3900 Ind. Eng. Chem. Res.

2003, 42, 3900-3913

A New Phase-Equilibrium Model for Simulating Industrial Nylon-6

Production Trains
Kevin C. Seavey, Neeraj P. Khare, and Y. A. Liu*
Honeywell Center of Excellence in Computer-Aided Design, Department of Chemical Engineering,
Virginia Polytechnic Institute and State University, Blacksburg, Virginia 24061

Thomas N. Williams
Honeywell International, Inc., 15801 Woods Edge Road, Colonial Heights, Virginia 23834

Chau-Chyun Chen
Aspen Technology, Inc., Ten Canal Park, Cambridge, Massachusetts 02141

This paper presents a thermodynamically consistent model for the phase equilibrium of water/
caprolactam/nylon-6 mixtures, based on the POLYNRTL (polymer nonrandom two-liquid)
activity-coefficient model. Our model predicts phase equilibrium for any binary or ternary mixture
containing water, -caprolactam, and nylon-6 at industrially relevant temperatures and pressures
with an average error of 1%. This paper also demonstrates its application to simulate a melt
train and a bubble-gas kettle train for industrial production of nylon-6. Prior literature makes
simplifying assumptions about liquid-phase (molten polymer) activities of water and -capro-
lactam; these assumptions are shown to be unrealistic.

1. Introduction facturing technologies to demonstrate the critical impact

of phase equilibrium. We then outline prior simulation
It is hard to overestimate the importance of nylon in studies that seek to understand and optimize these
the development of polymer science and in the com- processes and their deficiencies. Lastly, we discuss cur-
mercial growth of polymer applications. Nylon was rent problems facing the simulation of nylon-6 phase
discovered by Wallace Hume Carothers in 1935 and equilibrium and show how to address these problems.
produced commercially by DuPont and IG Farben 1.1. Commercial Nylon-6 Manufacturing Pro-
beginning in 1939.1 Nylon’s combination of strength, cesses. Figure 1 depicts a conventional nylon-6 produc-
toughness, and high melt temperature made it the first tion scheme.
engineering thermoplastic, capable of myriad uses. It The reactor feed typically contains fresh caprolactam
quickly found application in synthetic fibers, which monomer, recycled monomer, water, and desired addi-
continue to dominate its current production of 12.5 tives,9 such as chain-terminating agents, fire-retarding
billion lbs/year.2,3 Other high- volume applications agents, and delustrants. The reactor section hydrolyzes
include household articles, automotive parts, electrical the monomer, converts it to a polymer, and builds up
cable, and packaging. the polymer molecular weight.
There are two major types of nylon: (1) nylon-6,6 Because the molten polymer is in equilibrium with
made commercially by the condensation polymerization water and monomer, the byproduct water must be
of adipic acid with hexamethylenediamine; (2) nylon-6, removed by vaporization to increase the molecular
made from the ring opening of -caprolactam and its weight. However, vaporization of the water may also
subsequent polycondensation. While nylon-6,6 domi- remove a significant amount of caprolactam, which is
nates production in the U.S., nylon-6 is dominant in recovered and recycled.
Europe and Asia. Nylon-6 production accounts for In conventional processes, the polymer is then pel-
approximately 7 billion lbs/year of fibers and resins. letized and leached with hot water to remove low-
Considering the economic scale of nylon-6 production, molecular-weight extractables, such as residual mono-
it is not surprising that a number of polymerization mer and cyclic oligomers. The polymer is then dried and
models based on first principles have appeared.4-8 These shipped elsewhere or used directly to spin fibers.10
models attempt to apply knowledge of kinetics, physical By contrast, in direct melt processes, the extraction
and thermodynamic properties, mass and energy trans- and drying steps are avoided by devolatilizing the
port, and phase equilibrium to simulate a manufactur- polymer of unreacted monomer. Molten polymer can
ing process. We can apply an accurate model to improve then be fed directly to spinning or pelletizing. However,
the product quality, increase the production rate, and melt polymers are typically modified (terminated with
reduce production costs. monofunctional amines and carboxylic acids) to limit the
This paper addresses inaccuracies in phase equilib- molecular weight produced at the low levels of water
rium that limit the accuracy of these existing models. achieved when removing unreacted caprolactam. There-
In particular, we describe commercial nylon-6 manu- fore, direct melt processes have a limited product range
compared with conventional processes.
* To whom correspondence should be addressed. Tel.: (540) Within these two process variants, i.e., conventional
231-7800. Fax: (540) 231-5022. E-mail: [email protected]. and direct melt, a number of commercial technologies
10.1021/ie030112+ CCC: $25.00 © 2003 American Chemical Society
Published on Web 07/18/2003
 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003 3901

literature contains an accurate, fundamental thermo-

dynamic model. This is because there is considerable
uncertainty in the phase-equilibrium data themselves.8
A possible exception is the modeling study by Loth et
al.;17 however, they give no details regarding their
phase-equilibrium model.
Experimental phase-equilibrium measurements are
Figure 1. Block diagram of a conventional nylon-6 polymerization difficult to make because the water/caprolactam/nylon-6
process. system is reactive. Furthermore, the mixture water/
caprolactam is wide-boiling, with the normal boiling
Table 1. Nylon-6 Production-Recycling Technologies point of caprolactam above 260 °C. Characterizing this
Practiced on a Commercial Scale and Their Associated system and analyzing the resulting data require a deep
Operating Conditions
understanding of the underlying kinetics and thermo-
industrial process temperature (K)/pressure (kPa) range ref dynamics.
batch-semibatch 533/101.325-490.332 8, 12 In an ideal world, we would have many sets of
bubble-gas kettle 528/101.325 13 thermodynamically consistent phase-equilibrium data
VK column 517-555/101.325 5 for water/caprolactam, water/nylon-6, and caprolactam/
melt process 473-573/0.67-648 10
solid state 383-478/101.325 14
nylon-6. These data would cover a wide range of
depolymerization 513-673/10-1400 15 temperatures and pressures. Unfortunately, we suffer
from a lack of data, and the literature data that we do
are available to polymerize caprolactam. These tech- have are inconsistent.
nologies include batch, bubble-gas kettle, melt (termi- 1.3. Objectives of the Current Study. We develop
nated and unterminated resins), traditional VK (Vere- a thermodynamically consistent model for nylon-6 phase
infacht Kontinuierliches), and solid-state processes. equilibrium using the appropriate information out of
Even the depolymerization of nylon-6 to recover capro- scarce and inconsistent literature data. This develop-
lactam is practiced on a commercial scale.11 Table 1 ment involves characterizing the binary interactions
summarizes these major technologies along with their between caprolactam, water, and nylon-6.
approximate temperature and pressure conditions. This fundamental model represents an important
The temperature for industrial nylon processes ranges advance in the ability to simulate nylon-6 polymeriza-
from 383 to 673 K (110-400 °C). The pressure ranges tions because it facilitates integrated process modeling.
from 0.67 to 1400 kPa (0.007-13.8 atm). This strikingly When we perform integrated process modeling, we build
wide range of processing conditions underscores the unified models of entire manufacturing trains, not just
need for a fundamental phase-equilibrium model that reactors. In nylon-6 integrated process modeling, we
adequately represents all of these processes throughout simulate reactors, flash units, and condensers, all of
the entire temperature and pressure ranges. which together cover wide ranges of pressures and
1.2. Previous Nylon-6 Process Simulation Re- temperatures.
search. We can better understand chemical manufac- We rely entirely on literature data to develop the
turing operations by developing simulation models that present model. First, we demonstrate that the available
are based on the underlying engineering fundamentals. data for water/caprolactam are thermodynamically in-
These simulation models can be used in a variety of consistent, and then we use only the temperature-
beneficial ways: they speed up process development and pressure-liquid-mole-fraction (T-P-x) data to regress
optimization and can facilitate new product develop- binary interaction parameters for water/caprolactam.
ment. Then, by simulating the reactive system, water/capro-
Researchers who recognized the value of simulation lactam/nylon-6, we are able to extract binary interaction
began publishing models of nylon-6 polymerization in parameters for water/nylon-6 and caprolactam/nylon-6
the 1970s, with Giori and Hayes4 among the first. Initial segments.
models for most commercial nylon polymerization tech- We then validate our phase-equilibrium model by
nologies have now been documented, including VK comparing its predictions with plant experience for a
tube,5 plug-flow,6 wiped-film,7 and semibatch reactors.8 condenser and two finishing reactors. Lastly, we dem-
In the modeling of any multiphase chemical manu- onstrate the utility of our thermodynamic model by
facturing process, it is important to properly model the developing integrated process simulations for a melt
phase equilibrium of the process.16 Specifically, we must train and a bubble-gas kettle train for industrial nylon-6
pay particular attention to phase equilibrium when production.
modeling nylon polymerizations. This is because the
water concentration in the liquid phase is the over- 2. Literature Review
whelming factor in determining product characteristics,
such as the production rate and molecular weight.8 We split our literature review into three sections:
Furthermore, properly representing the thermodynam- nylon-6 polymerization kinetics, phase-equilibrium mod-
ics is the basis for fundamentally modeling mass- eling, and water/caprolactam/nylon-6 phase-equilibrium
transfer resistances in step-growth finishing reactors. data from the literature. We report nylon-6 polymeri-
If we cannot adequately model phase equilibrium, then zation kinetics for two reasons. First, we simulate the
we will not be able to properly diagnose the appearance reactive system to obtain the binary interaction param-
of significant mass-transfer limitations or calculate eters for water/nylon-6 and caprolactam/nylon-6 seg-
interfacial equilibrium compositions. ments. Second, we use the existing kinetic model, along
While it is clear that properly modeling the phase with our thermodynamic and mass-transfer models, to
equilibrium in nylon polymerizations is important, it is build integrated models for commercial nylon-6 pro-
unfortunate that virtually no nylon-6 model in the cesses.
3902 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003

