A Heuristic Approach for the Economic
Optimization of a Series of CSTR‘s
Performing Michaelis-Menten Reactions
F. Xavier Malcata
Escola Superior de Biotecnologia, Universidade Catolica Portuguesa, Rua
DL Antonio Bernardino de Almeida, 4200 Porto, Portugal
Accepted for publication November 20, 1987
The estimation of the size o f each reactor of a series of
CSTR‘s performing a Michaelis-Menten reaction i n the
liquid phase can be obtained to advantage via an opti-
mization technique leading to the minimum overall capi-
tal cost. The cost scaleup is assumed to be described by
a power rule o n the equipment capacity. Various contri-
butions are lumped into the exponent, thus leading to
values above unity. The analytical development leading
to the optimal intermediate concentrations of substrate
according t o t he foregoing criterion i s presented. A
short-cut method based on an empirical expression that
approximates the numerical solution is reported. This
correlation is found to be exact at the asymptotic behav-
iors, and to give accurate results within an acceptable
error level for the range w i th physical interest. There-
fore, it is particularly useful during the predesign steps
of equipment for the biochemical industry.
INTRODUCTION
Much attention has been given to reactor systems con-
sisting of a series of well stirred tanks in series. Besides
the lower construction costs when compared to classical
tubular reactors, the efficient stirring of the reactor con-
tents ensures uniform temperature (thus avoiding local hot
spots), coupled with ease of access to the interior surface
for manutention’ and appreciable residence times. Opera-
tion is simple, and various degrees of molecular level mix-
ing can be attained, provided a variable speed agitator is
available. Ans2 and Levenspie13presented general concepts
of reactor design and optimization, having developed sev-
eral criteria for nth order reaction kinetics. The optimal de-
sign for fermentation using the Monod equation was reported
by B i ~ c h o f fwhereas
,~ Luyben and Tramper’ derived an ana-
lytical formula for the optimal design of CSTR’s in series
using Michaelis-Menten kinetics, assuming a constant ac-
tivity of biocatalyst in the reactors and using as definition
of the optimum the smallest total reactor size to perform a
specified conversion. This objective function does not, how-
ever, approach the actual minimum capital investment on
equipment, which proves to be a major constraint in reactor
design. Such a difficulty was overcome for particular cases
by a number of authors: Gulyas6 developed a mathematical
Biotechnology and Bioengineering, Vol. 33, Pp. 251-255 (1989)
0 1989 John Wiley & Sons, Inc.
method for a series of CSTR’s and optimized it using func-
tional analysis on the basis of the lowest investment cost;
Lee and co-workers7 applied a previous model to achieve
optimal reactor design for a cellulose hydrolysis reactor,
the sensitivity of the product price being tested against en-
zyme and capital investment costs. More recently, Malcata’
reported that only a few reactors should be considered in
the series if minimum total investment cost were to be
achieved, using a standard six-tenths factor rule for equip-
ment cost scaleup, after having pre-optimized the series by
minimizing the total reactor space time, in order to decrease
the relative enzyme denaturation to a low level, and skip
the high substrate concentration region (which proves
useful whenever higher-order side reactions tend to occur).
When estimating the size and the global investment on a
series of CSTR’s, for some biochemical plant, one impor-
tant problem arises: which intermediate concentrations
should be chosen, in order to achieve minimal investment,
taking the whole operating life of the process as the time
scale. Simple correlations allowing one to compute the op-
timum intermediate concentrations would therefore be of
much help, if only a quick sketch of the reactor sizes and
investments involved, were required. It is the purpose of
this Communication to present the mathematical basis for a
correlation developing strategy, convenient examples being
actually reported.
MATHEMATICAL ANALYSIS
Consider N continuous stirred tank reactors in series,
performing a simple, one-substrate reaction following
Michaelis-Menten kinetics. A mass balance to the reactant
over a general reactor i, assuming maximum mixedness
conditions’ yields
Da, =
(CZ, - C , * ) ( K * + C,*) ; i = 1 , 2 , . . ., N
c:
where C,! denotes the dimensionless, scaled concentration
at the outlet stream of reactor i ; Da, is a Damkholer num-
CCC 0006-3592/89/03025
1 -05$04.00
her for the ith reactor; and K * is a dimensionless kinetic
constant. These variables are defined by
c* = c,/c, (2)
Da, = v.v,/QC, (3)
K * = K,,,/C, (4)
where C , denotes the substrate concentration in the feed
stream to the first reactor; v,,, is the maximum kinetic rate;
