Вы находитесь на странице: 1из 9

Chemical Engineering Science 62 (2007) 7196 7204

www.elsevier.com/locate/ces

A multi-scale approach for CFD calculations of gasliquid ow within large


size column equipped with structured packing
L. Raynal , A. Royon-Lebeaud
IFP, BP 3, 69390 Vernaison, France
Received 19 April 2007; received in revised form 16 July 2007; accepted 12 August 2007
Available online 19 August 2007

Abstract
This work has been carried out in the framework of post-combustion CO2 capture process development. Considering the huge amount of
gases to be treated and the constraints in terms of pressure drop, it appears that the absorption column will be equipped with high efciency
high capacity packings such as structured packings. The present paper focuses on the CFD modellisation of the two-phase ow within this
complex geometry. For limited computational resources reasons, it is presently impossible to run computations at large scales taking into
account the gasliquid interaction and the real geometry of the packing and original approaches must be developed. In the present work, a
multi-scale approach is proposed. It rst considers liquidwall and liquidgas interaction at small scale via two-phase ow calculations using
the VOF method. Second, the latter results are used in three-dimensional calculations run at a meso-scale corresponding to a periodic element
representative of the real packing geometry. Last, those results are further used at large scale in three-dimensional calculations with a geometry
corresponding to a complete column. Results are compared with experimental data and with other CFD simulations in terms of liquid hold-up,
pressure drop and unit operation. Some suggestions are made for further development.
2007 Elsevier Ltd. All rights reserved.
Keywords: Structured packing; CFD; Two-phase ow; Packed column; CO2 capture; Gas treatment

1. Introduction
This work has been carried out in the framework of postcombustion CO2 capture process development. It focuses on
the capture of CO2 emitted by power plants using gas, fuel or
coal. This process consists in using an amine based solvent,
typically monoethanolamine (MEA), which removes CO2 from
post-combustion emitted gas within a packed tower, the absorber, and in regenerating the solvent in a second column, the
desorber, by heat regeneration (see e.g. Freguia and Rochelle,
2003). Post-combustion CO2 capture process is mainly characterised by three aspects. First, it deals with huge volume of
gas to be treated. For a typical 400 MW power station, the corresponding gas ux is approximately 1.106 Nm3 h1 . Second,
it is characterised by low CO2 partial pressure, in the range
of 0.040.13 bar which is much less than the values of up to
20 bar observed in natural gas treatment. This implies relatively
Corresponding author. Tel.: +33 4 78 02 25 27; fax: +33 4 78 02 20 08.

E-mail address: ludovic.raynal@ifp.fr (L. Raynal).


0009-2509/$ - see front matter 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2007.08.010

low liquid load and large gas supercial velocity. Last, since
this process is located just downstream the power plant operating at a pressure close to the atmospheric pressure, the capture
process must generate very low pressure drop. Since the maximum pressure drop across the CO2 absorber would be around
50100 mbar for packings bed height of almost 30 m, typical
values of admitted pressure drop per unit length would be in
the range of 1.5.3 mbar m1 . These three aspects call for high
capacity, high efciency and low pressure drop gasliquid contacting internals. Due to their geometric characteristics, high
specic geometric area and high void fraction, structured packings are good candidates to meet these three requirements.
Since the work carried out in the group of Pr. Fair at
University of Austin, many studies have been devoted to
structured packings in order to develop models for pressure
drop and liquid hold-up, the latter being further used in mass
transfer models (Bravo et al., 1985; Fair and Bravo, 1990).
Until recently, most of the experimental studies on structured
packings have been carried out for distillation. In most cases,
uids are water or light hydrocarbons and less work has been

L. Raynal, A. Royon-Lebeaud / Chemical Engineering Science 62 (2007) 7196 7204

carried out with liquids more viscous than water. Since global
warming and greenhouse gas emission reduction have became
a major worldwide issue, recent works have been focused on
amine based solvent for CO2 capture (see e.g. Aroonwilas et al.,
1999; Tobiesen and Svendsen, 2006). Such solvents have viscosities which range from 2 up to 10 cP and little work has
been done on liquid viscosity effects for such viscosity values. Besides, it might be interesting to develop new packings
that would be optimised for such process. The present study
aims at showing how CFD can be an interesting tool to complete experimental work for pressure drop and liquid hold-up
determination for CO2 capture process development. For such
development the two-phase ow within the packing has to be
well understood at local scale to be further used at large scale.
Petre et al. (2003) proposed a rst approach based on mesoscale CFD three-dimensional calculations within said representative elementary units, called REU. The obtained results
in REU are further used in a zero-dimensional model for dry
pressure drop calculations at large scale. In the latter work, it
is shown that most of the pressure drop is caused at the crisscrossing junctions within packing metal sheets. Such results
are of qualitatively interest but, as discussed in Raynal et al.
(2004), they may not be considered for quantitative purposes
since the considered REU geometry does not correspond to a
fully representative geometry. It is believed that in order to be
representative, the geometry must correspond to the smallest
periodic element which can be found in the real packing geometry. Besides those calculations were limited to wet pressure drop that is without liquid. More recently, Ataki and Bart
(2006) made CFD simulations for a metal structured packing,
the Rombopak 4M, with an interesting approach considering
fully periodic elements. The latter packing is made of lamellas and differs signicantly from common structured packings
made of continuous metal sheets. In the present study, the chosen geometry corresponds to the well-known Mellapak 250.Y
packing from Sulzer. Due to the complexity of the structured
packing geometry and the limited CPU resources of calculators, it is impossible to run CFD simulations at the scale of
a column while taking gasliquidwall interactions into account at the scale of the liquid lm. It is thus important to
propose original approaches that enable calculations at large
size with models representative of phenomena occurring at
small scales.
In the present study, CFD calculations have been carried out
in three steps at different scales for the Mellapak 250.Y structured packing. The calculation strategy is described in Section
2. Section 3 shows the results obtained at small scale, dealing
with gasliquidwall interactions at the liquid lm scale. Section 4 shows the results obtained at the so-called meso-scale
which are based on a geometry corresponding to the smallest
periodic element in a real three-dimensional packing. Section
5 is dedicated to calculations run at large scale which enable
discussion about internals and packing designs. Indeed, less
work has been made on gas distributor devices and their impact on the ow eld within the column. Recent studies have
shown that CFD could be an appropriate tool to determine the
gas ow eld above the distributor and below the packed bed

7197

(see Wehrli et al., 2003); however, CFD calculations in such


studies were done around the distributing devices only and not
within the packed bed. One of the objectives of this study is to
show that the present strategy is able to determine gas distributor effects on the ow eld within the layers of packings and
allows for internals design optimisation.
2. Calculation strategy
Fig. 1a shows a sketch of an industrial gas treating column
operating in a counter-current ow mode. It contains liquid and
gas distributors at top and bottom of the column, respectively.
In between distributors, the packed bed is made of layers of
packing elements which are turned 90 from each other as
shown in Fig. 1b. The column diameter for a CO2 absorber
may reach up to 10 m, the total height being about 2030 m.
Fig. 2 shows close views of the packing. