You are on page 1of 54

2 Thermodynamic models in process simulation

Without reliable physical properties, a process simulator is just an expensive


random number generator.
(A. Harvey and A. Laesecke, 2002)

The understanding of physical properties of fluid substances and their phase equilib-
ria is one of the main keys for a successful process engineering. Although there are
more exciting problems to solve in process engineering, almost every larger project
starts with the clarification of the physical property situation. Without having a sound
understanding of this, a reasonable process simulation cannot be carried out. Mainly,
the following properties are important for process engineering:
– thermodynamic properties:
– density;
– enthalpy;
– phase equilibrium;
– transport properties:
– viscosity;
– thermal conductivity;
– surface tension;
– diffusion coefficients.

For the mass and energy balance, only the thermodynamic properties are required,
whereas the transport properties play a major role for the equipment design, e.g. for
columns, heat exchangers, or pumps. For the thermodynamic properties, the calcula-
tions of the phase equilibria play the key role for the whole process simulation, as they
determine the separation steps in the process, which often cause 60–80 % of the total
costs of the process. In process simulators, a model has to be chosen which decides
in which way the phase equilibria are determined; also, it has an influence on the
evaluations of densities and enthalpies. It is not necessary but advantageous if one
model could be used for the whole process; each model change holds the risk that in-
consistencies are introduced (Chapter 2.11). Two kinds of models can be distinguished:
the equation-of-state models (φ-φ-approach) and the activity coefficient models (γ-φ-
approach). The difference and the advantages and disadvantages between these two
approaches should be known for a reasonable choice of the model. In the following
section, the essentials of the most important models are introduced and discussed
without any thermodynamic framework. For more details of these models, see [11].
The importance of the particular physical properties have been rated in Table 2.1
[12], where the relationships between accuracy of the particular properties and the
influence on the investment costs are listed. Certainly, this table should not be taken
as the absolute scientific truth but as an examplary case study. The outstanding item
26 | 2 Thermodynamic models in process simulation

is the large influence of the activity coefficient¹. The author would agree upon its im-
portance; however, in fact the costs are driven by the separation factors
p si γ i
α ij = . (2.1)
p sj γ j

The importance of its accuracy strongly depends on the case. When the separation fac-
tors are far away from unity, the influence on the accuracies of the activity coefficent
and vapor pressures is limited. When the separation factors are close to unity, their in-
fluence is incredibly high, and, moreover, they can even decide whether a separation
is possible at all.
The heat of vaporization is clearly proportional to the reboiler duty in a distillation
and therefore to the size of the reboiler; thus, the proportionality is quite in line with
the experience of the author. In comparison to the heat of vaporization, the influence
of specific heat capacity is smaller. Nevertheless, single pieces of equipment can be
strongly influenced (e.g. liquid-liquid heat exchangers), and errors occur pretty often
(Chapter 2.8). The transport properties, thermal conductivity and viscosity, have an
influence on the heat transfer coefficient in heat exchangers. The author would guess
that the influence of the viscosity is greater for large viscosities. Moreover, errors in
the viscosity occur quite frequently, especially for mixtures, whereas thermal conduc-
tivities do not vary too much for the particular liquids, with the exception of water
and glycols. Vapor viscosities and thermal conductivities are usually not measured
but estimated anyway.
The accuracy of estimations of physical properties is often of great interest. This
question is not easy to answer, as most of the authors are suspected to claim a higher
accuracy than they should. However, it is hard to imagine how objective criteria can
be set up. Clearly, the average deviation of the fit to the data available is an indica-
tor, but certainly the fit to unknown data will be worse. Other authors try to leave out

Table 2.1: Example of a relationship between physical property accuracy and investment costs [12].

Physical property % error % error capital cost


Thermal conductivity 20 % 13 %
Specific heat capacity 20 % 6%
Heat of vaporization 15 % 15 %
Activity coefficient 10 % 100 %
Diffusion coefficient 20 % 4%
Viscosity 50 % 10 %
Density 20 % 16 %

1 The activity coefficient γ will be explained in Chapter 2.1. In the context of this paragraph, it is de-
fined as a factor describing the deviations from Raoult’s law. It can be interpreted as a correction factor
for the concentration.
2.1 Phase equilibria | 27

Table 2.2: Accuracy of physical property predictions [15].

Physical property Error expected Error desired


Heat of formation 2.5–4 kJ/mol 4 kJ/mol
Liquid heat capacity > 10 % 10 %
Liquid density > 2% 2%
Vapor pressure > 10 % 10 %
Normal boiling point 6K 3K
Transport properties > 10 % 10–20 %
Heat of vaporization 15 % 15 %
Limiting activity coefficient > 10 % 10 %

certain data sets in the fitting process [13], however, one does not know according to
which criteria they are chosen. Another method is to predict new data sets before they
are integrated into the database [14]. This method can not deliver large amounts of
examples, and, moreover, it is not reproducible, as after testing they will certainly be
added to the database.
A thorough examination has been done in Table 2.2. The conclusions are as fol-
lows [15]:
– the accuracy of the methods are not at industrial target level;
– experimental data for thermal and transport properties are limited;
– group contribution methods seem to have reached their potential, there is hardly
room for improvement. To meet the industrial demand, new approaches are nec-
essary.

2.1 Phase equilibria


Life is too short to worry about phase equilibria.
(Based loosely on a comment of Georgios Kontogeorgis on electrolytes.)

The knowledge and understanding of the various phase and chemical equilibria is
the key for a successful process simulation. Two-phase regions have a great impor-
tance in technical applications. Even for a one-component system phenomena occur
which need to be discussed thoroughly. Exemplarily, Figure 2.1 illustrates the isobaric
vapor-liquid equilibrium of water when it is heated from t1 = 50 °C to t2 = 150 °C at
atmospheric pressure.
In the two-phase region, vapor and liquid coexist at the same temperature and
pressure [11]. In this case, both liquid and vapor are called saturated. If the saturated
liquid is further heated, the temperature does not change. Instead of a temperature
rise, the liquid is vaporized. After the last drop of liquid vanishes, the temperature
rises again. Figure 2.1 clearly indicates that much more heat is consumed for the evap-
oration of water (enthalpy of vaporization, ∆hv ) than for the 100 K temperature eleva-
28 | 2 Thermodynamic models in process simulation

150

125

100
t/°C

75

50

25

0
0 500 1000 1500 2000 2500 3000
q/J/g
p
p

p
p p

V V V

L L L

All liquid First vapor Phase Last liquid All vapor


bubble equilibrium drop

Fig. 2.1: Temperature change of water at p = 1.013 bar with respect of the heat added [11].
© Wiley-VCH Verlag GmbH & Co. KGaA. Reproduced with permission.

tion. The state of a pure substance in the phase equilibrium region is not characterized
by temperature and pressure as in the one-phase region. Temperature and pressure
are related; the relationship
ps = f(T) (2.2)
is the vapor pressure curve. For the complete determination of the two-phase system
the vapor quality x
n󸀠󸀠
x= 󸀠 (2.3)
n + n󸀠󸀠
is necessary, where the superscripts 󸀠 and 󸀠󸀠 denote the saturated states of liquid and
vapor, respectively. x = 0 means a saturated liquid, x = 1 means a saturated vapor. In
the two-phase region, the specific volume v, the specific enthalpy h and the specific
entropy s can be written as
2.1 Phase equilibria | 29

v = xv󸀠󸀠 + (1 − x)v󸀠 (2.4)


󸀠󸀠 󸀠
h = xh + (1 − x)h (2.5)
󸀠󸀠 󸀠
s = xs + (1 − x)s . (2.6)

Vapor pressure and enthalpy of vaporization of a pure substance are related by the
Clausius–Clapeyron equation
dps 󸀠󸀠
∆hv = T (v − v󸀠 ) . (2.7)
dT
For a multicomponent system, we must also consider that both phases have different
composition, and the distribution of each component is a central issue to be solved
by the thermodynamic model. As will be shown later, the behavior of a multicompo-
nent system is mainly described by binary subsystems. The best way to illustrate the
vapor-liquid equilibrium of a binary system is the pxy diagram at constant tempera-
ture. Figure 2.2 gives an example for the system ethanol/water.

80
Bubble point curve

70
Liquid
Tie-line
60
p/kPa

2-phase Vapor
region
50

40 Dew point curve


T = 343.15 K
30 Fig. 2.2: Example for a pxy diagram [11].
0 0.5 1 © Wiley-VCH Verlag GmbH & Co. KGaA.
x1, y1/mol/mol Reproduced with permission.

In the upper part of this diagram, there is the liquid region above the bubble point
curve. Below the dew point curve, there is the vapor region. Between bubble and dew
point curve, there is the two-phase region. For a specific pressure, a horizontal tie-line
connects the corresponding liquid and vapor points in phase equilibrium, referring
to their concentration on the abscissa. These phase equilibrium diagrams can be ob-
tained by correlating phase equilibrium data. There are several experimental options;
the most popular ones are to measure both the vapor and the liquid concentration
and the temperature at constant pressure, or to measure the bubble point at constant
temperature for certain liquid concentrations. The latter option should be preferred
in most cases; isothermal data are much more useful for the adjustment of model pa-
rameters [16]. Bubble and dew point line meet at the ordinates at x = 0 and x = 1,
indicating the corresponding pure component vapor pressures. If the temperature is
greater than the critical temperature of one of the components, the vapor pressure of
30 | 2 Thermodynamic models in process simulation

60
177.59 K 166.48 K
50
144.26 K
40 133.15 K
p/bar

30

20 110.93 K
10
Fig. 2.3: pxy diagram with one component
0 becoming supercritical [11].
0 0.5 1 © Wiley-VCH Verlag GmbH & Co. KGaA.
x1, y1/mol/mol Reproduced with permission.

this component does not exist, although it can take part in a phase equilibrium in the
dilute state. In Figure 2.3, the typical change in the shape of the pxy diagram is shown
for the system nitrogen/methane when nitrogen becomes supercritical.
Some other diagrams are useful for illustrating a binary vapor-liquid equilibrium.
The simple yx diagram shows the relationship between vapor and liquid concentra-
tion without indicating the pressure or temperature. For a better overview, the bisect-
ing line is usually depicted (Figure 2.4, first column). Also a Txy diagram is possible
for isobaric phase equilibria, as can be seen in the right column of Figure 2.4. In this
diagram, the upper line is the dew point curve, and the lower line is the bubble point
curve. Figure 2.4 also shows the various kinds of binary vapor-liquid equilibria.
The upper row shows an ideal mixture obeying Raoult’s law (Equation (2.40)).
There are no interaction forces between the molecules. The phase equilibrium is just
determined by the vapor pressures of both components. Typical examples for ideal
systems are benzene/toluene or n-hexane/n-heptane. In the pxy diagram, the boiling
point curve is a straight line. In the Txy diagram, usually no straight lines occur. The
activity coefficients (Chapter 2.3.1) are all equal to 1, i.e. their logarithm is 0.
Row 2 shows a system having small nonidealities. Molecules of the same kind
“prefer” to be together instead of mixing with molecules of a different kind. As a result,
greater pressure is built up than for the ideal mixture. The activity coefficients are
greater than 1. A typical example for such a system is methanol/water.
When the activity coefficients become larger, the system can exhibit an azeotrope
with a pressure maximum in the pxy diagram and a temperature minimum in the
Txy diagram (row 3). At the azeotropic point, the liquid and vapor concentrations
are identical. Therefore, an azeotrope cannot be separated by simple distillation. The
knowledge of azeotropes is essential for any process development. A typical exam-
ple for a homogeneous azeotrope is water/1-propanol. The occurrence of azeotropes
is also related to the vapor pressures; the closer together the vapor pressures of the
components are, the more probable is the occurrence of an azeotrope [11] (p. 46).
2.1 Phase equilibria | 31

1 4 140
0.8 380
3 120
0.6

p/kPa
100 370

ln γi

T/K
2
y1

0.4
80 360
0.2 1
60
0 0 350
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(1) x1 x1 x1, y1 x1, y1

1 4 350
80
0.8 340
3
0.6 60

p/kPa
ln γi

T/K
2 330
y1

0.4
40
0.2 1 320
20
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(2) x1 x1 x1, y1 x1, y1

1 4 130 355
0.8 120
3
350
0.6
p/kPa

110
ln γi

T/K
2
y1

0.4
100 345
0.2 1
90
0 0 340
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(3) x1 x1 x1, y1 x1, y1

1 4 30 345
0.8 25 340
3
335
0.6 20
p/kPa
ln γi

T/K

2 320
y1

0.4 15
315
0.2 1 10 310
0 0 5 305
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(4) x1 x1 x1, y1 x1, y1

1 0 120 340
0.8 110
335
0.6
p/kPa
ln γi

T/K

–1 00
y1

0.4
330
0.2 90
0 –2 80 325
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(5) x1 x1 x1, y1 x1, y1

Fig. 2.4: Kinds of binary vapor-liquid equilibria [11].


1 ideal mixture
2 small nonidealities
3 larger activity coeffcients, homogeneous azeotrope
4 heteroazeotrope
5 vapor pressures close to each other/strong negative deviations from Raoult’s law.

© Wiley-VCH Verlag GmbH & Co. KGaA. Reproduced with permission.


32 | 2 Thermodynamic models in process simulation

If the activity coefficients increase even more, the system exhibits a miscibility gap,
and the liquid splits into two phases. A heteroazeotrope occurs (row 4). A typical
example is water/n-butanol. Note that a miscibility gap and the occurrence of a het-
eroazeotrope is not necessarily coupled. If the vapor pressures largely differ, the
azeotrope does not occur, whereas the miscibility gap itself is not related to the vapor
pressures.
In addition, negative deviations from Raoult’s law can occur, where the activity
coefficients are lower than 1. This means that the molecules “like” each other and
prefer to be surrounded by molecules of a different kind. An example for such a system
is dichloromethane/2-butanone [11].
If vapor pressures are close to each other and the negative deviations are strong,
even an azeotrope with a pressure minimum in the pxy diagram and a temperature
maximum in the Txy diagram can occur (row 5). An example is acetone/chloroform.
It cannot happen that systems with negative deviations from Raoult’s law show a mis-
cibility gap.
In addition to these five types of phase equilibria there are others. For example,
methyl acetate/water shows a miscibility gap with a homoazeotrope [17], and ben-
zene/hexafluorobenzene has a double azeotrope, one with a pressure minimum and
one with a pressure maximum [18].
As mentioned, there are two different approaches for the description of phase
equilibria. In the following sections, they are explained on vapor-liquid equilibria.

