11564 Ind. Eng. Chem. Res.

2010, 49, 11564–11571

Heuristic Design of Reaction/Separation Processes

William L. Luyben*
Department of Chemical Engineering, Lehigh UniVersity, Bethlehem, PennsylVania 18015, United States

The dominant trade-off in chemical process design is between reactor size and recycle flow rate. Big reactors
require larger capital investments in vessels and catalyst, but they result in smaller recycle flow rates for a
given yield, which means lower capital investments and energy costs in the separation section of the process.
Small reactors have reverse effects. Therefore, an economic optimum exists that balances the costs of the
reaction section and the separation section of the process. This paper presents a heuristic approach to quickly
determine this optimum trade-off during preliminary conceptual design. The basic idea is to start with a very
large reactor and find the recycle flow rate required to meet some specified conversion/yield/selectivity criterion.
This is the minimum recycle flow rate. Then a heuristic similar to that used in distillation column design is
employed. The actual recycle flow rate is set equal to 1.2 times the minimum, and the reactor and separation
sections are designed with this recycle. The heuristic recycle ratio has some dependence on the phase equilibrium
(decreases as relative volatility decreases) and catalyst cost (increases as catalyst cost increases). The process
studied has a continuous stirred-tank reactor, two distillation columns, and one recycle stream. Two consecutive
reactions (A + B f C and A + C f D) produce a desired product C and an undesired product D. Achieving
high selectivity requires low concentrations of A and C, so there is a large recycle of mostly B. Relative
volatilities are assumed to be RA > RB > RC > RD, so there is one recycle from the overhead of the first
distillation column containing unreacted A and B. The second column separates C and D. A second process
is also studied involving an actual chemical system.

1. Introduction 2. Process Studied

In the design of distillation columns, there is a classical trade- Figure 1 shows the flowsheet of the process. The equipment
off between the number of trays and the reflux ratio required to sizes and conditions shown are the economic optimum devel-
achieve a specified separation. Simple heuristic optimization oped later in this paper for the base-case conditions: a selectivity
rules are widely and effectively used to quickly find an specification of 100 (kilomoles of product C produced divided
approximate economic design at the conceptual design stage by kilomoles of product D produced), kinetic parameters of k1
of the project. The two most often applied alternative heuristics ) 25 and k2 ) 1, relative volatilities between adjacent
are (1) to set the actual number of trays equal to twice the components (R) of 2, and catalyst price of $10 per kg.
minimum number and (2) to set the actual reflux ratio equal to 2.1. Reactor. The molar holdup of the continuous stirred-
1.2 times the minimum reflux ratio. tank reactor is 35 kmol. Two irreversible reactions occur with
In many processes, either of these two alternative rules specific reaction rates k1 and k2.
produces a design that is reasonably close to the real economic
optimum, which can be found by rigorously evaluating designs A+BfC
R1 ) VRk1zAzB (1)
with varying numbers of trays and finding the design that
minimizes total annual costs (capital plus energy). Using more
trays increases column height and capital investment in the A+CfD
vessel, but more trays reduce energy cost and capital investment R2 ) VRk2zAzC (2)
in heat exchangers.
Many chemical processes feature a reaction section coupled where zj ) mole fraction of component j in the reactor and VR
with a separation section through recycle streams. In the design ) reactor molar holdup.
of these multiunit processes, the dominant design issue is the This type of basic kinetic system is very commonly encoun-
classical trade-off between the size of the reactor and the amount tered in many real industrial processes. The desired product C
of recycle required to achieve a specified conversion, yield, or is formed by the first reaction, but it can react further to produce
selectivity in the overall process. Major capital and operating an undesired product D. Therefore, the concentrations of A and
costs are often in the separation section, but the performance C must be kept small to achieve the desired selectivity by
of the reaction section is usually critical because of the dominant running with an excess of B, thus requiring a large recycle.
economic effect of raw material costs. Two fresh feed streams of pure A and pure B are fed to the
It would be useful to have a simple heuristic approach to reactor (FA0 and FB0) in addition to the recycle stream from the
this trade-off that could provide some guidance for initial top of the first column (D1).
flowsheet development at the conceptual process design stage. Selectivity is defined as the moles of the desired component
The purpose of this paper is to propose such a simple heuristic C produced divided by the moles of the undesired component
method. The effect of recycle on the economics of the process D produced.
is explored.
moles of C xB1C
* E-mail: [email protected]. Phone: 610-758-4256. Fax: 610-758- Selectivity ) ) (3)
moles of D xB1D
5057.

10.1021/ie101509w  2010 American Chemical Society

Published on Web 10/04/2010
 Ind. Eng. Chem. Res., Vol. 49, No. 22, 2010 11565

Figure 1. Flowsheet (k1 ) 25, selectivity ) 100, $10/kg, R ) 2).

where xB1j ) mole fraction of component j in the bottoms stream estimating the capital costs (column shell, condenser, and
of the first distillation column C1. reboiler) and the energy cost (reboiler heat input at $7.78 per
The fresh feed of reactant A is fixed at FA0 ) 100 kmol/h in GJ). Standard column sizing and cost relationships are used.1,2
all cases. The fresh feed of B is calculated for each case by The capital cost of the column is based on the size of the vessel,
solving the nine nonlinear simultaneous algebraic equations getting the height from the number of theoretical trays with a
describing the reactor and column for the specified conditions 2-ft tray spacing and the diameter from Aspen’s tray sizing
(reactor size, desired selectivity, and kinetic parameters). option.
The equations describing the reactor are The equations describing the column are given below. They
are solved simultaneously with those of the reactor to find the
FA0 + D1xD1A ) F1zA + R1 + R2 (4) unknowns.
FB0 + D1xD1B ) F1zB + R1 (5) F1 ) D1 + B1 (9)

D1xD1C ) F1zC - R1 + R2 (6) F1zA ) D1xD1A (10)

0 ) F1(1 - zA - zB - zC) - R2 (7) F1zB ) D1xD1B + B1xB1B (11)

The composition of C in the D1 recycle is assumed to be F1zC ) D1xD1C + B1xB1C (12)

xD1C ) 0.001. The five specified variables are FA0, VR, xD1C, k1,
and k2. The eight unknown variables in these four equations The additional unknown variable not in the previous list of
are FB0, zA, zB, zC, D1, xD1A, xD1B, and F1. Note that the reactions eight is xB1C. Therefore, we have a system of nine unknowns.
are nonequimolar, so there is a reduction in the molar flows But eqs 4-7 and eqs 9-12 give us only eight equations.
into and out of the reactor. Each reaction consumes 2 mol of 2.3. Solution Method. The method for solving these equa-
reactants while making 1 mol of product. tions for the reactor and column linked together with the recycle
2.2. Column C1. The reactor effluent is fed to a distillation stream is to fix the total flow rate of the fresh feed of B plus
column whose job is to recycle unreacted A and B back to the the recycle D1.
reactor and send a mixture of products C and D downstream to
the second distillation column. Constant relative volatilities are Total ) FB0 + D1 (13)
assumed with reactant A and B lighter than products C and D.
With the variable Total fixed, eq 13 provides an additional
RA > R B > R C > R D (8) equation. The Matlab function fsolVe was used to solve these
nine nonlinear simultaneous algebraic equations for a given case.
With these phase equilibria relationships, the separation Then an iterative procedure (interval halving) was used to
section has one recycle from the top of column C1. change Total until a desired specification (Selectivity) was
The sizing and cost analysis for the column use the Fenske achieved for the case (fixed VR, k1, and k2).
equation to find the minimum number of trays and the One of the reviewers of this paper suggested an alternative
Underwood equations to find the minimum reflux ratio. The solution method that avoids the iteration by using the definition
separation is between components B and C. All A is assumed of selectivity (eq 3) as one of the equations instead of the
to go overhead (xB1A ) 0). The impurity of B in the bottoms is equation for Total. This method was tested and works well.
set at xB1B ) 0.001. Figure 2A shows results from these calculations for the case
Then the number of trays is set at twice the minimum, and when a selectivity of 100 is specified with three different kinetic
the required reflux ratio is set at 1.2 times the minimum for parameters k1 ) 15, 25, and 50. The k2 is 1 in all cases. The
11566 Ind. Eng. Chem. Res., Vol. 49, No. 22, 2010

