Taylor 6/13/03 10:00 AM Page 28

Reactions and Separations

Real-World
Modeling of Distillation
Previously, simulations based on
nonequilibrium, or rate-based, models were
considered impractical due to their complexity.
Ross Taylor, However, with ever-increasing computing
Clarkson University power, these simulations are not only feasible,
and University of Twente
Rajamani Krishna, but in some circumstances they should be
University of Amsterdam regarded as mandatory.
Harry Kooijman,
Shell Global Solutions International

C HEMICAL ENGINEERS HAVE BEEN

solving their distillation problems using the
equilibrium stage model since Sorel first used
the model for the distillation of alcohol over 100 years
ago. Seader (1) has provided an elegant history of the
an entire column, but the most common approach is to
divide the column into a number of discrete “stages,”
as depicted in the third panel. Thus, the question to be
addressed first is: How do we model these stages?
The equations that model equilibrium stages are
first century of equilibrium stage modeling. Real distil- known as the MESH equations. MESH is an acronym
lation and absorption processes, however, normally do referring to the different types of equation that are used
not operate at equilibrium. in the model:
In recent years, it has be-
come more common to simu-
The Stage Concept
late distillation and absorption C
as a mass-transfer-rate-based A
operation, using what have be-
A
come known as nonequilibri-
um, or rate-based, models. This A,C
article presents a brief outline
of nonequilibrium modeling B V1 Reflux
and provides pointers to the
growing literature in this field. Stage 1
A, B, C
Modeling the V2 L1
old-fashioned way
To model a plant like the B
one shown in Figure 1, we de- Vj Lj-1
compose the entire plant into
smaller units. In this case, the ■ Figure 1. Stage j
plant contains a distillation col- Decomposition of a chemical
umn that is shown enlarged in plant into unit operations, and Vj+1 Lj
the center panel of the figure. decomposition of the distillation into stages.
There are many ways to model

28 www.cepmagazine.org July 2003 CEP

Taylor 6/13/03 10:00 AM Page 29

• M stands for material balances of what is referred to frequently as the rigorous model (with
• E stands for equilibrium relationships (to express the some disregard for semantic accuracy), is to employ effi-
assumption that the streams leaving the stage are in equi- ciencies. Several kinds of efficiency have been used in distil-
librium with each other) lation column modeling and design, including the overall,
• S stands for summation equations (mole fractions are Murphree, Hausen and vaporization efficiencies. The Mur-
perverse quantities and won’t sum to unity unless you phree efficiency (2) is arguably the most widely employed
force them to) by distillation engineers and is defined by:
• H stands for heat or enthalpy balances (processes con- yiL − yiE
serve energy, as well as mass). Ei, MV = (1)
There are few mathematical models in any branch of
*
yiL − yiE
engineering that are as well-suited to computer solutions
and that have prompted the development of as many differ- where the overbars indicate the average mole fraction in the
ent algorithms as have the MESH equations. It would not entering (E) and leaving (L) streams, as depicted in Figure 2.
be too far from the truth to claim that it is equilibrium For packed columns, we use something analogous to the
stage calculations that
brought computing into L = Liquid
Nomenclature chemical engineering —
V = Vapor
V,yL yE = Mole fraction in
c = number of components, and chemical engineers L
entering stream
dimensionless to computers (1). yL = Mole fraction in
ct = total concentration, mol/m3 leaving stream
The equilibrium stage
d = driving force for mass
transfer, m –1 model is so simple in
Di,k = Maxwell-Stefan diffusivity, m2/s concept, so elegant from
Ei,MV = Murphree tray efficiency, the mathematical view-
dimensionless point, the basis for so
f = proportionality coefficient many commercial col-
k = mass transfer coefficient, m/s umn simulation pro-
K = vapor-liquid equilibrium grams, and been used to
constant,
simulate and design so
Ni = molar flux of species i,
mol/m2-s
many real columns, that
it seems almost heretical L
P = pressure, Pa
p = partial pressure, Pa to mention that the V,yE
R = gas constant, J/mol-K model is fundamentally
t = time, s flawed. However, chemi-
T = temperature, K cal engineers have long
u = average velocity been aware of the fact ■ Figure 2. Idealized flow patterns on a distillation column tray.
x = mole fraction, dimensionless
that the streams leaving
y = mole fraction, dimensionless
a real tray or section of a stage efficiency called the HETP (Height Equivalent to a
Greek letters packed column are not in Theoretical Plate). In practice, efficiencies and HETPs
κ = mass transfer coefficient of equilibrium with each often are estimated simply from past experience with simi-
binary pair in multicomponent
other. In fact, the separa- lar processes. However, for new processes, this approach is
mixture, m/s
µ = Chemical potential, J/mol
tion actually achieved of no use whatsoever (and often fails even for old ones).
η = distance along diffusion path, depends on the rates of Chemical engineers have, therefore, devoted a great deal of
dimensionless mass transfer from the effort to devising methods for estimating efficiencies and
Subscripts
vapor to the liquid phas- HETPs (3, 4).
i = component index es, and these rates de- These different kinds of efficiencies all attempt to repre-
I = referring to interface pend on the extent to sent the extent to which the real trays in a tray column (or
j = stage index which the vapor and liq- the entire column itself) depart from equilibrium. The
k = alternative component index uid streams are not in HETP is a number that is easy to use in column design.
m = reaction index equilibrium with each However, there are several drawbacks to employing effi-
t = total other. The next question ciencies and HETPs in a computer simulation based on the
Superscripts is: What have we done equilibrium stage model:
F = referring to feed stream about this fundamental • There is no consensus on which definition of efficien-
I = referring to interface weakness? cy is best (although many distillation experts will admit to
L = referring to liquid phase
The conventional way a preference for Murphree-type efficiencies).
V = referring to vapor phase
around this shortcoming • The Murphree vapor-phase efficiency is not the same