K , is the standard Michaelis-Menten constant; V, is the ef-
fective volume of the ith reactor; and Q is the volumetric
flow rate through the system.
The fluid flow pattern for a CSTR is complex and largely
influenced both by the geometry of the system and the
characteristics of the stirring devices. Although the equa-
tions of continuity and motion for the system can be writ-
ten (see ref. lo), they cannot be generally integrated. For
this reason, extensive use of models has been made."-I4
The fundamental equation obtained from dimensional analy-
sis for one liquid phase takes the Pn = Pn (Re, Fr)
where the power number (Pn), the Reynolds number, and
the Froude number are here defined as
Pn = P/pw3d5 (5)
where P is the power input to the stirrer; p is the mass
density of the fluid; w is the angular velocity of the agita-
tor; p is the viscosity of the fluid; g is the acceleration of
gravity; and d is the impeller diameter. The dependence on
the Froude number can be relaxed as long as no vortices
are allowed to develop in the fluid: this is usually achieved
by baffling the inner surface of the reactor. Moreover,
viscous swirl should be avoided, for it leads to very low
mixing rates. The usual range of operation is characterized
by Reynolds numbers above lo4, where the form drag"
dominates over the friction drag on the tank wall. In this
turbulent range, the power number tends to be Reynolds
independent, as long as the ratio of impeller diameter to
tank diameter, D i , remains the same.17Miller and Mann'*
concluded that small and large mixers give equal mixing
rates at equal power inputs per unit volume (so, the larger
the reactor size, the larger the power consumption in stir-
ring, for a given volumetric flow rate). Mechanical and
building constraints usually lead to a constant value for the
ratio of diameter to length of an actual cylindrical reactor.
The angular velocity can then be computed through
which means that larger reactors require lower angular ve-
locities, for a given mixing rate. Oldshue" reported that
the capital cost of the stirrer for reactor i, Is,, is propor-
tional to the torque applied raised to the 0.8th power; since
the torque is the power divided by the tangential speed,
eq. (8) finally enables one to write
252 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 33, JANUARY 1989
(9)
Other possible expressions for the stirrer capital invest-i
ment as obtained from experimental data have been re-
ported.I9 The optimum ratio of impeller diameter to vessel
diameter for a given power input is strongly influenced by
the nature of the agitation pattern required. For the case of
reactions in the liquid phase, which depend on high circu-
lation rates rather than steep velocity gradients, large,
slow-moving impellers should be used.20 As emphasized
by Oldshue," the major cost of a mixer is quite often related
to the drive unit. If an economic criterion is to be followed
in order to find the optimum size for each of the reactors in
a series of CSTR's, for a given volumetric throughput,
then both equipment and operating costs (extending to the
whole useful industrial life of the process) should be taken
into account. A number of author^^'-^' suggest the utiliza-
tion of a fixed-factor exponential rule on the capacity ratio,
for scaling up of equipment construction costs. On the other
hand, assuming geometric similarity to be kept during scal-
ing up of cylindrical reactor sizes (as already stated), the
available area for heat transfer to or from the reacting mix-
ture does increase less than the reactor actual volume: there-
fore, larger reactors usually require extended areas for heat
exchange, as long as isothermal operation is concerned.
Another problem arising from larger reactor volumes is the
increasing first order time constant for the reactor dynam-
ics, which leads to longer and longer transients. These two
factors can then be ascribed a fictitious cost, expressed by
a higher value for the exponential factor, in the mentioned
exponential rule. This reasoning can be eventually extended
to the capital investment on the stirring device and the
operating costs themselves. The foregoing arguments sug-
gest a simple functional dependence of the whole invest-
ment on the size of the individual N reactors in the series,
expressed by
where all the listed factors were lumped together in
one single exponential factor, f,which is likely to assume
values above unity. Variable I * is the ratio of the global
investment on the series normalized by the required invest-
ment on one sole reactor, able to perform the same conver-
sion under the same operating conditions.
The optimal intermediate concentrations can be found
from the condition for an extremum of I * , as reported
elsewhere?
i = 1,2, . . . ,N - 1
Some algebraic work on eq. (1 1) yields amount of numerical work. The following equation meets

