Fluid Phase Equilibria 228229 (2005) 213221

Advanced equation of state method for modeling TEGwater

for glycol gas dehydration
Chorng H. Twu
a
, Vince Tassone
b
, Wayne D. Sim
b
, Suphat Watanasiri
c
a
Aspen Technology, Inc., 2811 Loganberry Court, Fullerton, CA 92835, USA
b
Aspen Technology, Inc., Suite 900, 125 9th Avenue SE, Calgary, Alta., Canada T2G 0P6
c
Aspen Technology, Inc., Ten Canal Park, Cambridge, MA 02141, USA
Abstract
An advanced equation of state has been developed for modeling triethylene glycol (TEG)water system for glycol gas dehydration process.
The dehydration of natural gas is very important in the gas processing industry. It is necessary to remove water vapor present in a gas stream
that may cause hydrate formation at low-temperature conditions that may plug the valves and ttings in gas pipelines. In addition, water
vapor may cause corrosion difculties when it reacts with hydrogen sulde or carbon dioxide commonly present in gas streams. The most
effective practice to remove water from natural gas streams is to use TEG in the gas dehydration process. In modeling such a process, it is
crucial that the phase behavior of the TEGwaternatural gas system is correctly modeled, with methane being the predominant component
in natural gas. Of the three binaries, methanewater, methaneTEG and TEGwater, the methane binaries can be adequately modeled by
an equation of state, e.g. [J.R. Cunningham, J.E. Coon, C.H. Twu, Estimation of aromatic hydrocarbon emissions from glycol dehydration
units using process simulation, in: Proceedings of the 72nd Annual Gas Processors Association Convention, San Antonio, TX, March 1517,
1993]. For the TEGwater binary, the Parrishs empirical hyperbolic correlation [W.R. Parrish, K.W. Won, M.E. Baltatu, Phase behavior of the
triethylene glycolwater system and dehydration/regeneration design for extremely low dew point requirements, in: Proceedings of the 65th
Annual GPA Convention, San Antonio, TX, March 1012, 1986] is recommended by GPSA [GPSA Engineering Data Book, 10th ed., First
Revision, Gas Processors Suppliers Association, Tulsa, OK, 1994] and is currently widely used in the industry. In this work, we applied the
TST (TwuSimTassone) equation of state to model this binary system. A methodology was also developed to determine the water dew point
and calculate water content for this system. The TST equation of state is shown to accurately represent the activity coefcients of TEGwater
solutions as well as water dew point temperatures and water content of gas over the entire application range of temperature, pressure and
concentration encountered in a typical TEG dehydration unit.
2004 Elsevier B.V. All rights reserved.
Keywords: Triethylene glycol; Water; Water dew point; Water content; CEoS; TST; Cubic equation of state; Excess energy mixing rule; Activity coefcient;
Gas dehydration
1. Introduction
The dehydration of natural gas is an important operation
in the gas processing industry. The standard method for nat-

Corresponding author. Tel.: +1 403 303 1000; fax: +1 403 303 0914.
E-mail addresses: chorng.twu@aspentech.com (C.H. Twu),
vince.tassone@aspentech.com (V. Tassone), wayne.sim@aspentech.com
(W.D. Sim), suphat.watanasiri@aspentech.com (S. Watanasiri).
ural gas dehydration is by absorption of water using TEG.
Glycol dehydration units typically consist of a contactor, a
ash tank, heat exchangers and a regenerator. The lean TEG
liquid streamenters at the top of the absorber or the contactor
while the natural gas stream containing water to be removed
(wet gas) enters at the bottom of the absorber. The lean TEG
liquid absorbs water as it progresses toward the bottomof the
column. A dry gas exits at the top of the absorber. The rich
TEGstreamis sent to the regenerator where water is removed
and the lean TEG liquid is returned to the absorber.
0378-3812/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.uid.2004.09.031
214 C.H. Twu et al. / Fluid Phase Equilibria 228229 (2005) 213221
The most important aspect of modeling any dehydration
unit is to correctly model the methaneglycolwater ternary
[1]. This ternary controls the predicted glycol circulation
rates, purities of the lean glycol and the water content of the
dry gas. Of the three binaries, methanewater, methaneTEG
andTEGwater, the methane binaries canbe adequatelymod-
eled by an equation of state (e.g. [1]). At one time, only
graphical data from vendors were available in the literature
to model the TEGwater binary. These data often were in
disagreement at low water concentration portion resulting in
confusion.
Since the concentration of water in the natural gas is typi-
callylow, less than0.2 mol%, andthe concentrationof TEGin
the lean TEG solution is high, normally higher than 98 wt.%,
highly concentrated TEG solutions higher than 99.50 wt.%
are usually required if the water concentration in the efuent
gas stream is specied to be very low. Therefore, in order to
have an accurate design of a dehydration unit, vaporliquid
equilibrium data for TEGwater need to be accurate, espe-
cially in the dilute region of water. Parrish et al. [2] gave
an extensive review of the available equilibrium data. Based
on the data reviewed, they found the data of Herskowitz and
Gottlieb [4] to be the most reliable.
Herskowitz and Gottlieb [4] measured the activity coef-
cients of water in TEG at two temperatures, 297.60 and
332.60 K. The lowest mole fraction of water for which ac-
tivities were measured was 0.1938 and 0.2961 at 297.60 and
332.60 K, respectively. They t their measured activity coef-
cients to the van Laar equation. They did not measure data
in the innite dilution region. In order to predict the equilib-
riumbehavior in the innite dilution region, most researchers
simply extrapolated the measured data at low water concen-
trations to innite dilution using an activity coefcient model
such as van Laar. However, extrapolating the van Laar, or any
other activity coefcient model will yield erroneous results
for the innite dilution activity coefcients.
To help better dene the TEGwater system, Parrish et
al. [2] measured activity coefcients at innite dilution as a
function of temperature. These data, shown in Fig. 1, were
used to evaluate the existing data. The data of Herskowitz
and Gottlieb [4] were found to be in good agreement with the
measured innite dilution activity coefcient data.
Bestani and Shing [5] subsequently measured activity co-
efcients of water in TEG at innite dilution, but their data
are 1317% higher than those of Parrish as shown in Fig. 1.
Based on the Bestani extrapolation method to 477.15 K, the
predicted water activity coefcient is above 1.0. A value of
water activity coefcient that is greater than unity at 477.15 K
implies that TEGwouldbe a poor dehydratingagent at around
this temperature, which is contrary to plant experience [6].
Therefore, Bestani data are not used in this work.
Parrish et al. [2] combined their innite dilution activity
coefcients data with the nite-concentration activity coef-
cients of Herskowitz and Gottlieb and then t them to the
activity coefcient model of a four-sufx Margules equation
over the entire range of composition at each temperature for
Fig. 1. Innite dilution activity coefcient of water in waterTEG system:
() Exp., Parrish et al. [2]; () Exp., Bestani and Shing [5]; () this work;
(- - -) Parrish et al. [2].
the TEGwater system. However, since the Margules activity
coefcient model is unable to t the activity coefcients over
an extended temperature range for process design calcula-
tions, Parrish et al. proposed an empirical hyperbolic equa-
tion to predict the activity coefcients of TEG (1) and water
(2) over the entire range of temperatures and compositions:
ln
1
=
B
2
ln[cosh()]
A

x
2
Btanh()
x
1
Cx
2
2
(1)
ln
2
= B[tanh() 1] Cx
2
1
(2)
where
=
Ax
2
Bx
1
(3)
and tanh and cosh are hyperbolic tangent and cosine func-
tions. The subscript number represents component: 1for TEG
and 2 for water. A, B, and C are temperature-dependent pa-
rameters:
A = exp(12.792 +0.03293T) (4)
B = exp(0.77377 0.00695T) (5)
C = 0.88874 0.001915T (6)
where T is the temperature in Kelvin.
Both Herskowitz and Gottlieb, and Parrish used activity
coefcient models to t the activity coefcient data as a func-
tion of mole fraction and temperature. However, using an
activity coefcient model to describe the liquid phase still
requires the use of an equation of state to handle the non-
ideality of the gas phase and a Poynting correction factor to
account for the effect of pressure. In addition, the standard
state fugacity as a function of temperature ranging from the
temperatures below 273.15 K to critical temperature need to
be correlated for water for the TEG gas dehydration. Since
different models are used for vapor and liquid, the approach
has limitations. For example, no critical conditions can be
calculated, the K-values near critical region is not reliable,
C.H. Twu et al. / Fluid Phase Equilibria 228229 (2005) 213221 215
other thermodynamic properties such as density cannot be
calculated, modeling light gases requires the use of Henrys
law constants, etc. Besides, the results of extrapolation to
high temperatures and high pressures may not be reliable.
For example, using the van Laar equation with the given bi-
nary interaction parameters fromHerskowitz and Gottlieb for
extrapolation to 477.15 K, the van Laar predicts a water activ-
ity coefcient of 1.1. As mentioned earlier in the analysis of
Bestani and Shing data, a value of water activity coefcient
that is greater than unity is not realistic. Using the empirical
hyperbolic function of Parrish, a water activity coefcient of
0.9477 is obtained at 477.15 K, which is below unity, but is
still too high. The activity coefcient of water at this tem-
perature is expected to be below 0.80 [6]. Furthermore, Eqs.
(1) and (3) contain mole fraction of component 1, x
1
, in the
denominator. These equations will break down when x
1
ap-
proaches zero. Although the TEG mole fraction is unlikely
to be zero in practical natural gas dehydration process, it is
a notable deciency that engineers should be aware of. The
Parrishs correlation is recommended by the GPSA [3] and
is currently widely used in the industry.
In this work, we applied the TST (TwuSimTassone)
equation of state model [7,8] to describe the phase behavior
of the TEGwater system. We also presented a methodology
to determine water dew point and calculate water content for
this system. The TST equation of state represents accurately
the activity coefcients for water and TEG, the water dew
point temperatures and water content over the entire range
of temperature, pressure and composition encountered in a
typical TEG dehydration unit.
2. Advanced TST (TwuSimTassone) equation of
state model
Twu, Simand Tassone (TST) recently developed CEoS/A
E
mixing rules [7,8] that permit a smooth transition of the mix-
ing rules to the conventional van der Waals one-uid mixing
rules. They also proposed a cubic equation of state for bet-
ter handling of polar and heavy components and a G
E
model,
which when combined with the CEoS/A
E
mixing rules allows
both a van der Waals uid and highly non-ideal mixtures to
be described over a broad range of temperatures and pres-
sures in a consistent and unied framework. It is extremely
desirable to have the CEoS/A
E
mixing rules reduce to the
classical quadratic mixing rules because the classical mixing
rules work very well for nonpolar and slightly polar systems.
Introducing this capability into an excess energy model en-
sures that the binary interaction parameters for the classical
mixing rules available in many existing databanks for sys-
tems involving hydrocarbons and gases can be used directly
in the new excess energy mixing rules. In other words, it
allows the equation of state to describe some binaries in a
multi-component mixture using the van der Waals one-uid
mixing rules, while other pairs with more non-ideal interac-
tions are described by the excess energy mixing rules.
The TST cubic equation of state is represented by the
following equation:
P =
RT
v b