CEP July 2003 www.cepmagazine.org 29

Taylor 6/13/03 10:00 AM Page 30

Reactions and Separations

as the liquid-phase efficiency on the same tray (the Hausen are to be estimated from a mathematical model (3, 4)). The
efficiency does not share this property). molar fluxes at a vapor liquid interface may be expressed as:
• The generalized Hausen efficiencies (sometimes
known as Standart efficiencies (5)) are the most fundamen- NiV = ciV kiV ( yiV − yiI ) (2)
tally sound, but are impractically complicated to calculate
and are never used in practice. NiL = ciL kiL ( xiI − xiL ) (3)
• Vaporization efficiencies, favored by some in the past
because they are easy to include in computer programs, are where ciV and ciL are the molar densities of the superscript-
not often used today. ed phases, yiV is the mole fraction in the bulk vapor phase,
• Efficiencies vary from component to component, and xiL is the mole fraction in the bulk liquid phase, and xiI and
from tray to tray, in a multicomponent mixture. Very rarely yiI are the mole fractions of species i at the phase interface.
is this fact taken into account in a simulation model that kiV and kiL are the mass-transfer coefficients for the vapor
uses efficiencies. and liquid phases.
• Efficiencies vary from stage to stage in a tray column. The inclusion in the model of the mass transport equa-
HETPs are a function of height in a packed column. These tions introduces the mole fractions at the interface, some-
behaviors of efficiencies and HETPs are often not account- thing we have not had to deal with so far, at least not explic-
ed for in conventional column simulation software. itly. It is common to assume that the mole fractions at the in-
These weaknesses of the standard model have been terface are in equilibrium with each other. We may, there-
known for a long time (6). Thus, our third question is: How fore, use the very familiar equations from phase equilibrium
should we deal with the shortcomings of the standard model? thermodynamics to relate the interface mole fractions:

Modeling in the real world

yiI = Ki xiI (4)
In recent years, a new approach to the modeling of dis-
tillation and absorption processes has become available — where the superscript I denotes the interface compositions
the so-called nonequilibrium, or rate-based, models. These and Ki is the vapor-liquid equilibrium ratio for component
models treat these classical separation processes as the i. These K-values are evaluated at the interface composi-
mass-transfer-rate-governed processes that they really are. tions and temperature using the same thermodynamic mod-
The building blocks of the nonequilibrium model shown els used in conventional equilibrium stage simulations. The
in Figure 3 are sometimes referred to as the MERSHQ interface composition and temperature must, therefore, be
equations, where: computed during a nonequilibrium column simulation. In
• M represents material balances equilibrium stage calculations, the equilibrium equations
• E represents energy balances are used to relate the composition of the streams leaving
• R represents mass- and heat-transfer rate equations the stage and the K-values are evaluated at the composition
• S represents summation equations of the two exiting streams and the stage temperature (usu-
• H represents hydraulic equations for pressure drop ally assumed to be the same for both phases).
• Q represents equilibrium equations.
Some of these equations are also used in building equi- VL, LE,
librium stage models; however, there are crucial differences yL xE Y
in the way in which the conservation and equilibrium equa- X
Interface

Vapor
tions are used in the two types of model. In a nonequilibri- T
um model, separate balance equations are written for each
distinct phase. Figure 3 shows that the material balance for
each phase includes terms to represent the mass transferred Mass
Transfer Vapor Liquid
from one phase to the other. For the equation used in the Film Film
equilibrium stage model, the sum of the phase balances Energy
yields the material balance for the stage as a whole. The en- Transfer
Liquid
ergy balance is treated in a similar way — it is split into VE, LL,
two parts, one for each phase, each part containing a term yE xL
for the rate of energy transfer across the phase interface.
Modeling distillation and related operations as the rate- LE = Liquid entering stream xE = Liquid mole fraction in entering stream
based processes that they really are requires us to face up to LL = Liquid leaving stream xL = Liquid mole fraction in leaving stream
T = Temperature yE = Vapor mole fraction in entering stream
the challenge of modeling interfacial mass and energy trans- VE = Vapor entering stream yL = Vapor mole fraction in leaving stream
fer in tray and packed columns. This is something that we do VL = Vapor leaving stream
not do in the conventional equilibrium stage model (al-
though we face essentially the same problem if efficiencies ■ Figure 3. Schematic diagram of a nonequilibrium stage.

30 www.cepmagazine.org July 2003 CEP

Taylor 6/13/03 10:00 AM Page 31

Physical Property Requirements Model Requirements: Equations

Activity Coefficients Mass Balances

Vapor Pressures
Activity Coefficients Fugacity Coefficients Phase Mass Balances Energy Balances
Vapor Pressures Densities
Fugacity Coefficients Enthalpies Phase Energy Balances Equilibrium Eqs.
Densities
Enthalpies Equilibrium Eqs. Summation Eqs.

Diffusivities ■ Figure 4. Physical property Summation Eqs.