c,*)l-l
the foregoing requirements:
c:,w* + C,*)(C,*_,
[Cf(K* + C,*,,)(C,*--c,*,,>
clTopl [ -5.148f + 7.207f2 - 2.059f3
lOOK* - 5.148f + 7.207f - 2.059f3 I
-L
C:,(K*C,*_, + CT2) =

C,*2(K*+ C),:
The second derivative takes the form
x Crkopt,K’+O + exp
19.444f - 27.66f2 + 8.216f3
lOOOK *
x C:opI.K’+m
I(17)
The actual values for the overall investment can be easily
obtained from eq. (lo), by replacing the suitable values for
the intermediate concentrations, as obtained via eq. (17).
The dimensionless investment is bounded by the invest-
ments required for the same number of stages, in the
asymptoptic situations, according to
x [(f- 1)[1 + I$*+= < I * < (18)
where Z$+m and I:**o are obtained from eqs. (lo), (15),
2K*C,*_, and (16), to give
+ C,*’(C,*_,- C,*)(K* + C,*)

which proves to be always positive for f above unity, as

long as the physical constraint of having C,* between C ,:
Some values for the actual intermediate concentrations, to-
and C,: is taken into account. Therefore, eq. (12) has
gether with the corresponding values obtained by using the
only one solution in that range, which corresponds to a
reported correlation, are shown in Tables I, 11, and 111. In
minimum of I * .
each table, two different parameters were fixed, and the
If parameter f tends to unity, then eq. (12) reduces to remaining two were allowed to take values in the range of
c*
c,*2=c* 1-1 1+1 ( 14) interest, in order to enable a rough estimate of the leading
causes for the deviations between the actual and the corre-
as already obtained by Luyben and Tramper.5 The asymp- lated values to be obtained.
totic behaviors can be approached by making K* tend to
-
zero (K* + C,* C,*, or pseudo-zero-order kinetics), or DISCUSSION
-
to infinity ( K * + C,* K*, or pseudo-first-order kinet-
ics). The former condition applied to eq. (12) leads to The suggested correlation actually allows steps 2, 3, 4,
and 4a of the foregoing flow-scheme, to be skipped, thus
Z
C:opt,K*+O = 1 - - (1 -
N
c3 (15)
Table I. Intermediate optimal concentrations for a series of two, three,
and four CSTR’s, assuming C,* = 0.1 andf = 1.1. The first line con-
whereas the latter yields tains the actual values, the second contains the values obtained using the
correlation
C;opt,K*+m= G I ” (16)
N = 2 N=3 N=4
Solving eq. (12) in its generalized form for a given set of
values of K*, N, and C,* is a cumbersome task. A possible K* C: Ct C: c: CT c:
solving strategy is presented in the following flow-scheme:
0.001 0.5315 0.6913 0.3853 0.7711 0.5423 0.3144
1) DefineN.
0.5188 0.6799 0.3727 0.7682 0.5188 0.3037
2) Arbitrate C:,. 0.01 0.4354 0.6431 0.3171 0.7427 0.4922 0.2635
3) Compute each C,*, for i = 2 , 3 , . . . ,N [by a Newton 0.4145 0.5825 0.2885 0.6881 0.5145 0.2368
method using eq. (12)l. 0.1 0.3362 0.5184 0.2369 0.6404 0.3718 0.1985
4) Compare the obtained value for C,* to the given C,*. 0.3295 0.4804 0.2253 0.5798 0.3295 0.1858
1 0.3189 0.4714 0.2181 0.5736 0.3233 0.1803
4a) If the error is above a predefined convergence limit,
0.3176 0.4658 0.2165 0.5642 0.3176 0.1787
iterate from step 2, with a new estimate for C Z , (by a 10 0.3166 0.4651 0.2158 0.5634 0.3169 0.1781
Golden Search Method). 0.3164 0.4643 0.2155 0.5625 0.3164 0.1779
4b) If the error is acceptable, iterate from step 1, with 100 0.3163 0.4629 0.2155 0.5624 0.3162 0.1778
the next integer value for N. 0.3162 0.4642 0.2154 0.5624 0.3162 0.1778
loo0 0.3163 0.4642 0.2155 0.5623 0.3161 0.1777
A suitable correlation for the quick estimation of the in-
0.3162 0.4642 0.2154 0.5623 0.3162 0.1778
termediate optimal concentrations would readily reduce the