2.2 φ-φ-approach

You cannot just learn thermodynamics. You must love it.


(Recommendation to desperate students)

Generally, an equation of state is a relationship between the pressure p, the abso-


lute temperature T, and the specific volume v. In a first step only pure substances are
regarded. It makes sense to focus on equations of state which are pressure-explicit.
Hence, the general form of an equation of state can be written as

p = f(T, v) . (2.8)

The simplest equation of state is the ideal gas equation


RT
p= . (2.9)
v
Equation (2.9) is exact for a gas where the molecules have no volume and do not
exert interaction forces on each other. It is a good approximation for gases at low
pressures. At increasing pressures, Equation (2.9) becomes increasingly inaccurate.
2.2 φ-φ-approach | 33

A well-known modification of Equation (2.9) is the virial equation:

RT B(T) C(T) D(T)


p= (1 + + 2 + 3 + ⋅⋅⋅) . (2.10)
v v v v

Equation (2.10) is the so-called Leiden form; correspondingly, there is also a volume-
explicit Berlin form expressed in form of a polynomial of the pressure p. The virial
coefficients B, C, D, … account for the deviations from the ideal gas equation. Al-
though it has a theoretical background, the virial equation is not suitable for practical
applications. Usually, it has to be truncated after the second term, as the third virial co-
efficient C and the ones following are hardly ever known. The consequence is that the
virial equation can only be used up to moderate densities (rule of thumb: ρ = 0.5ρc ).
The most widely used equations of state in technical applications are modifica-
tions of the van-der-Waals equation
RT a
p= − 2 (2.11)
v−b v
Invented in 1873 [19]², the van-der-Waals equation was the first equation of state valid
for both the vapor and the liquid phase that could at least qualitatively explain the
pvT behavior of a pure substance, illustrated by the pv diagram in Figure 2.5.

Melting curve
Crystallization curve

p Critical point

pc
Dew point
C B curve
ps (T)
A T>
Boiling Tc
point curve Tc
T<T
c
Fig. 2.5: The pvT behavior of a pure
substance [11].
© Wiley-VCH Verlag GmbH & Co. KGaA.
vc v Reproduced with permission.

2 Just like all authors, I give this citation. However, I will admit that I have not read it: The language
is Dutch, and even if I could speak it I would guess that the way it was used in the 19th century would
not be comprehensible for a nonnative.
34 | 2 Thermodynamic models in process simulation

The pv diagram for a pure substance is dominated by the large vapor-liquid equilib-
rium region, where the isotherms are horizontal, e.g. between points B and C, indi-
cating that the pressure during condensation or evaporation remains constant. At the
left-hand side of the vapor-liquid equilibrium region, the isotherms are very steep,
meaning that large pressures are needed to lower the specific volume. This is the re-
gion of the liquid phase, and the steep slope means that compressing a liquid does
not result in a major volume change. At the right-hand side there is the vapor region,
where it is easier to compress the substance. At low pressures, the isotherms obey to
the ideal gas law equation (2.9). At higher pressures, the ideal gas law becomes more
and more inaccurate. In the two-phase region, the boiling point line and the dew point
line are connected by horizontal isotherms as mentioned, giving the saturated vapor
volume v󸀠󸀠 (e.g. point B) and the saturated liquid volume v󸀠 (e.g. point C). With increas-
ing temperature, v󸀠󸀠 and v󸀠 are getting closer to each other. At the critical point, vapor
and liquid become identical. Above the critical temperature (and, correspondingly,
the critical pressure), no phase equilibria between vapor and liquid exist.
Remarkably, one can get from the vapor to the liquid region without crossing the
two-phase region, i.e. a vapor can be gradually transformed into a liquid and vice
versa. E.g. starting from point C, the saturated liquid can be isochorically (i.e. at v =
const.) heated up to a temperature above the critical one, where simultaneously the
pressure is increased to a value above the critical pressure. Then, the substance can
be heated isobarically, and finally, it can be cooled down isochorically to point B. As a
result, a liquid has been turned into a vapor without ever having both phases existing
simultaneously.
When the typical application of an equation of state, the calculation of v at given
T and p, is performed, Equation (2.11) gives a third-degree polynomial:

RT 2 a ab
v3 − (b + )v + v− = 0. (2.12)
p p p

Therefore, the van-der-Waals equation (2.11) is called a cubic equation of state. The
advantage is that it can be solved analytically [11, 20]. One obtains either one real and
two complex or three real solutions. In the latter case, the largest solution corresponds
to a vapor specific volume, and the smallest one corresponds to a liquid volume. The
middle one has no physical meaning. In contrast to Equation (2.9) or the virial equa-
tion (2.10) truncated after the second term, the cubic equations can be used both for
the vapor and for the liquid phase. To decide whether the liquid or the vapor solution
is valid, the vapor pressure is needed.
Of course, an equation of state defined by a continuous function like the van-der-
Waals equation (2.11) cannot reproduce the dogleg at the transition from the one-phase
to the two-phase region. Thermodynamics points out that the equation of state can
evaluate the phase equilibrium using the Maxwell criterion (Figure 2.6).
Liquid and vapor phase are in equilibrium at p = ps if the hatched areas between
p = ps and the equation of state in Figure 2.6 are equal. Analytically, it can be deter-
2.2 φ-φ-approach | 35

35
v′ v″
p = ps
p / bar

10
0.5 1 1.5 2
–15
v / kmol/m3
–40

–65 Fig. 2.6: Application of the Maxwell criterion


(propylene, t = 20 °C, Peng–Robinson equa-
–90 tion).

mined by the equation


v󸀠󸀠
󸀠󸀠 󸀠
ps (v − v ) = ∫ p dv . (2.13)
v󸀠

The curvature between v󸀠 and v󸀠󸀠 has no physical meaning, including the negative
values obtained for the pressure. Equation (2.13) is equivalent to the formulation

f 󸀠 = f 󸀠󸀠 , (2.14)

where f is the so-called fugacity, which can be interpreted as a corrected pressure. For
a pure substance, the fugacity is the product of the pressure and the fugacity coeffi-
cient φ:
pφ󸀠 = pφ󸀠󸀠 , (2.15)
where the pressure cancels out.
However, while its theoretical value is unquestioned, the van-der-Waals equation
had limited success in practical applications, where quantitatively correct results are
required. In the second half of the 20th century, several modifications of the van-der-
Waals equation have been developed. The most successful and most widely-used ones
are the Soave–Redlich–Kwong equation (SRK or RKS)
RT a(T)
p= − , (2.16)
v − b v(v + b)
with
R2 Tc2
a(T) = 0.42748 α(T) (2.17)
pc
2
α(T) = [1 + (0.48 + 1.574 ω − 0.176 ω2 )(1 − Tr0.5 )] (2.18)
RTc
b = 0.08664 , (2.19)
pc

and the Peng–Robinson equation (PR)


RT a(T)
p= − , (2.20)
v − b v(v + b) + b(v − b)
36 | 2 Thermodynamic models in process simulation

where
R2 Tc2
a(T) = 0.45724 α(T) (2.21)
pc
2
α(T) = [1 + m(1 − Tr0.5 )] (2.22)
m = 0.37464 + 1.54226 ω − 0.26992 ω2 (2.23)
RTc
b = 0.0778 . (2.24)
pc
It is remarkable that both the Soave–Redlich–Kwong and the Peng–Robinson equa-
tion need only three substance-specific parameters, i.e. the critical temperature Tc ,
the critical pressure pc and the acentric factor ω, which is defined as
ps
ω = −1 − lg ( ) . (2.25)
pc T=0.7 Tc
Essentially, ω represents the vapor pressure at T = 0.7Tc . For most substances, this
is a temperature close to the normal boiling point. Thus, the critical point and one
specified point of the vapor pressure curve are the only substance-specific information
used. This characterization of a substance is called the three-parameter corresponding
states principle (Tc , pc , and ω). Equations of state using only this input information
are called generalized equations of state.
Equations of state valid for both the vapor and the liquid state can provide all
thermodynamic properties needed for process calculations. Its original well-known
purpose was to be a relationship between p, v, and T in the vapor phase. As seen,
cubic equations of state can also be used to calculate the specific volume in the liq-
uid phase, and as seen in Equations. (2.13) and (2.15), the vapor pressure for a given
temperature can be evaluated. From thermodynamics, an expression for the enthalpy
using a pressure-explicit equation of state can be derived: [11]:
T v
∂p
h(T, v) = h0 + ∫ cid
p dT + ∫ [T ( ) − p] dv + pv − RT . (2.26)
∂T v
T0 ∞

Inserting the equation of state into Equation (2.26), an expression is obtained which
can calculate the specific enthalpy of the substance at any state. The only additionally
required input is the specific isobaric heat capacity in the ideal gas state. The enthalpy
of vaporization for a temperature T is then easily determined by

∆h v = h(T, v󸀠󸀠 ) − h(T, v󸀠 ) . (2.27)

For the Peng–Robinson equation, the most important calculation equations for pure
substances are given in the following:
– Cubic equation for the determination of the volume:
RT 2 a 2bRT RT 2 ab
v3 + (b − ) v + ( − 3b2 − ) v + b3 + b − = 0. (2.28)
p p p p p
2.2 φ-φ-approach | 37

– Specific enthalpy [11]:

1 da v + (1 + √2)b
(h − hid (T, v)) = RT(Z − 1) − (a − T ) ln ( ), (2.29)
√8b dT v + (1 − √2)b
with
pv
Z= (2.30)
RT
and
da R2 Tc2 m
= −0.45724 [1 + m(1 − Tr0.5 )] . (2.31)
dT pc √ TTc
– Fugacity coefficient [21]:

bp a v + (1 + √2)b
ln φ = Z − 1 − ln (Z − )− ln ( ). (2.32)
RT √
2 2bRT v + (1 − √2)b

The success of this route is remarkable, taking into account that only three substance-
specific parameters are used (Tc , pc , ω), but limited. The three-parameter correspond-
ing states principle is especially valid for nonpolar compounds. The generalized cubic
equations of state have become standard tools in the oil and gas industries, where the
compounds involved are usually nonpolar. In these cases, the vapor densities and the
vapor pressures are reproduced very well, especially in the high-pressure region. This
can be explained by the fact that the critical point is necessarily reproduced exactly;
the closer the distance to the critical point is, the better the results will be. For polar
compounds (e.g. water, methanol, ethanol), the results are usually remarkably good
for the specific volume in the vapor phase, but the vapor pressure is not reproduced re-
liably enough for process calculations. The enthalpies of the vapor are usually a good,
but not exact, estimate. For the enthalpy of vaporization the same result as for the va-
por pressure is obtained; it is remarkably good for nonpolar substances, but hardly
reliable for polar ones. The enthalpies of a liquid are usually poor for both polar and
nonpolar compounds; the reason for this is explained in Chapter 2.8. The specific vol-
ume of the liquid phase is not even intended to be correct. In most cases, only the
order of magnitude is met. For technical applications, the liquid volume calculated
by cubic equations of state is by far not accurate enough. Instead of the correct liquid
volume, it should only be taken as an auxiliary quantity for the calculation of the vapor
pressure and the liquid enthalpy. In the result view of process simulators, the liquid
volume is usually overwritten by default with results from a liquid density correlation
(Chapter 2.3).
To overcome these limitations, the α-functions of the generalized equations
of state can be replaced by individual ones. An example is the PSRK (Predictive
Soave–Redlich–Kwong) equation, where the generalized α-function (2.18) has been
replaced by
2
α(T) = [1 + c1 (1 − Tr0.5 ) + c2 (1 − Tr0.5 )2 + c3 (1 − Tr0.5 )3 ] . (2.33)
38 | 2 Thermodynamic models in process simulation

The adjustable parameters c1 , c2 , c3 are usually fitted to vapor pressure data. Tr (re-
duced temperature) is an abbreviation for T/Tc . This way, polar components can be
described as well. In the volume-translated Peng–Robinson equation (VTPR) the α-
function of the Peng–Robinson equation (2.22) has been replaced by the flexible Twu-
α-function
N(M−1)
α(Tr ) = Tr exp [L(1 − TrMN )] , (2.34)

where the adjustable parameters L, M, N can again be fitted to vapor pressure data.
Also, data for cLp can simultaneously be adjusted so that the enthalpy of the liquid can
also be described by the equation [22]. With the volume translation, an attempt was
made to solve the remaining problem with the bad reproduction of the liquid density.
The approach is that the specific volume v in the original equation is replaced by a
term v + c, where c is a volume translation:
RT a(T)
p= − . (2.35)
v + c − b (v + c)(v + c + b) + b(v + c − b)
Therefore, it remains a cubic equation of state, but the results for the specific volumes
for both the vapor and the liquid phase are shifted by the value of c. While this is al-
most negligible for the vapor phase, it causes an improvement for the specific volume
of the liquid phase. The parameter c can be fitted to liquid density data or, if not avail-
able, calculated by a generalized function. However, an acceptable improvement is
restricted to low pressure data, and the densities of liquids can still not be reproduced
with the accuracy required in technical applications. Therefore, in spite of the use of
volume translations it is still recommended to make use of the option to overwrite it
with a liquid density correlation.
Besides the cubic ones, a lot of other equations of state are in use. Only a few of
them with a special importance for process engineering purposes can be mentioned
here.
In process engineering, cases occur where a much higher accuracy in the physical
properties is required. Examples are the power plant processes, the heat pump process
or the pressure drop calculation in a large pipeline. For the description of the complete
pvT behavior including the two-phase region several extensions of the virial equation
were suggested. All these extensions have been derived empirically and contain a large
number of parameters, which have to be fitted to experimental data. A large database
is necessary to obtain reliable parameters. One of the first approaches to equations of
state with higher accuracies has been made by Benedict, Webb, and Rubin (BWR) in
1940 [23, 24], who used 8 adjustable parameters. Their equation allows reliable cal-
culations of pvT data for nonpolar gases and liquids up to densities of about 1.8ρc .
Bender [25] extended the BWR equation to 20 parameters. With this large number
of parameters it became possible to describe the experimental data for certain sub-
stances over a large density range in an excellent way. In the last two decades, the
so-called technical high-precision equations of state have been developed [26–28].
2.2 φ-φ-approach | 39