Viscosities needs of equilibrium (right)
Surface Tension Mass-Transfer in ■ Figure 5. Equations used in
and nonequilibrium (left)
Thermal Conductivities Vapor Phase equilibrium (right) and
models.
nonequilibrium (left) models.
Mass-Transfer Coefficients Mass-Transfer in
Heat-Transfer Coefficients Liquid Phase
Interfacial Areas
Energy Transfer

Physical properties Equipment design

Figure 4 identifies the major physical property require- The estimation of mass-transfer coefficients and inter-
ments. It is obvious that nonequilibrium models are more facial areas from empirical correlations nearly always re-
demanding of physical property data than are equilibrium quires us to know something about the column design. At
stage models (except when tray-efficiency or HETP and the very least, we need to know the diameter and type of
equipment-design calculations are carried out, but those internal (although usually we need to know more than
are done after a simulation and are not needed to carry out that, since most empirical correlations for mass-transfer
the column simulation). The only physical properties re- coefficients have some dependency on equipment design
quired for an equilibrium stage simulation are those needed parameters, such as the weir height of trays). This need for
to calculate the K-values and enthalpies. Those same prop- complete equipment design details suggests that nonequi-
erties are needed for nonequilibrium models as well. librium models cannot be used in preliminary process de-
Mass-transfer coefficients and interfacial areas must sign (before any actual equipment design has been carried
be computed from empirical correlations or theoretical out). However, this is not true. Column design methods
models. There are many correlations for mass-transfer are available in the literature, as well as in most process
coefficients in the literature (3, 4). These coefficients simulation programs. It is straightforward to simultane-
depend on the column design, as well as its method of ously solve equipment sizing calculations and stage-equi-
operation. librium calculations (8). This does not add significantly to
We do not believe that the need for additional physi- the difficulty of the calculations, and it allows nonequilib-
cal properties should be a reason not to use a nonequilib- rium models to be used at all stages of process simulation,
rium stage model. Estimation methods are available for including preliminary design, detailed plant design and
these properties, although they are typically much less simulation, troubleshooting and retrofitting. In fact,
accurate than methods for evaluating thermodynamic nonequilibrium models can be particularly valuable in
properties (7). However, these properties are needed only troubleshooting and retrofitting, even to the point of help-
in so far as they are required to estimate mass-transfer ing identify what particular equipment design detail might
coefficients. In fact, the sensitivity of these coefficients be responsible for a column failing to do what it was de-
to any of these properties is not that large, and the fact signed to do.
that we do not always have accurate estimation methods
should not act as a deterrent to their use. Rather, it Solving the model equations
should serve as a spur to more research and to the devel- There has been so much work done on developing com-
opment of better methods for transport property predic- putational methods for solving the equilibrium stage model
tion and estimation in much the same way as the need for equations that we may essentially use the same approaches
reliable phase equilibrium models has served as motiva- to solve the nonequilibrium model equations (8). The equa-
tion for the development of methods to predict thermo- tions required by the two kinds of model are summarized
dynamic properties. in Figure 5. The fact that the nonequilibrium model in-

CEP July 2003 www.cepmagazine.org 31

Taylor 6/13/03 10:00 AM Page 32

Reactions and Separations

nonequilibrium model, it must be re-

28 Sieve Trays Properties: membered, does not use efficiencies.
p = 5.5 to 6 bar Peng Robinson McCabe-Thiele diagrams (9) can be
constructed from the results of a
nonequilibrium simulation (Figure 8),
and are just as useful for understanding
column behavior as they are for binary
distillation. Note how the triangles do
not touch the equilibrium line.
Example 2: A not-so-simple ab-
C3: 1.5 sorber. Consider the simple packed
i-C4: 56.5 column depicted in Figure 9. The rich
n-C4: 4.5
C3: 1.5 ammonia and air mixture enters at the
i-C5: 1.9
-1 i-C4: 56.5
C6: 2.9 mol s bottom where the ammonia is ab-
n-C4: 4.5
C7: 4.9 sorbed. The enthalpy of absorption is
i-C5: 0.03
C8: 34.3
C9: 3.9
released, causing the temperature of
i-C5: 1.86 the liquid to rise. As a result, water
C6: 2.9
C7: 4.9
evaporates. The mass transfer process
C8: 34.3 in the gas therefore involves three
C9: 3.9 species — ammonia, water and (essen-
tially stagnant) air. Toward the top of
Downcomer Area the column, the gas encounters cold
10% 11% entering water. Therefore, water vapor
Weir Length condenses near the top of the column,
1.9m 2.1m and we now have co-diffusion of am-
Hole Area % of Active monia and water through air. We
12% 10% should not ignore water vaporization at
the bottom and condensation at the top
in the analysis. The resulting tempera-
0.6m

■ Figure 7. Murphree efficiency profiles (predict-

■ Figure 6. Debutanizer adapted from Example 9.1 in Ref. 9. ed) for the debutanizer shown in Figure 6.
The simulation program created the tray design.
1.2
volves more equations is not a concern. In our experience,
the equations of both models are about equally simple (or
difficult) to solve. 1
Murphree Efficiency

Numerical solution of the nonequilibrium model equa-

tions provides the chemical engineer with all of the quan-
tities normally associated with the conventional equilibri-
um stage model — temperatures, flowrates, mole frac- 0.8
tions, etc. Nonequilibrium-model calculations also pro-
vide a great deal of additional information, such as physi-
cal and transport property profiles, and equipment design
0.6
and operating data.
Example 1: A simple debutanizer. Consider the
simple debutanizer shown in Figure 6. The flowrate and
composition profiles do not differ to any significant ex- 0.4
tent from the results that you would obtain with a con- 5 10 15 20 25
ventional equilibrium stage model (although the number
Stage Number
of stages and feed stage location would be different).
However, a nonequilibrium model can also provide con- Propane n-Butane n-Hexane n-Octane
siderable additional information, such as mass-transfer Isobutane Isopentane n-Heptane n-Nonane
rates and predicted efficiency profiles (Figure 7). The