Table 2. Nylon-6 Hydrolytic Polymerization Reaction

Mechanismsa
reaction name equilibrium reaction
k1
ring opening of caprolactam W + CL y\ z P1
k ′ 1

k2
polycondensation Pm + Pn y\ z Pm+n + W
k ′ 2

k3
polyaddition of caprolactam CL + Pn y\ z Pn+1
k ′ 3

k4
ring opening of cyclic dimer W + CD y\ z P2
k ′ 4
Figure 2. Nylon-6 molecules of degree of polymerization n
k5 existing as two types: unterminated (above) and terminated by
polyaddition of cyclic dimer CD + Pn y\ z Pn+2
k ′ AA (below). B-ACA represents nylon-6 repeat segments, T-NH2
5
represents amine end groups, T-COOH represents carboxylic acid
a P is aminocaproic acid. P is a nylon-6 molecule with a degree
end groups, and T-AA represents AA end groups.
1 n
of polymerization n.
(T-AA). Table 4 presents the reactions in Table 2
expressed using segment notation.
2.1. Nylon-6 Polymerization Kinetics. Arai et al.18 Table 4 also gives the associated reaction rates for
have presented the accepted standard regarding the each equilibrium reaction. Table 5 gives the species
chemistry and kinetics of the hydrolytic polymerization conservation equations corresponding to the reaction
of -caprolactam. Their reaction mechanism includes the rates of Table 4.
ring opening of caprolactam (CL), polycondensation, POLYMERS PLUS estimates the concentrations of
polyaddition of CL, ring opening of a cyclic dimer (CD), oligomers of degree of polymerization 2 and 3 below:
and polyaddition of a CD. Table 2 lists these five
equilibrium reactions.
Table 3 gives the accompanying rate constants ki,
again from Arai et al.18
[P2] ) [T-COOH] ( [T-NH2]
[B-ACA] + [T-NH2] ) (2)

( )
We add one more equilibrium reaction to this kinetic
scheme: the termination reaction with a monofunctional [B-ACA]
[P3] ) [T-COOH]
acid, such as acetic acid (AA). [B-ACA] + [T-NH2]

AA + Pn y\
k
k2′
z Pnx + W
2
(1) ( [T-NH2]
[B-ACA] + [T-NH2] ) (3)

We assume that this reaction follows the same kinetics We compute the number-average degree of polymeri-
as the polycondensation reaction, as in ref 5. zation, DPn, by considering the distribution of polymer
We ignore the analysis of cyclic oligomers higher than chain lengths. This distribution contains m different
dimers for the sake of simplicity. chain lengths, with each chain length i characterized
We use Aspen Technology’s (Cambridge, MA) com- by a population of Ni chains with degree of polymeri-
mercial simulation package POLYMERS PLUS to simu- zation DPi:
late nylon-6 polymerizations. This package implements
m

∑
the above reaction-kinetics model using the segment-
based methodology detailed in the patent by Barrera NiDPi
i)1
et al.19 This methodology tracks the polymer concentra- DPn ) (4)
tion and number-average degree of polymerization by m
tracking the concentrations of their constitutive seg-
ments. Figure 2 shows the segmental breakdown of
∑
i)1
DPi
nylon-6 molecules that can be either unterminated or
terminated by AA. In terms of segments, the numerator of eq 4 represents
Nylon-6 segments include the nylon-6 repeat segment the total concentration of segments that count as repeat
(B-ACA) and the end groups terminal amine (T-NH2), units, or ([B-ACA] + [T-COOH] + [T-NH2] + [P1]). The
terminal carboxylic acid (T-COOH), and terminal AA denominator represents the concentration of polymer

Table 3. Rate Constants for the Equilibrium Reactions in Table 2

rate constant
expression ki ) A0i exp -( ) E0i
RT ( )
+ Aci exp -
Eci
RT
[T-COOH]

equilibrium constant
expression Ki )
ki
ki′
) exp (
∆Si - ∆Hi/T
R )
i A0i (kg/mol‚s) E0i (J/mol) Aci (kg2/mol2‚s) Eci (J/mol) ∆Hi (J/mol) ∆Si (J/mol‚K)
1 1.66 × 102 8.32 × 104 1.20 × 104 7.87 × 104 8.03 × 103 -33.01
2 5.26 × 106 9.74 × 104 3.37 × 106 8.65 × 104 -2.49 × 104 3.951
3 7.93 × 105 9.56 × 104 4.55 × 106 8.42 × 104 -1.69 × 104 -29.08
4 2.38 × 108 1.76 × 105 6.47 × 108 1.57 × 105 -4.02 × 104 -60.79
5 7.14 × 104 8.92 × 104 8.36 × 105 8.54 × 104 -1.33 × 104 2.439
 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003 3903

Table 4. Nylon-6 Polymerization Reactions Written in Segment Notationa

equilibrium reaction reaction rate
k1
Ring-Opening of Caprolactam (W + CL y\ z Pn)
k ′ 1
k1
CL + W \
yk ′)k /K z P1 R1 ) k1[CL][W] - k1′[P1]
1 1 1
k2
Polycondensation (Pm + Pn y\ z Pm+n + W)
k ′ 2
k2
P1 + P1 y\
k ′)k /K
z T-COOH:T-NH2 + W R2 ) k2[P1]2 - k2′[P2][W]
2 2 2

( )
k2 [B-ACA]
P1 + T-COOH y\
k ′)k /K
z T-COOH:B-ACA + W R3 ) k2[P1][T-COOH] - k2′[W][T-COOH]
2 2 2 [B-ACA] + [T-NH2]

T-NH2 + P1 y\
k ′)k /K 2
k2
z T-NH2:B-ACA + W
2 2
R4 ) k2[T-NH2][P1] - k2′[W][T-NH2] ( [B-ACA]
[B-ACA] + [T-COOH] )
( )
k2 [B-ACA]
T-NH2 + T-COOH y\
k ′)k /K
z B-ACA:B-ACA + W R5 ) k2[T-NH2][T-COOH] - k2′[W][B-ACA]
2 2 2 [B-ACA] + [T-NH2]
k3
Polyaddition of Caprolactam (CL + Pn y\ z Pn+1)
k ′ 3

k3
P1 + CL y\
k ′)k /K
z T-NH2:T-COOH R6 ) k3[P1][CL] - k3′[P2]
3 3 3

T-NH2 + CL y\
k ′)k /K
z T-NH2:B-ACA
3
k3

3 3
R7 ) k3[T-NH2][CL] - k3′[T-NH2] ( [B-ACA]
[B-ACA] + [T-COOH] )
k4
Ring-Opening of Cyclic Dimer (W + CD y\ z P2 )
k ′
k4 4

CD + W \
yk ′)k /K z T-COOH:T-NH2 R8 ) k4[CD][W] - k4′[P2]
4 4 4

k5
Polyaddition of Cyclic Dimer (CD + Pn y\ z Pn+2)
k ′ 5
k5
P1 + CD y\
k ′)k /K
z T-NH2:B-ACA:T-COOH R9 ) k5[P1][CD] - k5′[P3]
5 5 5

T-NH2 + CD y\
k ′)k /K
z B-ACA:B-ACA:T-NH2
5
k5

5 5
R10 ) k5[T-NH2][CD] - k5′[T-NH2] ( [B-ACA]
[B-ACA] + [T-COOH] )
k3
Polycondensation of Acetic Acid (Pn + AA y\ z Pn,T-AA)
k ′ 3

P1 + AA y\
k ′)k /K
2
k2
z T-AA:T-COOH + W
2 2
R11 ) k2[AA][P1] - k2′[W][T-AA] ( [T-COOH]
[T-COOH] + [B-ACA] )
T-NH2 + AA y\
k ′)k /K
z B-ACA:T-AA + W
2
k2

2 2
R12 ) k2[AA][T-NH2] - k2′[W][T-AA] ( [B-ACA]
[T-COOH] + [B-ACA] )
a A colon represents a covalent bond between segments.

