 24, 807 (1951).
4. Eldridge, J. W., and E. L. Piret, Chem.
Eng. Progr., 46, 290 (1950).
5. James, H. M., N. B. Nichols, and R. S.
Phillips, "Theory of Servomechanisms,"
McGraw-Hill Book Company, Inc.,
New York (1947).
6. Kramers, Hans, and G. Alberda, Chem.
Eng. Sci., 2, 173 (1953)
7. Mason, D. R., and E. L. Piret, Znd. Eng.
Chem., 42, 817 (1950); 43, 1210 (1951).
or
vi-l = M,Vi (3)
(') with
(2)
Therefore, the relation between product
and feed variations in composition is
V,=(M,M, . . . M i. . * M n ) - l v o (5)
This formula gives the desired result,
namely that the product composition
varies as a function of feed-composition
variations.
0 Xi-1
1 + jwO, Yi-1
June, 1957
Illustration 4
The foregoing formula will be applied to
the case of a first-order reaction
A-+B
performed isothermally in a chain of
reactors of equal size. The rate of the
reaction is
INPUT OUTPUT feed- and product-composition variations
for a chain of n stirred-ffow reactors. It
m a y be interesting t o examine what
happens when t h e number of reactors
FINAL increases indefinitely, the total holding
CONTROL COMROLLER INSTRUMENT
ELEMENT time remaining constant. This limiting
case is represented b y a tubular reactor
Fig. 4. Block diagmm of a control system. with uniform conditions i n a cross section.

R(x) = dB dA
- = -- = kx with a 9570 conversion of compound A . Limiting Case of the Tubular Reactor
at dt The inlet and outlet concentrations are
All holding times are assumed equal
Therefore R,'(z) = 12 and the quantity compound A and r = no is the tubular-reactor holding
OR,' = Ok is constant for all reactors. All 20= 1.00 mole/liter, X, = 0.05 mole/liter time. Therefore
the matrices M , are therefore the same, and product B
Formula (5) may be written as v n = M-"vO yo = 0.00 mole/liter, yn = 0.95 mole/liter
nr
"1

'1 + elc + jwe, (6)

where the increment AT is thus defined.
Then Expression (4) of t h e matrix Mi
- ek:, may be written .
Matrix multiplication and division yields
I + ek + j w e , The operation is isothermal, with a rate

1
constant
1 + jwe lc = 19.0 hr.-'
Ari
1 The holding time, 0, for a reactor of the -RZi', jw
-1 + elc + j w e y
- (I
0 chain is found from a mass balance over
the system, given by Neglecting second-order terms in AT;
gives

the term Q being The maximum amplitude ratio is attained

X=n-1 at vanishing frequency. I t s value, calculated
Q = elc C (1 + elc + jwe)A-n from Formula (9) for w = 0, is

Developing matrix expression (6) shows

the relation between variations in concen- = 1- Xn
- = 0.95
tration of A in the feed and product streams Xn
to be Passing t o t h e limit gives
It is therefore a constant, independent of
the number of reactors.
While vanishing frequency results are the
easiest to calculate, more information about
and the variation in the conoentration of B system behavior may be derived from the
i n the product stream Y,, caused by con- values of the phase angle and amplitude
centration variations X,, IToin the feed ratio of the concentration variations a t a
stream, frequency physically related to system (11)
Y, = 9x0 + (I+jw8)"
1 yo
properties, that is, for instance, such as
w n 8 = 1. Such results, calculated by means The exponential form of the transfer func-
of Equation ( Q ) , are presented in the accom- tions characterizing t h e behavior of
Usually Y o = 0, no product being present panying table.
i n the feed stream, and thus
Y, = QXO, POLAR-PLOT
DATAFOR A REACTOR
CHAIN,FOR wnO = I
or Amplitude ratio
X=n-1
Number of Unit holding Total holding I Yn/XoI Phase angle
Y, = elc CX=O
+ ek: + jwe)"-"
(I
reactors, n time, O time, nO for wn8 = 1 for wnO = 1, deg.
1 1,000 hours 1.000 hours 0.671 -48
(I + jwe)-'-'x0 (9) 2
3
0.183
0.090
0.366
0.271
0.763
0.813
-56
-57
or, summing the geometric series, 5 0.043 0.216 0.861 -58
10 0.018 0.184 0,904 -58
Yn
- = (1
xo +j d - " - (I + jwe + ek)-" m 0.000 0.158 0,950 -57.3
These values give an indication of t h e
This is an interesting result, as it relates transformation of t h e shape of t h e systems with dead time may be recog-
the variation8 in product composition to Nyquist plot with reactor number. From nized. The matrices
those in feed composition for a reactor chain. a nearly circular plot touching t h e imagi-
To illustrate this relationship bether, a nary axis a t t h e origin for a single reactor,
numerical example is presented. A chain t h e plots evolve toward a circle centered
of n equal-sized reactors operates on the
fir&-order reaction at t h e origin for the limiting case of t h e
tubular reactor.
A-B Formula ( 5 ) gives t h e relation between