Their significant improvement was possible due to progress in measurement tech-


niques and the development of mathematical algorithms for optimizing the structure
of equations of state. With respect to the accuracy of the calculated properties, their
extrapolation behavior and their reliability in regions where data are scarce, these
equations define the state-of-the-art representation of thermal and caloric properties
and their particular derivatives in the whole fluid range. There is also an important de-
mand to get reliable results for derived properties (cp , cv , speed of sound). Technical
high-precision equations of state are a remarkable compromise between keeping the
accuracy and gain in simplicity. Furthermore, these equations should enable the user
to extrapolate safely to the extreme conditions often encountered in industrial pro-
cesses. For example, in the LDPE³ process ethylene is compressed to approximately
3000 bar, and it is necessary for the simulation of the process and the design of the
equipment to have a reliable tool for the determination of the thermal and caloric prop-
erties. The complexity and limited availability is no longer an issue. Currently, there
are approximately 80 substances for which the data situation has justified the devel-
opment of a technical high-precision equation of state, e.g. water, methane, argon,
carbon dioxide, nitrogen, ethane, n-butane, isobutane, and ethylene. In the FLUID-
CAL software [29], these equations of state have been made applicable for users with-
out special knowledge. The most serious drawback of the high-precision equations of
state is that the concept can hardly be applied to mixtures. So far, only for natural gas
mixtures an equation of state of similar quality has been developed [30].
Table 2.3 illustrates the accuracy demand for technical equations of state.
The φ-φ-approach can be extended to mixtures. Thermodynamics says that the
equilibrium condition Equation (2.14) becomes
px i φ󸀠i = py i φ󸀠󸀠
i (2.36)
for each component involved. Again, the pressure cancels out. The cubic equations of
state can be transformed to mixture applications by mixing rules for the parameters
a and b. The most common ones are
a = ∑ ∑ z i z j (a ii a jj )0.5 (1 − k ij ) (2.37)
i j

b = ∑ zi bi (2.38)
i

Table 2.3: Accuracy demand for technical high-precision equations of state.

ρ(p, T) w ∗ (p, T) cp (p, T) ps (T) ρ󸀠 (T) ρ󸀠󸀠 (T)


p < 30 MPa 0.2 % 1–2 % 1–2 % 0.2 % 0.2 % 0.2 %
p > 30 MPa 0.5 % 2% 2%

3 low density polyethylene.


40 | 2 Thermodynamic models in process simulation

for both the Peng–Robinson and the Soave–Redlich–Kwong equation of state. As the
mixing rules (2.37) and (2.38) refer to both the vapor and the liquid phase, the neutral
variable z is used for both the vapor and the liquid mole fraction. k ij is an adjustable
binary interaction parameter. It is symmetric (k ij = k ji , k ii = k jj = 0) and has usually
small values (−0.1 < k ij < 0.1). Nevertheless, it has a significant influence on the cal-
culation of phase equilibria and cannot be neglected. The impact on the results for
the liquid and vapor volumes is comparably small. Using Equations (2.37) and (2.38),
the calculation routes for pure components can be applied to mixtures. For the phase
equilibrium calculations, Equation (2.36) looks very easy but is in fact a complicated
equation, as the determination of the fugacity coefficients ends up in long equations,
e.g. for the Peng–Robinson equation [21]

bi p
ln φ i = (z − 1) − ln [ (v − b)]
b RT
a 2 bi v + (1 + √2)b (2.39)
− ( ∑ z j a ij − ) ln [ ].
2√2bRT a j b v + (1 − √2)b

The mixing rules (2.37) and (2.38) have had considerable success as long as only non-
polar substances were involved. When polar compounds are regarded, poor results are
obtained. For an adequate description of systems with polar components using the φ-
φ-approach, the so-called g E mixing rules have to be applied (Figure 2.7). They will be
explained in Chapter 2.7, as understanding them is not possible without knowledge
of the γ-φ-approach.

120 120

100 100
333K 333K
80 80
p/kPa

p/kPa

60 323K 60 323K

40 40

20 20

308K 308K
0 0
0 0.5 1 0 0.5 1
x1, y1 / mol/mol x1, y1 / mol/mol

Fig. 2.7: Experimental and calculated VLE data for the system acetone (1)/water (2) using the
Peng–Robinson equation with k ij (left-hand side) and g E mixing rules (right-hand side) [11].
© Wiley-VCH Verlag GmbH & Co. KGaA. Reproduced with permission.
2.3 γ-φ-approach | 41

2.3 γ-φ-approach

2.3.1 Activity coefficients

The activity coefficient is the factor by which I have to miscalculate, to get the correct result al-
though I use the wrong equation.
(Student in his thermodynamics exam. He passed easily.)

The phase equilibrium condition (2.14) can also be elaborated in a different way [11],
using different approaches for the vapor and the liquid phase. The simplest solution
is Raoult’s law:
x i p si = y i p , (2.40)
where the bubble point curve in the pxy diagram is a straight line. Only few binary
systems obey this law, an example is the system benzene/toluene (Figure 2.4, upper
row).
If deviations from Equation (2.40) occur, the activity coefficient γ i is introduced,
which depends on the concentration as well as on the temperature. One can take the
activity coefficient as a factor which corrects the concentration. The equilibrium con-
dition, still not the final one, is then written as

x i γ i p si = y i p . (2.41)

This equilibrium condition is valid in the low pressure region, as the vapor phase is
regarded as an ideal gas. While it is still often used in academics, it has become more
and more outdated in process simulation applications, as the enthalpy calculations
should not be performed using the ideal gas law for the vapor phase (Chapter 2.8).
Instead, a real vapor phase is considered using an equation of state:

φ i (T, p, y i )
x i γ i p si Poyi = y i p pure , (2.42)
φi (T, p si )

with Poyi as the Poynting factor

vLi (p − p si )
Poyi = exp . (2.43)
RT
pure
φ i (T, p, y i ) accounts for the nonideality of the vapor phase, whereas φ i (T, p si ) refers
to the liquid fugacity coefficient at the vapor pressure. Fortunately, at p = p si the liq-
uid fugacity coefficient is equal to the vapor one (Equation (2.15)); thus, in contrast
to Equation (2.36), the equation of state does not need to be valid for both the vapor
and the liquid phase. Therefore, the virial equation (2.10) truncated after the second
term can also be applied. Nevertheless, usually generalized cubic equations of state
are used, as they are the most powerful tools for this purpose with easily accessible
input parameters (Tc , pc , and ω). The Poynting factor corrects the small error caused
42 | 2 Thermodynamic models in process simulation

by evaluating the fugacity of the pure liquid at its vapor pressure p si instead of the
system pressure p. It can usually be neglected.
The heart of Equation (2.42) is the activity coefficient γ i . The formal character of
the γ i is a correction of the molar concentration, which, however, hardly explains
anything. In fact, the activity coefficient accounts for the intermolecular interactions
between the molecules and the entropic effects. Figure 2.8 shows the typical isother-
mal concentration dependence of the activity coefficient.

4
γi

Fig. 2.8: Typical isothermal concentration


1 dependence of the activity coefficient [11].
0 0.5 1
© Wiley-VCH Verlag GmbH & Co. KGaA.
x1 Reproduced with permission.

Its maximum values (respectively minimum values for systems with negative devia-
tions from Raoult’s law) occur when the component is extremely diluted; at the con-
centration x i → 0. This value is called the activity coefficient at infinite dilution (γ∞
i )
and is characteristic for the illustration of the nonideal behavior.
Nowadays, mainly three equations for the correlation of the γ i are in use; the Wil-
son [31], the NRTL⁴ [32], and the UNIQUAC⁵ equation [33]. They are all based on the
so-called principle of local compositions, which is explained by means of the Wilson
equation in the following paragraph:
In Figure 2.9 it can be seen that in both cells the concentrations of the molecules 1
and 2 are x1 = 37 and x2 = 47 . Due to the intermolecular forces, the local concentrations
can be different. In the left cell, there is a molecule of kind 1 in the center of the cell.
Around this molecule, the concentration of the molecules of kind 1 is x11 = 26 , and the
concentration of the molecules of kind 2 is x21 = 46 . Similarly, in the cell on the right
hand side with a molecule of kind 2 in the center the concentrations are x12 = x22 = 36 .
The concentrations around a molecule and the total concentrations are assumed to be

4 nonrandom two-liquid.
5 universal quasi-chemical.
2.3 γ-φ-approach | 43

2 1
1 1 1 2
1 2
2 2 2 2
2 1

Fig. 2.9: Sketch for the explanation of the Local Composition Models [11].
© Wiley-VCH Verlag GmbH & Co. KGaA. Reproduced with permission.

related by Boltzmann factors:


x ji x j exp (−λ ji /(RT))
= , (2.44)
x ii x i exp (−λ ii /(RT))
where the λ terms account for the intermolecular forces. Defining the local mole frac-
tion as
x ii v i
ξi = , (2.45)
∑j x ji v j
γ i can be introduced as a correction factor of the total concentration:

γ i = ξ i /x i . (2.46)

After several mathematical transformations [11] the following expression for the activ-
ity coefficients is obtained:
x k Λ ki
ln γ i = 1 − ln ∑ x j Λ ij − ∑ , (2.47)
j k
∑j x j Λ kj

with Λ as an abbreviation representing the intermolecular forces, which are tempera-


ture-dependent
T
Λ ij = exp (A ij + B ij /T + C ij ln + D ij T) (2.48)
K
The parameters A, B, C, D can be adjusted to experimental binary phase equilibrium
data. The structure of Equation (2.48) often leads to misunderstandings. The first pa-
rameters to be adjusted are the B ij . To account for the temperature dependence, the
A ij parameters can be additionally fitted if corresponding data are available, i.e. phase
equilibrium data at significantly different temperatures or h E data. For better illustra-
tion, Equation (2.48) should be rewritten as
B ij + A ij T + C ij T ln KT + D ij T 2
Λ ij = exp . (2.49)
T
The C ij or D ij parameters are used only if it is justified by the data situation, which is
not often the case. It should be noted that the specific volumina introduced in Equa-
tion (2.45) do not occur in the final equations (2.47) and (2.48). The term A ij is equal
to
vj
A ij = ln , (2.50)
vi
44 | 2 Thermodynamic models in process simulation

so the ratio of the volumina just represents the term for the linear temperature depen-
dence. In process simulation, it would be awkward anyway if the binary parameters
would depend on the pure component properties which could be subject to change any
time. A more detailed derivation of the Wilson equation can be found in [11]. The NRTL
and UNIQUAC equations are more complicated [11], but for application it is sufficient
to understand their structure:
– Wilson:
ln γ i = f(x i , Λ ij ) , (2.51)
with Λ ij as explained above.
– NRTL:
ln γ i = f(x i , α ij , τ ij ) , (2.52)
with τ ij as a temperature function describing the molecular interactions
T
τ ij = exp (A ij + B ij /T + C ij ln + D ij T) , (2.53)
K
analogous to Λ ij in the Wilson equation and α ij as an additional symmetric (α ij =
α ji ) adjustable parameter for complicated concentration dependences. Usually,
α ij is set to 0.3, a reasonable range of values is α ij = 0.2–0.5. In extreme cases,
this range of values can be exceeded or α ij can be made temperature-dependent.
The coefficients A ij , B ij , C ij , D ij , and α ij are called binary interaction parameters
(BIPs). They play the key role in the thermodynamic model.
– UNIQUAC:
ln γ i = f(x i , τ ij ) , (2.54)
with τ ij analogous to the NRTL equation.

The remarkable issue is that it takes only binary parameters to describe a multicom-
ponent system. Therefore, the effort for the model development is at least limited.
Without this simplifying tool, process simulation would still not be possible up to
now. A famous example from Gmehling [34] indicates that it would take approx.
37 years to carry out the measurements necessary for a model capable to describe a
10-component-system.
There has been a lot of discussion whether the binary interaction parameters have
a physical meaning or not. This is a question which cannot be answered simply with
“yes” or “no”. The tendency of the author is to say “no”. The local composition models
can be mathematically derived [11, 35], where the binary parameters have the charac-
ter of defined energies for removing single molecules from a cell. However, this energy
is not measured; instead, the model with all its assumptions is adjusted to measured
phase equilibrium data. The results obtained are not unique; the parameters strongly
intercorrelate. It can be seen that various completely different parameter pairs are pos-
sible to obtain more or less the same results. This means that the value of a single
parameter cannot have any physical significance without the other parameter in the
2.3 γ-φ-approach | 45

pair. Therefore, a single parameter B ij has no significant physical meaning. However,


it can be shown mathematically that any equation describing the activity coefficients
must obey the Gibbs–Duhem equation [11]:

∑ x i d ln γ i = 0 (2.55)
i

The established equations like Wilson, NRTL, or UNIQUAC and several other do so;
thus, they are in some way physically justified. They should be taken as empirical ap-
proaches to fulfill the Gibbs–Duhem equation, meaning that they are able to represent
the correct shape of the course of the activity coefficient as a function of concentration.

0.6

0.55

0.5

0.45
p/bar

0.4

0.35

0.3

0.25

0.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
xR134a′ yR134a /mol/mol

Fig. 2.10: The azeotropic system R134a–R218 at T = 220 K [36].

An azeotrope occurs when the vapor pressure curves intersect.


(Completely wrong but often successful explanation for the occurrence of azeotropes.)

Activity coefficients can be used to obtain an understanding how azeotropes occur.