Table 5. Species Conservation Equations for the lymerization, nor do we compute higher moments of the
Reaction Rates in Table 4 distribution of the DP.
functional 2.2. Polymer Equilibrium. 2.2.1. Previous At-
group time rate of change tempts at Developing a Phase-Equilibrium Model
W d[W]/dt ) R2 + R3 + R4 + R5 + R11 + R12 - (R1 + R8) for Water/Caprolactam/Nylon-6. Researchers typi-
CL d[CL]/dt ) -(R1 + R6 + R7) cally use one of three phase-equilibrium models in
CD d[CD]/dt ) -(R8 + R9 + R10) simulating nylon-6 polymerizations: the Jacobs-Sch-
AA d[AA]/dt ) -(R11 + R12) weigman model,20 the Fukumoto model,21 or the Tai et
P1 d[P1]/dt ) R1 - (2R2 + R3 + R4 + R6 + R9 + R11) al. model.12 These models only predict water concentra-
B-ACA d[B-ACA]/dt ) R3 + R4 + 2R5 + R7 + R9 + 2R10 + R12
T-NH2 d[T-NH2]/dt ) R2 + R6 + R8 + R9 - (R5 + R12) tions in reacting mixtures of water, caprolactam, and
T-COOH d[T-COOH]/dt ) R2 + R6 + R8 + R9 + R11 - (R5) nylon-6.
T-AA d[T-AA]/dt ) R11 + R12 The Jacobs-Schweigman model20 is the most sim-
plistic of the three models. It consists of an empiricism
chains, or ([T-COOH] + [P1]). Therefore, we rewrite the based on experimental equilibrium data for a VK
number-average degree of polymerization below: (Vereinfacht Kontinuierliches) tube reactor. The model
predicts the concentration of water [W] as a function of
[B-ACA] + [T-NH2] + [T-COOH] + [P1] temperature T:
DPn ) (5)
[T-COOH] + [P1]
1.76 - 0.006T (°C)
[W] (mol/kg) ) (6)
We do not compute the weight-average degree of po- 1.8
3904 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003

Table 6. Liquid-Phase, Water Mole-Fraction Predictions 2. POLYNRTL takes advantage of the already exist-
at 1 atm for the Jacobs-Schweigman,20 Fukumoto,21 and ing database of NRTL binary interaction parameters.
Tai et al.12 Phase-Equilibrium Models
On the basis of these suggestions, we choose to model
temp (K) Jacobs-Schweigman Fukumoto Tai et al. nylon-6 phase equilibrium using the POLYNRTL prop-
473 0.006 0.017 0.103 erty method. This method uses (1) the polymer NRTL
493 0.004 0.008 0.050 activity-coefficient model for the liquid phases, (2) the
513 0.003 0.004 0.025 Redlich-Kwong26 equation of state for the vapor phase,
533 0.002 0.002 0.013 (3) the van Krevelen27 model for the liquid properties
(enthalpy, entropy, Gibbs free energy, heat capacity, and
Because this expression does not contain pressure as a molar volume), and (4) Henry’s law for any supercritical
free variable and was developed using VK tube equi- components.
librium data, it should not be used at pressures deviat- The POLYNRTL activity-coefficient model combines
ing from atmospheric pressure. the traditional NRTL model with the FH description for
Adding the system pressure P to the Jacobs-Sch- configurational entropy. It essentially calculates the
weigman model results in the Fukumoto model21 based Gibbs free energy of mixing in a polymer solution as
on vapor-liquid equilibrium (VLE) data: the sum of two contributions: (1) the entropy of mixing

( )
8220 from the FH activity-coefficient model; (2) the enthalpy
[W] (mol/kg) ) P (mmHg) exp - 24.0734 (7) of mixing from the NRTL activity-coefficient model.28
T (K) These activity coefficients include binary parameters to
The third model is the Tai et al. model,12 which model the interactions between two components. POLYN-
explicitly considers the partial pressures of volatile RTL represents the temperature dependence of the
components. It consists of a system of three equations binary interaction parameters (τij) by eq 11. We set the
to solve for the mole fraction of water, xW, given the
bij
temperature T and total pressure P: τij ) aij + + cij ln T (11)
T
log
[ xW
PW (mmHg) ] )
3570
T (K)
- 11.41 (8) nonrandomness factor Rij to 0.3, as suggested by Praus-
nitz et al.29
Therefore, we model binary interactions in equilibri-
4100
log[PCL (mmHg)] ) - + 9.6 (9) um mixtures by specifying the coefficients aij through
T (K) cij. We typically regress equilibrium data or use a
predictive model, such as UNIFAC, to obtain the values
P ) PW + PCL (10) of these parameters.
2.3. Equilibrium Data for Water/Caprolactam/
PW is the partial pressure of water, and PCL is the Nylon-6. There are two sources of phase-equilibrium
partial pressure of caprolactam. data that we use for regression. The first source
All three models are empirical. Therefore, we cannot characterizes the binary interactions between water and
be confident in their phase-equilibrium predictions at caprolactam. The second characterizes the interactions
conditions that deviate from those in which they are in the reactive, ternary system nylon-6/caprolactam/
based. water. All of the data sets appear in the Supporting
Significantly, these three models give conflicting Information.
predictions for identical process conditions. Consider, Maczinger and Tettamanti30 give four sets of low-
for example, the water liquid-mole-fraction predictions pressure, isobaric phase-equilibrium data for the binary
at atmospheric pressure for a mixture of water, capro- water/caprolactam. Tables 15-18 in the Supporting
lactam, and nylon-6 at a range of relevant processing Information show their data.
temperatures of 473-523 K (Table 6). Giori and Hayes31 present one set of isothermal
The Fukumoto model21 predicts mole fractions that phase-equilibrium data for the ternary system water/
are about twice those of the Jacobs-Schweigman20 caprolactam/nylon-6. They carry out nylon-6 polymer-
predictions. Furthermore, the Tai et al. model12 predicts izations at 543 K in a laboratory reactor and charac-
mole fractions that are an order of magnitude higher terize the resulting phase equilibrium. Table 19 in the
than both models. This difference casts considerable Supporting Information shows their experimental data.
doubt on the validity of at least two of these models.
2.2.2. Polymer Nonrandom Two-Liquid (POLYN- 3. Methodology
RTL) Model. Bokis et al.16 give an excellent description
of how to choose an appropriate thermodynamic equi- We first characterize the binary interactions between
librium model when simulating polymer processes. They water and caprolactam using the first four data sets30
suggest using activity-coefficient models, instead of (Tables 15-18 in the Supporting Information).
equations of state, for processes that involve low-to- We then simulate the Giori and Hayes experiments31
moderate pressures (pressure < 1 × 106 Pa) and/or to quantify the binary interactions between water/
nonideal components (e.g., polar compounds such as nylon-6 and caprolactam/nylon-6.
alcohols, water, and ketones). 3.1. Characterizing the Phase Equilibrium of
Furthermore, they state that the POLYNRTL activ- Water/Caprolactam. Here, we explain how to obtain
ity-coefficient model22 has the following advantages over the binary parameters for water/caprolactam. Our
the Flory-Huggins (FH)23,24 and universal quasi-chemi- handling of the data consists of testing the data for
cal functional group (UNIFAC)25 activity-coefficient thermodynamic consistency. We use two consistency
models: tests for our isobaric data sets: a Van Ness et al. test32
1. POLYNRTL covers large ranges of temperatures and a Wisniak test.33 The Wisniak test can be used in
and compositions accurately. a point or area mode; however, we use only the point
 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003 3905

test because of the difficulty in accurately of computing mentally based and preferable to a combined Redlich-
integrals of “functions” of our experimental data. Kister34/Herington method35 for testing the consistency
3.1.1. Consistency Test One: Van Ness et al.32 In of isobaric phase-equilibrium data.
this test, we first obtain a redundant data set comprised For each data point i at system temperature Ti, we
of temperature, pressure, liquid-mole-fraction, and compute two functions Li and Wi:
vapor-mole-fraction measurements (T-P-x-y). We then
regress only the T-P-x data. After regressing these
data, we are able to predict the original pressure data Li )
∑T0kxk∆s0k - T (17)
i
∆s
(using T-x data) and vapor-mole-fraction data (using