a
v
2
+2.5bv 1.5b
2
(7)
Eq. (7) can be rewritten in another form as
P =
RT
v b

a
(v +3b)(v 0.5b)
(8)
The values of a and b at the critical temperature are found
by setting the rst and second derivatives of pressure with
respect to volume to zero at the critical point resulting in
a
c
= 0.470507
R
2
T
2
c
P
c
(9)
b
c
= 0.0740740
RT
c
P
c
(10)
Z
c
= 0.296296 (11)
where subscript c denotes the critical point. It is noted
that the values of Z
c
from the SoaveRedlichKwong [9]
and PengRobinson [10] models are both larger than 0.3
(0.333333 and 0.307401, respectively), but that for TST is
slightly below 0.3, which is closer to the typical value of Z
c
for most compounds.
The parameter a is a function of temperature. The value
of a at any temperature a(T) can be calculated from
a(T) = (T)a
c
(12)
where the alpha function, (T), is a function only of reduced
temperature, T
r
=T/T
c
. We use the Twu alpha function [11]:
= T
N(M1)
r
e
L(1T
NM
r
)
(13)
Eq. (13) has three parameters, L, M and N. These parameters
are unique to each component and are determined from the
regression of pure component vapor pressure data. Table 1
lists the L, M and N parameters for TEG and water for use
with the TST equation of state in this paper. The values of L,
M and N for N
2
, CO
2
, H
2
S and light hydrocarbons in natural
gas from methane to n-decane are also included in the table
for future applications.
The TST zero-pressure mixing rules for the mixture a and
b parameters are
a

= b

_
a

vdw
b

vdw
+
1
Cr
_
A
E
0
RT

A
E
0vdw
RT
__
(14)
b = b
vdw
(15)
The parameters a
*
and b
*
in Eq. (14) are dened as
a

=
Pa
R
2
T
2
(16)
b

=
Pb
RT
(17)
216 C.H. Twu et al. / Fluid Phase Equilibria 228229 (2005) 213221
Table 1
L, M and N parameters of Twu function with the TST CEoS
ID Component T
c
(K) P
c
(kPa) L M N
1 TEG 769.50 3320.00 0.196667 0.863521 5.10947
2 H
2
O 647.13 22055.00 0.430058 0.870932 1.67211
3 N
2
126.20 3400.00 0.0649944 0.892385 2.34000
4 CO
2
304.21 7383.00 0.945951 0.888652 0.65000
5 H
2
S 373.53 8962.90 0.231877 0.784346 1.12000
6 CH
4
190.564 4599.00 0.0813821 0.905296 2.13000
7 C2 305.32 4872.00 0.147335 0.879706 1.98500
8 C3 369.83 4248.00 0.172517 0.879570 2.20000
9 NC4 425.12 3796.00 0.515633 0.846523 1.02632
10 NC5 469.70 3370.00 0.385772 0.817594 1.35710
11 NC6 507.60 3025.00 0.119904 0.858552 3.17252
12 NC7 540.20 2740.00 0.658164 0.829578 1.11729
13 NC8 568.70 2490.00 0.486147 0.809629 1.49823
14 NC9 594.60 2290.00 0.477371 0.796573 1.58848
15 NC10 617.70 2110.00 0.436564 0.800707 1.82508
Note that the temperature-independent van der Waals mixing
rule b
vdw
is used for the b parameter in Eq. (15). The expres-
sionof b
vdw
is givenbelowinEq. (22). The TSTzero-pressure
mixing rules assume that A
E
0vdw
, the excess Helmholtz energy
of van der Waals uid at zero pressure, can be approximated
by A
E
vdw
, the excess Helmholtz energy of van der Waals
uid at innite pressure:
A
E
0vdw
RT
=
A
E
vdw
RT
= C
1
_
a