Figure 2.10 shows the well-defined azeotrope of the system R134a(1)–R218(2) at T =
220 K. Neglecting the influence of the fugacity coefficients in Equation (2.42), one can
add up Equation (2.41) for both components in the following form:

p = x1 γ1 p s1 + x2 γ2 p s2 . (2.56)

This is the equation of the boiling point line, i.e. the upper one in Figure 2.10. As
an azeotropic system it shows a clear maximum at the azeotropic point at approx.
x1 = y1 = 0.3. Both pure component vapor pressures are lower than the azeotropic
46 | 2 Thermodynamic models in process simulation

pressure, while R218 is the light end with the higher vapor pressure. On the right hand
side of the diagram, it is expected that the addition of R218 to R134a increases the pres-
sure, as R218 is the low-boiling component. It is more surprising that the addition of
the high-boiler R134a to the low-boiler R218 on the left hand side of the diagram gives
a pressure increase as well. This is a necessary condition, as starting from both pure
components the maximum in the azeotropic pressure must be reached. Evaluating
Equation (2.56) at the pure R218, one gets for x1 → 0:

p = x1 γ∞
1 p s1 + (1 − x 1 ) ⋅ 1 ⋅ p s2 > p s2 (2.57)

or
γ∞
1 p s1
> 1. (2.58)
p s2
Equation (2.58) is the azeotropic condition: If the product of vapor pressure and activ-
ity coefficient at infinite dilution of the high-boiler is larger than the vapor pressure of
the low-boiler, an azeotrope with a pressure maximum will result.
Similarly, for systems with negative deviations from Raoult’s law one gets the fol-
lowing: if the product of vapor pressure and activity coefficient at infinite dilution of
the low-boiler is smaller than the vapor pressure of the high-boiler, an azeotrope with
a pressure minimum will result.
Solely by the continuity of the boiling point line, the occurrence of an azeotrope
can be predicted just by slopes of the boiling point line caused by the addition of traces
to the pure components. However, the first step for checking whether an azeotrope
occurs should be a look into the database. There are even special monographs where
azeotropic data are collected in a systematic way [37].
A question often asked is which of the three equations, Wilson, NRTL, and UNI-
QUAC, is the best one. In fact, it is not much use to compare large amounts of fitting
results; no clear advantage can be observed. A description of the characteristics might
be more helpful.
The Wilson equation cannot describe systems with a miscibility gap due to math-
ematical reasons [11]. For a system without a miscibility gap the Wilson equation is an
adequate model. The NRTL equation can describe a miscibility gap. The third param-
eter α is a valuable tool if complicated systems have to be correlated. The comparison
with Wilson and UNIQUAC is not fair in these cases, as these equations do not have
this opportunity. Nevertheless, sometimes the third parameter is perceived to be use-
ful and makes NRTL the favorite choice. In a project, all binary subsystems must be
described with the same model; if NRTL gives a significant improvement for one of the
important subsystems, the decision for NRTL is probable. Furthermore, NRTL has a
very popular extension for electrolytes (ELECNRTL), which is fully consistent with the
simple NRTL equation for conventional systems. Thus, if electrolytes occur or even if it
is possible that they might occur, it is usually a good choice to take NRTL. UNIQUAC is
clearly the equation with the strongest physical background. For application, it is not
so popular as it requires the van-der-Waals surfaces and volumina as additional pure
2.3 γ-φ-approach | 47

component parameters, which cannot be assigned for all components due to missing
subgroups (e.g. CO2 ). An advantage of UNIQUAC is that it has a combinatorial part,
which takes into account the behavior of molecules differing in size.
In comparison with the φ-φ-approach, the γ-φ-approach has the disadvantage
that supercritical components cannot be treated with this concept. According to Equa-
tion (2.42), as there is no reference point in the liquid phase for a pure supercritical
component. As a workaround, a supercritical component can be treated as a Henry
component, with its reference state at infinite dilution in a pure solvent. The phase
equilibrium condition for such a component is

x i H ij = y i pφ i (T, p, y i ) , (2.59)

with H ij as the so-called Henry coefficient of Henry component i in solvent j. Equa-


tion (2.59) can be applied for low concentrations of the Henry component in the liquid
phase (rule of thumb: x i < 0.03), otherwise, further corrections are necessary, clearly
favoring the φ-φ-approach. The Henry coefficient has the character of a vapor pres-
sure; its meaning becomes clear when Equation (2.59) is applied to a subcritical com-
ponent at low pressure (i.e. φ i (T, p, y i ) = 1) at infinite dilution of the Henry component
in the liquid phase. Compared with Equation (2.41), the Henry coefficient is equal to

H i = p si γ∞
i . (2.60)

The Henry coefficient is a temperature function; the temperature dependence is usu-


ally described using a function like
Hi B
ln = A + + CT + DT 2 , (2.61)
p0 T
where p0 is an arbitrary pressure unit, necessary to make the argument of the loga-
rithm dimensionless.
It should be noted that in contrast to the vapor pressure the Henry coefficient is
not necessarily a function monotonically rising with temperature. It can exhibit well-
defined maxima (e.g. oxygen in water [11, 35]).
For mixed solvents, a mixing rule like
H ij
H i ∑j x j ln p0
ln = (2.62)
p0 ∑j x j

can be applied. In process simulators, often even more complicated mixing rules are
used. It should be mentioned that the averaging in the mixing rules is only performed
with the solvents where Henry coefficients are available. The index j refers only to the
solvent components where a Henry coefficient is given. One must be careful, as these
solvents are not always representative for the whole solvent.
It is often remarked that mixing rules like Equation (2.62) are arbitrary and empir-
ical. From the physical point of view, the application of equations of state, especially
48 | 2 Thermodynamic models in process simulation

with g E mixing rules (Chapter 2.7), is more justified. Nevertheless, one must realize
that in process calculations the whole gas solubility calculation has the character of
an estimation, giving only a reasonable order of magnitude. The solubility of gases in
liquids usually takes a lot of time to reach equilibrium. For experimental setups, sev-
eral hours are scheduled to get a data point; much more than a gas has in a process
step. Therefore, even relatively large errors in the Henry coefficient are not relevant for
the target of the calculation. For example, if the correct solubility of a gas component
is 100 ppm, an overestimation of 20 % of the Henry coefficient in Equation (2.59) would
yield a solubility of 83 ppm, which is a fully acceptable result in a process calculation.
In principle, excess enthalpies (enthalpies of mixing, see Glossary) can also be
calculated if a correlation for the activity coefficients is available. h E depends on the
temperature derivatives of the activity coefficients:

∂ ln γ i
h E = −RT 2 ∑ x i ( ). (2.63)
i
∂T

However, the physical background of the equations for the activity coefficient is not
sufficient so that application of Equation (2.63) usually gives wrong results unless the
parameters are temperature-dependent and have been fitted to both phase equilib-
rium and excess enthalpy data.
The great advantage of the γ-φ-approach is that all properties are described by
individual equations that have no influence on each other. The phase equilibrium is
described only by the activity coefficients, which do not need to be varied if any other
property is changed. Independent and highly accurate correlations are available for
each property.

2.3.2 Vapor pressure and liquid density

In contrast to the φ-φ-approach, in the γ-φ-approach the liquid density, the vapor
pressure and the specific enthalpy are determined independently with separate corre-
lations, giving the opportunity of yielding better accuracies and having a more conve-
nient workflow, as changes in one property do not affect the others. While the enthalpy
calculation needs an own chapter (Chapter 2.8), the correlations and estimations for
the liquid density and the vapor pressure can be briefly discussed here.
The density of a pure substance or a mixture is a fundamental quantity in any
process calculation. The vapor density as a function of temperature and pressure is
determined by the equation of state chosen in Equation (2.42). The liquid density of
pure components is treated only as a function of temperature, the pressure effect on
the density can usually be neglected. Appropriate correlations and their coefficients
are the Rackett equation
A
ρL = D
(2.64)
B1+(1−T/C)
2.3 γ-φ-approach | 49

and the PPDS⁶ equation


ρL ρc
= + A(1 − Tr )0.35 + B(1 − Tr )2/3 + C(1 − Tr ) + D(1 − Tr )4/3 , (2.65)
kg/m3 kg/m3

which is usually slightly more accurate.


It is important to mention that the liquid density mixing rule is based on the spe-
cific volume, i.e. the reciprocal value of the density:
1 xi
=∑ . (2.66)
ρL i
ρ L,i

Thermodynamics says that there is a so-called excess volume (see Glossary), i.e. a sys-
tematic deviation from Equation (2.66). If the liquid density is calculated using an
equation of state, as it is the case in the φ-φ-approach (Chapter 2.2), this excess vol-
ume is accounted for automatically. Nevertheless, no advantage can be taken, as it
is usually not described quantitatively correct, and in most cases the density of the
pure substances is badly reproduced so that even with a correction Equation (2.66)
yields better results (p. 38). The maximum error caused by neglecting the excess vol-
ume can be quantified regarding the system exhibiting the largest excess volume: To
the knowledge of the author, it is in fact ethanol-water, and the maximum excess vol-
ume observed is approx. 3.5 % [11].
Liquid densities can be estimated e.g. with the COSTALD equation [38]. It can in
principle be written as
1
= v∗ ⋅ f (T, Tc , ω) , (2.67)
ρL
meaning that the liquid density can be estimated with a generalized function depend-
ing only on Tc and ω. Additionally, the parameter v∗ is used, which can be adjusted
to one or more data points. In this case, the procedure is very accurate; up to tem-
peratures below 0.95 Tc the error is usually below 2 %. If no data point is available,
one can use the critical volume for v∗ , which of course increases the risk but often
yields surprisingly good results. It should be mentioned that the COSTALD equation
has weaknesses if polar components are involved.
The vapor pressure is the most important quantity in thermodynamics. It is de-
cisive especially in the simulation of distillation columns. Furthermore, it is directly
related to the enthalpy of vaporization via the Clausius–Clapeyron equation (Chap-
ter 2.8). The vapor pressure is an exponential function of temperature, starting at the
triple point and ending at the critical point. It comprises several orders of magnitude,
therefore, a graphical representation usually represents only part of its characteris-
tics. Figure 2.11 shows two diagrams, with linear and logarithmic axes for the vapor
pressure of propylene. The linear diagram makes it impossible to identify even the

6 physical property data service


50 | 2 Thermodynamic models in process simulation

50 5
45
40 0

ln (Ps/bar)
35
30 –5
Ps/bar

25
20 –10
15
10 –15
5
0 –20
80 130 180 230 280 330 80 130 180 230 280 330
(a) T/K (b) T/K

Fig. 2.11: Typical vapor pressure plots as a function of temperature [11].


© Wiley-VCH Verlag GmbH & Co. KGaA. Reproduced with permission.

1
0.8
0.6
0.4
Δps/ps/%

0.2
0
–0.2
–0.4
–0.6
–0.8
150 200 250 300 350
T/K

Fig. 2.12: Deviation plot for the fit of a vapor pressure equation [11].
© Wiley-VCH Verlag GmbH & Co. KGaA. Reproduced with permission.

qualitative behavior at low temperatures, whereas in the logarithmic diagram only


the order of magnitude can be identified on the axis. At least, logarithmic diagrams
allow the comparison between vapor pressures of different substances, e.g. deciding
whether they intersect or not. On the other hand, many process calculations require
extremely accurate vapor pressure curves, e.g. the separation of isomers by distilla-
tion. For the visualization of a fit of a vapor pressure curve, special techniques must
be applied, the simple graphical comparison between experimental and calculated
data in a diagram is not possible. For this purpose, a deviation plot is useful (Fig-
ure 2.12).
Most companies use parameter databanks. When a parameter database is set up,
e.g. the one in a commercial process simulation program, it cannot be known in ad-
vance at which conditions exact vapor pressures will be required. For distillation ap-
plications, a high accuracy is often required, especially when components with similar
vapor pressures have to be separated in distillation columns. Simple vapor pressure
equations cannot be applied for the whole temperature range from the triple point to
the critical point. More capable vapor pressure correlations are needed, the most pop-
2.3 γ-φ-approach | 51

ular ones are the extended Antoine equation and the Wagner equation. In principle,
the extended Antoine equation is a collection of various useful terms:

ps B T T T G
ln =A+ + D + E ln + F ( ) , (2.68)
p0 T/K + C K K K

where p0 is an arbitrary reference pressure, usually the pressure unit. A very useful
correlation is the Wagner equation
ps 1
ln = [A(1 − Tr ) + B(1 − Tr )1.5 + C(1 − Tr )3 + D(1 − Tr )6 ] , (2.69)
pc Tr

where Tr = T/Tc (reduced temperature). It can correlate the whole vapor pressure
curve from the triple point to the critical point with an excellent accuracy. The Wag-
ner equation has been developed by identifying the most important terms in Equa-
tion (2.69) with a structural optimization method [39]. The extraordinary capability of
the equation has also been demonstrated by Moller [40], who showed that the Wagner
equation can reproduce the difficult term ∆h v /(z V − z L ) reasonably well. Eq. (2.69) is
called the 3-6-form, where the numbers refer to the exponents of the last two terms.
Some authors [41] prefer the 2.5-5-form, which is reported to be slightly more accu-
rate. For the application of the Wagner equation, accurate critical data are required.
As long as the experimental data points involved are far away from the critical point
(e.g. only points below atmospheric pressure), estimated critical data are usually suffi-
cient. As the critical point is automatically met due to the structure of the equation, the
Wagner equation extrapolates reasonably to higher temperatures, even if the critical
point is only estimated. However, like all vapor pressure equations it does not extrapo-
late reliably to lower temperatures. Sometimes, users of process simulation programs
calculate vapor pressures beyond the critical point, although it is physically meaning-
less. If the Wagner equation is applied above the critical temperature, it will yield a
mathematical error. Therefore, the simulation program must provide an extrapolation
function that continues the vapor pressure line with the same slope. For the particular
parameters of the Wagner equation, the following ranges of values are reasonable for
both forms:
A = −9 . . . −5
B = −10 . . . 10
(2.70)
C = −10 . . . 10
D = −20 . . . 20 .
If these ranges are exceeded, one should carefully check the critical data used and
the experimental data points for possible outliers. Coefficients for the Wagner equa-
tion can be found e.g. in [41–43]. For a vapor pressure correlation, average deviations
should be well below 0.5 %. Data points with correlation deviations larger than 1 %
should be rejected, as long as there are enough other values available. Exceptions can
be made for vapor pressures below 1 mbar, as the accuracy of the measurements is
52 | 2 Thermodynamic models in process simulation

lower in that range. The structure of the deviations should always be carefully inter-
preted. A guideline is given in [11].
Despite this high accuracy demand for vapor pressures, there is also a need for
good estimation methods. Often, a lot of components is involved in a distillation
process. Not all of these components are really important, however, one should know
whether they end up at the top or at the bottom of a distillation column. In many
cases, a measurement would even be not possible, as the effort for the isolation and
purification of these components might be too large. Estimation methods are mostly
applied to medium and low pressures for molecules with a certain complexity, as
small molecules usually have well-established vapor pressure equations, whereas
large molecules often have a volatility which is so low that a purification of the
substance for measurement by distillation is not possible. The estimation of vapor
pressures is one of most difficult problems in thermodynamics. Due to the exponen-
tial relationship between vapor pressure and temperature, a high accuracy must not
be expected. Deviations in the range of 5–10 % have to be tolerated. Thus, estimated
vapor pressure correlations should not be used for a main substance in a distillation
column to evaluate the final design, however, they can be very useful to decide about
the behavior of side components without additional measurements.
Different estimation methods are discussed in [11]. At least, one data point, usu-
ally the normal boiling point, should be known. Most methods use a vapor pressure
equation with two adjustable parameters; therefore, a second piece of information is
necessary. It can be generated by a group contribution method (e.g. Rarey method
[40]) or by a second data point, which can be either a genuine data point or a point
based on an estimation method itself, e.g. the critical point. An example for the latter
case is the application of the Hoffmann–Florin equation [44]

ps 1 T T
ln = α + β[ − 7.9151 ⋅ 10−3 + 2.6726 ⋅ 10−3 lg − 0.8625 ⋅ 10−6 ] , (2.71)
p0 T/K K K

with the adjustable parameters α and β. It has the advantage that it can easily be trans-
formed to the extended Antoine equation (2.68) by

