Computers and Chemical Engineering 23 (1999) 479 – 495

A case study for reactor network synthesis: the vinyl chloride

process
A. Lakshmanan, W.C. Rooney, L.T. Biegler *,1
Chemical Engineering Department, Chemical Mellon Uni6ersity, Pittsburgh, PA 15213, USA

Received 11 February 1998; received in revised form 3 July 1998

Abstract

A key objective of the integrated reactor network synthesis approach is the development of waste minimizing process flowsheets
(Lakshmanan & Biegler, 1995). With increasing environmental concerns in process design, there is a particularly strong need to
maximize conversion to product and avoid generation of wasteful byproducts within the reactor network. This also avoids
expensive treatment and separation costs downstream in the process. In this study, we present an application of the mixed integer
nonlinear programming (MINLP)-based reactor network synthesis strategy developed by Lakshmanan and Biegler (1996a). Here
we focus on applying these reactor network synthesis concepts to the vinyl chloride monomer production process. Vinyl chloride
is currently produced by a balanced production process from ethylene, chlorine and oxygen with three separate reaction sections:
oxychlorination of ethylene; direct chlorination of ethylene; and pyrolysis of ethylene dichloride. The hydrogen chloride produced
in the pyrolysis reactor is used completely in the oxychlorination reactor. Byproducts such as chlorinated hydrocarbons and
carbon oxides are generated by these reaction sections. These are studied using reaction kinetic models for the three reaction
sections. The case study results in optimal reactor networks that improve the conversion of ethylene to vinyl chloride and
minimize the formation of byproducts. These results are used to generate an improved flowsheet for the production of vinyl
chloride monomer. Moreover, an overall profit maximization, that includes the effect of heat integration, is presented and a set
of recommendations that improve the selectivity of vinyl chloride production are outlined. Finally, the optimal reactor structures,
overall conversion and annual profit are shown to be only mildly sensitive with respect to small changes in the kinetic parameters.
© 1999 Elsevier Science Ltd. All rights reserved.

1. Introduction reactor network synthesis strategy that combines princi-

In most chemical processes the selective conversion of
raw materials in the reactor determines the composition
and the amount of waste products produced, and the
reactor network has a significant influence on the recycle
structure and downstream processing steps. Hence, re-
action systems and reactor design often determine the
character of the flowsheet and better performance of the
reactor system improves the performance of the entire
process. In the past decade, several researchers have
worked on problems of reactor network synthesis and
reactor design. This problem involves nonlinear reaction
models, uncertain rate laws and numerous possible
reactor types and networks. Hence, researchers in this
area have achieved limited success. Motivated by these
thoughts Lakshmanan and Biegler (1996a) proposed a
* Corresponding author.
1
Present address: Aspen Technology, Cambridge, MA02141, USA.
ples developed for MINLP optimization of prepostu-
lated reactor superstructures and geometric techniques
for establishing the reactor network.
In particular, geometric techniques construct an at-
tainable region, which represents the limits that can be
achieved by the processes of reaction and mixing. This
region includes the performance of any reactor network
structures in terms of conversion of raw materials to
desired products and the amount of waste products that
are produced. Glasser, Crowe and Hildebrandt (1987)
and Hildebrandt, Glasser and Crowe (1990) proposed
constructive techniques to map the attainable region for
reactor network synthesis problems in two or three
dimensions. These techniques are useful to develop new
reactor structures and to understand the reactor net-
work synthesis problem, but it is difficult to map these
regions in more than three dimensions by using the
geometric approach. On the other hand, the optimiza-
tion based approach of Lakshmanan and Biegler

0098-1354/99/$ - see front matter © 1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 9 8 - 1 3 5 4 ( 9 8 ) 0 0 2 8 7 - 7
480 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495
(1996b) overcomes this dimensionality problem and
allows for easy integration of the reactor network syn-
thesis problem with other process subsystems such as
separation, heat integration and waste treatment. This
paper applies this integrated reactor network synthesis
strategy to the vinyl chloride monomer production
process.
Vinyl chloride monomer is one of the largest com-
modity chemicals produced in the United States. One of
the first patented processes to produce vinyl chloride
involved reacting hydrogen chloride and acetylene in
the presence of a mercuric chloride catalyst. Nowadays,
it is produced commercially by pyrolytic decomposition
of ethylene dichloride (EDC). EDC may be produced
by the direct chlorination of ethylene or oxychlorina-
tion of ethylene in the presence of oxygen and hydrogen
chloride. Pyrolysis of EDC produces vinyl chloride and
an equal amount of hydrogen chloride as a co-product,
which is one of the reactants in the oxychlorination of
ethylene. Hence, the hydrogen chloride produced in the
pyrolytic reactor is recycled to the oxychlorination re-
actor where it is completely consumed. The component
processes of direct chlorination, EDC pyrolysis and
oxychlorination are combined to develop a balanced
process for the production of vinyl chloride from
ethylene and chlorine with no net consumption or
production of hydrogen chloride. Apart from the main
products, a variety of byproducts are produced in these
reactors. Many of these byproducts are now classified
as hazardous and are expensive to treat and dispose.
Hence, there is a significant need for performance im-
provement in the reactor sections. This paper focuses
on developing optimal reactor configurations for the
processes of direct chlorination and oxychlorination of
ethylene and pyrolysis of EDC.
In the next section we present a brief literature survey
of the area of reactor network synthesis. Section 3
outlines the mixed integer nonlinear programming
(MINLP) based reactor network synthesis algorithm
(Lakshmanan & Biegler, 1996a). Section 4 presents a
detailed discussion of the vinyl chloride process and
Section 5 applies the algorithms outlined in Section 3 to
the vinyl chloride monomer production process and the
results of this case study are presented. A candidate
flowsheet and a set of recommendations that improve
the conversion of ethylene to vinyl chloride product are
also proposed. Section 5 also presents an economic
optimization of the process flowsheet and highlights
changes in the reactor structure when the objective is to
maximize the profit rather than conversion to the de-
sired product. Here a brief network sensitivity analysis
to kinetic parameters is also included. In addition, the
process flowsheet is energy integrated and this results in
significant cost savings. Finally, Section 6 summarizes
the results and presents some ideas to further improve
the candidate flowsheet presented here.
2. Literature review on reactor network synthesis
Academic research to develop systematic optimiza-
tion techniques to synthesize reactor networks has its
roots in the work of Aris (1960, 1961), Horn (1964),
Horn and Tsai (1987) and Jackson (1968). Recent
approaches to reactor network synthesis may be
classified under two broad categories: superstructure
optimization and attainable region targeting. In super-
structure optimization, a fixed network of reactors is
first postulated based on heuristics and physical in-
sights. The optimal subnetwork which maximizes the
performance index is derived from this superstructure
using mixed integer nonlinear programming (MINLP)
techniques (see Viswanathan & Grossmann, 1990).
In attainable region targeting, an attempt is made to
find an achievable bound on the performance index of
the system irrespective of the actual reactor configu-
ration. A general functional representation is used
to model the entire variety of reaction and mix-
ing states. Bounds are then derived based on limits
posed by reaction kinetics on the space of concen-
trations achievable by the processes of reaction and
mixing.
2.1. Superstructure optimization
Researchers have attempted to develop optimal reac-
tor networks by postulating a variety of reactor super-
structures. Chitra and Govind (1981, 1985) studied
PFR systems with a recycle stream from an intermedi-
ate point along the reactor and optimized the recycle
ratio and the point of recycle. This reactor system could
model PFRs, CSTRs and recycle reactors. Later, they
classified different reaction mechanisms and postulated
a superstructure consisting of a series of recycle reactors
with bypass streams and heat exchangers at the reactor
inlets. Kokossis and Floudas (1990, 1991) consid-
ered a large superstructure of isothermal and non-
isothermal networks of PFRs and CSTRs, by modeling
a PFR as the limit of a large number of sub-CSTRs
(CSTRs of equal size). Their MINLP formulation was
free of differential equations. Later, the technique was
extended to handle stability of reactor networks and
integration with recycle systems. Recently, Smith and
Pantelides (1995) proposed a synthesis technique for
reaction and separation networks, using detailed unit
operation models. Complete connectivity among the
units, both forward and recycle, was assumed in the
superstructure. However, these approaches may be sub-
optimal since the solution obtained is only as rich as
the initial superstructure chosen and it is difficult to
ensure that all possible networks are included in the
initial superstructure.
 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495
2.2. Attainable region targeting
Targeting for reactor network synthesis is based on
the concept of the attainable region (AR) in concentra-
tion space, first suggested by Horn (1964). The attain-
able region is the convex hull of all of the
concentrations that can be achieved starting from the
feed point using only reaction and mixing. Recently,
Glasser et al. (1987), Hildebrandt et al. (1990) devel-
oped a geometric technique to map the entire region in
the concentration space that is attainable from a given
feed concentration by the processes of reaction and
mixing. Alternate plug flow reactor (PFR) and continu-
ous stirred tank reactor (CSTR) trajectories were drawn
to cover the attainable region and derive an optimal
reactor network. However, these geometric techniques
are difficult to apply beyond three dimensions. To
overcome this limitation, Omtveit, Tanskanen and Lien
(1994a) considered higher dimensional problems by us-
ing the principle of reaction invariants (Fjeld, Asb-
jornsen & Astrom, 1974) and by introducing system
specific constraints (for example, coking) to project the
problem into two dimensions. On the other hand, Bal-
akrishna and Biegler (1992a,b) adapted the geometric
technique for targeting to a mathematical programming
based framework. The Balakrishna and Biegler al-
gorithm is a constructive technique which optimizes the
objective at each stage. However, this is a sequential
approach which may exclude some structures consid-
ered by the AR approach.
In addition to developing techniques purely for reac-
tor network synthesis, several researchers have devel-
oped systematic approaches to simultaneously
synthesize the entire process. Pibouleau, Floquet and
Domenech (1988) considered a network of CSTRs and
distillation columns, with limited connectivity between
the reaction section and the separation section. Kokos-
sis and Floudas (1991) considered synthesis of isother-
mal reactorseparator-recycle systems with almost
complete connectivity between the reactor network
(consisting of CSTRs) and the separation network.
Omtveit, Wahl and Lien (1994b) and Glasser et al.
(1987) also applied the geometric targeting technique to
the synthesis of reactor-separator systems. Balakrishna
and Biegler (1993) and Lakshmanan and Biegler (1995,
1996a) integrated the synthesis of reactor networks with
simultaneous flowsheet synthesis. They demonstrated
their technique on a simplified process flowsheets for
the production of allyl chloride and the Williams Otto
process flowsheet. Lakshmanan and Biegler (1996b)
extended this approach to consider simultaneous reac-
tion, mixing and separation by integrating the reactor
network synthesis algorithm with mass exchange net-
work synthesis. Along the same lines, Papalexandri and
Pistikopoulos (1996) proposed a generalized modeling
framework based on mass/heat transfer principles and
481
the representation potential of this framework was
demonstrated with examples on specific unit operations
like reactive distillation.
3. Algorithm for reactor network synthesis
Lakshmanan and Biegler (1996a) proposed a combi-
nation of superstructure and attainable region targeting
techniques to synthesize optimal reactor networks. This
approach incorporates attainable region properties
derived from Feinberg and Hildebrandt (1997). Some
of these properties are: (i) the boundary of the attain-
able region for reaction and mixing consists of plug
flow reactor trajectories and straight lines; (ii) recycle
streams need not be considered to map the boundary;
and (iii) only continuous stirred tank reactor (CSTR),
plug flow reactor (PFR) and differential sidestream
reactor (DSR) trajectories make up the entire boundary
of the attainable region.
The reactor network synthesis technique considers
multiple reactor paths at each stage by targeting the
attainable region using reactor modules. A reactor
module consists of a differential sidestream reactor
(DSR) and a CSTR. The sidestream entering the DSR
in the Ith reactor module could be the feed stream to
the DSR or it could be from any of the reactor module
exits or any combination thereof (i.e. from any point
within the candidate AR). The DSR sidestream
flowrate is determined in the optimization and may be
set to zero, hence, a separate PFR model need not be
considered in the reactor module. A typical reactor
module is shown in Fig. 1. A binary variable is associ-
ated with each reactor path in a module. If the DSR is
chosen in the Ith reactor module, the binary variable
associated with it, YId, is set equal to 1.
Additional reactor modules are added successively
and the superstructure is developed in a constructive
manner by solving a sequence of improving MINLP’s.
The attainable region properties ensure that the
MINLP is compact. The steps involved in the construc-
tive stagewise algorithm are: (i) Solve a segregated flow
model to obtain a lower bound on the solution. For
Fig. 1. Ith Reactor module.
482 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495