MALCATA: CSTR’S PERFORMING MICHAELIS-MENTEN REACTIONS 253

Table 11. Intermediate optimal concentrations for a series of two, three, practice. As parameterf increases, the variation of the cor-
and four CSTR’s, assumingf = 1.2 and K * = 0.1. The first line con- responding intermediate concentrations with K * tends to
tains the actual values, the second contains the values obtained using the
show a steeper and steeper behavior. The parameters for
correlation.
the correlation were found by non-linear fitting to a large
N = 2 N=3 N=4 number of cases, ranging from very low to very high val-
-
ues for K * . The validity conditions require that f be com-
c; Cf Cf C,* Cf c: c: prised between 1 and 2 (which covers a large number of
0.10 0.3516 0.5492 0.2508 0.6704 0.3794 0.2100 situations). Extrapolation beyond this limit is, then, to be
0.3422 0.4959 0.2347 0.5963 0.3422 0.1934 made with the necessary precautions.
0.20 0.4951 0.6623 0.3879 0.7519 0.5278 0.3401
The investment on the series of CSTR’s is a monotoni-
0.4733 0.6158 0.3624 0.7018 0.4733 0.3164
0.30 0.5942 0.7302 0.4916 0.8002 0.6127 0.4434 cally decreasing function of the number of reactors. When
0.5745 0.7055 0.4701 0.773 I 0.5745 0.4248 this number tends to infinity, both eqs. (19) and (20) tend
0.40 0.6708 0.7810 0.5782 0.8368 0.6804 0.5336 to zero, therefore the actual investment on the series of
0.6603 0.7683 0.5667 0.8285 0.6603 0.5247 CSTR’s also tends to zero [see eq. (18)], for the standard
0.50 0.7354 0.8236 0.6558 0.8681 0.7399 0.6167
case of Michaelis-Menten kinetics. Industrial practice has,
0.7363 0.8260 0.6559 0.8746 0.7363 0.6188
0.60 0.7932 0.8620 0.7282 0.8966 0.7951 0.6960 however, for long proved that only a few reactors are eco-
0.8053 0.8766 0.7395 0.9144 0.8053 0.7085 nomically feasible. The departure of the concIusions drawn
out through the analysis presented, from the commonly ac-
cepted heuristic rules is mainly due to the lack of con-
Table 111. Intermediate optimal concentrations for a series of two,
three, and four CSTR’s, assuming C; = 0.20 and K * = 0.01. The first stancy of parameterf, when the final desired volume is
line contains the actual values, the second contains the values obtained very different from the volume on which the estimation of
using the correlation. f was performed. Moreover, a large number of reactors
N=2 N=3 N=4
usually brings about other operating problems, such as the
- difficulty in cleaning the inside surfaces (scaling is likely
f c: c: c: c: c: c: to occur when organic solutions are handled) and the slug-
1.1 0.5586 0.7125
gishness of the system response to feedback control actions
0.4368 0.7880 0.5791 0.3783
0.5433 0.6981 0.4174 0.7890 0.5433 0.3630 (overdamped behavior).26Therefore, for large numbers of
1.2 0.5743 0.7208 0.4479 0.7926 0.5872 0.3862 reactors, the correlation presented loses its validity.
0.5900 0.7515 0.4543 0.8443 0.5900 0.3941 Additional restraints to the final design, such as catalyst
1.3 0.5809 0.7238 0.4524 0.7943 0.5902 0.3895 deactivation and nonidealities in the mixing pattern as-
0.6122 0.7754 0.4720 0.8678 0.6122 0.4088
1.4 0.5839 0.7252 0.4545 0.5914
sumed, must be taken into account if an accurate design is
0.7949 0.3908
0.6222 0.7849 0.4802 0.8760 0.6222 0.4154 required. In particular, the disadvantages associated with
1.5 0.5857 0.7260 0.4557 0.7954 0.5921 0.3916 having reactors of different sizes, such as the costs of engi-
0.6264 0.7880 0.4838 0.8777 0.6264 0.4181 neering design, fabrication and inventory of spare parts’
1.6 0.5870 0.7266 0.4566 0.7957 0.5927 0.3923 have to be carefully weighed. For the cases where a rough
0.6284 0.7890 0.4856 0.8775 0.6284 0.4195
prediction of the reactor sizes leading to minimum overall
investment over the useful life of the reactors is enough,
speeding up the calculating procedure. Simple inspection the correlation presented proves useful: being straightfor-
allows one to find that the correlation satisfies the asymp- ward, applicable, and being especially suitable for compu-
totic conditions expressed by eqs. (14), ( 1 9 , and (16). tational designing strategies (as a source of initial estimates
The required dimensionless investment for Michaelis- for the iterative numerical routines) are probably their most
Menten standard kinetics is lower bounded by the invest- remarkable features.
ment required if the reaction could be assumed first-order
kinetics, and upper bounded by the investment if zero-order NOMENCLATURE
behavior were approached. Although the totaI investment
on a series of CSTR’s monotonically decreases with the C concentration at the outlet of the reactor (mol/m3)
C* dimensionless concentration at the outlet of the reactor
number of stages selected, this trend is much faster for
(CICO)
larger kinetic constants. Another interesting fact is that the d impeller diameter (m)
functional dependence of I * on K * for the asymptotic be- D Reactor diameter (m)
haviors actually disappears, according to eqs. (19) and (20). Da Damkholer number for the reactor (Vvm/QCo)
In the functional form selected for the correlation, a f exponential factor for capacity-based cost scaling
Fr Froude number (02d/g)
compromise solution was taken between accuracy of the I* reactor overall investment on the series, scaled by the in-
prediction and formal simplicity, in order to make it suit- vestment required for one single reactor
able for predesign steps. The average relative error associ- Is capital cost of the stirrer (currency units)
ated to the correlated values is about 4%, a maximum error Km Michaelis-Menten kinetic constant (mol/m3)
of 11% being expected to hold for the largest conversions, K* dimensionless Michaelis-Menten kinetic constant ( K , /Co)
8 acceleration of gravity (m/s2)
associated with the highest values for parameterf. Even in N number of reactors in the series
the worst cases, the error associated with the correlated P power input to the stirrer (W)
values falls within acceptable limits for common industrial Pn Power number ( P / p o 3 d 5 )