Vol. 3, No. 2 A.1.Ch.E. Journal Page 257

 CONVERSION, I - *e
Fig. 5. Fractional control quality against
conversion for integral control of a first-
order reaction in a C.S.T.R.
commute; therefore Expression (11) niay
be written
or
8, = exp { - j w ~ }
Developing this matrix expression one
finds the formula for a tubular reactor
(with any order reaction) whirh cor-
responds to Expression (9) for a reactor
chain :
.exp { - j u ~ ] X , ,
The Nyquist diagram, which represents
I’,/Xo, is a circle centered a t the origin
of the complex plane.
Second Case: Variation in Concentration and
Temperature with a Steady Heat Input
The treatment for this case is similar
to that above, the equations to be con-
sidered being
(1+ a8R,’ + jw8)X + oc8Rz.’Z
- 8R,’X + (1 + j w 8 ) Y
- 8RT’Z
Page 258
1.0 , I I I I I
CONVERSION. I- +e CONVERSION, I- %
Fig. 6. Fractional control quality against Fig. 7. Fractional control quality against
conversion for proportional control of a first- conversion for derivative control of a first-
order reaction in a C.S.T.R. order reaction in a C.S.T.R.
-aQ8R,’X + (1 + jw8 is obtained, which must be satisficd in
the case of a highly exothermic reaction
- aQ8ET’)Z = Z, (Q > Rzt’/RTc‘)in order to have stable
operation. This is again an example of
obtained from the basic expressions (17), the restrictions which control require-
(18), (19) of Part I. ments may impose on reactor design.
The matrix M , similar to (4) is now
I 1
and the relation between feed and product
variations for a chain of n reactors is as
before REACTOR CONTROL
In the preceding page only the calcu-
lation of the frequency response of
continuously stirred tank reactors has
been considered. This section will com-
bine for illustrative purposes, the fre-
quency-response functions thus obtained
with different controller-response func-
tions to find the quality of the control
V,=(M,M,. . . M ; . .Mn)-‘V, (15) circuit formed by reactor and controller.
The control system considered may be
where the vectors Ti, are defined now as represented by Figure 4.
A variable in the reactor outlet stream,
V “= I x,,y , , z,I for instance product concentration, is
Numerical calculations on Formula (15) measured and is compared with the
are best carried out step by step. desired value in the controller which
(13) The determinant of Matrix (14) is sends out a signal acting on the controlled
variable in the reactor inlet stream, for
D, = (I + jw8,)’[1 + a8,(R,,’ instance feed concentration, through a
final control element. Detailed informa-
- QRz.,’) + jw8,I tion about such diagrams and control
theory in general may be found in
The factor
references 1 and 2.
1 + ae,(R,,’ - +
Q R ~ , ’ ) jwe, A negligible lag is assumed in measure-
ment and in the final control element, in
is recognized as one that appeared in order to isolate the effect of controller-
Expression (24) of Part I. response characteristics on the system.
The stability of the reactor chain is For present purposes the three usual
obtained when all units of the chain are idealized types of controllers arc ex-
= X, stable. Following the discussion in Part I
the condition on the holding times
amined ( 1 ) . Actual controllers often
approach combinations of these and can
be treated in the same manner; such a
case will be illustrated in the analogue
= Y,, example.
A.1.Ch.E. Journal June, 1957
 0 006
Fig. 8. R.E.A.C. circuit for
temperature control in a
C.S.T.R. Data of
illustration 3.
G,(jw) is the frequency-response function
for the reactor chain without control.
For the controlled system (Figure 4)
the input to the rcactor is equal to the
input to the system Xuminus the output
of the final controller. The input to the
controller is the output of the reactor
Y,. Hence if G(ju) represents the fre-
quency-response function for the con-
troller, thp output of the controller is
G(jw)Y,; the input to the reactor is
Xu - G(jw)Y,; and the output of the
reactor is Y , = G,(jw)[Xa - G(ju)Y,].
Sol\Ting this for Y , yields
I ,.