32 www.cepmagazine.org July 2003 CEP

Taylor 6/13/03 10:00 AM Page 33

1
d1 = f12 x1 x 2 (u1 − u2 ) (5)
YC4
YC4 + YC5 where d1 is the driving force for diffusion and ui is the av-
0.8 erage velocity of species i.
This expression may be derived using nothing more
complicated than Newton’s second law — the sum of the
forces acting on the molecules of a particular species is di-
0.6 rectly proportional to the rate of change of momentum
(Ref. 11 provides a more complete derivation). The rate of
change of momentum between different species is propor-
tional to the concentrations (mole fractions) of the different
0.4 species and to their relative velocity. In Eq. 5, f12 is the co-
efficient of proportionality and is related to a friction fac-
tor. Eq. 5 is more often written in the form:
x1 x 2 (u1 − u2 )
0.2 d1 = (6)
XC4 D12
XC4 + XC5
where D12 is the MS diffusion coefficient.
The MS equations are readily extended to multicompo-
0
0 0.2 0.4 0.6 0.8 1 nent systems simply by adding similar terms on the right-
hand side to account for momentum exchanged between
each pair of differing types of molecules. For a ternary
■ Figure 8. McCabe-Thiele diagram for the debutanizer shown in Figure 6.
mixture, for example, we would have two terms on the
right, one of momentum exchange between molecules of
ture profiles along the column show a pronounced bulge types 1 and 2, and a second term for momentum transfer
near the bottom (Figure 9). between molecules of types 1 and 3:

The Maxwell-Stefan approach x x (u − u2 ) x1 x3 (u1 − u3 )

Equations 2 and 3 are included in all basic mass trans- d1 = 1 2 1 − (7)
D12 D13
fer texts and chemical engineering handbooks, and are
taught to all chemical engineers in undergraduate chemi- with the equations for species 2 and 3 obtained by rotating
cal engineering degree programs. Strictly speaking, these the subscripts.
equations are valid only for binary
systems and under conditions where Water
the rates of mass transfer are low.
Most industrial distillation and absorp-
tion processes, however, involve more
than two different chemical species.
The most fundamentally sound way Gas Temperature
to model mass transfer in multicompo- Condensation of Water
Absorption of Ammonia
nent systems is to use the Maxwell-Ste-
Height

fan (MS) theory (11–13). In our opin-

ion, the MS approach to mass transfer Liquid
should be what is taught to students, Temperature
but rarely is that done, even at the grad- Evaporation of Water
Absorption of Ammonia
uate level; most texts give little or no
serious attention to the matter of mass
transfer in systems with more than two
components (exceptions include the 20 25 30 35
texts by Seader and Henley (9) and Temperature, ˚C
Benitez (14)). Air Ammonia
The MS equation for diffusion in a
binary ideal gas mixture is: ■ Figure 9. Ammonia absorber adapted from Example 8.8 in (10).

CEP July 2003 www.cepmagazine.org 33

Taylor 6/13/03 10:00 AM Page 34

Reactions and Separations

The generalization of this expression to mixtures with

xi dµ i
any number of different species is: di = (12)
RT dz
c xi x k (ui − uk )
d1 = − ∑ (8) The difference approximation of this expression is
k =1 Dik
somewhat more involved, since we have to include the
which is more familiar to us in the form: derivative of the activity (or fugacity) coefficient (13).
Example 3: The need for rigorous Maxwell-Ste-
x N − x k Ni
c fan-based nonequilibrium models. The differences
d1 = − ∑ i k (9) in column composition profiles predicted by a rigorous
k =1 ct Dik
nonequilibrium model that incorporates the MS equa-
where we have replaced the velocities with the molar flux- tions may differ significantly from those predicted by
es Ni = ciui. an equilibrium stage model. Consider the experimental
For an ideal gas mixture, the driving force is the partial work of Springer et al. (15) on the distillation of water
pressure gradient: (1), ethanol (2) and acetone (3) carried out in a 10-tray
column operated at total reflux. The residue curve map
1 dpi dxi for this system is shown in Figure 10a. This system
d1 = = (10) shows a binary minimum boiling azeotrope between
P dz dz water and ethanol; an almost-straight distillation
boundary connects the azeotrope with pure acetone.
Solving the MS equations might involve the computation A measured composition profile, carried out in the re-
of various matrices and functions thereof (11). In practice, gion to the left of the distillation boundary, is shown in
we most often employ a simple film model for mass transfer Figure 10b. Simulations of the column, starting with the
with a simple difference approximation to the MS equations: vapor composition at the column top, are also shown. It is
evident that the nonequilibrium model is able to follow
cx i N k − x k Ni the experimentally observed column trajectories much
∆xi = − ∑ (11) better than the equilibrium model. The differences in the
k =1 ct κ ik
column composition trajectories are due to differences in
the component Murphree efficiencies (Figure 10c).
where x-i is the average mole fraction over the film. The MS Differences in component efficiencies could have a
mass-transfer coefficients κij can be estimated from exist- significant impact on a column design that aims for a spe-
ing correlations. For a nonideal fluid, the driving force is cific purity at either ends of the column. For example, for
related to the chemical potential gradient: the water (1), ethanol (2) and acetone (3) system operat-

a b c
Component Murphree Efficiency, Ei

1.0 1.0 2.0

Ethanol Composition

Ethanol Composition

0.8
1.5
0.8
0.6
1.0
0.4
0.6
0.5
0.2

0.0 0.4 0.0

0.0 0.2 0.4 0.6 0.8 1.0 0.02 0.04 0.06 2 4 6 8 10

Water Composition Water Composition Stage Number

Residue Curve Lines Nonequilibrium Model Water

Azeotrope Equilibrium Model Ethanol
Distillation Boundary Experimental Data Acetone

■ Figure 10. Distillation of water (1), ethanol (2) and acetone (3) in a bubble cap tray column: (a) residue curve map; (b) experimental composition trajecto-
ry for Run 6, compared with the nonequilibrium and equilibrium simulations; and (c) component Murphree efficiencies for Run 6 (15).