vdw
b

vdw

i
x
i
a

i
b

i
_
(18)
With this assumption, the zero-pressure mixing rule transi-
tions smoothly to the conventional van der Waals one-uid
mixing rule. The C
1
in Eq. (18) is a constant and is dened
as
C
1
=
1
(w u)
ln
_
1 +w
1 +u
_
(19)
where u and w are equation-of-state-dependent constants
used to represent a general two-parameter cubic equation of
state. For the TST equation of state, u is 3 and w is 0.5 as
shown in Eq. (8).
C
r
in Eq. (14) is a function of a parameter r, which is the
reduced liquid volume at zero pressure:
C
r
=
1
w u
ln
_
r +w
r +u
_
(20)
The value of r =1.18 is recommended by Twu et al. [12].
Using r =1.18, C
r
=0.518850 is used in this work.
A
E

and A
E
0
in above equations are the excess Helmholtz
free energies at innite pressure and zero pressure, respec-
tively. The subscript vdw in A
E
vdw
and A
E
0vdw
denotes that
the properties are evaluated from the cubic equation of state
using the van der Waals mixing rule for its a and b parame-
ters, a
vdw
and b
vdw
:
a
vdw
=

j
x
i
x
j

a
i
a
j
(1 k
ij
) (21)
b
vdw
=

j
x
i
x
j
_
1
2
(b
i
+b
j
)
_
(22)
Since A
E
0
in these equations is at zero pressure, its value is
identical to the excess Gibbs free energy G
E
at zero pressure.
Therefore, any activity model such as the NRTL equation
can be used directly for the excess Helmholtz free energy
expression A
E
0
in Eq. (14).
The TST zero-pressure mixing rule assumes that the ex-
cess Helmholtz free energy of the van der Waals uid (A
E
0vdw
,
Eq. (18)) is independent of pressure. This approximation is
required to allow a smooth transition to the conventional van
der Waals one-uid mixing rule. Therefore, a binary interac-
tion parameter k
ij
is introduced in Eq. (21) to correct for this
approximation. In this work, k
ij
is not needed to adequately
t the TEGwater VLE data and is set equal to zero.
Twu et al. [7,8] proposed a multi-component equation for a
liquid activity model for use in the TSTexcess energy mixing
rules:
G
E
RT
=
n

i
x
i

n
j
x
j

ji
G
ji

n
k
x
k
G
ki
(23)
Eq. (23) has the same functional form as the NRTL equa-
tion, but there is a fundamental difference between them.
NRTL assumes that A
ij
, A
ji
and
ij
are the parameters of the
model, but the excess Gibbs energy model proposed by Twu
et al. [7,8] assumes that
ij
and G
ij
are the binary interaction
parameters. More importantly, any appropriate temperature-
dependent function can be applied to
ij
and G
ij
. For example,
to obtain the NRTL model,
ji
and G
ji
are calculated as usual
from the NRTL parameters A
ji
, A
ij
and
ji
:

ji
=
A
ji
T
(24)
G
ji
= exp(
ji

ji
) (25)
In this way, the NRTLparameters reported in the DECHEMA
Chemistry Data Series can be used directly in our mixing
C.H. Twu et al. / Fluid Phase Equilibria 228229 (2005) 213221 217
Table 2
Binary interaction parameters for use in TST zero-pressure mixing rule
Binary TEG(1)/H
2
O(2)
A
12
141.490
A
21
158.166
B
12
0.254489
B
21
5.83380

12
0.278879
rule and there is no difference between NRTL model and our
model in the prediction of phase equilibrium calculations.
We also note that Eq. (23) can recover the conventional
van der Waals mixing rules when the following expressions
are used for
ji
and G
ji
:

ji
=
1
2

ij
b
i
(26)
G
ji
=
b
j
b
i
(27)
where

ij
=
C
1
RT
_
_
a
i
b
i

a
j
b
j
_
2
+2k
ij

a
i
b
i

a
j
b
j
_
(28)
Eqs. (26) and (27) are expressed in terms of cubic equation
of state parameters a
i
and b
i
, and the binary interaction pa-
rameter k
ij
. The above discussion demonstrates that Eq. (23)
is more generic in its form than the NRTL model. Both the
NRTL and van der Waals one-uid mixing rule are special
cases of our excess Gibbs free energy function.
3. Correlation of activity coefcients
The TST equation of state and mixing rules described in
the previous section are used to correlate the innite dilution
activity coefcient data of Parrish et al. [2] and the nite-
concentrationactivitycoefcients of Herskowitz andGottlieb
[4]. To cover the entire application range of temperature, Eq.
(24) is modied to include a temperature-dependent binary
interaction parameter B
ji
as follows:

ji
=
A
ji
+B
ji
T
T
(29)
where T is the temperature in K. The unit of A
ji
is in K and
B
ji
is dimensionless.
Table 2 lists the values of the binary interaction parameters
A
ij
, A
ji
, B
ij
, B
ji
and
ij
obtained for the TEGwater system.
Table 3 compares measured and calculated innite dilution
activity coefcients of water in TEG as a function of temper-
ature. Also shown in the table are the innite dilution activity
coefcients calculated from Parrishs empirical hyperbolic
equation (Eq. (2)) for comparison. Data in Table 3 are also
shown graphically in Fig. 1. Results shown in Table 3 and
Fig. 1 indicate that the TST equation of state can accurately
correlate the innite dilution activity coefcients of water in
TEG covering a wide range of temperature with an average
Table 3
Comparison of Parrish et al. [2] measured innite dilution activity coef-
cients of water (2) in TEG (1) solution with those calculated using TST
CEoS and Parrishs hyperbolic equation
T (K)

2
(measured) This work Parrish et al. [2]

2
(calc.) Devi%

2
(calc.) Devi%
300.43 0.5510 0.5565 0.99 0.5587 1.40
311.71 0.575 0.5773 0.41 0.5826 1.32
323.26 0.5900 0.5979 1.34 0.6072 2.91
333.76 0.6170 0.6159 0.19 0.6296 2.04
343.43 0.6240 0.6318 1.25 0.6503 4.21
355.93 0.6360 0.6515 2.44 0.6772 6.47
364.93 0.6690 0.6651 0.58 0.6966 4.13
378.32 0.6920 0.6845 1.08 0.7257 4.87
AAD% 1.03 3.42
error of 1.03%. Parrishs model systematically overpredicts
the data with a positive average error of +3.42%. Fig. 1 also
indicates that Eq. (2) represents data at lowtemperatures bet-
ter than those at higher temperatures. The TST EoS, on the
other hand, more accurately ts the data over the entire tem-
perature range.
Table 4 shows the nite-concentration activity coefcients
of water in TEG measured by Herskowitz and Gottlieb as
functions of temperature and composition. For comparison,
the table again includes results fromthis work and those using
Eq. (2). Data in Table 4 are also shown graphically in Fig. 2.
Reviewing Fig. 2, it is observed that the data of Herskowitz
and Gottlieb are not quite consistent with the data of Parrish.
Due to the curvature of Herskowitz data, the extrapolation to
water mole fraction of zero (i.e., TEG mole fraction of 1.0)
will give innite dilution activity coefcients of water in TEG
of about 0.65 at 297 K. This is much larger than the innite
dilution activity coefcient data of Parrish at the same tem-
perature (<0.55). Since both the TST EoS and Eq. (2) t the
innite dilution activity coefcient data well, the t to Her-
skowitz and Gottlieb data at high TEG concentration is quite
Fig. 2. Activity coefcient of water vs. liquid mole fraction TEG. Exp.,
Herskowitz and Gottlieb [4]: () T=297.6 K, () T=332.6 K; () this
work; (- - -) Parrish et al. [2].
218 C.H. Twu et al. / Fluid Phase Equilibria 228229 (2005) 213221
Table 4
Comparison of Herskowitz and Gottlieb [4] measured activity coefcients of water (2) in TEG (1) solution with those calculated using TST CEoS and Parrishs
hyperbolic equation
T (K) X (1)
2
(measured) This work Parrish et al. [2]