yield or selectivity objectives and fixed feeds and tem- 1=YIc + YId (14)
peratures, this formulation leads to a linear program Nmod
with a global solution; (ii) Initialize the first reactor FIf XI = % FIj XI (15)
j=I
module with the solution obtained from step 1 and
optimize it with respect to a specific objective. The inlet Xexit = XNmod exit 05 FIc 5 UYIc 05 FId 5 UYId
conditions to this reactor module are the feed condi-
YIc {0, 1} YId {0, 1}
tions. This yields an initial target to the attainable
region. In addition, this solution does not eliminate Here, all flowrates are volumetric and concentrations
other network paths; the binary variables associated are based on mass, with
with each path are merely set to one or zero; (iii)
Extend the reactor module with an additional reactor Fif is the flowrate at the inlet of the Ith
module. The feed to the second module is the exit of reactor module.
the first module or a combination of the fresh feed and FkI−1, XkI−1 are the flowrate and concentration
the exit of the first module. If the extension improves from the exit of the kth module which
the objective then further extensions need to be consid- is an inlet stream to the Ith module
ered, else the optimal network is assumed to have been (k= 0, I−1)
found; (iv) Ensure that the feed to the Ith reactor Xif is the concentration at the inlet to the
module extension may be the exit of any one or a Ith reactor module.
combination of the exits of the previous I − 1 modules FIc and Fid are the flowrates of the streams passing
or the initial feed; and (v) Account for bypass streams through the CSTR and DSR in the Ith
by ensuring that the exit from the Ith reactor module reactor module. The product of F *X If If
plus the bypasses from the exits of any one or a is in mass flow units.
combination of the previous (I− 1) reactor modules, XIc in, Xic are the concentration at the inlet and
forms the inlet to the (I+1) th reactor module. exit of the CSTR and DSR XId in, Xid,
The reactor network synthesis problem is formulated respectively in the Ith reactor module.
as an MINLP as in (Eq. (P1)): XI exit is the concentration at the exit of the
Ith reactor module
maxJ(Xexit, t) (P1)
q, f, T YIc, YId are the binary variables associated with
XCSTR exitI = R(XCSTRI, TCSTRI )tcI +XIf (1) the CSTR and DSR in the Ith reactor
module.
dXDSRI /daI =R(X(aI ), T(aI )) XI side sidestream composition of the Ith DSR
+ (q(aI )QsideI /Q(aI ))(XsideI −X(aI )) (2) reactor module. It may be specified or
determined from any stream in the re-
XDSROI =XIf (3)
&
actor network.
 QI side sidestream flowrate of the Ith DSR re-
XDSR exitI = f(aI )X(aI ) daI (4) actor module.

&
0

1= f(aI ) daI (5) Eqs. (1)–(8) constitute the reactor module at stage I

&
0 made up of a CSTR (Eq. (1)) and a DSR (Eqs.
 (2)–(8)). The inlet conditions to the module and the
1= q(aI ) daI (6) individual reactors are given by (Eqs. (9)–(12)). Eq.

&&
0
(13) represents the exit from the Ith module and Eq.
 a
t= (q(a %)Q (14) ensures that only one among the two reactors is
I sideI /QexitI −f(a %))
I da %I daI (7)