34 www.cepmagazine.org July 2003 CEP

Taylor 6/13/03 10:00 AM Page 35

1.0
stages are needed to reach the specified 96% ethanol pu-
rity at the top, whereas the equilibrium model indicates
that only 25 stages are needed. In this case, the nonequi-
0.8
librium model takes “the scenic route” to reach the de-
sired top purity. Ignoring the differences in component ef-
Ethanol Composition

ficiencies may lead to severe underdesign.

Columns operating close to the distillation boundary
0.6
may experience much more exotic differences in the col-
umn composition trajectories predicted by the nonequi-
librium and equilibrium models. For operation with the
0.4
Nonequilibrium Model same water, ethanol and acetone system, Figure 12a
Equilibrium Model + shows that the experiments cross the straight-line distil-
60% Efficiency
lation boundary (15), something that is forbidden by the
0.2
NEQ = 39 Stages equilibrium model (16). The nonequilibrium model is
EQ, 60% Efficiency = 25 Stages able to retrace this boundary-crossing trajectory, whereas
0.0 the equilibrium model remains on one side of the distil-
0.00 0.04 0.08 lation boundary. The nonequilibrium model predicts that
the column gets progressively richer in water as we pro-
Water Composition ceed down the column to the reboiler, whereas the equi-
■ Figure 11. Comparison of nonequilibrium and equilibrium models for librium model anticipates that the column gets enriched
distillation of water (1), ethanol (2) and acetone (3) in a bubble cap tray col- in ethanol as the reboiler is approached. The root cause
umn with the objective of reaching 96% ethanol purity at the top. of this behavior lies with the differences in the efficien-
cies of the individual species (Figure 12b); the compo-
nent efficiency of ethanol varies sig-
a b nificantly from tray to tray. Compar-
Component Murphree Efficiency, Ei

ing the component efficiency values

1.0 1.4 Water in Figures 10c and 12b reveals that
Ethanol Composition

1.2 Ethanol even though the mass transfer param-

0.8 1.0
Acetone eters used in the nonequilibrium
model are identical for these two runs,
0.8 the calculated component efficiency
0.6 values bear no resemblance to one an-
0.6
Ethanol-Water
Binary Azeotrope
other. This underlines the difficulty of
0.4 0.4
trying to emulate the performance of
0.2 the nonequilibrium model by fudging
0.2 0.0 component efficiency values. There is
0.0 0.1 0.2 2 4 6 8 10
no way that this can be achieved.
Water Composition Stage Number
Other applications
Nonequilibrium Model The principles outlined above are
Experimental Data applicable to a wide range of related
Distillation Boundaries processes. Below, we very briefly
Equilibrium Model
consider some of these applications.
Three-phase distillation. Three-
■ Figure 12. Distillation of water (1), ethanol (2) and acetone (3) in a bubble
cap tray column: (a) experimental composition trajectory for Run 26, com- phase distillation remains relatively poorly understood
pared with the nonequilibrium and equilibrium simulations; and (b) Compo- compared to conventional distillation operations involv-
nent Murphree efficiencies for Run 26 (15). ing just a single liquid phase. Simulation methods cur-
rently in use for three-phase systems employ the equi-
ing in the region to the left of the distillation boundary, librium stage model (16). It is important to be able to
let us demand a purity of 96% ethanol at the top of the correctly predict the location of the stages where a sec-
column. For a specified feed composition and reflux ond liquid phase can form (to determine the appropriate
ratio, the column composition trajectories for the location for a sidestream decanter, for example). The
nonequilibrium model and the equilibrium model (assum- limited experimental data available suggest that effi-
ing 60% efficiencies for all components) are presented in ciencies can be low and highly variable. Clearly, a
Figure 11. The nonequilibrium model suggests that 39 model based on the assumption of equilibrium on every

CEP July 2003 www.cepmagazine.org 35

Taylor 6/13/03 10:00 AM Page 36

Reactions and Separations

Heterogenous
Ternary Azeotrope
0.6

Vapor Liquid l

Transfer
Transfer

Cyclohexane Composition
Liquid-Liquid
0.4
Phase
Splitting

Transfer
Liquid ll
0.2

■ Figure 13. Schematic representation of a three-phase nonequilibrium stage.

0.0
0.04 0.08 0.12 0.16
stage cannot predict column performance. Springer and
Water Composition
others (17) stress the limitations of simulation models
Ethanol-Water
assuming equal Murphree efficiencies for all compo- Nonequilibrium Model Binary Azeotrope
nents in the mixture. Experimental Data
It is straightforward in principle to extend the ideas Distillation Boundaries
Equilibrium Model
that underlie nonequilibrium models to systems with
more than two phases, as first shown by Lao and Tay-
lor (18). A complete nonequilibrium model for the sys- ■ Figure 14. Distillation of water (1), cyclohexane (2), ethanol (3) and ace-
tem depicted in Figure 13 contains three phase bal- tone (3) in a bubble cap tray column: experimental composition trajectory,
ances, each of which contains terms for mass transfer compared with the nonequilibrium and equilibrium simulations (17).
to or from both of the other two phases. In addition,
the model contains up to six sets of the MS equations, for reactive distillation processes are considerably more
two for each phase boundary (vapor–liquid I, complex than those of either conventional reactors or
vapor–liquid II, and liquid I–liquid II). Three sets of conventional distillation columns. The introduction of
equilibrium equations, one for each possible interface, an in situ separation function within the reaction zone
complete the model. In practice, it is quite likely that leads to complex interactions between vapor-liquid
the vapor phase and a dispersed liquid phase see only a equilibrium, vapor-liquid mass transfer, intra-catalyst
continuous liquid phase, thereby considerably simpli- diffusion (for heterogeneously catalyzed processes) and
fying the model (17). chemical kinetics. For such systems, the chemical reac-
Example 4. Heterogeneous azeotropic system. tion influences the efficiencies to such an extent that the
Sometimes the curvature of the distillation boundary is concept loses its meaning (19).
such that its crossing by the equilibrium stage model is Building a nonequilibrium model of a reactive sepa-
allowable (16). This is illustrated in Figure 14 for the ration process is not as straightforward as building an
water (1), cyclohexane (2) and ethanol (3) system. For a equilibrium stage model, in which we simply add a
column operating at total reflux with the top composi- term to account for reaction to the liquid-phase material
tion corresponding to the heterogenous ternary balances. It must be recognized that no single nonequi-
azeotrope, the equilibrium model has no difficulty librium model can deal with all possible situations.
crossing the curved distillation boundary from the con- Separate models are needed depending on whether the
vex side, moving in the direction of high water compo- reaction takes place within only the liquid phase or if a
sitions and proceeding down the column. However, the solid phase is present to catalyze the reaction. Refer to
experimental data of Springer et al. (17) show that the Refs. 16, 19 and 20 for further discussion.
boundary is not crossed in practice and the column com- Gas absorption. Efficiencies in gas absorption tend
position trajectories are anticipated very well by a to be much lower than in distillation, sometimes as low
nonequilibrium model. as 5%. In addition, many important gas absorption pro-
Reactive distillation. The design and operation issues cesses involve chemical reactions. It does not seem to