Figure 2. (A) Required recycle and (B) other variables for selectivity ) 100.

ordinate is the recycle flow rate D1 (Total - FB0). The abscissa The curves shown in Figure 2A are similar to those seen in
is reactor molar holdup. the classical plots of reflux ratio versus trays in distillation
More recycle is needed to achieve a specified selectivity for design. At any point on the curve, the products from the process
a fixed reactor size as the specific reaction rate k1 is reduced. are exactly the same, but equipment (reactor and column) and
Small values of k1 lead to larger concentrations of A in the energy (reboiler heat input) are different. So where is the
reactor (zA) but smaller concentrations of C (zC) and D (zD). optimum point on the curve? We address this question next for
The key feature of these plots (and the basis for the heuristic a number of different cases to see if there is some simple
proposed) is that the required recycle flow rate levels out at relationship between the economic optimum recycle flow rate
some minimum value as reactor size is made very large. We and the minimum. The minimum recycle flow rate can be easily
define the asymptotic recycle flow rate as the minimum recycle. determined by running a case with a very large reactor.
It can be expressed as a ratio to the fresh feed of A to put it in
dimensionless form. For example, for the k1 ) 25 case, the 3. Economic Optimization
minimum recycle is 425 kmol/h for a fresh feed of FA0 ) 100
kmol/h. So the minimum recycle ratio Rmin is 4.25. One of the famous “Douglas1 doctrines” is that energy and
Figure 2B gives other important variables. Notice that the capital can usually be spent to improve yield and selectivity.
reactor compositions all level out as reactor size is increased. This result is easily demonstrated by comparing the annual cost
 Ind. Eng. Chem. Res., Vol. 49, No. 22, 2010 11567
of raw materials and the annual value of products with annual Table 1. Sizing and Economics for Selectivity ) 100
capital and energy costs. The former are typically orders-of- k1 ) 15 h-1
magnitude larger than the latter. This translates into designs for
VR (kmol) 50 55 60
most processes that feature very high conversion, yield, and recycle D1 (kmol/h) 842.3 818.0 799.8
selectivity. We cannot afford to waste raw materials and incur reactor (106 $) 0.2652 0.2814 0.2971
expensive waste disposal problems by producing undesired catalyst (106 $) 1.250 1.375 1.500
products, so yields of 99% and selectivities of 100 are typical. column diameter (m) 4.315 4.275 4.245
The precise “best” values for these criteria are strongly column (106 $) 1.185 1.175 1.168
heat exchangers (106 $) 1.078 1.065 1.055
dependent on market prices for raw materials and finished total capital (106 $) 3.777 3.896 4.020
products. Estimates of market prices are often far from reliable. total energy (106 $/yr) 2.752 2.701 2.663
The predictive success of marketing analysts is usually lower TAC (106 $/yr) 4.011 4.000 4.003
than that of the meteorologist’s success at forecasting tomorrow
k1 ) 25 h-1
weather or the stock market analyst’s success at predicting what
the stock market will do in the next year. VR (kmol) 30 35 40
In order to avoid these uncertainties, we take the approach recycle D1 (kmol/h) 529.8 500.8 482.6
reactor (106 $) 0.1930 0.2124 0.2308
that the designer will specify a reasonable criterion, such as a catalyst (106 $) 0.7498 0.8747 1.000
selectivity ) 100. Then a plant can be designed that meets this column diameter (m) 3.452 3.394 3.356
criterion with the minimum total annual cost (TAC). Pricing of column (106 $) 0.9312 0.9176 0.9090
feed streams and products is avoided. heat exchangers (106 $) 0.8063 0.7886 0.7774
total capital (106 $) 2.680 2.793 2.917
Total annual cost is the sum of the energy cost plus the annual total energy (106 $/yr) 1.761 1.702 1.665
cost of the capital investment using a payback period. In this TAC (106 $/yr) 2.655 2.633 2.637
work, a payback period of 3 years and the installed cost of
equipment are used. k1 ) 50 h-1
VR (kmol) 15 20 25
Capital Installed Investment recycle D1 (kmol/h) 310.3 264.8 246.3