&
0 0 chosen in the Ith module. Finally, the exit from the Ith
 reactor module forms the inlet to any one or a combi-
TDSR exitI = f(aI )T(aI ) daI (8) nation of modules I+ 1 to N as shown in Eq. (15). The
0
feed conditions (FIf, XIf ) to Ith module may be the exit
I−1
FIf = % FkI − 1 (9) conditions of any one or a combination of the previous
k=0 I− 1 modules (Eqs. (9) and (10) illustrate this fact).
I−1 Hence, when the Ith reactor extension is considered,
FIf XIf = % FkI − 1Xk (10) bypasses from the exits of the previous I− 2 modules
k=0
and the feed are automatically considered. Similarly,
XIf =XIc in =XId in (11) parallel reactor structures up to the I− 1 th module are
also accounted for. This structure keeps the construc-
FIf =FIc +FId (12)
tive MINLP algorithm as compact as possible. Inside
FIf XI exit = FIcXCSTR exitI +FIdXDSR exitI (13) the Ith reactor module, either the CSTR or the DSR
 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495
may be chosen. Thus, the exit conditions from the Ith
reactor module (XI ) are determined appropriately (Eqs.
(9) – (14)). The exit stream from the Ith reactor module
forms the inlet stream to any one or a combination of
reactor modules I + 1 to Nmod (Eq. (15)). Lakshmanan
and Biegler (1996a) discuss the advantages and disad-
vantages of this strategy. They suggest an initialization
procedure, a solution technique and remedies for possi-
ble problems due to the bilinear Eqs. (10) and (13).
The Lakshmanan and Biegler (1996a) algorithm for
reactor network synthesis is a versatile algorithm that
allows the designer to incorporate additional attainable
region properties when necessary. In addition, it may be
easily incorporated with techniques for simultaneous
synthesis of heat exchange networks and mass exchange
networks (Lakshmanan & Biegler, 1996b) to simulta-
neously synthesize waste minimizing process flowsheets.
4. Vinyl chloride monomer production process
The principal operating steps used in the ethylene
based balanced process for the production of vinyl
chloride are shown in the reactions below (Cowfer &
Magistro, 1983). Ethylene is chlorinated to EDC by the
processes of oxychlorination and direct chlorination.
EDC is purified in the EDC purification section and fed
to the pyrolysis reactor where vinyl chloride monomer is
produced. Vinyl chloride is purified in the vinyl chloride
purification section and the recovered hydrogen chloride
and EDC are recycled. The main reactions involved are:
Direct chlorination: C2H4+Cl12 “C2H4Cl12
Oxychlorination: C2H4+2HCl+1/2O2
“C2H4Cl2+H2O
Ethylene dichloride 2C2H4Cl2 “2C2H3Cl
pyrolysis:
+ 2HCl
Overall reaction: 2C2H4+Cl2+1/2O2
“2C2H3Cl+H2O
The kinetics for these reactions are described below
and detailed rate expressions are given in Appendix A.
Apart from the main products shown in the equations
above, a variety of byproducts are produced in these
reactors. Some of the more significant byproducts are
1,1,2 trichloroethane, chloral (trichloroacetaldehyde),
cis and trans- 1,2 dichloroethylenes, mono-, di-, tri- and
tetrachloromethanes and carbon oxides in the oxychlo-
rination reactor; in the direct chlorination network,
1,1,2 trichloroethane is also produced. Acetylene,
ethylene, butadiene, methyl chloride, vinyl acetylene,
chloroprene, vinylidene chloride, benzene, trichloro-
ethylene, tri- and tetrachloromethane and other chlori-
nated products are produced in the pyrolysis reactor.
483
All of these by-products cause problems and their
production should be minimized to improve conversion
of raw materials to desired products, lower raw material
costs, decrease the difficulties in the EDC purification
section, prevent coking and fouling of the pyrolysis
reactor and reduce waste handling and treatment costs.
For example, trace quantities of impurities in the pyrol-
ysis reactor feed makes the EDC pyrolysis process
susceptible to fouling and inhibition. The feed to the
pyrolysis reactor should be at least 99.5% pure dry
EDC. Also, the EDC recovered from the exit of the
pyrolysis reactor contains an appreciable number of
impurities, some of these like chloroprene polymerize to
a rubbery material which can seriously foul the EDC
purification section. Similarly, trichloroethylene can
form an azeotrope with EDC. If this byproduct is not
removed in the EDC recycle from the pyrolysis section,
it can lead to reduced cracking rates and increased
fouling in this reactor. (McPherson, Starks and Fryar,
1979). Hence, there is a significant need to improve the
performance of these reactors.
Fig. 2 shows a simplified process flowsheet for the
production of vinyl chloride monomer (VCM). This
flowsheet is based on the PRO/II Casebook (1992) on
the vinyl chloride monomer plant with a capacity of 125
million lbs/year (or 56.7× 106 kg/year) of VCM. The
three reaction sections are integrated into a balanced
process flowsheet with the separation sections (EDC
purification and VCM purification) by recycling the
hydrogen chloride recovered from the VCM purification
section to the oxychlorination section. In addition, the
EDC recovered from the VCM purification section is
recycled to the EDC purification section.
Direct chlorination of ethylene to EDC is a homoge-
neous catalytic reaction in the liquid phase. Commer-
cially, the exothermic reaction is conducted in a liquid
phase reactor by intimately mixing ethylene and chlo-
rine in liquid EDC. The reaction is catalyzed by ferric
chloride catalyst. The heat produced in the reaction is
removed by cooling water or by operating the reactor at
the boiling point of EDC and allowing the pure product
to vaporize. Here, the latter case is considered. We
modeled the direct chlorination reaction using kinetics
derived from Wachi and Morikawa (1986) and pre-
sented in Appendix A. On the other hand, in the
oxychlorination process ethylene and oxygen react with
dry hydrogen chloride gas produced from the pyrolysis
of EDC. Commercially the reaction is conducted in
either a fixed bed (230–300°C, 1.5–14 atm gauge pres-
sure) or a fluidized bed (220–235°C, 1.5–5 atm gauge
pressure) reactor containing a copper chloride catalyst
impregnated on a porous alumina support (Cowfer &
Magistro, 1983). It is a highly exothermic reaction and
good temperature control is essential. Increasing reac-
tor temperature results in increased byproduct forma-
484 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495
Fig. 2. Vinyl chloride monomer production process flowsheet.
tion and catalyst deactivation. The reaction is mod-
eled using the kinetics given in Appendix A (obtained
from a series of papers by Gel’perin, Bakshi, Avetisov
& Gel’bshtein, 1979; Gel’perin, Bakshi, Avetisov &
Gel’bshtein, 1983; Gel’perin, Bakshi, Avetisov &
Gel’bshtein, 1984).
Trace quantities of impurities in the pyrolysis reac-
tor feed can cause fouling of the reactor. Hence,
EDC from three different sources: direct chlorination,
oxychlorination and EDC recovered and recycled
from the pyrolysis section, must be purified to at
least 99 wt.% pure EDC. The EDC produced in the
oxychlorination section is washed in caustic to re-
move unreacted hydrogen chloride. This stream is
mixed with the EDC produced in the direct chlorina-
tion section and fed to the EDC purification section.
The EDC purification section consists of a decanter
followed by two distillation columns. The first column
is the lights column where 99.9% of the water is re-
moved as top product. The bottom product of this
column is fed into the heavies column. The top
product of the heavies column is 99.9 wt.% pure
EDC which is fed to the pyrolysis section, as ex-
plained in the next paragraph. The VCM produced in
the pyrolysis section is separated from the HCl and
EDC in the VCM purification section. The first
column in the VCM purification section, called the
HCl column, receives mixed phase feeds containing
mainly EDC, VCM and HCl with small amounts of
reaction byproducts. HCl distills off at the top; half
of it is recycled to the oxychlorination section and
the rest is refluxed back to the column. The bottom
product is fed to the VCM column, where purified
VCM is separated out as the overhead product and
the recovered EDC is recycled to the EDC purifica-
tion section.
Finally, vinyl chloride monomer is produced by
thermal cracking of EDC. The endothermic reaction
is carried out commercially in tubular reactors at
temperatures of 480–530°C and reactor gauge pres-
sures of 6–35 atm (Ranzi, Grottoli, Bussani Che &
Zahng, 1993). The main reaction which yields vinyl
chloride monomer and hydrogen chloride is a homo-
geneous, first-order free-radical chain mechanism. Sev-
eral byproducts are produced in trace amounts
through related free-radical and molecular mecha-
nisms. The EDC pyrolysis process is very selective to
vinyl chloride but it is difficult to verify the forma-
tion and amount of the byproducts due to propri-
etary kinetics. Also, literature models for the kinetic
parameters of the reactions involved are not compre-
hensive. The pyrolysis reaction kinetics used in this
study were compiled from various literature sources
(Kurtz, 1972; Weissman & Benson, 1984; Karra &
Senkan, 1988; Ranzi, Dente, Tiziano, Mullick & Bus-
sani, 1990) and described further in Appendix A.
 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495 485

5. Results of the vinyl chloride case study Primary Problems (P2)

Max VCM Min W(x, y)
Eq. (P1) as well as the flowsheet connectivity and Yield (x, y)
separators shown in Fig. 2 were modeled in GAMS x, y x, y
(Brooke, Kendrick & Meeraus, 1988). The resulting h(x, y)= 0 h(x, y)=0
model is a MINLP (which we refer to as P1f) with g(x, y)50 g(x, y)50
4236 constraints, 4995 continuous variables and eight W(x, y) W(x, y)
binary variables. Four binaries were used in each of 5 Wallowed 5 Wallowed
the direct chlorination and oxychlorination sections to xX x X
synthesize the optimal reactor structures. Based on
literature reports, a tubular reactor (PFR) in a fuel ¤Rm yY ¤Rm yY
fired furnace was used to model the pyrolysis reactor. = {0, l} q
= {0, l} q