A = α − 7.9151 ⋅ 10−3 β
B=β
D = −0.8625 ⋅ 10−6 β
(2.72)
2.6726 ⋅ 10−3
E= β
ln 10
C = F = G = 0.

If no single data point is available, one must estimate even the normal boiling point
[11, 45]. In this case, one can hardly rely on the results obtained, however, as long as
no better information is available, there is no choice. When a vapor pressure has to be
estimated, one should also have a look at vapor pressure curves of components which
2.3 γ-φ-approach | 53

have a similar structure. Just by defining a constant vapor pressure ratio between the
two components, one can at least obtain a reasonable vapor pressure curve.
Vapor pressures play the most decisive role in distillation if isomers have to be
separated by distillation. In this case, the separation factor depends only on the ra-
tio between the vapor pressures as the activity coefficients between isomers can be
set to 1 as an approximation⁷. If the vapor pressures of the isomers are close, a large
number of theoretical stages are necessary, and its determination is very sensitive to
the separation factor. In these cases, it is strongly recommended not to rely on data
from the literature, not even on good data. Instead, the vapor pressure of the isomers
should be measured as accurately as possible in the same apparatus and on the same
day to avoid any systematic measurement errors.

2.3.3 Association

Another advantage of the γ-φ-approach is that substances showing association in the


vapor phase can be described. These substances are the carboxylic acids like formic
acid, acetic acid⁸ or propionic acid, which form dimers in the vapor phase, and hydro-
gen fluoride, which forms hexamers. These substances are involved in many chemical
processes, and the deviations from the ideal gas law are significant even at low pres-
sures. For example, the compressibility factor of acetic acid at the normal boiling
point, which is expected to be close to 1, is ZNBP = 0.6. Up to now, no equation of
state valid for both the vapor and the liquid phase has been available in this case. The
corresponding equation of state does not need to be valid for both the vapor and the
liquid phase; with the γ-φ-approach, it is sufficient to cover only the vapor phase. The
formation of associates is treated as a chemical reaction in equilibrium. As an illustra-
tion, formic acid as a substance forming only dimers is regarded. The association can
be described with the law of mass actions:
z2
K2 = , (2.73)
z1 (p/p0 )
2

where K is the equilibrium constant for the reaction

2 HCOOH 󴀕󴀬 (HCOOH)2 .

z1 is the true concentration of the monomer, while z2 denotes the true concentration
of the dimers in the mixture. p0 is simply the pressure unit. The equilibrium constant
can be correlated by
B2
ln K2 = A2 + . (2.74)
T

7 which is, however, not always correct.


8 For acetic acid, an additional tetramer formation is often considered to obtain more accurate results.
54 | 2 Thermodynamic models in process simulation

The sum of true mole fractions z is equal to one

z2 + z1 = 1 (2.75)

Combining Equations (2.73) and (2.75), z1 can be determined by

√1 + 4K2 (p/p0 ) − 1
z1 = . (2.76)
2K2 (p/p0 )
Assuming that the ideal gas equation is valid, the specific volume can be determined
by
RT 1
v= . (2.77)
p z1 + 2z2
Figure 2.13 illustrates the difference to conventional equations of state like the cubic
ones. Case (a) represents the situation in an ideal gas phase; there are no forces
between the molecules. Case (b) represents a real vapor phase, where the molecules
attract or repulse each other by intermolecular forces. Vapor phases like this are
typically modeled with the equations of state described in Chapter 2.2 like the cubic
equations. Case (c) denotes the association in the vapor phase. The model (Equa-
tions (2.73)–(2.77)) takes into account that associates are formed, but it is still assumed
that no intermolecular forces are exerted. The nonideality in case (c) is only achieved
by the formation of associates. This approximation is sufficient for low pressures.
For higher pressures, a good model for substances showing vapor phase association
would have to take the intermolecular forces into account as well (case (d)). However,
an appropriate model for this situation has not yet been introduced.
For associating substances, the heat capacity of the vapor phase shows a well-
defined maximum (Figure 2.14). At low temperatures, the dimerization as the exother-
mic reaction is preferred in the equilibrium, and all molecules are dimerized. When

(a) (b)

(c) (d)

Fig. 2.13: Illustration of the modeling of vapor phases.


2.4 Electrolytes | 55

6 p = 1.013 bar

5
cp/J/gK

4
p = 6 bar
3

2 p = 0.332 bar
1
ideal, p = 0
0 Fig. 2.14: Specific isobaric heat capacity
350 450 550 650 750 of acetic acid vapor at different pressures [11].
T/K © Wiley-VCH Verlag GmbH & Co. KGaA.
–0.332 bar –1.013 bar Reproduced with permission.

the temperature rises, the dimers are split. For this endothermic reaction, energy is
required which is not used for increasing the temperature. Therefore, the heat capac-
ity increases drastically. With increasing temperature, the number of dimers which
can be split decreases. The heat capacity passes a maximum and comes down to the
normal value of the ideal gas. For the sizing of heat exchangers, this effect must be con-
sidered. The curvature is even more dramatic for HF with its hexamers, where the peak
of the maximum can be up to 40 times higher than the ideal gas heat capacity [46].

2.4 Electrolytes

Water is one of the strangest substances that occurs in chemical engineering. Its phys-
ical properties are not comparable to the behavior of other substances. With a low
molecular weight of 18 g/mol, its normal boiling point of tb = 100 °C is incredibly
high, as well as its critical point (tc = 374 °C, pc = 221 bar). The well-known maxi-
mum in the density of liquid water at t = 4 °C is technically not important, but the
course of the liquid density as a function of temperature is extraordinarily flat. Be-
tween 0 °C and 50 °C, the density decreases just by 1 %. For comparison, n-heptane
with a similar normal boiling point (98.4 °C) has a density decrease of more than 4 %
in the same temperature range. When water crystallizes, it expands significantly by
almost 10 %, whereas other substances reduce their density as expected. The specific
heat capacity of liquid water (approx. 4.2 J/(g K)) is about twice as large as for typical
organic substances, but the most remarkable physical property is the enthalpy of va-
porization. At t = 100 °C, it is approx. 2250 J/(g K), more than seven times larger than
for n-heptane as a typical organic substance.
The reason for this behavior is the strong polar character of the molecule. It is not
linear; the two bonds between oxygen and hydrogen form an angle of approx. 105°.
56 | 2 Thermodynamic models in process simulation

The oxygen atom is a strongly electronegative, it attracts the electrons of the bond
with their negative charges so that they concentrate at the oxygen atom, while the
hydrogen atoms become positively charged. Therefore, the water molecules arrange
in a kind of lattice.
Moreover, the strong polarity of the water molecule is the reason why it is an ex-
cellent solvent for many electrolytes. An electrolyte is a substance which conducts
electric current as a result of its dissociation into positively and negatively charged
ions in solutions or melts. Ions with a positive charge are called cations, ions with
negative charge anions, respectively. The most typical electrolytes are acids, bases,
and salts dissolved in a solvent, very often in water.
The total charge of an ion is a multiple of the elementary charge (e = 1.602 ⋅
10 −19 C), given by the number z. Examples are:

H3 O+ z=1

Cl z−1
Ca 2+
z=2
SO2−
4 z = −2 .

In a macroscopic solution, the sum of charges is always zero, since the solution is
always neutral. Otherwise, an electric current would occur.
For electrolyte solutions, the particular ions are formed by dissociation reactions
like
NaCl → Na+ + Cl−
CaSO4 → Ca2+ + SO2−
4
H3 PO4 + H2 O → H3 O+ + H2 PO−4
H2 PO−4 + H2 O → H3 O+ + HPO2−
4
+
4 + H2 O → H3 O + PO4 .
3−
HPO2−
The H+ ion cannot exist as a pure proton; it is always attached to a water molecule
H2 O, giving H3 O+ . Strong and weak electrolytes can be distinguished. While strong
electrolytes like HCl, H2 SO4 , HNO3 , NaCl, or NaOH dissociate almost completely, weak
electrolytes do so only to a small extent. Sometimes, their electrolyte character plays a
secondary role and can often be neglected. Examples are formic acid (HCOOH), acetic
acid (CH3 COOH), HF, H2 S, SO2 , NH3 , or CO2 [11].
The molecular structure of an electrolyte solution is significantly determined by
the electrostatic interactions between the charged ions (Coulomb-Coulomb interac-
tions) and by the long-range interactions of the charged ions and the dipole moments
of the solvent (Coulomb-dipole-interactions). Figure 2.15 illustrates the schematic dis-
tribution of water as a strongly polar solvent around a cation and an anion.
As mentioned, the oxygen atom in the water molecule has a negative partial
charge due to its high electronegativity. Therefore, the water molecule in the vicinity
of an ion is arranged in a way that the oxygen atom is directed towards the positively
2.4 Electrolytes | 57

(–)
(+) (+) (+)
(+)
(–)

(+) (–) (+) (+)


+ (–) (+) –
(–) (–)
(+) (+) (+)
(+)
(+)
(–) (+)

(+) (–)
(+)

Fig. 2.15: Structure of an aqueous electrolyte solution [11].


© Wiley-VCH Verlag GmbH & Co. KGaA. Reproduced with permission.

charged cations. Vice versa, the hydrogen atoms in the water molecule are partially
positively charged and oriented towards the anions. Around the ions, a shell of solvent
molecules is formed.
The corresponding procedure is called solvation. It is an exothermic process. On
the other hand, the dissociation of the electrolyte is an endothermic process, because
the ionic lattice has to be destroyed, which is connected to a need of energy. The
overall heat of solution is the sum of both contributions. It is usually dominated by
the solvation and therefore remains exothermic. However, there are many exceptions.
Electrolytes in their dissociated form are not volatile and remain completely in the liq-
uid phase. However, it can happen that liquid drops containing electrolytes are subject
to entrainment (Chapter 9).
The process simulation programs offer models for electrolyte solutions which de-
scribe the arrangement above. Not necessarily the best, but the most widely used one
is the electrolyte NRTL model, often also called Chen model [47, 48]. Its obvious ad-
vantage is the compatibility with the conventional NRTL activity coefficient model.
This is the most important property of an electrolyte model, often even more impor-
tant than its accuracy. During a project, process simulation might start in a part of
the plant where electrolytes are not involved. Later, when it is extended, electrolyte
components occur, and defining these components simply as heavy ends is always
tried first⁹. Although of course not state of the art, this is often a feasible approach
as long as the electrolyte concentration is low and not a decisive issue in the process.

9 This can simply be accomplished by assigning the same properties to the component as water, and
then overwriting the molecular weight with the correct value for maintaining the mass balance and,
also, overwriting the vapor pressure with an equation giving negligible values over the whole concen-
tration range.
58 | 2 Thermodynamic models in process simulation

When it finally turns out that electrolytes occur in a way that their character must be
correctly described, it is useful if at least the BIP matrix does not need to be revised.
The NRTL electrolyte model consists of the theoretically derived Debye–Hückel
term for the long ranging interactions due to the charges of the ions and an NRTL
based term for the short ranging interactions. Still, the local composition concept is
applied. The model is not restricted to systems of electrolytes just with water, other
solvents and side components are possible. However, the database is mainly based on
aqueous systems. Different kinds of parameters occur:
– the normal binary parameters between molecules;
– pair parameters between ion pairs and molecules;
– pair parameters between different ion pairs.

Nowadays, the possible electrolyte reactions and the species produced are generated
by the process simulation program. It is up to the user to neglect them or take them
into account. For instance, the dissociation reaction of ammonia

NH3 + H2 O → NH+4 + OH−

can easily been neglected if just the system NH3 /H2 O is regarded, as only a small
fraction of the ammonia dissociates, which is negligible. In a caustic environment,
even this hardly ever happens. However, in an acid environment, the ammonia will
dissociate, and in fact completely at low pH values. A famous example is the system
NH3 /CO2 /H2 O, where both NH3 and CO2 are weak electrolytes, but they keep the other
component in the solution as NH3 is a caustic and CO2 is an acid component.
In the process simulation program, the equilibrium constants for the generated
dissociation reactions are usually provided automatically, as well as the pair parame-
ters. Note that the pair parameters really refer to a pair of ions and not to single ones.
An example for the application of the NRTL electrolyte model is given in [11].
There are two options for the notation of the electrolytes. In the true component
approach, the ions are listed as ions. The advantage is that it is listed what really hap-
pens, the disadvantage might be that it is more difficult to keep the overview. In the
apparent component approach, the ions are recombined to components for the result
list. This is easier for discussion but does not always work in a plausible way. It can
happen that according to the balance components like NaOH and HCl coexist in the
aqueous solution, which one can hardly imagine.