The frequency-response function for the

system with control is therefore

Control quality may then be determined

0.020 TIME from formula (IS), where Y(jw) is
0 replaced by its expression (20).
-100 The case of a single reactor will be
examined in more detail. The expression
of control quality for a single reactor
with different controllers or no control are,
Frequency response from the foregoing development,
Controller G(+)
integral control
Integral -P/E = l/jwTl
Proportional -P/E = K
Derivative -P/E = jwrl;
I n t h e preceding expressions of con-
troller response to a cyclic disturbance of proportional control
frequency w , E designates the error
signal and P the controller output. The
quantities T I , K , T z are constants char-
acteristic of controller operation.
Control quality may be expressed by
the quantity derivative control

I = 1"1 W ( t ) 1' d t (16)

where W ( t ) is the system response to a

unit impulse applied to systcm input a t These expressions are valid only for When there is no control, that is to say,
time zero. The choice of this criterion stable systems. The value of I would be
takes into consideration the importance infinite for an unstable system, as in -1= K = T - 0
of the deviation amplitude caused by the this case response W ( t ) is undamped or 2 -
perturbation and also the length in time TI
ever increasing in amplitude.
of the deviation. This expression will be these integrals reduce to
used in the study of control quality.
Example
The integral 1 may be evaluated in
terms of the frequency-response function A control loop will be considered on the
of the controlled system Y ( j w ) . From basis of measurement of product con-
reference 2 centration in the outlet stream and -
- ek2
control of feed concentration in the inlet 2(1+ ek)~)(2+*
W(t)= -
1
1
2a.
/ "+m

-m
dwe'"'Y(jw) (17) stream. Reactor operation is assumed to
be isothermal and data of Illustration 4 The integrals expressed above are
are used. Thus the frequency-response calculated by integration in the complex
Therefore function for a chain of n equal-sized plane or use of tables in the appendix of
reactors is given by Expression (9) : reference 9. The results have been
I ='S,- 1 W ( t ) l a dt cl(jw)= --
Y, = ek C
X=n-1
(1 + ek
expressed here in terms of the system with
no control in fractional form:
X0 X-0
= /m Y ( j w ) Y ( - j w ) du (18)
a G O

Vol. 3, No. 2 A.1.Ch.E. Journal Page 259

which better shows the gain in control These equations, written for an appro- dY
quality due to the controller. The value priate control loop, may be associated e- - e R L f X+ Y
dt
of A when the control loop is broken is with the linearized equations describing
zero and reaches unity for perfect control. - BRTfZ = Y o (22)
reactor behavior: [Equations (12), (13),
The expressions for the respective frac-
tional control qualities are for integral
control
(14) of Part I]. dZ
e- - aQOR,fX
dt
+ (1 - a&enTf)z
e dt + (1 + aewx
dX = eAH + Zo (23)
The system thus obtained may be solved
+ a % R T f Z= X , (21) to obtain the response of the controlled

for proportional control

c_ I TIME , MINUTES QOI 1 Y, MOLE /LITER

p -001 -0.01
X . MOLE/LITER TIME , MINUTES
for derivative control
TIME , MINUTES
T,lc 1 2 , *K 0 10 20
AD =
2 + + rF2%
elc
.K 1.0 0 7