The purification sections are modeled using split frac- Secondary Problems
tions, which were based on the data published in the Min VCM yield
PRO/II Casebook (1992). Each of the reactors in the x, y
model was initialized with profiles obtained from h(x, y)= 0
solving segregated flow initial value problems. The g(x, y)=0
MINLP model was solved using DICOPT W(x, y)=o
(Viswanathan & Grossmann, 1990). x X R m yYR q
o[Wmin, Wallowed]
5.1. Vinyl chloride yield maximization
where x is the set of continuous flowsheet parameters,
We first consider the flowsheet where we maximize y is the set of binary variables and h(x, y) and
the overall conversion of ethylene to vinyl chloride g(x, y) are the constraints from (P1f) and o is a
monomer. Model (P1f) was solved in 8.12 s on a parameter varied between Wmin and Wallowed. See
HP-UX 9000-720 workstation. The results of this Lakshmanan and Biegler, (1995) for more details.
flowsheet optimization are shown in Tables 1 and 2. From the solutions of these MINLP models, we next
In addition we can obtain trade-off curves between describe the results for each reaction section for the
the conversion of ethylene and the amount of waste yield maximization case.
allowed by solving the following sub-problems:

Table 1
Results of the vinyl chloride case study

Chlorination Purification

Direct Oxy- Pyrolysis EDC VCM

Raw material*
C2H4 0.533 0.481 7.0e−6 7.0e−3 7.0e−6
Cl2 0.528 1.6e−6 l.5e−3 1.3e−5
O2 0.241 1.6e−4
Intermediate*
EDC 0.521 0.480 1.67 1.0e−3 0.67
By-product*
C2H3Cl3 5.798e−3 5.582e−4 6.3e−3 1.6e−5
CO2 1.622e−4 1.6e−4
C2H2,
C4H6, 2.9e−5
C2H2Cl2
Main product
C2H3Cl 1

* Mol/mol of vinyl chloride product.

486 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495

Table 2
Reactor conditions that maximize conversion of ethylene to vinyl chloride

Reactor struc- Reactor conditions (residence time, pressure, temperature, conversion) Reactor feed (gmol/s)
ture

Direct chlorina- CSTR 9293 s, 1 atm, 333 K, 99.7% 16.00 C2H4 15.84 Cl2
tion
Oxy-chlorination PFR+ 324 s, 5.3 atm, 466.3–465.8 K, 96.2% 14.43 C2H4 7.26 O2
CSTR 5.6 s, 5.3 atm, 465.7 K, 96.24% 30.03 HCl
Pyrolysis PFR 0.35 s, 20 atm, 773 K, 60% 49.99

5.2. Direct chlorination of ethylene 0.00154 and 0.00015 kg, respectively, using the reactor
network (PFR+ CSTR) and conditions shown above.
In this scheme (Wachi & Morikawa, 1986, see Ap- Hence, a waste reduction of one and two orders of
pendix A), EDC is formed by an addition reaction magnitudes is realized.
between ethylene and chlorine. The main byproduct is Also, using MINLP (P2) we can plot the noninferior
1,1,2 trichloroethane which is formed by substitution curve, shown in Fig. 4. between conversion of ethylene
and addition reactions between chloride radicals, chlo- to EDC and the amount trichloroethane produced in
rine and EDC. The temperature range considered was the oxychlorination reactor. The optimal reactor struc-
30 – 75°C, the pressure range was 0 – 2 atm. A slight tures for this curve remained a PFR+ CSTR combina-
excess of ethylene was used. The optimal reactor struc- tion for the conversion maximization case.
ture determined by using the reactor network synthesis
algorithm (Eq. (P1)) was a CSTR (9293.48 s residence 5.4. Pyrolysis of EDC
time). This yielded a 99.70% conversion of the limiting
component (chlorine) and 99% selectivity to EDC. Vinyl chloride monomer is produced by thermal
cracking of EDC and this endothermic reaction is
5.3. Oxychlorination of ethylene carried out commercially in tubular reactors at temper-
atures of 480–530°C and gauge pressures of 6–35 atm
Here EDC is formed through a heterogeneous cata- (Ranzi et al., 1993). The byproducts considered in the
lytic cycle where the main chlorinating agent is cop- kinetic model include acetylene, ethylene, butadiene,
per(II) chloride. The byproducts considered are trichloroethane and vinyl acetylene. Each of these
trichloroethane and carbon oxides. The recycled hydro- byproducts are harmful and could foul the reactor, if
gen chloride stream from the top of the vinyl chloride recycled. A plug flow reactor was used to model the
monomer purification section is the limiting component pyrolysis section using the kinetics in Appendix A.
in this reaction. The optimal reactor network deter- This choice was based on solutions obtained from
mined by using the reactor network synthesis al- solving initial value problems, which show very high
gorithm (P1) was a PFR with a residence time of 324 s selectivity to vinyl chloride in a plug flow reactor and
and 96.20% conversion of the rate limiting component
(hydrogen chloride), followed by a CSTR with a resi-
dence time of 5.6 s and 96.24% conversion of the
limiting component, with a selectivity of ethylene to
EDC of 99.8%. Note that the small CSTR is needed
for the additional conversion, optimization with the
PFR alone leads to a conversion of only 96.205%. The
optimal temperature profile in the PFR is shown in
Fig. 3 and the reactor pressure was found to be 5.34
atm.
To compare our reactor network to that used in the
vinyl chloride monomer plant PRO/II (1992) case
study, we consider their reactor network, a PFR at
290°C and 6.122 atm, modeled using the kinetic
scheme in Appendix A. The results for this base case
reactor showed that the amount of trichloroethane
produced per kg of EDC formed is 0.016 kg and the
amount of carbon oxides produced per kg of EDC
formed is 0.03 kg, as compared to our results of Fig. 3. Temperature profile in the PFR-oxychlorination section.
 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495 487

Table 4
Prices of utilities

Utility Cost

Cooling water (298–319 K) 1.865e−5 $/kg

300 psig steam 7.694e−3 $/kg
Steam Condensate
Fuel Oil 1.9162e−8 $/J
Refrigerant 5e−8 $/J

structures and conditions that maximize the conversion

Fig. 4. Noninferior curve between ethylene conversion and byproduct
formation in the oxychlorination process.
from literature reports on the EDC pyrolysis process.
The temperature and pressure of the PFR were 500°C
and 20 atm, respectively. The selectivity of EDC to
vinyl chloride in this PFR was found to be 99.9% with
a single pass conversion of 60%.
The results of this study are summarized in Tables 1
and 2. In Table 1, the values shown in the intermediates
section represent the composition of products of oxy-
chlorination and direct chlorination and the composi-
tion at the inlet to the pyrolysis reactor (moles of
component/mole of vinyl chloride produced). In the
byproducts section, the values shown are the composi-
tions of these byproducts produced in the reactors. The
compositions of acetylene, butadiene and vinylidene
chloride produced in the pyrolysis reactor are summed
up in the last column. The values listed under the
purification sections represent compositions of streams
leaving the process. The results shown in Table 1 match
published industrial data (McPherson et al., 1979) with
almost equal amounts of EDC produced in the oxy-
chlorination and direct chlorination sections. This con-
forms to stoichiometric requirements since the process
flowsheet considered in this study is a balanced produc-
tion process flowsheet and for every mole of EDC
pyrolyzed to vinyl chloride monomer, a mole of HCl is
formed. This HCl is completely consumed in the oxy-
chlorination reactor. Table 2 gives the optimal reactor
of ethylene to vinyl chloride. The overall conversion of
ethylene to vinyl chloride is 98.7% (see Table 1;
1/(0.533 + 0.481)).
So far we have assessed the performance of the
reactor networks based on conversion to the desired
product. In addition, we synthesized maximum yield
reactor networks for a given waste limit. This allowed
us to establish an optimal trade-off curve of conversion
of raw material versus waste generated, as shown in
Fig. 4. However, an economic optimization of this
process could be of more interest to the designer. The
next section presents a preliminary cost analysis of the
proposed process flowsheet and highlights changes in
reactor structures when the objective is to maximize
profit. This analysis is then extended to consider energy
integration of the flowsheet.
5.5. Economic optimization of the 6inyl chloride
monomer production process
The aim of this study is to analyze changes in reactor
structures when the objective is to maximize pre-tax
profit instead of conversion. The vinyl chloride
monomer flowsheet shown in Fig. 2 is used in this
study. The objective function, the pre-tax profit, in-
cludes the product and raw materials as well as utility
and equipment costs. The substitution of the objective
function in Eq. (P1) forms the MINLP (P1e).
The prices of the product and raw materials, heat
capacity data and utility prices are given in Tables 3
and 4. The expressions used for the installed costs of
the reactors and columns are given in Table 5 in $/yr
according to data from Turkay and Grossmann (1996).
The installed costs of the reactors are based on the