TAC ) Energy Cost + reactor (106 $) 0.1253 0.1499 0.1723
Payback Period
catalyst (106 $) 0.3749 0.5000 0.6248
(14)
column diameter (m) 2.659 2.544 2.496
column (106 $) 0.6980 0.6724 0.6620
For the flowsheet considered in Figure 1, we are concerned heat exchangers (106 $) 0.5743 0.5421 0.5290
with the reactor and the first column. The second column does total capital (106 $) 1.772 1.864 1.988
not change from case to case for a specified selectivity. Capital total energy (106 $/yr) 1.045 0.9564 0.9209
TAC (106 $/yr) 1.636 1.578 1.584
investment is required for the reactor vessel, catalyst, column
vessel, reboiler, and condenser. Energy cost depends on reboiler
duty. fixed. The yield is reduced from 98 to 96%, since more of
Table 1 gives sizing and costing results for several cases with reactant A ends up in undesired product D instead of in desired
selectivity fixed at 100, which corresponds to a yield of 98%. product C. Optimum reactor sizes do not change much, but
The cost of catalyst is $10 per kg, and the relative volatility recycle flow rates are much lower. However, minimum recycle
between all adjacent components is 2. Three values of k1 (15, flow rates are also smaller, so how is the ratio of optimum to
25, and 50 h-1) are explored. In all cases, the required recycle minimum affected?
decreases as reactor size increases. This reduces energy cost,
With the selectivity set at 50, the minimum recycle flow rates
but the larger reactor increases total capital cost.
for the three values of k1 (as shown in Figure 3A) are 360
The optimum reactor size and the optimum recycle flow rate
kmol/h for k1 ) 15 h-1, 220 kmol/h for k1 ) 25 h-1, and 110
decrease as k1 increases. The optimum designs are indicated in
kmol/h for k1 ) 50 h-1. The economic optimum recycle flow
Figure 2A by the stars. Notice that the minimum recycle flow
rates are 447.2 kmol/h for k1 ) 15 h-1, 277.0 kmol/h for k1 )
rates for the three values of k1 are 720 kmol/h for k1 ) 15 h-1,
25 h-1, and 145.8 kmol/h for k1 ) 50 h-1. The three ratios of
425 kmol/h for k1 ) 25 h-1, and 210 kmol/h for k1 ) 50 h-1.
optimum-to-minimum are 1.24, 1.26, and 1.33. The ratio
The economic optimum recycle flow rates are 818 kmol/h for
increases slightly as selectivity is lowered.
k1 ) 15 h-1, 500.8 kmol/h for k1 ) 25 h-1, and 264.6 kmol/h
Figure 4 shows what happens when the selectivity is increased
for k1 ) 50 h-1. The three ratios of optimum-to-minimum are
from 100 to 150 with the other parameters fixed. The yield is
1.14, 1.18, and 1.26.
increased from 98 to 99%. Optimum reactor size does not
These results suggest that the proposed heuristic of 1.2 may
change much, but recycle flow rates are much higher. However,
be valid for preliminary conceptual design.
minimum recycle flow rates are also larger, so how is the ratio
of optimum to minimum affected?
4. Other Cases
With the selectivity set at 150, the minimum recycle flow
A number of other cases were explored to see the impact of rates for the three values of k1 (as shown in Figure 4A) are
other design parameters on the heuristic ratio. We would expect 1100 kmol/h for k1 ) 15 h-1, 640 kmol/h for k1 ) 25 h-1, and
that a more expensive catalyst would shift the optimum to 320 kmol/h for k1 ) 50 h-1. The economic optimum recycle
smaller reactors and larger recycle flow rates. We would expect flow rates are 1189 kmol/h for k1 ) 15 h-1, 709.1 kmol/h for
that more difficult separations (smaller relative volatilities) k1 ) 25 h-1, and 364.3 kmol/h for k1 ) 50 h-1. The three ratios
would shift the optimum to large reactors and smaller flow rates. of optimum-to-minimum are 1.11, 1.08, and 1.14. The ratio
These are studied below. First we see if the specified selectivity decreases slightly as selectivity is raised.
has any effect on the optimum-to-minimum recycle ratio. These results indicate that a heuristic of 1.2 should be fairly
4.1. Selectivity. Figure 3 shows what happens when the adequate for initial estimates in conceptual design over a wide
selectivity is reduced from 100 to 50 with the other parameters range of selectivities.
11568 Ind. Eng. Chem. Res., Vol. 49, No. 22, 2010