Fractional control quality has been a5

represented for the different modes of
control A,, Ap, A, against conversion in 0.0 -2.0 H, .K/MINUTE
Figures 5, 6, and 7. Curves are given for 0
different values of the characteristic CONTROLLER SETTINGS = K = I , T = 1/10 MIN.
control parameters T I , K , and TO.
Conversion in this case is V
I--=-----
X ek Fig. 9. Example of C.S.T.R. temperature control (underdamped case).
z0 1 + ek
where z,,,J: are the concentrations of
reactant A in the feed and the product.
The plots of Figures 5, 6, and 7 show W
0.01 1 TIME, MINUTES 0.01 I Y, M O L E l L l T E R

an increase in control quality by the

same instrument when the conversion W

:.
-0.01
decreases, which is evident a priori, X . MOLE/LITER TIME ,MINUTES

since a t low conversion a disturbance on TIME, MINUTES

the feed will have only a small effect on 10 20
K
.
outlet product concentration. The plots
are characteristic of the three simplified
modes of control examined. Idealized
components have been used in this loop
analysis and the results in practice must TIME ,MINUTES
be tempered with a knowledge of the 00 -20 I H I 'K/MINUTE
performance of the components and
system actually available. CONTROLLER SETTINGS : K 14 , T, = 114MIN.

Fig. 10. Example of C.S.T.R. temperature control (correct controller settings).

ANALOGUE SIMULATION OF
CONTROLLED SYSTEMS

The equations describing the operation

of the three types of controllers which
have becn considered in this study are 0.01 I TIME , MINUTES QOll I Y, MOLElLlTER

Integral control
d -aoi
X, MOLElLlTER TIME, MINUTES

TIME, MINWES
*K 1.0 , r\ 2 , .K O
0r 10 20
Proportional control
1
E = --p 0.5 I/ \
K
Differential control " \- g-2.01
TIME, MINUTES

10 20
H, .K / MINUTE

dE CONTROLLER SETTINGS: K = 5 , T, = 114 MIN.

p = -T -
dt Fig. 11. Example of C.S.T.R. temperature control (overdamped case).