(∑ )
T-P-x data). Consistency test one comes by examining
( xk ln γk) - w
the extent of corroboration between the measured and Wi ) RTi (18)
predicted vapor mole fractions. This test is an indirect ∆s
application of the Gibbs-Duhem equation because the
POLYNRTL model obeys the Gibbs-Duhem equation. For every species k, at the system pressure, there is a
The fundamental phase-equilibrium relation for this pure-component boiling point T0k, a liquid mole fraction
analysis is xk, and entropy of vaporization ∆s0k. In eq 18, ∆s is the
entropy of mixing of the mixture.
yiφiP ) xiγi(xi,T) Psat
i (T) exp [
vi
RT
(P - Psat
i ) ] (12)
We compute the boiling temperature T0k for each
component k by setting the pressure to the system
pressure in eq 13 and then backing out the temperature.
We compute the activity coefficient γi using the POLYN- We calculate the entropy of vaporization from the
RTL model; it is a function of the liquid composition xi enthalpy of vaporization ∆h0k:
and temperature T. Psat

( )
i is the vapor pressure, which
B+C(T0k/Tc,k)+D(T0k/Tc,k)2
we compute using an Antoine form: T0k
Ak 1 -
Psat
i ( Bi
) exp Ai + + Ci ln T + DiTEi
T ) (13) ∆s0k )
∆h0k
T0k
)
Tc,k
T0k
(19)

We have neglected vapor-phase nonidealities (fugacity The critical temperature Tc,k as well as the constants
coefficient φi equals 1) because most of the data are at Ak through Dk are tabulated in standard reference
vacuum conditions; however, one can easily estimate the sources (such as Daubert and Danner36) and differ for
degree of vapor-phase nonideality occurring at high each chemical species.
pressures using the Redlich-Kwong equation of state. We find the activity coefficients from the experimental
Furthermore, the Poynting pressure correction is neg- data using eq 12:
ligible (the exponential term is 1).
We eliminate the need for vapor-phase composition γi ) yiP/xiPsat
i (T) (20)
data by summing eq 12 over all species; for a binary
system containing species i and j, we have We compute the mixture entropy of vaporization for the
mixture ∆s using the following mixing rule:
P ) xiγi(xi,T) Psat
i (T) + xjγj(xj,T) Psat
j (T) (14)

Using the T-P-x data, we manipulate the six POLYN-

∆s ) ∑xk∆s0k (21)
RTL binary interaction parameters (aij through cij for Last, we find the w term appearing in eq 18 using
water/caprolactam and caprolactam/water; eq 11) to
minimize the following sum-of-squares error (SSE) for
each data point k:
w) ∑xk ln(yk/xk) (22)

( )
We perform the point test by computing the ratio of
Pobsd - Pcalcd 2

∑k
k k Li and Wi and plotting this as a function of liquid mole
SSE ) (15) fraction. Inconsistent data contain ratio values that are
Pobsd
k much different than 1 and do not scatter randomly
about 1.
Once we have the binary interaction parameters, we 3.1.3. Determining Binary Interaction Param-
predict the vapor composition using the original T-P-x eters from Phase-Equilibrium Data. The Van Ness
data and eq 12. Last, we compute the percent difference et al. test32 effectively regresses binary interaction
in vapor-mole-fraction data vs prediction for each liquid- parameters using T-P-x data only. If the data are
composition data point: judged inconsistent, Prausnitz et al.37 suggest that we

( )
assume that the vapor-phase composition data are
ycalcd - yobsd incorrect for two reasons:
percent deviation ) 100 (16)
yobsd (i) Accurately measuring y is typically more difficult
than measuring x and pressure.
We plot the percent deviation vs the liquid mole fraction (ii) Composition data at the ends of the composition
to characterize the consistency of the observed phase- scale (y ∼ 0 or 1) are likely to be the least accurate.
equilibrium data. If the data are inconsistent and we throw out the
3.1.2. Consistency Test Two: Wisniak.33 The vapor-mole-fraction measurements, then we have al-
second consistency test that we use, due to Wisniak, is ready obtained binary interaction parameters using the
based on the bubble-point equation for mixtures and data that we are most confident in (T-P-x data).
utilizes the Clausius-Clapeyron equation. It is funda- However, if the data are consistent, then we can use
3906 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003

Table 7. Physical Property Constants for Water and

Caprolactama
constant water caprolactam
vapor pressure
Ai 7.3649 × 101 7.4172 × 101
Bi -7.2582 × 103 -1.0469 × 104
Ci -7.3037 -6.8944
Di 4.1653 × 10-6 1.2113 × 10-18
Ei 2.0000 6.0000
enthalpy of vaporization 5.2053 × 107 8.3520 × 107
Ai
Bi 3.1990 × 10-1 3.7790 × 10-1
Ci -2.1200 × 10-1 0.0000
Di 2.5795 × 10-1 0.0000
critical temperature (K) 806.00 647.13
a The vapor pressure is computed using eq 13, with units of Pa

and temperature units of K. The enthalpy of vaporization is

computed using eq 19, with units of J/kmol and temperature units
of K.
Figure 3. Percent error in vapor-mole-fraction prediction vs liquid
mole fraction for caprolactam for all water/caprolactam data sets
all of the data to compute activity coefficients by eq 20 (Van Ness et al. test32).
and re-regress the binary interaction parameters.
3.2. Characterizing the Ternary, Reactive Sys-
tem Water/Caprolactam/Nylon-6. By completing the
first step of this study, section 3.1, we obtain the binary
interaction parameters for the system water/caprolac-
tam. We then simulate the Giori and Hayes experi-
ments31 in order to characterize the binary interactions
between water/nylon-6 and caprolactam/nylon-6 seg-
ments. We use POLYMERS PLUS to model the batch
step-growth polymerizations. However, our methodology
is applicable in any polymer process simulator because
the kinetics and thermodynamics models are reproduced
from open literature sources.
We simulate a 0.000 077 m3 (77 mL) reactor using the
Arai et al. kinetic scheme18 and the binary interaction
parameters for water/caprolactam. We maintain a vapor-
liquid system throughout the polymerization. After Figure 4. L/W ratio vs caprolactam liquid mole fraction for all
water/caprolactam data sets (Wisniak test33).
reaction equilibrium is reached in the closed vessel, we
predict the final liquid and vapor compositions, as well
as the final pressure. We give our simulation procedure property constants for water and caprolactam, found in
below, with specifications given in Giori and Hayes.31 Daubert and Danner.36 We need these parameters to
1. Charge the batch reactor of a specified size with a compute the vapor pressure and enthalpy of vaporiza-
specified amount of water and caprolactam. tion.
2. Set the reactor to a specified temperature and Figure 3 shows the results of the Van Ness et al. test32
enforce VLE using the POLYNRTL activity-coefficient for thermodynamic consistency: a plot of the percent
model, thereby predicting the liquid-vapor composition error in vapor mole fraction vs liquid mole fraction for
and reactor pressure. caprolactam.
3. Allow the nylon-6 polymerization reaction to pro- Consistent data show small, randomly distributed
ceed until reaction equilibrium is established. errors around the x axis. Inconsistent data give large,
4. Report the final liquid and vapor compositions, as nonrandomly distributed errors. It is clear from Figure
well as the final reactor pressure. 3 that all data sets are inconsistent.
Because we only have five sets of batch reaction data Similarly, for the Wisniak test,33 we show a ratio plot
from Giori and Hayes,31 we must minimize the number in Figure 4.
of free parameters in fitting the reaction data. We Consistent data would reveal a ratio that is more-or-
regress the bij parameters for water/nylon-6 and capro- less 1 for each liquid-mole-fraction data point. However,
lactam/nylon-6 segment interactions. We fix the aij and all of the data show a bias, with the ratio being 2 or
cij parameters to zero. higher for all sets. This again suggests the presence of
We vary these four binary interaction parameters, two a systematic error in the phase-equilibrium data.
for each binary interaction, until we match the outlet Unfortunately, consistent tests never tell us which
compositions and pressures from the data of Giori and data are not constant. As stated previously, we assume
Hayes.31 The resulting binary interaction parameters that the vapor-phase composition data are incorrect;37
characterize the systems, water/nylon-6 and caprolac- unfortunately, no error analysis was performed in the
tam/nylon-6 segments. works of Maczinger and Tettamanti30 and Giori and
Hayes.31
4. Results and Discussion Table 8 summarizes the regressed binary interaction
parameters using only T-P-x data.
4.1. Binary Interaction Parameters for Water/ Figure 5 shows the plot of the POLYNRTL param-
Caprolactam. Table 7 contains the pertinent physical eters τij vs temperature.
 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003 3907