Figure 3. (A) Required recycle and (B) other variables for selectivity ) 50.

In the next section we explore the effect of other design that the optimum-to-minimum recycle ratio increases from 1.2
parameters that are expected to cause some shift in the heuristic up to 1.4 as the catalyst increases in cost by a factor of 5. So
ratio. Note that the minimum recycle flow rate depends on the the heuristic value will depend on the cost of the catalyst.
values of k1 and selectivity specified and not on catalyst price 4.3. Relative Volatility. A relative volatility of 2 between
or relative volatility. adjacent components has been used in all the previous cases
4.2. Catalyst Cost. A catalyst cost of $10 per kg has been (RA ) 8, RB ) 4, RC ) 2, RD ) 1). Decreasing relative
used in all the previous cases. Increasing the price of the catalyst
volatilities will make separations more difficult and the separa-
should make the reaction section more expensive, so the
tion section more expensive, so the economic optimum would
economic optimum would be expected to shift to somewhat
higher recycle flow rates. Figure 5 shows that this is what occurs. be expected to shift to somewhat smaller recycle flow rates.
These results use the base-case values of selectivity ) 100, k1 Figure 6 gives results for a number of cases in which the
) 25, and relative volatility of 2. relative volatilities are modified. The abscissa is relative
The optimum reactor size decreases and the recycle flow rate volatility between adjacent components. For example, a relative
increases as the catalyst becomes more expensive. Naturally, volatility of 1.2 means that RA ) 1.728, RB ) 1.44, RC ) 1.2,
TAC also increases. The lower right graph in Figure 5 shows and RD ) 1.
 Ind. Eng. Chem. Res., Vol. 49, No. 22, 2010 11569

Figure 4. (A) Required recycle and (B) other variables for selectivity ) 150.

Optimum reactor sizes, energy consumption (a direct indica- heuristic ratio should be used (1.1 instead of 1.2). The next
tion of recycle flow rate), and TAC all increase as relative section considers such a case.
volatilities are made smaller. The lower right graph in Figure 6
shows that the optimum-to-minimum recycle ratio decreases 5. Real Example
from 1.2 down to less than 1.1 as relative volatility decreases
from 2 to 1.2. Therefore, a lower value for the heuristic should Hypothetical chemical examples have been considered in
be used when separations are more difficult. previous sections of this paper. To see if the proposed heuristic
4.3. Separation Sequence. The flowsheet shown in Figure method works on a real chemical example, the process to make
1 features a single recycle stream coming from the first column the gasoline additive 2-methoxy-2-methylheptane (MMH) is
because the volatilities of the reactants A and B are bigger than studied. The optimum design of this process was presented in
the volatilities of the products C and D. In some systems, the a recent paper.3 The chemistry involves the liquid-phase
phase equilibria are such that one of the products has a volatility reversible reaction of methanol with 2-methyl-1-heptene (MH)
that is larger than or intermediate between the volatilities of to form MMH. However, methanol and MH also undergo an
the two reactants. In that situation, two recycle streams would undesirable reaction to form dimethyl ether (DME) and 2-meth-
be required and more distillation columns would be needed. yl-2-heptanol (MHOH). Thus the system has two undesirable
The separation costs would therefore be larger and a lower products.
11570 Ind. Eng. Chem. Res., Vol. 49, No. 22, 2010