36 www.cepmagazine.org July 2003 CEP

Taylor 6/13/03 10:00 AM Page 37

in the liquid phase). Models of this sort

have been used with some success in
the modeling of amine-based gas treat-
ing processes (22).
Distillation column dynamics. One
of the key points of this article is that
nonequilibrium models should be used
when efficiencies are unknown, cannot
be reliably predicted, and are low
and/or highly variable. Efficiencies in
any process depend strongly on the
properties of the mixture, whether or
Tray not chemical reactions are involved,
Section and (last, but by no means least in im-
portance) the type of column employed
and the way in which it is operated. If a
column is not at steady state, then effi-
ciencies vary with time as a result of
changes to flowrates and composition.
Cell Liquid Vapor Thus, equilibrium stage models with ef-
ficiencies should not be used to model
Packed the dynamic behavior of distillation and
Section absorption columns. Nonequilibrium
models for column dynamics are de-
scribed in Refs. 23–25.
Nonequilibrium cell model. An issue
that is not adequately addressed by most
models is that of vapor and liquid flow
patterns on distillation trays or maldis-
tribution in packed columns. Since reac-
tion rates and chemical equilibrium con-
stants depend on the local concentra-
tions and temperature, they may vary
along the flow path of liquid on a tray,
or from side to side in a packed column.
For such systems, the residence time
distribution could be very important.
To deal with this shortcoming of ear-
lier models, nonequilibrium cell models
■ Figure 15. The nonequilibrium cell model. have been developed (26–28). The dis-
tinguishing feature of this model is that the stages are
make a great deal of sense to employ an equilibrium divided into a number of contacting cells (Figure 15).
stage model for systems so far removed from equilibri- These cells describe just a small section of the tray or
um. In fact, although equilibrium stage models for packing, and by choosing an appropriate set of cell con-
such systems are used, it has long been more common nections, one can very easily study the influence of flow
to use mass-transfer-rate-based models to design gas patterns on the distillation process.
absorption processes (21). Nonequilibrium models Flow patterns on distillation trays are modeled by
apply more or less unchanged in principle to gas ab- choosing an appropriate number of cells in each flow
sorption (with or without reaction). The only differ- direction. A column of cells can model plug flow in the
ences between the models are the inclusion of different vapor phase, and multiple columns of cells can model
sub-models for the reaction kinetics and thermodynam- plug flow in the liquid phase as depicted in Figure 15.
ic properties. Many absorption processes involve dilute Backmixing may also be taken into account by using
mixtures, and the rate relationships in Eqs. 2 and 3 suf- an appropriate number of cells. Flow patterns in
fice (the latter modified by the inclusion of an en- packed columns are evaluated by means of a cell flow
hancement factor to account for any chemical reaction model (27).

CEP July 2003 www.cepmagazine.org 37

Taylor 6/13/03 10:00 AM Page 38

Reactions and Separations

Available software drop and interfacial area. RateFrac can use any of the ther-
AspenTech developed RateFrac, in collaboration with modynamic packages that exist within AspenPlus, and can
Koch Engineering, Inc. This implementation is based large- model columns with sidestreams, interstage heaters/coolers
ly on the nonequilibrium model described in the original and pumparounds. Complex specifications can designated
papers by Krishnamurthy and Taylor (29, 30), with the im- for product purity or internal streams. RateFrac is especially
portant additional capability of being able to handle sys- useful for modeling columns with chemical reactions that
tems with chemical reactions. The influence of reaction on influence the separation. Illustrations of the use of RateFrac
mass transfer is modeled by means of enhancement factors. are described in Seader and Henley (9). For more informa-
RateFrac has one mass-transfer coefficient model for each tion, visit www.aspentech.com/includes/product.cfm?Indus-
type of column internal, but it has the facilities to add user tryID=0&ProductID=110
models for the calculation of transfer coefficients, pressure CHEMCAD from Chemstations, Inc. (www.chemsta-