Page 260 A.1.Ch.E. Journal June, 1957

reactor to a given disturbance. The CYCLING IN SYSTEMS WITH DEAD TIME
1
analytical solution will, however, in If the number of stirred flow reactors
- In K(1 - e-kr) <o
many cases be cumbersome and in in a chain is increased indefinitely, the
practice, especially when more complex or
over-all holding time remaining constant, K(l - e+) < 1
performance functions are needed, the at the limit the case of a tubular reactor
use of an analogue computer is of great is obtained. The frequency-response func-
advantage. An example will be solved Thus a limitation exists on the values of
tions for the isothermal tubular reactor the proportional factor K which may be
with the analogue technique. without volume change have been ob-
Considered here is the case of a tem- used. If controller action is too powerful
tained in Expression (12). For the case (high K ) , cycling may develop.
perature-control loop with a controller of a first-order reaction the expression of
combining the proportional and integral the frequency-response function relating
control modes. A constant feed compo- product-concentration variations Y , to COMMENT
sition is assumed, that is to say Xo = feed concentration variations Xo has
Y o = 0, and a disturbance caused by a been found:
It should be realized that in the prac-
change in feed temperature, Zo. The tice of automatic control the results of
equations which describe this system are frequency-response analyses can today
(21), (22), (23) and the controller equlttion be used to advantage principally for
for the temperature-control loop becomes The presence of the imaginary expo- guidance rather than for the exact pre-
nential in the expression of the frequency- diction of the performance to be expected
response function is characteristic of of a practical control system. I n the
systems with dead time. Thus a measuring first place the disturbances occurring in
instrument placed a t the reactor outlet an industrial operation will not usually be
With the values of Illustration 2 for a will perceive feed perturbations only a regular or periodic, although of course the
second-order exothermic reaction time after they happen and any corrective typically random disturbance will excite
action will be delayed by that much. This all frequencies. Furthermore, the differ-
2A-+B+D dead time in the reactor system increases ences found between theory and practice
are often the result of deviations from the
with A H R = -50,000 cal./mole of A with the number of reactors in the chain simplified characteristicswhich are usually
these equations take the following form: and approaches a maximum time in a assumed for the sensing and the control
tubular reactor, and so in the latter
design control may be more difficult than elements. Furthermore largely indeter-
dX
+ +
0.33 - 16.50X 0.12452 = 0
dt in a continuously stirred tank reactor minate lags often occur in industrial
(C.S.T.R.). Also control by stages may installations. Nevertheless the equations
more readily be applied to the selected obtained as a result of the application of
dY
0.33 - - 7.750X
dt
+ Y elements of a chain. frequency response theory offer real ad-
Dead time may easily lead to cycling vantages in simplicity of treatment and
- 0.06222 = 0 of the controlled system if the control is
clarity of exposition, have proved ex-
improperly designed. Only an example tremely helpful in understanding the
dZ of such a case will be given, for propor- control process itself, and have guided the
0.33 - - 775.X - 5.2242 tional control applied to a tubular engineer to practical solutions to auto-
dt
reactor by measurement of product matic-control problems. As more precise
= 0.33AH + 2, concentration a t the outlet and control characterizations of modern commercial
instruments and of lag systems become
of inlet-feed concentration.
dH clZ 0.33 If A ( t ) designates the disturbance on available, better predictions are to be
0.33 - = -0.33K - - __ Z expected.
dt dt T, the feed concentration a t the reactor inlet,
the disturbance on product concentration Numerical examples have been worked
The analogue circuit which represents out in this paper and especially in Part I
a t the outlet is (1 - e-kr)A(t - T ) and
these equations with a time scale changed the equation of the proportional-control in order to demonstrate clearly the
from hours to minutes is given as an loop is method of application of frequency-
illustration on Figure 8. The symbols of response analysis to continuous-flow
be found in reference 3.
+
the circuit diagram are standard and may A(t) K ( 1 - e-k‘)A(t - T ) = 0 (25) reactions. These should also serve to
clarify the characteristics of the chemical
This circuit was used to determine the where K is the characteristic constant of reactor control process itself. It was
response of the controlled reactor to a the proportional controller. Equation (25) shown for example that such important
change in feed temperature is of the difference type, and methods of factors as temperature-sensitive rates of
solution are well known. Thus introducing reaction can be readily analyzed by these
2, = 1°C. a solution methods.
for three different settings of the con-
troller characteristics K and T I . The A ( t ) = e*‘
NOTATION
results are given on Figures 9, 10, and 11. into Equation (25) results in a charac-
Figure 9 represents a case with low teristic equation which gives the values See page 256.
damping ( K = 1, T I = 1/10 min.). The of s for which e * t is a solution of (25):
system oscillates around the steady state. LITERATURE CITED
Figure 10 represents a more damped case
( K = 4, T I = K min.), which would be
+
1 K ( l - e-kT)e-sT= O 1. Eckman, D. P., “Principles of Industrial
acceptable in practice. The system It is found that Control,” John Wiley and Sons, Inc.,
represented in Figure 11 is overdamped New York (1948).
2. James, H. M., N. B. Nichols, and R. 8.
( K = 5, T I = M min.) but may still be Phillips, “Theory of Servomechanisms,”
acceptable in practice. Such examples McGraw-Hill Book Company, Inc.,
could be multiplied for various combina- New York (1947).
tions of controller types. Discussion of where X may take any integer value. The 3. Korn, G. A., and T. M. Korn, “Elec-
them is easy, and they will not be system is stable if the exponential e” is tronic Analog Computers,” McGraw-Hill
developed further here. damped : Book Company, Inc., New York (1952)

Vol. 3, No. 2 A.1.Ch.E. Journal Page 261