Table 8. Regression Results for POLYNRTL Binary Table 9. Comparison of Model Predictions with Giori
Interaction Parameters for Water/Caprolactam (Based and Hayes Polymerization Data31
on T-P-x Data Only)
caprolactam liquid water liquid water vapor total pressure
POLYNRTL binary value (temperature mole fraction mole fraction mole fraction (Pa)
interaction parameter units are K)
Data
awater/CL -0.313 0.096 0.012 0.74 571 000
aCL/water 0.628 0.097 0.0249 0.85 974 000
bwater/CL -15.4 0.099 0.0387 0.886 1 400 000
bCL/water -13.7 0.101 0.0515 0.89 1 830 000
cwater/CL 0.0495 0.106 0.065 0.91 2 160 000
cCL/water -0.0898 Model Prediction
0.087 0.012 0.69 537 000
0.086 0.025 0.84 1 010 000
0.086 0.039 0.90 1 440 000
0.085 0.052 0.93 1 830 000
0.085 0.066 0.95 2 180 000
% Error
-9.38 0.00 -6.76 -6.02
-11.34 0.40 -1.18 4.02
-13.13 0.78 1.58 3.25
-15.84 0.97 4.49 0.28
-19.81 1.54 4.40 0.99

we shall see below, the ternary system water/caprolac-

tam/nylon-6 shows moderate deviations from ideal
behavior.
4.2. Binary Interaction Parameters for Water/
Nylon-6 and Caprolactam/Nylon-6 Segments. With
these binary interaction parameters, we turn to simu-
lating the Giori and Hayes data.31 We manipulate the
second binary interaction parameters bij for small
Figure 5. Predicted value of the POLYNRTL binary parameter molecule/nylon-6 segment interactions. We obtain model
τij for water/caprolactam and caprolactam/water. predictions and compare them with the Giori and Hayes
data in Table 9.
The binary interaction parameters are
bwater/nylon-6 segment ) 297, bnylon-6 segment/water ) -601,
bcaprolactam/nylon-6 segment ) 265, and bnylon-6 segment/caprolactam
) 207 (temperature units in K).
The average model prediction error for water compo-
sition is 0.74%. Note that we are most interested in this
composition because it significantly impacts the final
polymer properties, such as polymer molecular weight
and conversion. The model underpredicts the final
caprolactam liquid composition by about 13.9%. The
average error for the final pressure prediction is -0.51%.
Table 10 compares the Giori and Hayes data31 with
the three previous literature models, along with ideal
vapor-liquid predictions (fugacity-activity coefficients
equal 1).
Figure 6. Water/caprolactam T-x-y diagram. The solid lines The Jacobs-Schweigman model20 is ill-applied; sys-
represent the POLYNRTL activity coefficient model predictions tem pressures for the Giori and Hayes data31 range from
for the liquid mole fraction, while the dotted lines represent the 5 to 21 atm. The predictions for this model are about
POLYNRTL activity coefficient model predictions for the vapor
mole fraction.
an order of magnitude different from the experimental
data. This unacceptable level of error illustrates the
The binary interaction parameter τij is not signifi- inappropriateness of the Jacobs-Schweigman model for
cantly different from zero over a wide temperature modeling general phase equilibrium.
range. This suggests that the energy of interaction does The Fukumoto model predictions21 are the best out
not largely differ between unlike interactions and like of the three literature models: the average error is
interactions (i.e., near-ideal solution behavior). -31%. The Tai et al. model12 is surprisingly inaccurate,
We plot the experimental data vs POLYNRTL model given that it is the most advanced out of the three. The
calculations for each of the four isobars in Figure 6. average error of the Tai et al. model is 317%. Lastly
The POLYNRTL model accurately correlates the neglecting nonideality results in water liquid composi-
phase-equilibrium data. Furthermore, assuming ideal tion errors of 7.52%. Because water-concentration terms
behavior in the vapor and liquid phases gives nearly in the kinetic expressions are proportional to water mole
identical predictions. Therefore, we may safely neglect fractions in the liquid phase, we can expect that this
nonidealities in water/caprolactam solutions under error would cause errors of ca. 8% when computing
vacuum. This means that we can set the fugacity and reaction-rate terms. A likely explanation as to why ideal
activity coefficients to 1 when simulating water/capro- behavior approximates the Giori and Hayes data31 is
lactam mixtures under vacuum conditions. However, as that the excess Gibbs energy of mixing is low when the
3908 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003

Table 10. Predictions for Liquid-Phase Water Mole Fractions of Previous Literature Models and the Ideal Model and
Comparison with Experimental Data of Giori and Hayes31
ideal (fugacity and
liquid-phase water activity coefficients
mole-fraction data Jacobs-Schweigman20 Fukumoto21 Tai et al.12 set to 1)
0.0119 0.0014 0.0101 0.0607 0.0129
0.0249 0.0014 0.0181 0.1100 0.0269
0.0387 0.0014 0.0248 0.1509 0.0411
0.0515 0.0014 0.0323 0.1974 0.0554
0.0650 0.0014 0.0382 0.2338 0.0698

liquid-phase water Jacobs-Schweigman Fukumoto Tai et al. ideal error

mole-fraction data error (%) error (%) error (%) (%)
0.0119 -88 -15 410 8.47
0.0249 -94 -27 342 7.92
0.0387 -96 -36 290 6.21
0.0515 -97 -37 283 7.64
0.0650 -98 -41 260 7.33

composition is at extreme values (the polymer fraction

is near 0 or 1).
By way of comparison, our new phase-equilibrium
model generates an average prediction error for the
water liquid mole fraction of 1%. More importantly, it
allows us to make phase-equilibrium predictions in any
binary or ternary mixture that primarily contains water,
caprolactam, and nylon-6 at any specified temperature
and pressure.
We note that we do not have enough batch reaction
data to determine a unique set of binary interaction
parameters for water/nylon-6 and caprolactam/nylon-6
segments. Therefore, we have regressed the minimum
number of binary interaction parameters possible, set-
ting aij and cij to zero and only regressing bij. The fit to
the available data is good; however, because of the
limited data, we caution against extrapolation far away
from 270 °C and outside of the pressure range of 570-
2200 kPa.
Figure 7. Mass flow rates exiting the condenser: model predic-
tion vs plant experience (all values have been normalized using
5. Validation of Regressed Binary Interaction
the largest flow rate).
Parameters
We consider two model predictions to be good indica- semiterminated nylon-6. The molecular weight of the
tors of the quality of our binary interaction param- incoming polymer is 11.5 kg/mol. The reactor is operat-
eters: condenser performance and devolatilization in ing at ca. 520 K and 4 kPa with a residence time of 600
finishing reactors. Here, we perform exploratory simu- s.10
lations of a commercial condenser and two wiped-film By way of validation, we are looking at the devola-
evaporators. tilization behavior of small molecules, namely, capro-
We start with the analysis of the condenser. The feed lactam. From prior experience,38 we know that devola-
contains a mixture of caprolactam and water. The tilization of caprolactam is mass-transfer-limited. There-
condenser operates at a temperature of 483 K and a fore, if we ignore mass-transfer limitations in the model,
pressure of 53 kPa. When we simulate the condenser we should devolatilize too much caprolactam and there-
as a single flash unit, we get predictions for the liquid fore underpredict the mass fraction of caprolactam in
and vapor stream flow rates. Figure 7 compares our the exiting melt stream.
predictions with plant experience. Our model predicts a caprolactam mass fraction in
The vapor-phase flow rates of caprolactam and water the exit stream of 1.1%, while plant data show 2.9% for
are predicted with an average error of about 4%. The similar operating conditions. This is an underprediction
model underpredicts the liquid-phase component flow by 62%. This shows that our phase-equilibrium predic-
rates of caprolactam and water by about 9%. tions are in the right direction considering mass-transfer
It is significant to note the split fractions: the data limitations. What would not be reasonable is an equi-
show that about 49% of the caprolactam is vaporized, librium prediction for the caprolactam percentage that
while 99% of the water is vaporized. Our phase- is actually more than the mass-transfer-limited case.
equilibrium predictions nearly match the data, showing Regarding the caprolactam mass percentage in the
53% for caprolactam vaporization and 99% for water exiting melt, the equilibrium case should always be
vaporization. lower than the mass-transfer-limited case.
In our model of the first wiped-film evaporator, we Now we simulate a second wiped-film evaporator.
simulate a vapor-liquid plug-flow reactor, with a feed Once again, we consider a vapor-liquid plug-flow reac-
stream containing 2.00 × 10-3 parts by weight water, tor, with a feed stream containing 2.41 × 10-3 parts by
1.54 × 10-1 parts caprolactam, and 8.44 × 10-1 parts weight water, 1.4 × 10-1 parts caprolactam, and 8.58
 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003 3909