Literature Cited
1. Seader, J. D., “The B. C. (before computers) and A. D. of Equilibri- 21. Cornelisse, R., et al., “Numerical Calculation of Simultaneous
um-Stage Operations,” Chem. Eng. Educ., 19 (2), pp. 88–103 Mass Transfer of Two Gases Accompanied by Complex Reversible
(Spring 1985). Reactions,” Chem. Eng. Sci., 35, pp. 1245–1260 (1980).
2. Murphree, E. V., “Rectifying Column Calculations with Particular 22. Pacheco, M. A., and G. T. Rochelle, “Rate-Based Modeling of Re-
Reference to n-component Mixtures,” Ind. Eng. Chem., 17, pp. active Absorption of CO2 and H2S into Aqueous
747–750 (1925). Methyldiethanolamine,” Ind. Eng. Chem. Res., 37, pp. 4107–4117
3. Kister, H. Z., “Distillation Design,” McGraw-Hill, New York (1992). (1998).
4. Lockett, M. J., “Distillation Tray Fundamentals,” Cambridge Uni- 23. Kooijman, H. A., and R. Taylor, “A Nonequilibrium Model for
versity Press, Cambridge, MA (1986). Dynamic Simulation of Tray Distillation-Columns,” AIChE Journal,
5. Standart, G., “Distillation. V. Generalized Definition of Theoretical 41, pp. 1852–1863 (1995).
Plate or Stage of Contacting Equipment,” Chem. Eng. Sci., 20, pp. 24. Baur, R., et al., “Dynamic Behaviour of Reactive Distillation
611–622 (1965). Columns Described by a Nonequilibrium Stage Model,” Chem. Eng.
6. Seader, J. D., “The Rate-Based Approach for Modeling Staged Sepa- Sci., 56, pp. 2085–2102 (2001).
rations,” Chem. Eng. Prog., 85, pp. 41–49 (1989). 25. Gunaseelan, P., and P. C. Wankat, “Transient Pressure and Flow
7. Poling, B. E., et al., “The Properties of Gases and Liquids,” 5th Edi- Predictions for Concentrated Packed Absorbers Using a Dynamic
tion, McGraw-Hill, New York (2001). Nonequilibrium Model,” Ind. Eng. Chem. Res., 41, pp. 5775–5788
8. Taylor, R., et al., “A 2nd Generation Nonequilibrium Model for (2002).
Computer-Simulation of Multicomponent Separation Processes,” 26. Higler, A., et al., “Nonequilibrium Cell Model for Multicomponent
Comput. Chem. Eng., 18, pp. 205–217 (1994). (Reactive) Separation Processes,” AIChE Journal, 45, pp.
9. Seader, J. D., and E. J. Henley, “Separation Process Principles,” 2357–2370 (1999).
John Wiley, New York, NY (1998). 27. Higler, A , et al., “Nonequilibrium Cell Model for Packed Distilla-
10. Treybal, R. E., “Mass-Transfer Operations,” 3rd Edition, McGraw- tion Columns — The Influence of Maldistribution,” Ind. Eng. Chem.
Hill, New York, NY (1980). Res., 38, pp. 3988–3999 (1999).
11. Taylor, R., and R. Krishna, “Multicomponent Mass Transfer,” John 28. Baur, R., et al., “Dynamic Behaviour of Reactive Distillation Tray
Wiley, New York, NY (1993). Columns Described with a Nonequilibrium Cell Model,” Chem.
12. Krishna, R., and J. A. Wesselingh, “The Maxwell-Stefan Approach Eng. Sci., 56, pp. 1721–1729 (2001).
to Mass Transfer,” Chem. Eng. Sci., 52, pp. 861–911 (1997). 29. Krishnamurthy, R., and R. Taylor, “A Nonequilibrium Stage
13. Wesselingh, J. A., and R. Krishna, “Mass Transfer in Multicompo- Model of Multicomponent Separation Processes. Part I: Model De-
nent Mixtures,” Delft University Press, Delft (2000). scription and Method of Solution,” AIChE Journal, 31, pp. 449–456
14. Benitez, J., “Principles and Modern Applications of Mass Transfer (1985).
Operations,” John Wiley, New York, NY (2002). 30. Krishnamurthy, R., and R. Taylor, “A Nonequilibrium Stage
15. Springer, P. A. M., et al., “Crossing of the Distillation Boundary in Model of Multicomponent Separation Processes. Part III: The Influ-
Homogeneous Azeotropic Distillation: Influence of Interphase Mass ence of Unequal Component Efficiencies in Process Design Prob-
Transfer,” Ind. Eng. Chem. Res., 41, pp. 1621–1631 (2002). lems,” AIChE Journal, 31, pp. 1973–1985 (1985).
16. Doherty, M. F., and M. F. Malone, “Conceptual Design of Distilla- 31. Kooijman, H. A., and R. Taylor, “The ChemSep Book,” Books on
tion Systems,” McGraw-Hill, New York, NY (2001). Demand, Norderstedt, Germany (2001).
17. Springer, P. A. M., et al., “Composition Trajectories for Heteroge- 32. Lewis, W. K., and K. C. Chang, “Distillation. III. The Mechanism
neous Azeotropic Distillation in a Bubble-cap Tray Column: Influ- of Rectification,” Trans. Am. Inst. Chem. Eng., 21, pp. 127–138
ence of Mass Transfer,” Chem. Eng. Res. Des., 81, pp. 413–426 (1928).
(2003). 33. Krishna, R., “A Unified Theory of Separation Processes Based on Irre-
18. Lao, M. Z., and R. Taylor, “Modeling Mass-Transfer in 3-Phase versible Thermodynamics,” Chem. Eng. Commun., 59, pp.33-64 (1987).
Distillation,” Ind. Eng. Chem. Res., 33, pp. 2637–2650 (1994).
19. Taylor, R., and R. Krishna, “Modeling Reactive Distillation,”
Chem. Eng. Sci., 55, pp. 5183–5229 (2000). Further Reading
20. Sundmacher, K., and A. Kienle, “Reactive Distillation. Status and
For further reading, visit www.chemsep.org/publications
Future Directions,” Wiley-VCH Verlag, Weinheim, Germany (2003).