Table 3
Prices and heat capacity data of the products and raw materialsa

C2H4 Cl2 HCl O2 C2H3Cl C2H4Cl2 H2O

Prices @ $/kg 0.5346 0.1928 1.5891 0.0275 0.4624 0.3749

DHf(298) J/gmol 5.234e4 0 −9.236e4 0 3.517e4 −1.298e5 −2.42e5
Cpvap J/gmol-K 3.806 26.93 30.67 28.11 5.949 20.49 32.24

a
Chemical and marketing reporter, 12, 1996. Reid, Prausnitz and Poling, 1986. DHf(T) =DHf,298.2+Cp (T−298.2); DHr×n =SDHf(products)–
SDHf(reactants).
488 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495

reactor volumes (flowrate ×residence time) and fixed Table 6

Summary of economic optimization
values of g(P) and d(T) were assumed for each of the
reactors depending upon their temperature and pressure. No heat integration With heat integration
The installed costs of the column are determined based
on the throughput through the column. In Fig. 2, the Reactor structure
direct chlorination and oxychlorination reactors are (res. time s, press.
atm, temp. K)
exothermic, while the pyrolysis reaction is endothermic.
Direct chlorination CSTR(9293, 1, 333) CSTR(9427, 1, 333)
The heat produced in the direct chlorination reactors is Oxychlorination PFR(324, 5.2, 466.3–465.7) PFR(324, 5.l, 466.3–465.7)
removed by operating the reactor close to the boiling Pyrolysis PFR(0.35, 20, 773) PFR(0.35, 20, 773)
point of EDC and allowing the pure product to vaporize. Reactor installed 0.01116 0.01122
The heat loads of the rest of the process are supplied by costs 10 6 $/year
the utilities shown in Table 4. Hence, the heat produced
Column installed 0.00456 0.00456
in the oxychlorination reactor is removed by circulating costs 10 6 $/year
cold water and the effluent stream from the reactor is
Product & raw ma-
cooled in the oxychlorination aftercooler before entering terial flows kg-
the EDC purification section. In the EDC purification mol/year
section the heat required by the reboilers is supplied by C2H3Cl 907.135 907.135
fuel oil in the lights column and 300 psig steam in the C2H4 919.862 919.786
heavies column and cold water is used to cool the Cl2 463.772 463.697
O2 226.295 226.281
condensers. A constant temperature difference is as-
sumed between the reboiler and the condenser (An- O6erall con6ersion 98.62% 98.63%
of C2H4 to
drecovich & Westerberg, 1985). The reflux ratios shown
C2H3Cl
in Fig. 2 are control specifications. The top product of
the heavies column, which is about 99.5% pure dry EDC, Process utility costs
10 6 $/year
is compressed and heated to 500°C (by using 300 psig 300 psig steam 0.8603 0.3975
steam and fuel oil) in the pyrolysis preheater. The heat Fuel oil 0.8852 0.6995
required by the endothermic pyrolysis reaction is sup- Cooling water 0.0856 0.0508
plied by fuel oil and the exit stream from the pyrolysis Refrigerant 1.7423 1.4777
reactor is cooled by cooling water in the after-cooler. Savings due to heat 0.9479
integration
Finally, the HCl column recovers hydrogen chloride
which is recycled to the oxychlorination section and the Profit* 10 6 $/year 2.327 3.276
VCM column produces pure vinyl chloride monomer as
* Profit =Product value — raw material costs — reactor and column
the top product. The condensers in both columns require installed costs–process utility costs.
a refrigerant while 300 psig steam is used to supply heat
to the reboilers.
The optimal reactor structures from the solution of
(P1e) were found to be: a CSTR for direct chlorination, them. As a result, the reactor networks are simplified.
a PFR for oxychlorination and a PFR for the pyrolysis The pre-tax profit realized by this flowsheet was $2.33
section. The reactor conditions are similar to those million/year (based on 8400 h of operation/year). The
obtained for the conversion maximization case. This optimization problem contained 4265 variables, eight
shows that the improvement in conversion which may be binary variables and 5024 constraints; it was solved in 5.6
achieved by using the additional reactor structure (the CPU s on a HPUX 9000-720 from a good initial point.
small CSTR) described in the previous section is not Results of this problem are summarized in Table 6 along
enough to offset the increased costs incurred by using with reactor designs and operating conditions.

Table 5 5.6. Sensiti6ity of the 6inyl chloride monomer

Installed costs of the reactors, columns and heat exchangersa production process
Reactor (MS/280) Fm 0.125 VR g(P) d(T)
1000 $/year In this section, we take a brief look at the sensitivity
Column 0.1512 (throughput)0.5 1000 $/year of the VCM flowsheet with respect to changes in the
Heat exchangers, heaters 1000+560[Area (m2)]0.6 $/year kinetic parameters in the direct chlorination and oxy-
and coolers chlorination reaction sections. In particular, we want to
a
show that even with reasonable changes in the values of
Turkay and Grossmann, 1996. MS =1031.8 for 1996; Fm =2.25
for SS; VR=reactor volume; g(P)= 1 for PB3.4 MPa; d(T)= 1 for the kinetic constants, overall conversion to VCM re-
T = 300–400 K, d(T) = 1.1 for T= 400–500 K, d(T)= 1.4 for T = mains quite high with the reported optimal reactor
700–1000 K. network and that the annual process profit is not
 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495
significantly affected. In addition, if the flowsheet is
reoptimized according to the new kinetic values, the
reactor structures in each section of the plant do not
change significantly.
We investigated six scenarios for the direct chlorination
section in which each rate constant in reactions DC1–
DC3 (see Appendix A) is allowed to change from its
nominal value by 9 10% while the other two were held
fixed. The flowsheet was simulated using the reactor
structures found in Table 2 with the new values of the
rate constants and the change in overall conversion of
ethylene to VCM was noted. For all cases, the overall
conversion to VCM changed less than one percent with
the largest decrease in conversion occurring when rate
constant for reaction DC1 was 10% lower than its nominal
value. If we consider an unfavorable change in the rate
constants with a 20% decrease in k1 in reaction DC1 (lower
rate of ethylene dichloride formation) coupled with a 20%
increase in rate constants in reactions DC2 and DC3
(increased byproduct formation), the overall conversion
to VCM is still quite high for the reactor structure in Table
2 with a slightly lower conversion of 98.1%.
For the oxychlorination section, the pre-exponential
factors for reactions O1 – O3 were changed by 9 10%. The
equilibrium constant for reaction O4 was also changed
by 910%. With these variations, the overall conversion
changed less than 1% when using the reactor structure
from Table 2, with the largest decrease in conversion
realized when the rate constant in reaction O1 was 10%
lower than its nominal value. Even with a 20% decrease
in the rate constant for reaction O1 and a 20% increase
in the rate constants for reactions O2 – O3 (we assume a
nominal value of the equilibrium constant in reaction O4
here), the overall conversion to VCM decreases less than
one percent to 98.2%. Thus, with small changes in the
direct chlorination and oxychlorination rate constants,
overall conversion is quite good for the optimal reactor
network found in Table 2.
We also reoptimized the flowsheet for the new values
of the kinetic parameters to see what changes would be
realized in the reactor networks. All fresh feeds to the
process were fixed to the values reported in Table 2. For
the oxychlorination section, the inlet temperature was
fixed to 466.3 K. For each case, the flowsheet was
initialized using the results from Table 2. In all of the
reoptimized flowsheets, the reactor structures remain the
same as reported in Table 2 with minor changes to the
reactor residence times.
With varying rate constants in the direct chlorination,
the residence times of the reactors in both the direct
chlorination section and oxychlorination section changed
for maximizing overall conversion to VCM. For the rate
constants in reactions DC2 and DC3, very little effect was
seen on the direct chlorination CSTR residence time. In
the worst case for these two reactions, the optimal
residence time increased or decreased by less than half
489
a percent, clearly a negligible change. In the oxychlorina-
tion, both the PFR and CSTR also changed less than a
half percent from the values reported in Table 2. With
a decrease of 10% in the rate constant for reaction DC1,
the optimal residence time in the CSTR increased to
10 314 s. This is reasonable since reaction DC1 produces
the desired product, EDC. The oxychlorination PFR
decreased to 292 s and the CSTR decreased slightly to
5.1 s. This new structure maintains a very high overall
conversion of ethylene to VCM of 98.6%. With a 10%
increase in the rate constant for reaction DC1, the optimal
direct chlorination CSTR residence time decreased to
8457 s while the oxychlorination PFR increased to 392
s followed by a CSTR with a residence time of 38 s. Overall
conversion to VCM for this scenario is 98.8%.
For the oxychlorination section, eight optimization
problems were solved corresponding to a 9 10% change
in each of the four kinetic parameters. The residence time
of the oxychlorination PFR increased to 396 s when the
rate constant for reaction O1 was 10% lower than its
nominal value. The oxychlorination CSTR also increased
slightly to 41.1 s while the CSTR in the direct chlorination
decreased to 9183 s. Overall conversion of ethylene to
VCM was 98.5%. For all other combinations, the size of
the oxychlorination reactors didn’t change by more than
2% from the values reported in Table 2. The largest
increase in residence time for the direct chlorination
CSTR was 8.8% to 10 115 s when the rate constant in
for reaction O3 was 10% larger than its nominal value.
For all other instances of varying oxychlorination kinetic
parameters, the direct chlorination CSTR changed less
than 3% from the value reported in Table 2.
For the economic optimization (without heat integra-
tion) of the VCM plant, the annual profit changes only
slightly with variations in the direct chlorination and
oxychlorination kinetic parameters. If the reactor net-
works are fixed to those reported in Table 6, the overall
profit decreases for all scenarios of the varying kinetic
parameters. The largest decrease in profit was 5.1% to
2.208*106 $/year when the rate constant in reaction DC1
was 10% lower than its nominal value. If the flowsheet
is reoptimized for this worst case, the new optimal
structure contains a CSTR for the direct chlorination
(10 314 s), a PFR in the pyrolysis section (0.35 s) and PFR
(292 s) in the oxychlorination section with nearly identical
operating conditions as reported for the nominal case. The
annual profit is 2.267*106 $/year.
These results show that the optimal reactor networks
found for ethylene conversion to VCM and annual
process profit are only mildly sensitive to changes in the
kinetic parameters.
5.7. Energy integration of the process flowsheet
Obviously, the heat content of the reactor streams and
hot streams in the aftercoolers and condensers may be
490 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495