2.5 Liquid-liquid equilibria

With increasing activity coefficients, two liquid phases with different compositions are
formed (miscibility gap). The concentration differences of the compounds in the dif-
ferent phases can be used for example for the separation by extraction. In distillation
processes, the liquid-liquid equilibrium (LLE) is often used when a decanter separates
2.5 Liquid-liquid equilibria | 59

the condensate of the top product into two liquid phases. The knowledge of the VLLE
(vapor-liquid-liquid equilibrium) behavior is of special importance for the separation
of systems by heteroazeotropic distillation. Many engineers are of the opinion that
the formation of an LLE does not take place in distillation columns. In fact, it does,
however, in contrast to a phase equilibrium arrangement the two phases do not sep-
arate due to turbulences (tray columns) or form thin layers which trickle down the
internals of the column (packed columns). In both cases, it is useful to treat the two
liquid phases as one homogeneous liquid for the determination of the overall liquid
composition and the physical properties of the liquid phase. For the phase equilib-
rium calculations, there is no other way than to consider the liquid phase split to get
the correct vapor composition. Liquid-liquid equilibria can be evaluated by the iso-
activity criterion:
(x i γ i )I = (x i γ i )II , (2.78)
as the product x i γ i is also called activity ai .
At moderate pressures, the liquid-liquid equilibrium behavior as a function of
temperature only depends on the temperature dependence of the activity coefficients.
For the calculation of LLE again g E -models like NRTL, UNIQUAC, or equations of state
with a g E mixing rule (Chapter 2.7) can be used, whereas Wilson is not appropriate [11].
The formation of two liquid phases can result in the formation of binary and higher
heteroazeotropes, which can for instance be used for the separation of systems like
ethyl acetate–water (Figure 2.18).
Ternary LLEs can be illustrated in a triangle diagram (Figure 2.16). It is more dif-
ficult to calculate reliable liquid-liquid equilibria (LLE) of a system containing three
or more components using binary parameters than to describe vapor-liquid or solid-
liquid equilibria (Chapter 2.6). The reason is that in the case of LLE the activity coeffi-
cients have to describe the whole concentration dependence correctly, whereas in the
case of the other phase equilibria (VLE, SLE) the activity coefficients primarily have to
account for the deviation from ideal behavior (Raoult’s law resp. ideal solid solubility).
This is the main reason why up to now no reliable prediction of a multicomponent
LLE behavior (tie lines) using binary parameters is possible. Fortunately, it is quite
easy to measure LLE data of ternary and higher systems at least up to atmospheric
pressure. Binary parameters can be fitted to ternary data as well, and in this way LLEs
can at least be correlated.
Also, even the fit to a binary mixture is often a bit more exhausting, as priorities
must be set. It hardly ever happens that a set of binary parameters can represent both
the vapor-liquid equilibrium of systems with an LLE (the so-called vapor-liquid-liquid
equilibrium VLLE) and the miscibility gap itself. It must be decided which capabili-
ties the parameters should have. Often, two different parameter sets are used, one for
LLE and one for VLLE. In most cases, the LLE data are more significant, especially for
systems with wide miscibility gaps.
60 | 2 Thermodynamic models in process simulation

Ternary diag for WATER/ETHANOL/ETAC

0.05 0.95
0.10 0.90
0.15 0.85
0.20 0.80
0.25 0.75
0.30 0.70
0.35 0.65
ate

0.40 0.60

Mo
cet

0.45 0.55

le f
yl a

rac
0.50
eth

0.50

eth
rac

0.55 0.45

an
le f

ol
0.60 0.40
Mo

0.65 0.35
0.70 0.30
0.75 0.25
0.80 0.20
0.85 0.15
0.90 0.10
0.95 0.05

0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.95
Mole frac water

Fig. 2.16: Liquid-liquid equilibrium of the ternary system ethyl acetate–water–ethanol at t = 50 °C.

Moreover, the binary parameters for LLEs cannot be simply transferred from binary to
ternary and multicomponent mixtures. Doing so, one obtains more or less an estima-
tion. One should be aware that for a good description of multicomponent LLE at least
data from ternary mixtures are necessary.

Often, when a colleague tells me that his simulation results look strange, the first thing I do is
check whether the VLLE option is actually used for systems with a miscibility gap. Forgetting it is
a very common error in the simulation of vapor-liquid separations. Simulation results may look
strange in many ways. For this case, the trigger is that the temperature is way too low, not just by
a few degrees but by 100 K or so.
(Jo Sijben)

Calculating a phase equilibrium, there are two options in a process simulator. One is
the common VLE calculation. If it is known that miscibility gaps occur in the system,
the option VLLE (3-phase equilibrium) must be chosen so that the simulator checks
2.6 Solid-liquid equilibria | 61

whether there is an LLE before the equilibrium with the vapor phase is evaluated. If
this is forgotten, one gets strange phase equilibrium diagrams (Figure 2.17). The boil-
ing and the dew point curve then appear to be complete nonsense. The correct and
reasonable result is obtained with the 3-phase flash (Figure 2.18). In a flash in a process
simulation, the error is not so easy to detect, but whenever the result is not plausible,
it should be checked whether the VLLE option is necessary and chosen. One could
easily say that VLLE should always be chosen, but if it turns out to be not necessary, a
lot of calculation time has been wasted, as the VLLE option is quite time-consuming
in comparison with a simple VLE.
Generally, it can be said that systems with water and nonpolar organic substances
have strong intermolecular interactions and often form miscibility gaps. Therefore, at
least all the BIPs with water should be assigned in any project (Chapter 2.9).

3
2.5
2
p / bar

1.5
1
t = 80°C
0.5
0 Fig. 2.17: Calculation of the system
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
ethyl acetate–water at t = 80 °C with
x, γETAC / mol/mol the 2-phase flash.

1.6
1.4
1.2
1
p/bar

0.8
0.6
0.4 t = 80°C
0.2
0 Fig. 2.18: Calculation of the system
0 0.2 0.4 0.6 0.8 1
ethyl acetate–water at t = 80 °C with
x, γETAC/mol/mol the 3-phase flash.

2.6 Solid-liquid equilibria

Solid-liquid equilibria (SLE) are used for the synthesis and design of crystallization
processes, but taking them into account is also important to avoid undesired solids for-
mation. Information about solid-liquid equilibria can also be used for the adjustment
of binary parameters. SLEs are more complicated than VLEs or LLEs. Different types of
SLEs have to be distinguished, depending on the mutual solubility of the components
in the solid and in the liquid phase. However, the most important one, the simple
62 | 2 Thermodynamic models in process simulation

eutectic system, is comparably easy, and it is the only one which does not require a
specialist.
The eutectic system is characterized by the total immiscibility of the components
in the solid phase. This is advantageous for crystallization, as the crystallized phase
has a high purity. One theoretical stage is sufficient to obtain the pure compounds [11].
Usually, there are liquid mechanical inclusions so that the crystallization must at least
be repeated once, but from thermodynamics alone a pure solid phase is generated.
Fortunately, about 80 % of the systems behave in this way. Figure 2.19 shows the solid-
liquid equilibrium of the eutectic system benzene–naphthalene. Both solid phases
crystallize in pure form. Generally, solids are formed at a low temperature. Consider a
mixture with x1 = 0.5. When it is cooled, it reaches the liquidus line at T ≈ 320 K. The
first naphthalene crystals are formed. Cooling down further, the amount of solids in-
creases according to the lever rule, while the concentration of the liquid phase moves
towards the eutectic one. At 300 K, it is approx. x1 = 0.7. When the eutectic tempera-
ture is reached, both components crystallize forming pure solid phases. The eutectic
temperature is lower than the melting points of the participating pure components.
Thermodynamics does not give information about the shape of the crystals and the
amount of liquid included. It is the art of crystallization to form the crystals in the
desired way (Chapter 7.3).

360

340

320
T/K

300

280
Fig. 2.19: Solid-liquid equilibrium of the
260 eutectic system benzene–naphthalene [11].
0 0.5 1
© Wiley-VCH Verlag GmbH & Co. KGaA.
x1 / mol/mol Reproduced with permission.

For eutectic systems, the equilibrium condition can be written as [11]


∆h m,i T cLp,i − c Sp,i T m,i − T T m,i
ln (x i γ i ) = − (1 − )+ ( − ln ). (2.79)
RT T m,i R T T
The heat capacities of the solids are not often known, and, fortunately, the difference
in the corresponding term has a tendency to cancel out. Therefore, Equation (2.79) is
usually simplified to
∆h m,i T
ln (x i γ i ) = − (1 − ). (2.80)
RT T m,i
2.7 φ-φ-approach with g E mixing rules | 63

For an evaluation of Equation (2.80), only the activity coefficient and melting tempera-
ture and enthalpy of fusion as pure component properties must be known. Solid-liquid
equilibria are not very sensitive to pressure but much more to temperature. As the ac-
tivity coefficient γ i of the component i is strongly concentration-dependent, the mole
fraction x i in the liquid phase must be evaluated iteratively.
In process simulation programs, special solid components can be introduced
which do not take part in other phase equilibria. For crystallization, reaction blocks
can be introduced where the liquid component is transformed into the solid or vice
versa. The “reaction equilibrium” is calculated according to Equation (2.79) or (2.80),
respectively.

2.7 φ-φ-approach with g E mixing rules

Nowadays, in process simulation in the chemical industry the requirement for accu-
racy and reliability on the one hand and the occurrence of polar components on the
other hand make clear that advanced cubic equations of state with g E mixing rule
and individual α-functions should be preferred to the generalized equations of state
like Peng–Robinson or Soave–Redlich–Kwong with the quadratic mixing rule (Equa-
tion (2.37)). The generalized ones might still be tolerable in hydrocarbon processes;
however, the advanced cubic equations of state have no drawback there. The g E mix-
ing rule adopts the concept of the activity coefficients for use in the mixing rule for
the parameter a. Examples are the Predictive Soave–Redlich–Kwong equation (PSRK)
and the Volume-Translated Peng–Robinson equation (VTPR). For PSRK, the g E mixing
rule is
a a ii 1 gE b
= ∑ xi − ( 0 + ∑ x i ln ) , (2.81)
bRT i
b i RT 0.64663 RT i
bi

where g0E denotes the Gibbs excess energy at p = 0, i.e. at low pressure:

g E = RT ∑ x i ln γ i . (2.82)
i

To calculate the γ i , any appropriate equation like Wilson, NRTL, or UNIQUAC or a pre-
dictive approach (Chapter 2.10) can be used.
With the g E mixing rules, the general inability of the φ-φ-approach to describe
mixtures with polar components can be overcome.
There has been a great deal of discussion on whether the φ-φ-approach Equa-
tion (2.15) or the γ-φ-approach Equation (2.42) is the more favorable option for phase
equilibrium calculation, often in a more or less ideological way. The following items
point out the particular pros and cons:
– If supercritical components dissolve in the liquid phase to a considerable amount,
it is nowadays obligatory to use the φ-φ-approach, preferably with a g E -mixing
rule.
64 | 2 Thermodynamic models in process simulation

– Theoretically, the γ-φ-approach is valid in the whole subcritical region. However,


in the region close to the critical temperature of a component (approx. T > Tc −
15 K) the φ-φ-approach is more reliable.
– As long as supercritical components dissolve in the liquid phase to a minor ex-
tent, (e.g. nitrogen in water), there is no real disadvantage if Henry’s law with
the corresponding mixing rule is used instead of the φ-φ-approach. Although the
mixing rule Equation (2.62) is more or less arbitrary, it leads to the correct order
of magnitude of the gas solubility. As in processes these equilibria are usually not
reached, this information from process simulation is fully sufficient.
– If associating components are involved, the γ-φ-approach is obligatory. So far,
there is no established equation of state valid for both vapor and liquid phase for
associating components.
– One must be aware that with the φ-φ-approach all thermodynamic properties
(ρ L , p s , ∆h v , cLp ) are calculated just by the equation of state. This means that any
extension of the database for one property affects all other properties, giving a lot
of work. Furthermore, due to the limited number of parameters used, the accuracy
of the particular quantities is often not sufficient. An exception are, of course, the
high-precision equations of state for pure components. There is no option for set-
ting priorities in the enthalpy calculations (Chapter 2.8). With the γ-φ-approach,
all properties can be correlated separately and with the required accuracy, as long
as enough data are available.
– An often cited prejudice is that with the φ-φ-approach only data for a correlation
for cidp is required. In fact, this statement is not well founded. From the formal
point of view, all the thermodynamic properties as listed above can be calculated
without further data. However, in this case one has no information about the ac-
curacy of the equation of state. For process calculations, it is necessary to compare
the values obtained with experimental data and then probably adjust the param-
eters involved; a procedure with the disadvantage mentioned above that the pa-
rameters influence all quantities. Finally, a responsible use of the φ-φ-approach
requires presumably more preparation work than the γ-φ-approach.
– An argument against the γ-φ-approach is that it does not include the pressure
dependence of the activity coefficient. At high pressures, even small values of the
excess volume could have an influence on the results [49]. There are several ap-
proaches to encounter this argument. First, the fitting of data at the corresponding
temperatures and pressures should achieve some kind of error compensation so
that the data are still represented [49]. Second, a further correction taking the
excess volume into account could be included in the phase equilibrium condi-
tion (2.42) [50], which is, however, not considered in the process simulators. And
finally, it is clear that the φ-φ-approach takes the excess volume into account in
a formally correct way; but there is hardly any evidence that the excess volume is
represented correctly.
2.8 Enthalpy calculations | 65

– For the adjustment of the parameters, there is often a disagreement between the
vapor pressure equation used, which is usually well-founded, and the pure com-
ponent vapor pressure given as a data point in a binary vapor-liquid equilibrium
data set. It is a common and successful practice to replace the vapor pressure from
the correlation by the pure component vapor pressure given in the particular data
set just for the parameter fitting procedure to avoid inconsistencies with the rest
of the data and to get more reliable values for the γ∞
i -values (vapor pressure shift-
ing). In the process simulation, the vapor pressure correlation is then used again
together with the adjusted binary parameters [11]. In the γ-φ-formalism, this is an
easy change, whereas in the φ-φ-approach the individual α-function would have
to be manipulated, which has also an influence on the other quantities.
– It is a considerable disadvantage for the project administration that the g E mixing
rules use the equations for the determination of the activity coefficients in a differ-
ent context. The same parameters do not yield the same activity coefficients. This
leads to confusion when both approaches are used in a project simultaneously.