2
(calc.) Devi%
2
(calc.) Devi%
297.60 0.8062 0.6604 0.6299 4.62 0.6255 5.28
297.60 0.7838 0.6640 0.6394 3.71 0.6338 4.55
297.60 0.6731 0.6760 0.6875 1.71 0.6742 0.26
297.60 0.6731 0.6871 0.6875 0.06 0.6742 1.87
297.60 0.5809 0.7062 0.7290 3.23 0.7078 0.22
297.60 0.5449 0.7201 0.7456 3.54 0.7210 0.13
297.60 0.5001 0.7407 0.7664 3.47 0.7378 0.40
297.60 0.4967 0.7371 0.7680 4.19 0.7390 0.26
297.60 0.3996 0.7823 0.8138 4.02 0.7777 0.58
297.60 0.3898 0.7876 0.8184 3.91 0.7819 0.72
297.60 0.3057 0.8264 0.8585 3.89 0.8220 0.53
297.60 0.2505 0.8608 0.8849 2.80 0.8545 0.74
297.60 0.2187 0.8784 0.9000 2.46 0.8765 0.22
297.60 0.1894 0.8994 0.9139 1.62 0.8994 0.00
297.60 0.1657 0.9177 0.9252 0.82 0.9198 0.23
297.60 0.1575 0.9225 0.9292 0.72 0.9272 0.51
297.60 0.1157 0.9507 0.9494 0.13 0.9649 1.50
297.60 0.1077 0.9526 0.9534 0.08 0.9716 1.99
297.60 0.1051 0.9542 0.9547 0.05 0.9736 2.04
297.60 0.1026 0.9563 0.9559 0.04 0.9756 2.01
297.60 0.0657 0.9802 0.9745 0.58 0.9957 1.58
297.60 0.0641 0.9811 0.9753 0.59 0.9961 1.53
297.60 0.0549 0.9836 0.9799 0.37 0.9980 1.47
297.60 0.0541 0.9866 0.9803 0.63 0.9982 1.17
332.60 0.7039 0.8022 0.7254 9.58 0.7597 5.30
332.60 0.5107 0.8215 0.8029 2.26 0.8611 4.82
332.60 0.4970 0.8504 0.8085 4.93 0.8687 2.15
332.60 0.4562 0.8571 0.8252 3.72 0.8911 3.97
332.60 0.3759 0.8839 0.8581 2.92 0.9328 5.54
332.60 0.3614 0.8901 0.8640 2.93 0.9397 5.57
332.60 0.2886 0.9088 0.8934 1.69 0.9686 6.58
332.60 0.2853 0.9133 0.8947 2.03 0.9697 6.17
332.60 0.2429 0.9276 0.9115 1.74 0.9811 5.77
332.60 0.1993 0.9451 0.9283 1.77 0.9889 4.64
332.60 0.1562 0.9641 0.9447 2.01 0.9937 3.07
332.60 0.1212 0.9767 0.9580 1.91 0.9963 2.01
332.60 0.1170 0.9768 0.9596 1.76 0.9966 2.02
332.60 0.0808 0.9893 0.9737 1.58 0.9984 0.92
332.60 0.0796 0.9897 0.9741 1.57 0.9984 0.88
332.60 0.0454 0.9980 0.9877 1.04 0.9995 0.15
AAD% 2.27 2.23
poor. The predicted trend from the two correlations as the
mole fraction of TEG approaches unity will be very different
from what this set of experimental data would suggest. Fig. 2
also indicates that Eq. (2) was tted preferentially to the data
at 297.6 K while the t to the data at 332.6 K is signicantly
poorer. The TST EoS is less accurate at 297.6 K, but is more
accurate at the higher temperature of 332.6 K.
Another important observation can be made from Fig. 2.
The TST EoS predicts slightly lower activity coefcients
than the experimental data at higher temperature. On the
other hand, Parrishs model predicts higher activity coef-
cients at higher temperature. This difference in the predicted
trend can have signicant consequences in the extrapolation
of activity coefcients to high temperatures. To test the ex-
trapolation capability, both the TST equation of state and
the Parrish model are extrapolated to 204

C (477.15 K). The

ability to extrapolate the model to high temperatures to pre-
dict accurate activity coefcients is quite important for strip-
ping processes. Ref. [6] reports that at 204

C (477.15 K)
and 1.2 atm (121.59 kPa), experience shows approximately
1.2 wt.% water in the lean glycol. The TST equation of state
predicts 1.2 wt.% of water in the lean glycol at 477.15 K un-
der pressure of 123.91 kPa, which is very close to the ob-
served 121.59 kPa, a deviation of only 1.91%. This result
corresponds to a water activity coefcient of approximately
0.7960 at 204

C. At the same temperature of 204

C, the
Parrish model gives a value of 0.9477 for the water activity
coefcient. This activity coefcient value will predict a pres-
C.H. Twu et al. / Fluid Phase Equilibria 228229 (2005) 213221 219
sure well over the expected value of 121.59 kPa. The TST
EoS provides a clear advantage to the use of Eq. (2) because
it can (1) accurately correlate the innite dilution activity co-
efcients data over a wide temperature range, (2) correlate
nite concentration activity coefcients data with reasonable
accuracy and (3) extrapolate well to higher temperatures.
4. Prediction of water dew point and water content
VLE data for TEGwater system commonly are repre-
sented as charts of water dewpoint lines as a function of con-
tactor temperature and liquid TEG concentrations [3]. The
water dew point is the dew point of the gas, T
d
, which would
be obtained if the gas were brought to equilibrium with the
TEG solution at the contactor temperature, T. In this work,
the TST equation of state is used to predict the water dew
point and the water content of the TEGwater system as an
alternative to using the water dew point charts [3].
At phase equilibrium, the fugacities of TEG and water in
the liquid and vapor phases are the same:
f
L
TEG
= f
V
TEG
(30)
f
L
w
= f
V
w
(31)
where f
i
is the fugacity of component i in the solution. The
fugacity of component i is calculated from the TST equa-
tion of state presented in this work. Assuming that the liquid
compositions and the systemtemperature, T, are known, Eqs.
(30) and (31) can be used to determine the equilibrium pres-
sure P and vapor-phase compositions. In the context of gas
dehydration, the equilibrium temperature T is referred to as
the contact temperature (temperature of the top tray of the
contactor). The vapor mixture at the contact temperature is at
its dew point. However, the specication for the equilibrium
vapor in gas dehydration is normally in term of its water dew
point (T
d
), i.e. the temperature at which pure liquid water
may condense out of the gas phase. One can view this tem-
perature as a hypothetical temperature. In order to compute
T
d
, TEG is excluded from the gas phase during the dew point
temperature calculation. Since it is assumed that the rst drop
of liquid condensed from the vapor at the water dew point is
pure water, we have
f
0L
w
(T
d
) = f
V
w
(T
d
) (32)
where f
0L
w
(T
d
) is the liquid fugacity of pure water at the
water dew point temperature T
d
and equilibrium pressure P
and f
V
w
(T
d
) the vapor fugacity of water in the mixture without
TEG at the dew point temperature T
d
, equilibrium pressure
P and equilibrium vapor-phase compositions y
i
. Note that
when the dew point temperature is below the triple point of
water, 273.16 K, the condensed phase is either solid ice or
subcooled water. Following the work of Parrish et al. [2],
subcooled water was used as the standard state fugacity in
this work. The equilibrium P and y
i
are solved from Eqs.
(30) and (31) as discussed above. Rewrite Eq. (32) as
f
0L
w
(T
d
) = P
0L
w
(T
d
) = f
V
w
(T
d
) = y
w
P
V
w
(T
d
) (33)
or
y
w
=