× 10-1 parts semiterminated nylon-6. The molecular 6.1. Melt Train. We simulate a typical commercial
weight of the incoming polymer is 9.8 kg/mol. This melt train involving four reactors.10 The first two
reactor is operating at ca. 523 K and 4.67 kPa with a reactors carry out monomer conversion; the third and
residence time of 1740 s.10 fourth reactors carry out devolatilization and molecular
Our model predicts a caprolactam mass fraction in weight buildup. Figure 8 shows the process-flow dia-
the exit stream of 1.5%, while plant data show 2.7% for gram for this train.
similar operating conditions. This is an underprediction The makeup stream contains 5.35 × 10-5 kg/s of
by 44%. This again shows that our phase-equilibrium water, 0.0120 kg/s of caprolactam, and 2.01 × 10-5 kg/s
calculations are in the right direction considering mass- of AA terminator. We have assumed a production rate
transfer limitations. of about 0.0112 kg/s of polymer or 89 pounds per hour
Regarding the wiped-wall evaporator studies, one may (pph) and have calculated approximate flow rates for
ask the following valid question: may we approximate water and terminator based on refs 10 and 41-44.
the error of the phase-equilibrium model from these The first vessel is a continuous stirred-tank reactor
results, as in the condenser study? Unfortunately, this (CSTR), STAGE-1 in Figure 8. The residence time is
is not possible because we do not know to what extent 8100 s, the temperature is 488 K, and the pressure is
mass-transfer limitations are affecting the reactor 545 kPa.
performance. In other words, if there is no mass-transfer The second vessel is a plug-flow reactor (PFR),
limitation, the phase-equilibrium model is in error by STAGE-2 in Figure 8. The feed to this vessel enters a
-62 to -44%. If there is a strong mass-transfer limita- vapor headspace, which we simulate using a flash unit
tion, then the phase-equilibrium model may have little at reactor conditions (HD-SPACE in Figure 8). Vapor
or no error. from this flash enters the reflux condenser (RFLX-CND
in Figure 8). Most of the caprolactam is returned back
6. Simulating Integrated Industrial Nyon-6 to the reactor headspace, while a mixture of caprolactam
Production Trains and water goes to the train condenser (CONDENSE in
Figure 8). The reactor has a residence time of 28 800 s,
Previous attempts at nylon-6 integrated process mod- a temperature of 498 K, and a pressure of 66.7 kPa.
eling were hindered by the lack of a fundamental
The third reactor is a vapor-liquid plug-flow reactor,
thermodynamic phase-equilibrium model. For example,
EVAPORAT in Figure 8. The vapor phase is removed
Nagasubramanian and Reimschuessel39 simulated the
via a vacuum system (EVAP-FL in Figure 8) and enters
molecular weight buildup in the finishing stage by first
the train condenser CONDENSE. This reactor has a
“instantaneously” removing all water from the polym-
residence time of 1200 s, a temperature of 523 K, and a
erization mass while ignoring caprolactam devolatiliza-
pressure of 2.33 kPa.
tion. There is no corresponding unit operation that can
perform this separation in an actual plant. Further- The fourth reactor is a vapor-liquid plug-flow reactor,
more, this assumption has led Tirrell et al.40 to an FINISHER in Figure 8. The vapor phase is removed via
unusual conclusion: that it is preferable to perform a vacuum system (FIN-FL in Figure 8) and also enters
molecular weight buildup and monomer conversion in the train condenser CONDENSE. We used a sweep-
a single reactor. To have adequate molecular weight steam mass flow rate of ca. one-quarter of that of the
buildup, we need to enforce low water concentrations. polymer flow rate entering the reactor. This reactor has
However, to do this, we typically use severe operating a residence time of 7200 s, a temperature of 536 K, and
conditions, such as high vacuum in melt processes,10 to a pressure of 4 kPa.
remove most of the water. In doing so, we cannot avoid The train condenser CONDENSE recovers the unre-
losing a significant amount of the caprolactam to the acted monomer and sends it back to the inlet of the
vapor phase. Therefore, equilibrium thermodynamics train. Uncondensed species, mostly water, leave in a
clearly suggests that it is impractical to try to obtain waste stream. The condenser operates at an assumed
both molecular weight buildup and monomer conversion pressure of 101.325 kPa and an assumed temperature
in the same reactor. of 403 K.
Assuming an instantaneous and complete water Table 11 summarizes the vessel operating conditions.
removal was necessary in the past because researchers We model mass-transfer limitations in the evaporator
did not have access to a fundamental thermodynamic and finisher reactors using two-film diffusion theory.45
model. However, with our new phase-equilibrium model, The diffusion coefficient is relatively high for water in
we are able to simulate any unit operation that involves nylon-6 melts,38 and we assume that the evaporator and
caprolactam, water, and nylon-6 at multiphase condi- finisher reactors generate a large interfacial surface
tions. These include flash units and condensers. There- area. Therefore, we set the water mass-transfer coef-
fore, we simulate multiphase reactors with condensers ficient to 1 × 10-2 m/s and set the mass-transfer
that return unreacted monomer back to the inlet of the coefficient for caprolactam to 5 × 10-4 m/s.
train. In particular, we simulate two commercial tech- Table 12 shows the predicted mass flow rates for
nologies: a melt train and a bubble-gas kettle train. All water, caprolactam, cyclic dimer, and nylon-6 exiting
details such as the unit operation configuration, feed each reactor. Furthermore, the table reports number-
conditions, and operating conditions are available in the average molecular weight predictions (Mn).
patent literature. We see that all of the monomer conversion (about 81%
This section is not concerned with developing a in this case) takes place in the first two reactors.
validated simulation model for these processes; this However, because of elevated moisture levels, the mo-
endeavor is well beyond the scope of this paper. There- lecular weight only grows to about 10.7 kg/mol. In the
fore, we do not make detailed comparisons between third and fourth reactors, we remove nearly all of the
model predictions and plant data. We wish only to water and double the polymer molecular weight. Our
illustrate the utility of our phase-equilibrium model in final molten polymer product has a flow rate of 1.16 ×
simulating entire manufacturing trains. 10-2 kg/s, the polymer has a number-average molecular
3910 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003

Figure 8. Four-reactor nylon-6 manufacturing process. Most of the monomer conversion takes place in the first two reactors, while the
molecular weight buildup occurs in the third and fourth reactors. The second vessel has a reflux condenser, while the entire train has a
condenser to recycle the unreacted monomer.

Table 11. Operating Conditions for Each Reactor in the of monomer is devolatilized in the third and fourth
Four-Reactor Train Depicted in Figure 8 reactors (36% of incoming caprolactam). Previous analy-
residence ses, such as those by Nagasubramanian and Reims-
unit operation temp (K) pressure (kPa) time (s) chuessel39 and by Tirrell et al.,40 would assume that no
first reactor 488 545 8100 water is lost in the first two vessels (approximated
second reactor 498 66.7 28800 roughly by one vessel in their studies) and no caprolac-
reflux condenser 403 66.7 tam is lost in the third and fourth vessels (approximated
third reactor 523 2.33 1200 roughly by one vessel in their studies). Our simulation,
fourth reactor 536 4.00 7200
train condenser 403 101.325
which contains phase-equilibrium and mass-transfer
models, shows that these are poor assumptions for a
Table 12. Model Predictions for Liquid-Phase Mass Flow typical commercial melt train.
Rates Exiting Each Reactor (Process Flow Diagram in 6.2. Bubble-Gas Kettle Train. We now simulate a
Figure 8)
typical commercial bubble-gas kettle train.13 This train
cyclic has three agitated kettles as in Figure 9.
caprolactam water dimer nylon-6 Mn
reactor (kg/s) (kg/s) (kg/s) (kg/s) (kg/mol) A total of 2.08 × 10-3 kg/s of caprolactam enters the
1 2.85 × 10-4 9.36 × 10-3 7.39 × 10-6 5.34 × 10-3 2.9 first vessel, along with 7.73 × 10-4 kg/s of steam.
2 5.71 × 10-5 2.81 × 10-3 2.73 × 10-5 1.19 × 10-2 10.7 We model the first kettle, KETTLE1 in Figure 9, as
3 3.10 × 10-7 1.69 × 10-3 2.87 × 10-5 1.20 × 10-2 11.7 a vapor-liquid CSTR with exit liquid and vapor streams.
4 1.25 × 10-6 1.92 × 10-4 4.21 × 10-5 1.16 × 10-2 18.3
Steam and lactam are fed to this reactor. This kettle
weight of 18.3 kg/mol, and the caprolactam content is operates at 527 K and 579 kPa. This kettle has a
about 1.6%. residence time of 9300 s.
The molecular weight prediction is close to plant We assume that the second kettle, KETTLE2 in
experience of melt trains, about 18 kg/mol. The capro- Figure 9, is mass-transfer-limited. We again model it
lactam content is also close: plant experience ranges as a multiphase CSTR utilizing two-film theory to
from 0.7 to 1.5%. compute the evaporation rate of volatiles. A total of 4.91
Additional predictions now follow: 2.44 × 10-4 kg/s, × 10-5 kg/s of inert gas (assumed to be nitrogen) is
or 85%, of the incoming water is lost to the vapor phase bubbled through the reaction mixture to remove vola-
in the second reactor, while virtually all of the water is tiles. We assume that the mass-transfer coefficients are
vaporized in the third reactor. Almost no caprolactam 1 and 1 × 10-4 m/s for water and caprolactam, respec-
is lost in the second reactor because of the presence of tively. This kettle has a residence time of 13 400 s, a
the reflux condenser; however, about 2.83 × 10-3 kg/s temperature of 528 K, and a pressure of 101.325 kPa.
 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003 3911

Figure 9. Three-reactor nylon-6 manufacturing process. Lactam and water are fed to the first multiphase kettle, while inert gas flow
devolatilizes the second and third vessels.