2.8 Enthalpy calculations

For the evaluation of heating and cooling duties in a process, a correct description of
the specific enthalpies is decisive. As all components are more or less present in both
the liquid and the vapor phase, the difficulty is that a continuous enthalpy description
in both phases is necessary. The following quantities can contribute to the enthalpy:
f
– standard enthalpy of formation ∆h0 at t = 25 °C in the ideal gas state, used as
reference point for the specific enthalpy;
– ideal gas heat capacity cid p;
– enthalpy of vaporization ∆h v ;
– liquid heat capacity cLp ;
– enthalpy pressure correction of the vapor phase (h − hid );
– excess enthalpy h E .

First, it should be remembered that the absolute value of the enthalpy is normally
meaningless, only differences between specific enthalpies can be interpreted. A sin-
gle value for the enthalpy is only useful if a reference point is given. With an arbitrary
choice of the reference point (e.g. h = 0 for t = 0 °C), the calculation of chemical reac-
tions is awkward; it makes only sense if only pure components are involved. In process
simulation programs, the use of the standard enthalpy of formation as reference point
for the enthalpy makes sure that the enthalpies of reaction can be correctly calculated.
This is further explained in Chapter 10.1.
Only the standard enthalpies of formation and the ideal gas heat capacities are
explicitly necessary if equations of state are used, whereas the deviations from the
ideal gas can be calculated directly. Equations of state with generalized α-functions
66 | 2 Thermodynamic models in process simulation

are usually not accurate enough, therefore, individual α-functions of advanced cubic
equations of state with component-specific parameters have to be fitted to ∆h v , cLp , and
(h − hid ) to ensure that the equation of state works. It turned out that the adjustment
of cLp is sufficient to obtain good results [22].
For the γ-φ-approach, there are more options available, as the particular quan-
tities are not independent of each other. For a pure substance, knowing three of the
four quantities cid p , ∆h v , c p , (h − h ), the fourth one can be evaluated. Unfortunately,
L id

it turns out that this does not work for cid L


p and c p . The reason is that the slope of ∆h v
is not accurate enough if only values of this quantity are adjusted [51]. There are two
ways for the description of the enthalpy in process simulators, illustrated using hT di-
agrams (Figures 2.20, 2.21). For a pure substance, they show the bubble point line, the
dew point line and the ideal gas enthalpy at p = 0 for guidance.
(A) Vapor as the starting phase:
With this route, the cLp is the quantity which is determined indirectly (Figure 2.20).
Liquid enthalpies of the pure components are determined with considerable de-
viations via
T

h L,i (T) = ∆h0f,i + ∫ cid


p,i dT + (h − h )i (T, p si ) − ∆h vi (T) .
id
(2.83)
T0

–8600
T

–8700 ∫c id
p,i dT
Ideal gas enthalpy Δh0f
T0 (h − hid)(T, psi(T ))
–8800
Enthalpy of Δhvi Critical point
h/J/g

–8900 saturated vapor


Enthalpy of
–9000 saturated liquid

–9100
T0
–9200
–100 –75 –50 –25 0 25 50 75 100 125 150 175
t/°C

Fig. 2.20: Enthalpy description of a pure liquid using the vapor as the starting phase [11].
© Wiley-VCH Verlag GmbH & Co. KGaA. Reproduced with permission.

A way to overcome this difficulty is to fit the correlations for the enthalpy of va-
porization simultaneously to both enthalpy of vaporization and cLp data [51]. The
effect of pressure on the liquid enthalpy is considered to be small and therefore
neglected.
2.8 Enthalpy calculations | 67

–2200
T

∫c id
p,i dT
T0

(h − hid)(T, ps (T ))
–2400
(h − hid)(Ttrans, ps(Ttrans))
h/J/g

∫ c dT
L
p Δhvi (Ttrans)
–2600 T0

href (Tref) h, liquid


h, saturated
vapor
–2800 h, ideal gas
269 299 328 358
T/K

Fig. 2.21: Calculation of the enthalpy of saturated vapor of a pure substance using the liquid as the
starting phase [11].
© Wiley-VCH Verlag GmbH & Co. KGaA. Reproduced with permission.

(B) Liquid as the starting phase:


The reference enthalpy is set for a liquid state and adjusted in a way that the stan-
dard enthalpy of formation in the ideal gas state at t = 25 °C is met. The transition
to the vapor phase is performed at a certain arbitrary temperature, usually the
normal boiling point. Figure 2.21 illustrates the calculation route for an enthalpy
of a saturated vapor using route (B). The enthalpy of vaporization is calculated
indirectly with this route, the results are usually sufficiently accurate at low tem-
peratures. They become even qualitatively wrong in the vicinity of the critical
point [11]. Despite the fact that it is calculated directly, it can turn out that cLp
is again the problem quantity. It is often only measured for temperatures below
the normal boiling point, and its extrapolation to high temperatures is often bad.

For both routes, the change from pure components to the mixture is performed at the
system temperature. For gases, the mixing takes place in the ideal gas state, where the
excess enthalpy is zero. Liquid enthalpies are linearly mixed, the excess enthalpy is
often neglected.
Correlations for cid L
p , ∆h v , and c p supported by the various process simulation pro-
grams are discussed in [11]. The most important ones are as follows:
– for cidp:
– Aly-Lee equation:
2 2
C/T E/T
p = A+B(
cid ) +D( ) . (2.84)
sinh (C/T) cosh (E/T)
68 | 2 Thermodynamic models in process simulation

– PPDS equation:

p = RB + R(C − B)y [1 + (y − 1)(D + Ey + Fy + Gy + Hy )] ,


cid 2 2 3 4
(2.85)

with y = T/(A + T).


– polynomial:
p = A + BT + CT + DT .
cid 2 3
(2.86)

– for cLp :
– polynomial
cLp = A + BT + CT 2 + DT 3 + ET 4 + FT 5 (2.87)

– PPDS equation:

A
cLp = R ( + B + C(1 − Tr ) + D(1 − Tr )2 + E(1 − Tr )3 + F(1 − Tr )4 )
1 − Tr
(2.88)
– for ∆h v :
– DIPPR¹⁰ equation
∆h v = A(1 − Tr )B+CTr +DTr +ETr
2 3
(2.89)

– PPDS equation:

∆h v = RTc (A(1 − Tr )1/3 + B(1 − Tr )2/3 + C(1 − Tr ) + D(1 − Tr )2 + E(1 − Tr )6 )


(2.90)

In the particular equations, A, B, . . . , F are the adjustable parameters. R is the general


gas constant.
It should be mentioned that most data for ∆h v are based on the Clausius–Clapey-
ron equation (2.7). Using a good vapor pressure equation and an appropriate equation
of state for v󸀠󸀠 , an error of approximately 2 % can be expected. The application of the
Clausius–Clapeyron equation should be avoided at vapor pressures p s < 1 mbar and
in the vicinity of the critical point, as the description of v󸀠󸀠 becomes more and more
weak in this area [11].

2.9 Model choice and data management

Any problem, no matter how complicated it may be, has a simple, obvious and generally com-
prehensible wrong solution.
(Harald Lesch)

10 Design Institute for Physical Property Data


2.9 Model choice and data management | 69

Process simulators offer a vast amount of models, which often confuses the user, gen-
erating more disorientation than opportunities. The following paragraph should pro-
vide some clarification. Despite the fact that new ideas and solutions are frequently
presented in the literature, the number of models in use can in principle be restricted
to six. These models are the following.
– Standard activity coefficient model combined with an equation of state, e.g. NRTL-
PR:
A standard activity coefficient model should cover approx. 80–90 % of the cases.
An appropriate equation of state should be used for the vapor phase to make full
use of the capabilities of Equation (2.42). This equation can be used in principle
for temperatures up to the lowest critical point. The Peng–Robinson equation is
the favorite one of the author, definitely, there are other options. If single critical
points are by far exceeded (e.g. presence of an inert gas), the Henry concept can
be applied (Equation (2.59)). When they are exceeded only by a small extent, it
is theoretically wrong but pragmatic just to extrapolate the vapor pressure curve,
however, this should not be done for main components. It does not make sense
in process engineering to use the ideal gas equation for the vapor phase, even if
pressures are low. With the ideal gas equation, the enthalpy description is inaccu-
rate (Chapter 2.8). Moreover, in a later state of a project the pressure relief devices
must be designed (Chapter 14.2). In this case, the thermodynamic model is applied
at pressures far above the normal operation pressure, which then requires a rea-
sonably accurate equation of state for the vapor phase anyway. For the density, it
should be taken care that the mixing rule (Equation (2.66)) is applied. As well, the
correct representation of the enthalpies should be considered.
– Standard activity coefficient model combined with an equation of state for vapor
phase association, e.g. NRTL-ASS:
If carboxylic acids (formic, acetic, propionic, acrylic, butanoic, etc.) or HF play a
decisive role in the process, the use of an association model for the vapor phase
is important (Chapter 2.3.3). One should be aware that there are no models which
consider both association and the nonidealities caused by elevated pressures. The
problems concerning cLp via enthalpy route (A) should be considered [51].
– Electrolyte model, e.g. ELECNRTL:
Small amounts of electrolytes can be treated as heavy ends as a workaround.
If they play a major role, an electrolyte model considering the dissociation re-
actions and the interactions of the ions should be used [47, 48]. The electrolyte
model can as well be combined with equations of state for the vapor phase, e.g.
Peng–Robinson or Redlich–Kwong for high pressures or the association model in
case HF occurs.
– Equation of state model (PR, PSRK, VTPR):
If the process is mainly operated at high pressures, if supercritical components
play a major role with high concentrations in the liquid phase so that Henry’s
law cannot be successfully applied or if supercritical components change to the
70 | 2 Thermodynamic models in process simulation

subcritical state in one apparatus, the application of the φ-φ-approach with an


equation of state valid for both the vapor and the liquid phase is obligatory (Chap-
ter 2.2). Well-known examples are the polypropylene or natural gas conversion
processes. The standard generalized equations of state like Peng–Robinson [21],
Soave–Redlich–Kwong [52] or Lee–Kesler–Plöcker [53] have the disadvantage
that they perform well for nonpolar molecules (ethane, propane, propylene,
etc.), but the more the molecules have a polar character the more inaccurate they
become. A solution for this problem are the advanced cubic equations of state
like PSRK or VTPR [11]. They involve pure component parameters in the so-called
α-function (Equations (2.33), (2.34)). With this option, the vapor pressures, heat
capacities and enthalpies of vaporization of any components can be successfully
represented. There are still weaknesses in the liquid density description, even the
volume translation in the VTPR equation does not give satisfactory results.
The mixing rules of the advanced cubic equations of state are based on the ac-
tivity coefficient approaches (g E mixing rules). Combination of these advanced
equations of state with electrolyte models are available as well, however, as men-
tioned above, they can still not be combined with vapor phase association models.
– High-precision equation of state:
Unexpectedly, there are a lot of problems in process simulation which are related
to pure components, e.g. the utilities steam and cooling water, inertizations with
nitrogen or CO2 or major parts of the LDPE¹¹ process, where ethylene is the only
component. These problems can be covered very effectively by the use of the so-
called high-precision equations of state, which are equations adjusted individ-
ually for the particular components [26–28]. They represent all thermodynamic
properties within their experimental uncertainty over a wide range of pressures
and temperatures and should be used whenever possible. However, except for
natural gas no high-precision equations of state for mixtures are available. In
these cases, compromises have to be made¹². In the hydrocarbon processes, the
Lee–Kesler–Plöcker equation is a good approach, using accurate pure component
representations for hydrocarbons and applying corresponding mixing rules. For
polar components, this is not an option.
– Polymer model:
Polymer applications are often covered by representing the polymer as a high-
boiling component. A reasonable representation of the combinatorial part of
the activity coefficient, which often shows negative deviations from Raoult’s
law for molecules differing large in size, should be considered (Flory–Huggins,
UNIQUAC). There are models for polymers which are by far more complicated

11 low density polyethylene.


12 Copolymerization in LDPE is a good example. For instance, it has to be decided whether for a mix-
ture of 95 % ethylene and 5 % propylene at p = 100 bar it is more important to describe the pressure
effect sufficiently or to describe the deviation caused by the 5 % impurity.
2.9 Model choice and data management | 71

but represent the character and the effects in polymer mixtures accurately, e.g.
PC-SAFT¹³ [54].

The choice of a good model does not imply that simulation will be correct. It is even
more important to compile the binary interaction parameters (BIPs) for the various
possible combinations. For n components in a process, n(n − 1)/2 binary combina-
tions are possible. For a typical project in a chemical plant with 40 components this
makes 780 binary parameter sets. In hydrocarbon business, up to 200 components are
possible, giving 19 900 binary parameters sets. In both cases it is not possible to pro-
vide a perfect matrix in a reasonable time scale. For a chemical process, the following
procedure is recommended:
An EXCEL table is set up containing a matrix with all the components, showing a
color code for the possible options. Figure 2.22 gives an example.

Component H2 H2O BDO THF PAC PROH BUT-AC BUOH


H2
H2O
BDO
THF
PROP-AC
PROPANOL
BUT-AC
BUTANOL

not necessary just set experimental data


Simulator Database to be revised estimated

Fig. 2.22: Illustration of the binary interaction parameter matrix in a project.