0L
w
(T
d
)

V
w
(T
d
)
(34)
where y
w
is the vapor mole fraction of water (without TEG)
in equilibrium with the TEG solution at equilibrium temper-
ature (contact temperature) T. The vapor and liquid fugacity
coefcients of water are calculated using the TST equation
of state at T
d
and at the equilibrium pressure P. Since TEG
is removed from the gas mixture when the water dew point
is performed, the vapor mole fraction of water is normalized
to give y
w
= 1.0 in this case. Eq. (34) is used to solve for the
water dew point temperature T
d
. After that, the liquid fugac-
ity of pure water at T
d
, f
0L
w
(T
d
), can then be calculated from
the equation of state.
When the value of f
0L
w
(T
d
) at the water dew point temper-
ature T
d
is obtained, the water content in lb
H
2
O
/MMSCF at
standard condition of T
0
=60

F (288.71 K) and the equilib-

rium pressure P can be calculated from the following equa-
tion:
n
V
=
P
Z
0
RT
0
(35)
where Z
0
is the gas compressibility factor at T
0
and P, and
n the number of moles in vapor volume V. The number of
moles of water, n
w
, in the vapor then becomes
n
w
V
= y
w
P
Z
0
RT
0
(36)
where y
w
is the mole fractionof water inthe gas phase without
TEG. Rewrite Eq. (36) using Eq. (33) and assuming that the
vapor fugacity coefcient of water is 1.0,
n
w
V
=
y
w
P
Z
0
RT
0

f
V
w
Z
0
RT
0
(37)
Substituting Eq. (32) into Eq. (37), we obtain
n
w
V
=
f
0L
w
Z
0
RT
0
(38)
Given T
0
=60

F(288.71 K), the gas constant Rand assuming

Z
0
=1.0, we obtain the water content in lb
H
2
O
/MMSCF in
terms of f
0L
w
(T
d
) at the water dew point temperature T
d
,
Water content in lb
H
2
O
/MMSCF = 222 72.23f
0L
w
(T
d
) (39)
The water content in SI units is expressed in kg/10
6
scm:
Water content in kg/10
6
scm = 356 765.00f
0L
w
(T
d
) (40)
where f
0L
w
(T
d
) is in the pressure unit of kPa. Since the water
content of lb
H
2
O
/MMSCF is commonly used in the US and
in the eld, this unit is used in Table 6.
The typical application range is shown in Tables 5 and 6.
Table 5 shows the prediction of the water dew point temper-
atures of vapor in equilibrium with TEG solutions from the
220 C.H. Twu et al. / Fluid Phase Equilibria 228229 (2005) 213221
Table 5
Prediction of the water dew point temperature of vapor containing TEG and water in equilibrium with TEG solutions from TST equation of state and prediction
from GPSA recommended model
99.97 wt.% TEG
Contact temperature (K)
277.59 288.71 299.82 310.93 322.04 333.15 344.26
Water dew points of vapor (K)
GPSA 208.15 214.26 220.37 225.93 232.04 237.59 243.15
TST 208.86 214.99 221.00 226.91 232.74 238.52 244.30
99.50 wt.% TEG
Contact temperature (K)
277.59 288.71 299.82 310.93 322.04 333.15 344.26
Water dew points of vapor (K)
GPSA 232.04 239.82 247.59 255.93 263.15 270.93 278.71
TST 232.66 240.47 248.16 255.75 263.22 270.58 277.84
Fig. 3. Equilibrium water dew point vs. contact temperature at various TEG
concentrations in wt.%.
TST equation of state. The water dew points from GPSA are
also included for comparison. Both GPSA and TST equation
of state predict very similar water dew point temperatures,
generally within 1 K.
Table 6 shows the equilibrium water content in
lb
H
2
O
/MMSCF gas in equilibrium with 99.50 wt.% TEG.
The results predicted fromthe TSTequation of state are com-
pared with the reported data [13,14], which cover the range
of temperature from 222 to 277 K. The predicted water con-
tent fromthe equation of state model agrees with the reported
data very closely over the entire range of temperature. The
calculated equilibrium pressure is also included in Table 6.
Fig. 3 presents the water dewpoint predicted fromthe TST
equation of state developed in this work as a function of con-
Table 6
Prediction of the equilibrium water content in lb
H
2
O
/MMSCF from TST equation of state in equilibrium with 99.50 wt.% TEG
T dew (K) Reported by Predicted from TST equation of state
McKetta and Wehe [13] Bukacek [14] Water content Pressure (Pa)
277.59 390 396 393 838.0
266.48 170 176 174 370.0
255.37 70 72 71 151.0
244.26 28 27 26 56.1
233.15 9.2 9.1 9 18.7
222.04 2.4 2.8 2.6 6.0
tact temperature at various TEGconcentrations ranging from
95.00 to 99.99 wt.% TEG. The result from the TST equation
is quite interesting because it illustrates that the water dew
point is a linear function of the contact temperature at a con-
stant wt.% of TEG. This plot is similar to those presented
in Refs. [2,3] and is useful for practical use in estimating
the water dew point without actually solving the equation of
state.
5. Conclusions
The TST cubic equation of state has been developed suc-
cessfully to represent the TEGwater binary, which is an in-
dustrial important systemfor modeling TEGgas dehydration.
This work is an improvement over the empirical hyperbolic
equation of Parrish et al. We also present a methodology for
using the TST equation of state to calculate water dew point
and water content in natural gas systems. The innite-dilution
and nite-concentration activity coefcients of water in TEG,
water dew point temperatures and water content over the en-
tire application range of temperature, pressure and composi-
tion encountered in a typical TEG dehydration unit predicted
from the TST equation of state match the reported data very
closely.
List of symbols
a, b cubic equation of state parameters
A Helmholtz energy
A
ij
, A
ji
, B
ij
, B
ji
NRTL binary interaction parameters
C.H. Twu et al. / Fluid Phase Equilibria 228229 (2005) 213221 221
C
r
dened in Eq. (20)
C
1
dened in Eq. (19)
G Gibbs energy
k
ij
binary interaction parameter
L, M, N parameters in the Twus function
P pressure
r reduced liquid volume
R gas constant
T temperature
x
i
liquid mole fraction of component i
y
i
vapor mole fraction of component i
Greek letters
cubic equation of state alpha function