used to heat cold streams in the preheaters and with the max operator approximated with a smoothing
reboilers. Eight hot process streams (H1…H8) and nine function (see Balakrishna & Biegler, 1993). We augment
cold process streams (C1…C9) are shown in Fig. 2. As (P1e) with these equations to form the MINLP (P1h). A
mentioned earlier, all the heat produced in the direct minimum approach temperature (DTm) of 10 K is as-
chlorination reactor is used to vaporize the liquid EDC sumed for efficient heat exchange. For this energy
product. Hence, the streams flowing through this reactor integrated case, the pre-tax profit for the same vinyl
are not considered while energy integrating the process chloride production rate was found to be 3.27 million
flowsheet. The oxychlorination reactor is exothermic. $/year. This is mainly due to savings in the utility costs
The heat produced here may be used to heat other and slightly better conversions in the reactors. The results
process streams. Since the temperature drop across the of the optimization (4432 variables, eight binary vari-
oxychlorination reactor is small, the reactor is repre- ables, 5168 constraints, 6.96 CPU s on a HP-UX
sented by two hot streams (H1 and H2) (instead of two 9000-720) are summarized in Table 6. Note that heat
hot streams per finite reacting element). Hot stream H1 integration of the process flowsheet does not affect the
accounts for the heat of reaction (a 1 K temperature drop overall conversion significantly because the overall con-
is assumed) and H2 accounts for the hot reactor streams version is high even without heat integration. Also, as this
with a heat capacity equal to that of the reactor internal is a difficult problem to solve computationally, the
streams. The inlet temperature of streams H1 and H2 is solution times shown above are for a problem with a
the temperature at the inlet to the reactor and the exit good initialization.
temperature of stream H2 is equal to the temperature at Following the solution of (P1h), the resulting temper-
the reactor exit. The temperature at the reactor exit is atures and heat capacity flowrates are used to synthesize
optimized simultaneously with the rest of the process and a heat exchange network (HEN). This is determined
using the MINLP formulation proposed by Yee and
energy network constraints. Similarly, the endothermic
Grossmann (1990). The SYNHEAT2 program devel-
pyrolysis reactor is represented by two cold streams C6
oped by Bolio (1994) to implement this MINLP was
and C7, one of which (C6) accounts for the heat of
modified to solve this large HEN problem. The modifi-
reaction and the other stream (C7) may be heat inte-
cations include, relaxing the upper bound in the calcula-
grated with the rest of the process. Unfortunately, the
tion of approach temperatures and a user-supplied initial
temperature at which the pyrolysis reactor is operated,
set of stream matches. The MINLP model for the
773 K, is much higher than the temperature of the other
synthesis of the HEN involved 16 stages (7922 equalities
process streams. Hence, these streams require an external and inequalities, 7394 continuous variables). The heat
heating utility. Similarly, the lights column reboiler capacity data for the individual streams is given in Table
stream (cold stream C2) and the HC1 column condenser 7. The intermediate utilities (cold water, steam and steam
stream (hot stream H7) require an external heating and condensate) were considered as process streams. 9 hot
cooling utility, respectively. The rest of the streams may streams (H1–H6, H8, steam and steam condensate)
be energy integrated. and 7 cold streams (C1, C3–C5, C8, C9, cold water)
The simultaneous energy integration technique pro-
posed by Duran and Grossmann (1986) is used to Table 7
determine the minimum utility consumption of this Data for the HEN synthesis in the vinyl chloride case studya
process flowsheet. This requires the addition of the
Stream Tin (K) Tout (K) F (KW K-1)
following equations to MINLP formulation (P1e):
H1 466.32 465.32 849.71
QC =QH +ShoH(FCp)h[T in out
h −T h ]
H2 466.32 465.77 1.54
− ScoC(FCp)c[T out in
c −t c ]
H3 465.77 330.50 1.37
H4 394.07 393.07 608.14
QH ] z pH(x, y) H5 420.16 419.16 1600.24
H6 772.79 330.62 6.46
Here, (FCp) is the heat capacity-flowrate and T p corre- H7 241.80 240.80 484.48
sponds to the pinch candidates which are derived from H8 309.26 308.26 667.81
C1 298.20 466.32 1.19
T in in
h for the hot streams and t c +DTm for the cold
C2 481.07 482.07 12.32
streams. Also, we have C3 431.66 432.66 2560.38
C4 419.16 473.00 0.05
z pH(y)= %coC(FCp)c C5 473.00 772.00 0.06
C6 772.00 773.00 1206.47
C7 772.00 772.79 0.13
×[max {0; t out p
c −{T −DTm}} C8 359.80 360.80 2699.90
−max {0; t in p C9 430.47 431.47 1392.21
c −{T −DTm}}]
a
h =1 kW m−2 K−1 for all the streams. Cost of Heat Exchangers,
−%hoH(FCp)h[max {0; T in p out p
h −T } −max {0; T h −T }] Heaters and Coolers ($ year−l) =1000+560(Area(m2))0.6.
 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495 491

Fig. 5. Heat exchange network for the process streams.

were included in the MINLP model. Hot stream H7 was
matched with the refrigerant and cold streams C2, C6
and C7 were matched with heat from fuel oil. The
network shown in Fig. 5 indicates heat exchange between
the following streams labeled in Fig. 2.
As seen in Fig. 5, the HEN includes the following
matches:
“ oxychlorination reactor streams (H1) and the VCM
column reboiler stream (C9);
“ oxychlorination reactor streams (H2) and the oxy-
chlorination preheater stream (C1);
“ heavies column condenser stream (H5) and the HCl
column reboiler stream (C8);
“ pyrolysis after-cooler stream (H6) and the pyrolysis
preheater streams (C4, C5);
“ pyrolysis after-cooler stream (H6) and the heavies
column reboiler stream (C3); and
“ pyrolysis after-cooler stream (H6) and the VCM
reboiler stream (C9);
This network consumes about 16 KW more of the
heating and cooling utility as compared to the mini-
mum utility target found by (P1e). The capital costs of
this network were about $70 000/year (calculated by
the cost expressions shown in Table 5), which is consid-
erably smaller than the savings realized by the energy
integration.
6. Conclusions
The reactor network synthesis algorithm was applied
492 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495