Figure 5. Effect of catalyst price.

Figure 6. Effect of relative volatiliy.

Figure 7 shows the flowsheet with the economic optimum The yield of the process is defined as the flow rate of the
design conditions and equipment sizes. The first undesirable MMH product stream from the second column (D2) divided by
byproduct (DME) is very volatile, so it is removed in the first the fresh methanol feed. In the optimum design D2 is 49.02
column. Then the excess reactants are removed out the top of kmol/h, giving a yield of 0.9803 when the reactor is 12 m3.
the second column for recycle back to the reactor. The desired
Simulation runs were made with increasing larger reactors
product MMH is separated from the second undesired byproduct
(MHOH) in a third column. So the separation section is more while holding the yield of MMH at 0.9803 by varying the
complex and inherently more expensive than the simple system recycle flow rate. A reactor size of 15 m3 resulted in a lower
shown in Figure 1. Therefore, a somewhat small heuristic recycle flow rate (D2 ) 75.99 kmol/h). A reactor size of 20 m3
optimum-to-minimum recycle ratio is expected. resulted in an even lower recycle flow rate (D2 ) 73.69 kmol/
The economic optimum design has 50 kmol/h of fresh h). Increasing reactor size still further made no change in the
methanol and a recycle flow rate of 79.99 kmol/h with a 12 m3 required recycle flow rate. Therefore, the minimum recycle flow
reactor. The specifications for the various streams were described rate is about 74 kmol/h while the economic optimum recycle
in detail previously.3 flow rate is about 80 kmol/h.
 Ind. Eng. Chem. Res., Vol. 49, No. 22, 2010 11571

Figure 7. MMH flowsheet.

The correct optimum-to-minimum recycle ratio in the MMH design with plots of reflux ratio versus number of trays. Both
process is 1.08, which is very close to the 1.1 that the proposed cases demonstrate that energy is more costly than capital, so
heuristic method would have used in the light of the more the optimum designs are near the minimum energy regions of
complex separation section. the curves since both reflux and recycle represent primarily
energy costs.
6. Conclusion
Literature Cited
A simple heuristic approach to determine economic optimum
designs of chemical process with reaction and separation (1) Douglas, J. M. Conceptual Design of Chemical Processes; McGraw-
sections has been proposed and tested. The chemistry assumes Hill: New York, 1988.
(2) Turton, R., Bailie, R. C., Whiting, W. B., Shaelwitz, J. A. Analysis,
completing reactions with desired and undesired products being Synthesis and Design of Chemical Processes, 2nd ed.; Prentice Hall: New
formed that force a design with recycle to attain specified yield York, 2003.
or selectivity. The minimum recycle flow rate is determined by (3) Luyben, W. L. Design and control of the methoxy-methyl-heptane
using very large reactors. Then the actual recycle flow rate is process. Ind. Eng. Chem. Res. 2010, 49, 6164–6175.
set at 1.1 to 1.2 times the minimum, depending on the cost of
catalyst, the relative volatilities, and the complexity of the ReceiVed for reView July 14, 2010
separation section. ReVised manuscript receiVed September 6, 2010
Accepted September 19, 2010
The optima are at the knee of the curve of recycle versus
reactor size curve. This is analogous to results in distillation IE101509W