254 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 33, JANUARY 1989

 Q volumetric flow rate (m3/s) 5 A.M. Luyben and J. Tramper, Biotechnol. Bioeng., 24, 1217 (1982).
Re Reynolds number ( w d Z p / p ) 6 L. Gulyas, Magyar Kdm. Lapja, 38, 220 (1983).
V effective volume of the reactor (m3) 7 Y. H. Lee, M. M. Gharpuray, and L. T. Fan, Biotechno/. Bioengng.
"m maximum kinetic rate (mol/m3s) Symp., 12, 121 (1982).
8. F. X . Malcata, Can. J . Chem. Eng., 66, 168 (1988).
Subscripts
9. T. N. Zwietering, Chem. Eng. Sci., 11, 1 (1959).
1 if only one reactor were employed 10. R. B . Bird, W. E. Stewart, and E. N. Lightfoot, Tranport Phenomena
I referring to the ith reactor (Wiley, New York, 1960).
N referring to the last reactor in the series 11. J . H. Rushton, Chem. Eng. Prog., 47, 485 (1951).
0 referring to the inlet feed to the first reactor 12. J. H. Rushton, Chem. Eng. Prog., 44, 2931 (1952).
opt referring to intermediate optimal conditions, according to 13. W. Buche, 2. Ver. deut. Ing., 81, 1065 (1937).
the objective function 14. A. Brothman and H. Kaplan, Chem. Met. Eng., 46, 633 (1939).
K* -+ 0 when K* is very small compared to C* 15. R. E. Johnstone and M. W. Thring, Pilot Plants. Models, and Scale-up
K* + x when K* is very large compared to C* Methods in Chemical Engineering (McGraw-Hill, New York, 1957).
Greek letters 16. J. H. Rushton, E. W. Costich, and H. J. Everett, Chem. E n g . Prog.,
46, 395 (1950).
P viscosity of the fluid (PJ 17. V. W. Uhl and J. B. Gray, Mixing-Theory and Practice (Academic,
P mass density of the fluid (kg/m3) New York, 1966).
w angular velocity of the agitator (sf') 18. S. A. Miller and C. A. Mann, Trans. AIChE, 40, 709 (1944).
19. J. Y. Oldshue, Ffuid Mixing Technology (McGraw-Hill, New York,
1983).
20. W. L. McCabe and J. C. Smith, Unit Operations of Chemical Engi-
References neering (McGraw-Hill, New York, 1976).
21. C. A. Miller, Chem. Eng. Prog., 69, 77 (1973).
1. C . G. Hill, An Introduction to Chemical Engineering Kinetics and 22. J. D. Chase, Chem. Eng., 77 (1970).
Reactor Design (Wiley, New York, 1977). 23. G. J. Hoerner, Chem. Eng., 83, 141 (1976).
2. R. Aris, The Optimal Design of Chemical Reactors (Academic, New 24. G. Enyedy, Chem. Eng. Prog., 67, 73 (1971).
York, 1961). 25. M. S. Peters and K . D. Timmerhaus, Plant Design and Economics
3. 0. Levenspiel, Chemical Reaction Engineering, (Wiley, New York, for Chemical Engineers (McGraw-Hill, New York, 1980).
1962). 26. G. Stephanopoulos, Chemical Process Control- an Introduction to
4. K. B. Bischoff, Can. J . Chem. Eng.. 44, 281 (1966). Theory and Practice (Rentice-Hall, Englewood Cliffs, NJ, 1984).

MALCATA: CSTR'S PERFORMING MICHAELIS-MENTEN REACTIONS 255