Table 13. Operating Conditions for Each Kettle Depicted that exists for every realistic manufacturing process.
in Figure 9 This high level of modeling detail is possible using our
temp pressure residence steam/nitrogen fundamental phase-equilibrium model for water/capro-
kettle (K) (kPa) time (s) flow rate (kg/s) lactam/nylon-6.
1 527 579. 9300 7.73 × 10-4 (steam)
2 528 101.325 13400 4.91 × 10-5 (nitrogen) 7. Conclusions
3 528 101.325 13400 4.91 × 10-5 (nitrogen) We have advanced nylon-6 process-simulation tech-
Table 14. Predicted Liquid-Stream Compositions
nology by developing and documenting a fundamental
Exiting Each Kettle (Process Flow Diagram Depicted in phase-equilibrium model. It represents a step forward
Figure 9) toward the generation of a single, consistent model that
cyclic
describes the phase behavior of all commercially sig-
caprolactam water dimer nylon-6 Mn nificant nylon polymerization-depolymerization tech-
reactor (kg/s) (kg/s) (kg/s) (kg/s) (kg/mol) nologies.
1 4.83 × 10-4 4.77 × 10-5 6.45 × 10-6 1.51 × 10-3 5.2 In creating this model, we use phase-equilibrium data
2 2.37 × 10-4 3.99 × 10-6 8.71 × 10-6 1.67 × 10-3 12.4 to characterize the ternary system, water/caprolactam/
3 1.89 × 10-4 1.34 × 10-6 9.56 × 10-6 1.69 × 10-3 20.6 nylon-6. We illustrate adequate treatment for inconsis-
tent thermodynamic data, including the generation of
The third kettle, KETTLE3 in Figure 9, is identical a coherent set of POLYNRTL binary interaction pa-
to the second kettle regarding operating conditions. rameters. We then show how to simulate the reactive
Table 13 summarizes the vessel operating conditions. system, water/caprolactam/nylon-6, to extract binary
Table 14 shows the liquid-phase flow rates, composi- interaction parameters for water/nylon-6 and caprolac-
tions, and polymer molecular weights coming out of each tam/nylon-6 segments. Our model predicts the liquid
kettle. mole fraction of water with an average error of 1%;
With this train, we produce 1.69 × 10-3 kg/s of nylon previous literature models sometimes generate predic-
(81% conversion), with 10% caprolactam coming out of tions that are more than an order of magnitude in error.
the third kettle. The molecular weight of the polymer After generating a complete model for the phase
is 20.6 kg/mol. Both the caprolactam content and equilibrium for water/caprolactam/nylon-6, we validate
molecular weight predictions are close to plant experi- the interaction parameters by performing exploratory
ence of bubble-gas kettle trains: about 9% and 21 kg/ simulations of commercial manufacturing processes. We
mol for caprolactam and molecular weight, respectively. simulate a condenser and match the split fraction for
In the first kettle, 75% of the incoming water exits in water (99%). We predict a caprolactam split fraction of
the vapor stream, along with 4% of the incoming 53%, which approximates the real split fraction of 49%.
caprolactam. In the second kettle, the corresponding We also simulate two commercial wiped-film evapora-
percentages are 98% and 18%, respectively. Lastly, the tors and show that the models underpredict the mass
third kettle devolatilizes 80% of the incoming water and fraction of caprolactam in the exit polymer stream. This
34% of the incoming caprolactam. is in accordance with expectations: we did not simulate
This example again shows that an incomplete devola- mass-transfer limitations and, therefore, the amount of
tilization based on a more detailed analysis of the phase caprolactam predicted to be in the liquid phase should
equilibrium is more realistic. Using the previous as- be low when compared with plant data.
sumptions regarding nylon phase behavior would have Last, we demonstrate two applications of our new
resulted in severe prediction inaccuracies. phase-equilibrium model that simulate two integrated
6.3. Example Summary. These two applications nylon-6 production processes: a melt train and a bubble-
illustrate that we are no longer bound by unrealistic gas kettle train. We simulate wide ranges of tempera-
assumptions, such as no or instantaneous water re- ture and pressure conditions, with temperature ranging
moval, while ignoring caprolactam devolatilization. In from 403 to 536 K and pressure ranging from 2.33 to
fact, we obtain a high level of detail in our simulations, 545 kPa. We perform fundamental kinetic, thermody-
including an analysis of the monomer-recovery portion namic, and mass-transfer calculations to make detailed
3912 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003