for the improvement of the vinyl chloride monomer t residence time

production process. The approach has been successfully T(a) temperature inside the reactor at a
modeled for the three reaction sections (oxychlorination, TCSTR I temperature inside the CSTR in mod-
direct chlorination and pyrolysis) and the conversion of ule I
ethylene to vinyl chloride has been maximized. For this TDSR exit I temperature at exit of the DSR in
case, the three reaction sections were found to be: module I
Oxychlorination: PFR+CSTR (small); Direct chlorina- U the upper bound on the variable
tion: CSTR; and a PFR for the pyrolysis section. X(aI ) mass concentration vector at point a
Compared with the results of an industrial case study, inside the DSR in module I
these reactor structures perform better than the conven- XCSTR I mass concentration vector inside the
tional reactors used for these processes and enhance the CSTR in module I
conversion of raw materials to the desired products while XDSR I mass concentration vector inside the
minimizing waste production. Also, the noninferior DSR in module I
curve between one of the main waste products and the Xicin mass concentration vector at the inlet
conversion of ethylene was plotted and a simple process to the CSTR in module I
flowsheet which incorporates these results has been XIdin mass concentration vector at the inlet
proposed. to the DSR in module I
Moreover, an economic optimization of the proposed Xif mass concentration vector at the feed
process flowsheet was presented and changes in reactor to module I
structures while optimizing the pre-tax profit were high- Yic binary variable denoting the existence/
lighted. A simultaneous energy integration of the process nonexistence of the CSTR in module I
flowsheet was also performed to determine the minimum Yid binary variable denoting the existence/
utility consumption of the process and a heat exchange nonexistence of the DSR in module I
network was derived. The network indicated efficient use
of the oxychlorination reactor and pyrolysis aftercooler
streams to heat other process streams. This results in
significant cost savings. Future research will continue to Acknowledgements
address the sensitivity of reactor networks to process and
kinetic parameters used in the reaction schemes. Funding from the Engineering Design Research Cen-
ter, an NSF sponsored Engineering Research Center
and from the Pennsylvania Infrastructure Technology
7. Nomenclature Alliance (PITA) is gratefully acknowledged.

R(X(a), T(a)) residence time distribution in the PFR

f(aI ) residence time distribution in the PFR Appendix A. Kinetic schemes for the three reaction sections
or DSR in module I in the vinyl chloride monomer production process.
Fif inlet volumetric flowrate to the Ith re-
actor module A.1. Direct chlorination reaction kinetics
FIc, FId volumetric flowrate through the CSTR
and DSR in the Ith reactor module The kinetics for direct chlorination of ethylene to
F kI−1 exit volumetric flowrate from module EDC (at 60°C) were derived from (Wachi & Morikawa,
k which is an inlet to module I−1, 1986). EDC and trichloroethane are formed through
k= 0, I−1 molecular and radical addition and substitution
Ne the number of finite reacting elements reactions.
in the reactor discretization Reaction DC1:
k1
Nmod the number of reactor modules consid- C2H4 + Cl2 “ C2H4Cl2
ered in the synthesis of the reactor
R1 = k1[C2H4][Cl2]
network
q(aI ) fraction of Qside entering the DSR at k1 = 0.132 m3 mol − 1 s − 1
a in module I
Reaction DC2:
Q(aI ) volumetric flowrate through the PFR k
C2H4 + Cl2 “ C2H4 − Cl+Cl’
2
or DSR at a in reactor module I
k3
Qside side stream flowrate to the DSR Cl’ + C2H4Cl2 “ C2H3Cl2’ + HCl
R(X(a), T(a)) rate vector at a k
C2H3Cl2’ + Cl2 “ C2H3Cl3 + Cl’
4

R(Xij, Tij ) rate vector at element i, collocation k5

point j C2H3Cl2’ + Mi “ deactivate
 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495 493

Cl’ +Mi “ deactivate

k6
A.3. Pyrolysis reaction kinetics

R2 = kR2[C2H4][Cl2]2 The kinetics for the pyrolysis of EDC to vinyl chlo-

kR2 = 0.0239 m mol 6 −2
s −1 ride were derived from several papers (Kurtz, 1972;
Weissman & Benson, 1984; Karra & Senkan, 1988;
Reaction DC3: Ranzi et al., 1990). Vinyl chloride and several byprod-
kR 3 ucts are formed through molecular and radical addition
C2H4Cl2 +Cl2 “ C2H3Cl3 +HCl and substitution reactions. The main reactions which
follow elementary reaction kinetics are shown below.
R3 = kR3[C2H4Cl2][Cl2]
Reaction P1:
kR3 = 6.1210 − 9 m3 mol − 1s − 1 K1
1,2C2H4Cl2 “ C2H3Cl+ HCl

A.2. Oxychlorination reaction kinetics K1 = k1f /k1b

k1f = 1013.6 exp(− 58000/(1.987T)) s − 1
The kinetics for oxychlorination of ethylene to EDC
were derived from a series of papers by (Gel’Perin et k1b = 0.3109 exp(− 44000/(1.987T)) m3 kmol − 1 s − 1
al., 1979, 1983, 1984). EDC and several byproducts are
Reaction P2:
formed through molecular and radical addition and K2
substitution reactions. The main reactions are shown C2H3Cll C2H2 + HCl
below.
K2 = k2f /k2b
Reaction O1:
k1 k2f = 0.51014 exp(− 69000/(1.987T)) s − 1
C2H4 + 2HCl+1/2O2 “ C2H4Cl2 +H2O
k2b = 0.37109 exp(− 40000/(1.987T)) m3 kmol − 1 s − 1
k1 = a*1 exp(b1/RT) mol/(liter cat. h kPa1.5)
Reaction P3:
a1 = 104.2 K3
1,2C2H4Cl2 “ C2H4 + Cl2
b1 = − 40.1 kJ R=8.314 J/mol K
k3 = 1013 exp(− 72000/(1.987T)) s − 1
rC2H4 = k1pC2H4p 0.5
Cl2 mol/(liter cat. h)
Reaction P4:
K4
Reaction O2: C2H4 l C2H2 + H2
k2
C2H4Cl2 +HCl+1/2O2 “ C2H3Cl3 +H2O K4 = k4f /k4b
k2 = a*2 exp(b2/RT) mol/(liter cat. h kPa1.5) k4f = 0.11015 exp(82000/(1.987T)) s − 1
a2 = 1013.23 b2 = −128.04 kJ k4b = 0.8109 exp(− 38000/(1.987T)) m3 kmol − 1 s − 1

rC2H4Cl2 = k2pC2H4Cl2p 0.5

Cl2 mol/(liter cat. h)
Reaction P5:
K5
Reaction O3: C2H2 + C2H4 l C4H6
k3 K5 = k5f /k5b
C2H4 + 3O2 “ 2CO2 +2H2O
k5f = 0.15109 exp(− 32000/(1.987T)) m3 kmol − 1 s − 1
k3 = a*3 exp(b3/RT) mol/(liter cat. h kPa2.5)
k5b = 0.51013 exp(− 73000/(1.987T)) s − 1
a3 = 106.78 b2 = −112 kJ
Reaction P6:
r ox 0.5
C2H4 = k3pO2p Cl2pC2H4 mol/(liter cat. h) K6
C2H3Cl+ Cll CH2Cl CH Cl
Reaction O4:4
K K6 = k6f /k6b
4HCl +O2 l 2Cl2 + 2H2O
k6f = 0.21011 exp(− 0/(1.987T)) m3 kmol − 1 s − 1
K4 = k4f /krb k6b = 0.151014 exp(− 19000/(1.987T)) s − 1
k4f = 1000exp(17.13 −13000/(1.987T)) Reaction P7:
k7
k4b =exp(5.4+ 16000/(1.987T)) 1,2C2H4Cl2 “ CH2 Cl CH Cl+ HCl
rCl2 = K4pO2p − 1
Cl2 mol/(liter cat. h) k7 = 1010.8 exp(− 2900/(1.987T)) m3 kmol − 1 s − 1
494 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495