ij
NRTL binary interaction parameters

i
activity coefcient of component i

ij
characteristic of the interaction between molecules
i and j

i
the fugacity coefcient of component i in the mix-
ture
innite pressure
Subscripts
c critical property
i, j property of component i, j
ij interaction property between components i and j
vdw van der Waals
0 zero pressure
Superscripts
* reduced property
E excess property
L liquid phase
V vapor phase
References
[1] J.R. Cunningham, J.E. Coon, C.H. Twu, Estimation of aromatic hy-
drocarbon emissions from glycol dehydration units using process
simulation, in: Proceedings of the 72nd Annual Gas Processors As-
sociation Convention, San Antonio, TX, March 1517, 1993.
[2] W.R. Parrish, K.W. Won, M.E. Baltatu, Phase behavior of the triethy-
lene glycolwater system and dehydration/regeneration design for
extremely low dew point requirements, in: Proceedings of the 65th
Annual GPA Convention, San Antonio, TX, March 1012, 1986.
[3] GPSA Engineering Data Book, Tenth Edition, First Revision, Gas
Processors Suppliers Association, Tulsa, OK, 1994.
[4] M. Herskowitz, M. Gottlieb, J. Chem. Eng. Data 29 (1984) 173175.
[5] B. Bestani, K.S. Shing, Fluid Phase Equilib. 50 (1989) 209221.
[6] GPA Editorial Review Board, Recent developments in gas dehydra-
tion and hydrate inhibition, in: Proceedings of the Laurance Reid
Conference, Norman, Oklahoma, 1994.
[7] C.H. Twu, W.D. Sim, V. Tassone, Fluid Phase Equilib. 6574 (2001)
183184.
[8] C.H. Twu, W.D. Sim, V. Tassone, Fluid Phase Equilib. 194197
(2002) 385399.
[9] G. Soave, Chem. Eng. Sci. 27 (1972) 11971203.
[10] D.Y. Peng, D.B. Robinson, Ind. Eng. Chem. Fundam. 15 (1976)
5864.
[11] C.H. Twu, D. Bluck, J.R. Cunningham, J.E. Coon, Fluid Phase Equi-
lib. 69 (1991) 3350.
[12] C.H. Twu, J.E. Coon, D. Bluck, B. Tilton, M. Rowland, Fluid Phase
Equilib. 153 (1998) 3944.
[13] J.J. McKetta, A.H. Wehe, GPA Engineering Data Book Fig. 15-10,
Ninth Edition, Fourth Revision, Gas Processors Suppliers Associa-
tions, Tulsa, OK, 1979.
[14] R.F. Bukacekm, Equilibrium moisture content of natural gases, Re-
search Bulletin 8, Institute of Gas Technology, Chicago, IL, 1955.