predictions about reactor products. These predictions Literature Cited

could not be made in the past without a fundamental
(1) Kohan, M. I., Ed. Nylon Plastics; Wiley: New York, 1973.
phase-equilibrium model.
(2) Hajduk, F. Nylon Fibers. Chemical Economics Handbook;
SRI International: Menlo Park, CA, Mar 1-2, 2002.
Acknowledgment (3) Davenport, R. E.; Riepl, J.; Sasano, T. Nylon Resins.
Chemical Economics Handbook; SRI International: Menlo Park,
We thank Alliant Techsystems, Aspen Technology CA, Jul 1-4, 2001.
(particularly, Jila Mahalec, past Director of Worldwide (4) Giori, C.; Hayes, B. T. Hydrolytic Polymerization of Capro-
University Programs; Dustin MacNeil, current Director lactam. I. Hydrolysis-Polycondensation Kinetics. J. Polym. Sci.,
of Worldwide University Programs; Joseph Boston, Polym. Chem. Ed. 1970, 8, 335-349.
Senior Corporate Advisor and Past President; and Larry (5) Agrawal, A. K.; Devika, K.; Manabe, T. Simulation of
Hydrolytic Polymerization of Nylon-6 in Industrial Reactors: Part
Evans, Board Chairman), China Petroleum and Chemi-
I. Mono-Acid-Stabilized Systems in VK Tube Reactors. Ind. Eng.
cal Corp. (SINOPEC), China National Petroleum Corp., Chem. Res. 2001, 40, 2563-2572.
Honeywell Specialty Materials, and Honeywell Inter- (6) Gupta, S. K.; Kunzru, D.; Kumar, A.; Agarwal, K. K.
national Foundation for supporting the computer-aided Simulation of Nylon-6 Polymerization in Tubular Reactors with
design educational programs at Virginia Tech. We also Recycle. J. Appl. Polym. Sci. 1983, 28, 1625-1640.
thank Paul Mathias, David Tremblay, and Yuhua Song, (7) Amon, M.; Denson, C. D. Simplified Analysis of the Perfor-
Polymer Industry Business Unit, Aspen Technology, for mance of Wiped-Film Polycondensation Reactors. Ind. Eng. Fun-
dam. 1980, 19, 415-420.
their helpful input to this study.
(8) Gupta, A.; Gupta, S. K.; Ghandi, K. S.; Mehta, M. H.; Padh,
M. R.; Soni, A. V.; Ankleswaria, B. V. Modeling of Hydrolytic
Supporting Information Available: Tables of Polymerization in a Semi-Batch Nylon-6 Reactor. Chem. Eng.
phase-equilibrium data of the system water/caprolactam Commun. 1992, 113, 63-89.
at various pressures and of the system water/caprolac- (9) Twilley, I. C.; Roth, D. W. H.; Lofquist, R. A. Process for
tam/nylon-6. This material is available free of charge the Production of Nylon-6. U.S. Patent 3,558,567, 1971.
via the Internet at http://pubs.acs.org. (10) Yates, S. L.; Cole, C. J.; Wiesner, A. H.; Wagner, J. W.
Two-Stage Hydrolysis Process for the Preparation of Nylon-6. U.S.
Patent 4,310,659, 1982.
Nomenclature (11) http://www.oit.doe.gov/chemicals/pdfs/nylon.pdf (accessed
Sept 2002).
aij ) binary interaction parameter coefficient for the binary (12) Tai, K.; Arai, Y.; Tagawa, T. The Simulation of Hydrolytic
i-j Polymerization of -Caprolactam in Various Reactors. J. Appl.
A0i ) preexponential factor for uncatalyzed forward reac- Polym. Sci. 1982, 27, 731-736.
tion i, kg/mol‚s (13) Russell, W. N.; Wiesner, A. H.; Snider, O. E. Continuous
Aci ) preexponential factor for catalyzed forward reaction Polymerization of -Caprolactam. U.S. Patent 3,294,756, 1966.
i, kg/mol‚s (14) Gaymans, R. J.; Amirtharaj, J.; Kamp, H. Nylon-6 Polym-
erization in the Solid State. J. Appl. Polym. Sci. 1982, 27, 2513-
bij ) binary interaction parameter coefficient for the binary 2526.
i-j, K (15) Sifniades, S. Recovery of Caprolactam from Waste Carpets.
cij ) binary interaction parameter coefficient for the binary European Patent Application 0676395A1, 1995.
i-j (16) Bokis, C. P.; Orbey, H.; Chen, C. C. Properly Model
E0i ) activation energy for uncatalyzed forward reaction i, Polymer Processes. Chem. Eng. Prog. 1999, 95 (4), 39-52.
J/mol (17) Loth, W.; Vonend, J.; Lehnen, U.; Kory, G.; Hungenberg,
Eci ) activation energy for catalyzed forward reaction i, K. D. Simulation of the Entire Process for the Development and
Optimization of Polymer Process Concepts. DECHEMA Monogr.
J/mol
1998, 134, 545-556.
∆Hi ) enthalpy of reaction i, J/mol (18) Arai, Y.; Tai, K.; Teranishi, H.; Tagawa, T. Kinetics of
Ki′ ) equilibrium constant for reaction i Hydrolytic Polymerization of -Caprolactam: 3. Formation of
ki ) forward rate constant for reaction i, kg/mol‚s Cyclic Dimer. Polymer 1981, 22, 273-277.
ki′ ) reverse rate constant for reaction i, kg/mol‚s (19) Barrera, M.; Ko, G.; Osias, M.; Ramanathan, M.; Tremblay,
L ) Wisniak consistency test function, K D. A.; Chen, C. C. Polymer Component Characterization Method
n ) degree of polymerization and Process Simulation Apparatus. U.S. Patent 5,687,090, 1997.
P ) pressure, Pa (20) Jacobs, H.; Schweigman, C. Mathematical Model for the
Polymerization of Caprolactam to Nylon-6. Proc. Fifth Eur./Second
Pi ) partial pressure of species i, Pa
Int. Symp. Chem. React. Eng. 1972, B7, 1-26.
Pvap
i ) vapor pressure for species i, Pa (21) Fukumoto, O. Equilibria Between Polycapramide and
R ) ideal gas law constant, J/mol‚K Water. I. J. Polym. Sci. 1956, XXII, 263-270.
∆Si ) entropy of reaction i, J/mol‚K (22) Chen, C. C. A Segment-based Local Composition Model for
T ) temperature, K the Gibbs Energy of Polymer Solutions. Fluid Phase Equilib. 1993,
T0i ) boiling point for component i, K 83, 301-312.
w ) Wisniak consistency test function (23) Flory, P. J. Thermodynamics of High Polymer Solutions.
J. Chem. Phys. 1942, 10, 51-61.
W ) Wisniak consistency test function, K
(24) Huggins, M. L. Some Properties of Solutions of Long-Chain
xi ) liquid-phase mole fraction of species i Compounds. J. Phys. Chem. 1942, 46, 151-158.
yi ) vapor-phase mole fraction of species i (25) Fredenslund, A.; Jones, R. L.; Prausnitz, J. M. Group
Rij ) POLYNRTL nonrandomness factor for the binary i-j Contribution Estimation of Activity Coefficients in Nonideal Liquid
γi ) activity coefficient for species i Mixtures. AIChE J. 1975, 21, 1086-1099.
∆h0i ) enthalpy of vaporization at the boiling point for (26) Redlich, O.; Kwong, J. N. S. On the Thermodynamics of
component i, J/mol Solutions. V. An Equation of State. Fugacities of Gaseous Solu-
tions. Chem. Rev. 1949, 44, 233-244.
∆s ) mixture entropy of vaporization, J/mol‚K
(27) Van Krevelen, D. W. Properties of Polymers; Elsevier: New
∆s0i ) entropy of vaporization at the boiling point for York, 1990.
component i, J/mol‚K (28) Renon, H.; Prausnitz, J. M. Local Compositions in Ther-
τij ) POLYNRTL binary interaction parameter for the modynamic Excess Functions for Liquid Mixtures. AIChE J. 1968,
binary i-j 14, 135-144.
 Ind. Eng. Chem. Res., Vol. 42, No. 17, 2003 3913

(29) Prausnitz, J. M.; Lichtenthaler, R. N.; de Azevedo, E. G. (38) Nagasubramanian, K.; Reimschuessel, H. K. Diffusion of
Molecular Thermodynamics of Fluid-Phase Equilibrium, 3rd ed.; Water and Caprolactam in Nylon-6 Melts. J. Appl. Polym. Sci.
Prentice Hall: Englewood Cliffs, NJ, 1999; p 261. 1973, 17, 1663-1677.
(30) Maczinger, J.; Tettamanti, K. Phase Equilibrium of the (39) Nagasubramanian, K.; Reimschuessel, H. K. Caprolactam
System Caprolactam/Water. Period. Polytech. 1966, 10, 183. Polymerization. Polymerization in Backmix Flow Systems. J. Appl.
(31) Giori, C.; Hayes, B. T. Hydrolytic Polymerization of Ca- Polym. Sci. 1972, 16, 929-934.
prolactam. II. Vapor-Liquid Equilibrium. J. Polym. Sci., Polym. (40) Tirrell, M. V.; Pearson, G. H.; Weiss, R. A.; Laurence, R.
Chem. Ed. 1970, 8, 351-358. L. An Analysis of Caprolactam Polymerization. Polym. Eng. Sci.
(32) Van Ness, H. C.; Byer, S. M.; Gibbs, R. E. Vapor-Liquid 1975, 15, 386-393.
Equilibrium: Part 1. An Appraisal of Data Reduction Methods. (41) Wagner, J. W.; Haylock, J. C. Control of Viscosity and
AIChE J. 1973, 19, 238-244. Polycaproamide Degradation during Vacuum Polymerization. U.S.
(33) Wisniak, J. A New Test for the Thermodynamic Consis- Patent RE28,937, 1976.
tency of Vapor-Liquid Equilibrium. Ind. Eng. Chem. Res. 1993, (42) Wright, W. H.; Bingham, A. J.; Fox, W. A. Method to
32, 1531-1533. Prepare Nylon-6 Prepolymer Providing a Final Shaped Article of
(34) Prausnitz, J. M.; Lichtenthaler, R. N.; de Azevedo, E. G. Low Oligomer Content. U.S. Patent 3,813,366, 1974.
Molecular Thermodynamics of Fluid-Phase Equilibrium, 3rd ed.; (43) Boggs, B. A.; Balint, L. J.; Ager, P. W.; Buyalos, E. J.
Prentice Hall: Englewood Cliffs, NJ, 1999; p 247. Wiped-Wall Reactor. U.S. Patent 3,976,431, 1976.
(35) Herington, E. F. G. Tests for the Consistency of Experi- (44) Saunders, L. V. J.; Rochell, D. Polymer Finisher. U.S.
mental Isobaric Vapour-Liquid Equilibrium Data. J. Pet. Inst. Patent 3,686,826, 1972.
1951, 37, 457-470. (45) Whitman, W. G. The Two-Film Theory of Gas Absorption.
(36) Daubert, T. E.; Danner, R. P. Physical and Thermodynamic Chem. Metall. Eng. 1923, 29, 146-148.
Properties of Pure Chemicals. Data Compilation; Hemisphere Received for review February 6, 2003
Publishing Corp.: New York, 1989; Vol. 1. Revised manuscript received June 9, 2003
(37) Prausnitz, J. M.; Lichtenthaler, R. N.; de Azevedo, E. G. Accepted June 10, 2003
Molecular Thermodynamics of Fluid-Phase Equilibrium, 3rd ed.;
Prentice Hall: Englewood Cliffs, NJ, 1999; p 249. IE030112+