Reaction P8: since only trace amounts of these radicals were pro-
K8 duced and they caused numerical inconsistencies in the
1,2C2H4Cl2 + Cl“ CH2 CH2 Cl +Cl
model.
k8 = 0.21017 exp(−83000/(1.987T)) m3 kmol − 1 s − 1
Reaction P9: References
k9
C2H3Cl+third “ C2H3 +Cl + third
Andrecovich, M., & Westerberg, A. W. (1985). An MILP formula-
k9 = 0.251016 exp(−78000/(1.987T)) s − 1 tion for heat integrated distillation sequence synthesis. AIChE J.,
31, 363.
Reaction P10: Aris, R. (1960). Studies in optimization. I: the optimum design of
k10 adiabatic reactors with several beds. Chemical Engineering Sci-
CH2CH2Cl“ C2H4 + Cl ence, 12, 243.
Aris, R. (1961). The optimal design of chemical reactors. New York:
k10 =0.11015 exp(−19000/(1.987T)) s − 1
Academic Press.
Reaction P11: Balakrishna, S., & Biegler, L. T. (1992a). A constructive targeting
approach for the synthesis of isothermal reactor networks. I & EC
k11
C2H3 + third“ C2H2 +H + third Research, 31 (9), 300.
Balakrishna, S., & Biegler, L. T. (1992b). Targeting strategies for
k11 =0.81012 exp(−30600/(1.987T)) s − 1 synthesis and energy integration of non-isothermal reactor net-
works. I &EC Research, 31 (9), 2152.
Reaction P12: Balakrishna, S., & Biegler, L. T. (1993). A unified approach for the
k12
simultaneous synthesis of reaction, energy and separation systems.
C2H3Cl+H“ CH2CH2Cl Industrial and Engineering Chemistry, Research, 32 (7), 1372.
k12 =0.11011 exp(−0/(1.987T)) m3 kmol − 1 s − 1 Bolio, B. (1994). Guidelines for the SYNHEAT interface. Carnegie
Mellon University: Department of Chemical Engineering.
Reaction P13: Brooke, A., Kendrick, D., & Meeraus, A. (1988). GAMS: a user’s
k13 guide. Redwood City, CA: Scientific Press.
C2H3 + Cl+third “ C2H3Cl +third Chitra, S. P., & Govind, R. (1981). Yield optimization for complex
reactor systems. Chemical Engineering Science, 36, 1219.
k13 =0.11011 exp(+13000/(1.987T)) m3 kmol − 1 s − 1 Chitra, S. P., & Govind, R. (1985). Synthesis of optimal reactor
structures for homogenous reactions. AIChE Journal, 31 (2), 177.
Reaction P14: Cowfer, J. A., & Magistro, A. J. (1983). Vinyl chloride and poly(vinyl
k14 chloride). Kirk Othmer Encyclopedia, 23, 865.
C2H3 + C4H6 “ C6H9
Duran, M. A., & Grossmann, I. E. (1986). Simultaneous optimization
k14 =108.5 exp(− 3000/(1.987T)) m3 kmol − 1 s − 1 and heat integration of chemical processes. AIChE Journal, 32,
123.
Reaction P15: Feinberg, M., & Hildebrandt, D. (1997). Optimal reactor design from
k15 a geometric viewpoint. Chemical Engineering Science, 52 (10),
C6H9 “ C6H8 + H 1637.
Fjeld, M., Asbjornsen, O. A., & Astrom, K. J. (1974). Reaction
k15 =1013.5 exp(− 42200/(1.987T)) s − 1 invariants and their importance in the analysis of eigenvectors,
stability and controllability of CSTRs. Chemical Engineering Sci-
Reaction P16:
ence, 30, 1917.
k16
C6H8 “ C6H6 + H2 Gel’Perin, E. I., Bakshi, Y. M., Avetisov, A. K., & Gel’bshtein, A. I.
(1979). Kinetic model of the oxidative chlorination of ethane to
k16 =1013.5 exp(− 40000/(1.987T)) s − 1 vinyl chloride II. Kinetica I Kataliz, 20, 129.
Gel’Perin, E. I., Bakshi, Y. M., Avetisov, A. K., & Gel’bshtein, A. I.
Reaction P17: (1983). Kinetic model of the oxidative chlorination of ethane to
k17 vinyl chloride III. Kinetica I Kataliz, 24, 633.
CH2ClCHCl+Cl2 “ Cl +CH2ClCHCl2 Gel’Perin, E. I., Bakshi, Y. M., Avetisov, A. K., & Gel’bshtein, A. I.
(1984). Kinetic model of the oxidative chlorination of ethane to
k17 =108.8 exp(− 1000/(1.987T)) m3 kmol − 1 s − 1 vinyl chloride IV. Kinetica I Kataliz, 25, 842.
Glasser, D., Crowe, C., & Hildebrandt, D. (1803). A geometric
Reaction P18:
approach to steady flow reactors: the attainable region and opti-
K18
CH2ClCHCl2 l C2H2Cl2 +HCl mization in concentration space. Industrial Engineering Chemimal
K18 = k18f /k18b Research, 26 (9), 1987.
Hildebrandt, D., Glasser, D., & Crowe, C. (1990). The geometry of
k18f =0.21014 exp(− 58000/(1.987T)) s − 1 the attainable region generated by reaction and mixing with and
without constraints. Industrial Engineering Chemical Research,
k18b =0.3109 exp(−44000/(1.987T)) m3 kmol − 1 s − 1 29 (1), 49.
Horn, F. J. M. (1964). Attainable regions in chemical reaction
Reactions P9, P11, P12, P13, P14, P15 and P16 were technique. In The Third European Symposium on Chemical Reac-
not used in the GAMS model for the pyrolysis reactor, tion Engineering. London: Pergamon.
 A. Lakshmanan et al. / Computers and Chemical Engineering 23 (1999) 479–495
Horn, F. J. M., & Tsai, M. J. (1987). The use of adjoint variables in
the development of improvement criteria for chemical reactors.
Journal of Optimization Theory and Applications, 1 (2), 131.
Jackson, R. (1968). Optimization of chemical reactors with respect to
flow configuration. Journal of Optimization Theory and Applica-
tions, 2 (4), 240.
Karra, S. B., & Senkan, S. M. (1988). A detailed chemical kinetic
mechanism for the oxidative pyrolysis of CH3Cl. Industrial Engi-
neering Chemical Research, 27, 1163.
Kokossis, A. C., & Floudas, C. A. (1990). Optimization of complex
reactor networks – I. isothermal operation. Chemical Engineering
Science, 45 (3), 595.
Kokossis, A.C., & Floudas, C.A. Synthesis of non-isothermal reactor
networks. Annual AIChE meeting, San Francisco, CA (1991).
Kurtz, B. E. (1972). Homogeneous kinetics of methyl chloride chlori-
nation. Industrial Engineering Chemical Research, 11, 332.
Lakshmanan, A., & Biegler, L.T. (1995). Reactor network targeting
for waste minimization. In M. El-Halwagi and D. Petrides (eds)
Pollution Pre6ention 6ia Process and Product Modifications,
AIChE Symposium Series, Vol. 90, p. 128.
Lakshmanan, A., & Biegler, L. T. (1996a). Synthesis of optimal
chemical reactor networks. Industrial Engineering Chemical Re-
search, 35 (4), 1344.
Lakshmanan, A., & Biegler, L. T. (1996b). Synthesis of optimal
chemical reactor networks with simultaneous mass integration.
Industrial Engineering Chemical Research, 35 (12), 4523.
McPherson, R. W., Starks, C. M., & Fryar, G. J. (1979). Vinyl
chloride monomer… What you should know. Hydrocarbon Pro-
cessing, 75, 75.
Omtveit, T., Tanskanen, J., & Lien, K. (1994a). Graphical targeting
procedures for reactor systems. Computers and Chemical Engi-
neering, 18, S113.
Omtveit, T., Wahl, P. E., & Lien, K. M. (1994b). Decomposed
. .
495
algorithmic synthesis of reactorseparation-recycle systems. Com-
puters and Chemical Engineering, 18 (11 – 12), 1115 – 1124.
Papalexandri, K. P., & Pistikopoulos, E. N. (1996). Generalized
modular representation framework for process synthesis. AIChE
Journal, 42 (4), 1010.
Pibouleau, L. Floquet, & Domenech, S. (1988). Optimal synthesis of
reactor separator systems by nonlinear programming method.
AIChE Journal, 34, 163.
Pro/II Casebook. Vinyl Chloride Monomer Plant. Simulation Sci-
ences Inc. (1992).
Ranzi, E., Dente, M., Tiziano, F., Mullick, S., & Bussani, G. (1990).
Mechanistic modeling of chlorinated reacting systems. Chimica e
l’Industria (Milan), 72, 905.
Ranzi, E., Grottoli, M. G., Bussani, G., Che, S. C., & Zahng, G.
(1993). A new simulation program predicts EDC furnace perfor-
mances. Chimica e l’Industria (Milan), 75, 261.
Smith, E. M. B., & Pantelides, C. C. (1995). Design of reaction/sepa-
ration networks using detailed models. Computers Chemical Engi-
neering, 19, S83 – S88.
Turkay, M., & Grossmann, I.E. Structural flowsheet optimization
with complex investment cost functions. paper 90a. Annual AIChE
meeting, Chicago, IL, 1996.
Viswanathan, J. V., & Grossmann, I. E. (1990). A combined penalty
function and outer approximation method for MINLP optimiza-
tion. Computers and Chemical Engineering, 14, 769.
Wachi, S., & Morikawa, H. (1986). Liquid phase chlorination of
ethylene and 1,2-dichloroethane. Journal of Chemical Engineering
of Japan, 19, 437.
Weissman, M., & Benson, S. W. (1984). Pyrolysis of methyl chloride.
International Journal of Chemical Kinetics, 16, 307.
Yee, T. F., Grossmann, I. E., & Kravanja, Z. (1990). Simultaneous
optimization models for heat integration — II. heat exchanger
network synthesis. Computers in Chemical Engineering, 14, 1165.