38 www.cepmagazine.org July 2003 CEP

Taylor 6/13/03 10:00 AM Page 39

tions.net) contains a nonequilibrium model for both steady- HETPs and efficiencies have no physical meaning, as these
state and dynamic simulation. are also influenced by reaction.
ChemSep (31) incorporates some of the most recent de- Rigorous nonequilibrium models require the use of
velopments in nonequilibrium modeling. Many correla- the MS equations to properly describe mass transfer in
tions for the mass-transfer coefficients, interfacial area and multicomponent systems. These equations have, in fact,
flow models are built into ChemSep. It also contains a vari- been with us for much longer than has the equilibrium
ety of thermodynamic and physical property models. stage model (see Ref. 11 for original citations). The ap-
ChemSep can also provide a detailed design of the equip- plication of the MS equations to modeling mass transfer
ment selected for the simulation. This allows the program in distillation is also not all that recent. Lewis and Chang
to simulate columns for preliminary design purposes. It has (32), in a remarkably prescient paper that appears to
a limited component library but allows the user to add have been largely ignored, used the MS equations to in-
components with a databank manager. ChemSep is avail- vestigate the mechanism of rectification. They wrote:
able through CACHE (www.cache.org) for educational use “engineers generally are unfamiliar with them” — a situ-
only. Applications of ChemSep are discussed in Refs. 9, ation that has persisted until relatively recent times. Not
14, 31. For more information, visit www.chemsep.org. only do the MS equations allow us to model mass trans-
Many other models have been implemented primarily fer in conventional operations like distillation, absorp-
for research purposes and are not available to others. tion and extraction, they also describe transport in many
less common separation processes, such as membrane
Conclusion processes. Indeed, the MS formulation of mass transfer
Within the last two decades, a new way of simulating provides a rational basis for unifying the treatment of
multicomponent distillation operations has come of age. separation processes (33). CEP

These nonequilibrium, or rate-based, models abandon the

idea that the vapor and liquid streams in a distillation col-
umn ever are in equilibrium with each other. The idea of
modeling distillation as a mass-transfer-rate-based opera- ROSS TAYLOR is the Kodak Distinguished Professor of Chemical Engineering
tion is hardly new. Equations 2 and 3 (albeit in different at Clarkson Univ. in Potsdam, New York (taylor@clarkson.edu), where he
units) actually appear in the classic paper by E.V. Mur- has been since 1980. He currently serves as chair of the Dept. of Chemical
phree (2) that introduced us to efficiencies. Murphree went Engineering. He received his PhD degree from the Univ. of Manchester
Institute of Science and Technology in England. His research interests are
so far as to say: “the use of the general [mass-transfer] in the areas of separation process modeling, multicomponent mass
equation in rectifying column problems would cause the transfer, thermodynamics, and developing applications of computer
calculations to become very much involved, and it is there- algebra to process engineering (and cartography). He is a coauthor (with
fore not considered feasible for practical purposes.” Nowa- Krishna) of the textbook “Multicomponent Mass Transfer” (Wiley, 1993).
He also holds a joint appointment as Professor of Reactive Separations in
days, such calculations not only are feasible, there are cir- the Dept. of Chemical Technology at the Univ. of Twente in The
cumstances where they should be regarded as mandatory. Netherlands, and is a trustee of The CACHE Corp.
Of course, models based on equilibrium stage concepts
will not be abandoned, nor is there any need for us to do R. KRISHNA is a professor at the Univ. of Amsterdam (krishna@science.uva.nl).
He graduated in chemical engineering from the Univ. of Bombay and was
so. For design of new columns in which the column con- awarded a PhD in 1975 from the Univ. of Manchester. He then joined the
figuration is not fixed, it is best to start with the equilibri- Royal Dutch Shell Laboratory in Amsterdam, where he was engaged in
um model to determine the configuration, optimum reflux research, development and design of separation and reaction equipment.
,etc. (16). The final design should be checked against the After nine years of industrial experience, he returned to India to take over
the Directorship of the Indian Institute of Petroleum. Since 1990, he
nonequilibrium model because, as we have seen, it is pos- occupies the position of Professor of Chemical Reactor Engineering at the
sible for the predictions of the nonequilibrium model to Univ. of Amsterdam. His current research interests range from molecular
differ considerably from those of the equilibrium model. modeling, bubble and particle dynamics, and reactor scale-up to process
Nonequilibrium models are of great value in simulating synthesis. Krishna has co-authored three textbooks. His research
contributions have won him the Conrad Premie of the Royal Dutch
existing columns. No longer is it necessary to guess the Institution of Engineers in 1981, and the Akzo-Nobel prize in 1997.
number of equilibrium stages, the location of the feed and
any intermediate product streams, and the individual com- HARRY KOOIJMAN is a research distillation specialist at the Amsterdam
ponent efficiencies in order to try and model a column that laboratory of Shell Global Solutions International BV, The Netherlands
(kooijman@chemsep.org). He graduated from Delft Univ. of Technology
no longer is performing as intended. and received his PhD in 1995 from Clarkson Univ. He joined the BOC Group
Reactive distillation is an emerging application that has in 1996 as a senior research engineer, where he was involved with the
introduced additional complications. Here it is not uncom- development of structured packing for cryogenic distillation. In 1999, he
mon to assume equilibrium with regard to mass transfer, moved to Germany where he worked at science+computing as a consultant
in high-performance computing. He joined Shell Global Solutions in 2002,
but allow for finite reaction rates. This is fine for conceptu- where he focuses on the development of distillation tray technology and
al design. But for equipment sizing, the problem of deter- separation equipment.
mining column heights remains. For reactive distillation,

CEP July 2003 www.cepmagazine.org 39