To fill this matrix, first the BIPs given by the databanks of the process simulator should
be listed. Next, it should be checked which parameters are decisive for the process.
This is, of course, not an exact assessment, nevertheless, for components which oc-
cur together in any block with considerable concentrations, the corresponding BIP
should be important. For these cases, it is the responsibility of the user either to adopt
the parameters from the simulator or to adjust own ones from experimental data from
the established databanks [55, 56]. The latter is highly recommended, as the quality
of the parameters can be assured in this way. If the data situation turns out to be not

13 perturbated-chain statistical associating fluid theory.


72 | 2 Thermodynamic models in process simulation

sufficient, own phase equilibrium measurements can be initiated to overcome this sit-
uation. The adjustment of BIPs to experimental data is thoroughly described in [11].
Less important parameters can be estimated (Chapter 2.10). Parameters where the
current situation is not acceptable should be marked as “to be revised” or similar. It
must be clear that the BIPs for component pairs where both components occur only at
very low concentrations, e.g. in the ppm region, are not important, it has no influence
on the results if they are omitted. All BIPs with water should be assigned, either by
the simulation program, by adjustment to experimental data or by estimation, as the
nonidealities of water with organic components are usually large. Other parameters
can just be set; e.g., the BIPs for n-hexane–n-heptane can be assumed to be zero (ideal
mixture), as long as nothing better is available. In this way, a detailed overview about
the BIP situation can be obtained, and the quality of the simulation results is easier to
assess.
For the hydrocarbon business, the situation is different. The number of param-
eters involved is so large that a thorough check is not possible within a reasonable
time. Often, it cannot be distinguished between important and less important compo-
nents, as all components occur only in relatively low concentrations. As hydrocarbon
mixtures usually do not show major nonidealities and the interactions with water can
easily be covered, it might be an option to use Modified UNIFAC with an appropriate
equation of state or PSRK (Chapter 2.10).

In this case, process simulation can be skipped!


(Hans Haverkamp, after his boss suggested that he should vary the binary parameters until the
column behavior is met)

Fitting physical properties to reproduce operation results is something one should not
do. Process simulation which meets the data obtained from operation are a strong in-
dication that the process is understood, or it can help to detect errors. Fitting physical
properties to operation data will certainly reproduce the data, but the simulation is of
no use. Extrapolation to other operation states will simply not work.

2.10 Binary parameter estimation

BIPs can be estimated using group contribution methods, as illustrated in Figure 2.23
for the system ethanol–n-hexane. In group contribution methods it is assumed that
the mixture does not consist of molecules but of functional groups. Ethanol can be
divided in a CH3 -, a CH2 -, and an OH-group, whereas n-hexane consists of two CH3 -
and four CH2 -groups. It can be shown that the required activity coefficients can be
calculated as long as the interaction parameters between the functional groups are
known. Furthermore, if the group interaction parameters between the alkane and the
alcohol group are known, not only the system ethanol–n-hexane, but also all other
2.10 Binary parameter estimation | 73

–OH –CH3

–CH2 –CH2 –CH2

–CH2 –OH

Fig. 2.23: Illustration of the group contribution


Ethanol: –CH3 –CH2 –OH
concept [11].
© Wiley-VCH Verlag GmbH & Co. KGaA.
n-Hexan: –CH3 –CH2 –CH2 –CH2 –CH2 –CH3
Reproduced with permission.

alkane-alcohol or alcohol-alcohol systems can be predicted. The great advantage of


group contribution methods is that the number of functional groups is much smaller
than the number of possible molecules [11].
The UNIFAC (universal quasi-chemical functional group activity coefficients)
group contribution method has first been published in 1975 [57]. Like UNIQUAC,
it consists of two parts. The combinatorial part is temperature-independent and
takes into account the size and form of the molecules, whereas the residual part
is temperature-dependent and considers attractive and repulsive forces between the
groups.
The interacting groups are called main groups. They often consist of more than
one sub-group. For example, in the case of alkanes the subgroups are CH3 -, CH2 -,
CH-, and C-groups. The different sub-groups have different size parameters, the so-
called van-der-Waals properties, which represent the volume and the surface of the
groups. By definition, the group interaction parameters between groups belonging to
the same main group are equal to zero.
Meanwhile, the UNIFAC method has almost been replaced by the Modified
UNIFAC (Dortmund) method [58, 59]. Its main improvements are [11]:
– an empirically modified combinatorial part to improve the results for asymmetric
systems;
– temperature-dependent group interaction parameters;
– adjustment of the van-der-Waals properties;
– additional main groups, e.g. for cyclic alkanes or formic acid;
– an extension of the database, besides VLE data;
– activity coefficients at infinite dilution;
– excess enthalpy data;
– excess heat capacity data;
– liquid-liquid equilibrium data;
– solid-liquid equilibrium data of simple eutectic systems;
– azeotropic data.
74 | 2 Thermodynamic models in process simulation

Most important for the application of group contribution methods for the synthesis
and design of separation processes is a comprehensive and reliable parameter matrix.
Because of the importance of Modified UNIFAC for process development, the range of
applicability is continuously extended by filling the gaps in the parameter table and
revising the existing parameters with the help of new data. Since 1996, the further re-
vision and extension of the parameter matrix is carried out by the UNIFAC consortium,
and the revised parameter matrix is only available for its sponsors [60].
For the estimation of binary parameters, Mod. UNIFAC is used in a way that artifi-
cial data are generated, and the parameters of the current model are adjusted to them.
The advantage to the direct use as a model is that the well-known systems do not need
to be estimated with Mod. UNIFAC, and their accuracy is not lost.
Still, Mod. UNIFAC has some weaknesses:
– Isomer effects can not be predicted.
– Unreliable results are obtained for group contribution methods if a large number
of functional groups occurs in the molecule, as it is the case for pharmaceuticals.
Functional groups which are located closely together are often not represented
sufficiently, e.g. the configuration –C(Cl)(F)(Br) in refrigerants (proximity effect).
– Poor results are obtained for the solubilities and activity coefficients at infinite
dilution of alkanes or naphthenes in water [11].
– Systems with small deviations from Raoult’s law are difficult to predict, which is a
problem if also the differences in vapor pressures are small. In these cases, often
qualitatively wrong characteristics for the mixture are obtained.

The advanced cubic equations of state PSRK and VTPR use a g E mixing rule, which can
be used as a predictive tool if UNIFAC or Modified UNIFAC are used for the calculation
of g E . For PSRK, the original UNIFAC method can be used, whereas Modified UNIFAC
yields bad results in this combination. For VTPR, an own matrix has been built up [60,
61]. For both equations, additional groups have been introduced so that light gases can
be described as well. Certainly, as equations of state both can of course be used in the
supercritical region.
In recent years, the COSMO-RS model¹⁴ has been developed [62], which works
without adjusted parameters. Its accuracy compared to UNIFAC is a bit worse, but it
is applicable in any case. However, it should be used only by a professional user, as
well as Molecular Modeling. Molecular Modeling can generate data from quantum-
mechanical calculations of the interactions of the molecules. This is a fascinating
demonstration of how far these interactions are understood. However, in engineer-
ing applications more detailed justifications of the model are required, therefore, the
author is pretty convinced that the general structure of fitting models to experimental
data will remain.

14 COnductor like Screening MOdel for Real Solvents.


2.12 Transport properties | 75

2.11 Model changes

For the handling of enthalpies in a process simulation program, the change of a model
between two blocks is often critical. This problem has much to do with the enthalpy
description. Between the two blocks, the simulation program hands over the values for
p and h to describe the state of the stream. According to the particular models used in
the two blocks, the stream is assigned with two different temperatures that may differ
significantly [11].

Example

A vapor stream consisting of pure n-hexane (t1 = 100 °C, p1 = 2 bar) is coming out of a block which
uses the Peng–Robinson equation with the φ-φ-approach. It is transferred to a “nothing-happens-
block” (adiabatic, same pressure) working with the ideal gas law. Which error will be produced due to
the model change?

Solution

According to the Peng–Robinson equation of state, the specific enthalpy of the stream leaving the first
block is determined to be h1 = −1807.3 J/g at p1 = 2 bar. Using an activity coefficient model with an
ideal gas phase, the coordinates for p2 = p1 and h2 = h1 refer to the vapor state at t2 = 96.6 °C. Using
cLp ≈ 2 J/(g K), the corresponding enthalpy difference

J
∆h ≈ cLp ∆T = 2 ⋅ 3.4 K = 6.8 J/g
gK

will be missing in the energy balance.

Therefore, care must be taken when the thermodynamic model is changed in a flow-
sheet. It is recommended that a dummy heat exchanger is introduced between two
blocks operating with different models, which is defined in a way that inlet and outlet
states are the same in spite of the different models. Another option is to carry out a
model change where both models yield at least similar results, maybe in a block oper-
ating at low pressure. In general, one should stick to the most comprehensive model
in a flowsheet; nevertheless there are cases (e.g. association, electrolytes occurring
only in parts of the flowsheet) where a model change cannot be avoided.

2.12 Transport properties

The transport properties dynamic viscosity, thermal conductivity and surface tension
must be known for the design of many pieces of equipment, e.g. columns or heat
exchangers. Their correlations and mixing rules and further details are thoroughly
76 | 2 Thermodynamic models in process simulation

explained in [11] or [63]. Therefore, only general remarks are given here and the pit-
falls are explained.
– Dynamic viscosity of liquids:
The dynamic viscosity of liquids is probably the most important transport quan-
tity. Among others, it has an influence on heat exchangers, distillation and ex-
traction columns, and the pressure drop of pipes. Similar to the vapor pressure,
it can cover several orders of magnitude and it is similarly difficult to correlate
and extrapolate, although nowadays there are excellent correlation tools which
do this easily. The dynamic viscosity of liquids starts at high values at the melt-
ing point and then decreases logarithmically with increasing temperature. The
pressure influence is comparably small, but it should be considered at large pres-
sures p > 50–100 bar. Correlations refer to the dynamic viscosity at saturation
pressure. Correction terms for the pressure influence are available [11]. Also, just
recently the first reasonable estimation method has been developed [64]. How-
ever, the most crucial thing and probably the weakest part of process simulation
in general is the mixing rule. Several ones are available, but in most cases the
logarithmic relationship
ηM ηi
ln = ∑ x i ln (2.91)
η0 i
η 0

is used, where η0 is the unit used to make the argument of the logarithm dimen-
sionless. Equation (2.91) can only reproduce the order of magnitude of the result,
and in extreme cases not even that. It does not claim to give a good result, it is
essentially not more than a way to come to a number. Even simple systems like
methanol-water can exhibit large maxima, where the deviation of Equation (2.91)
can be up to 100 % [11]. The lucky circumstance is that most applications are not
strongly sensitive to errors in the calculation of small viscosities around or even
below 1 mPa s. Things become worse when larger viscosities become involved and
differences between the two components occur.

Example

Calculate the dynamic viscosity of a brine consisting of 40 mol% ethylene glycol (1,2-ethanediol, EG,
M = 62.068 g/mol) and 60 mol% water (W, M = 18.015 g/mol) at t = 20 °C.

Solution

The pure component viscosities are according to [42]

ηW = 1.01 mPa s ,
ηEG = 21.23 mPa s ,
2.12 Transport properties | 77

giving a calculated viscosity for the mixture according to Equation (2.91):


ηM
ln = 0.6 ⋅ ln 1.01 + 0.4 ⋅ ln 21.23 = 1.228 ⇒ η M = 3.415 mPa s .
mPa s
A more probable value can be taken from [65] after recalculation of the EG concentration from 40 mol%
to 69.7 mass%. The result is 7.08 mPa s, with a deviation of more than 100 % from the calculated value.
In fact, this is a deviation which might have a significant influence on equipment design.

– Thermal conductivity of liquids:


The thermal conductivity of liquids decreases with increasing temperature almost
linearly, except the region in the vicinity of the critical point. The order of mag-
nitude is approx. 0.1–0.2 W/(K m) for almost all liquids. An exception is water;
here the thermal conductivity is in the range 0.6–0.7 W/(K m), with a maximum
at approx. 150 °C. There is a mixing rule which does not produce major errors. The
influence of the pressure is similar to the one for the liquid viscosity; the values
usually refer to the saturation line, and it is a good approach to take it for pressure-
independent, but at p > 50–100 bar a correction term should be applied [11, 63].
– Dynamic viscosity of gases:
In contrast to the viscosity of liquids, the dynamic viscosity of vapors or gases
increases with increasing temperature almost linearly. The order of magnitude
is approx. 5–20 µPa s. The thermal conductivity of gases is hardly ever measured,
and most of the values are calculated ones from a well-defined estimation method
[11]. As for liquids, a pressure-correction should be applied when the pressure
exceeds p > 50–100 bar [11]. Given values usually refer to the ideal gas state at
low pressures. Mixing rules are available.
– Thermal conductivity of gases:
The qualitative behavior of the thermal conductivity of gases is similar to the
dynamic viscosity of gases. The typical order of magnitude is 0.01–0.03 W/(K m).
As well, for the pressure dependence the same statements hold. However, some
interesting remarks must be given: The thermal conductivity of hydrogen and
helium, the so-called quantum gases, is significantly higher; it is in the range
0.16–0.25 W/(K m) in the usual temperature range 0–250 °C for both substances.
This is the reason why helium or hydrogen are used as carrier gases in gas chro-
matography with thermal conductivity detectors. The high thermal conductivity
corresponds to the base line; when the sample passes the detector, the thermal
conductivity and therefore the heat removal from the detector decreases. The
detector temperature rises and gives a corresponding signal.
A second interesting item is that for very low pressures (p < 10−6 bar), the thermal
conductivity of gases is no more a physical property of the substance but depends
more on the dimensions of the vessel where the gas is located [11].
– Surface tension:
The surface tension occurs in various formulas used for the design of equipment
in process engineering, but the author is not aware that any of these equations are
78 | 2 Thermodynamic models in process simulation

sensitive to it. The surface tension refers to the phase equilibrium. A typical order
of magnitude is 5–20 mN/m. For water, the surface tension is significantly higher
(approx. 75 mN/m at room temperature). The surface tension decreases almost lin-
early with increasing temperature and becomes zero at the critical point. Mixing
rules are available, but for systems with water special ones must be applied.

You might also like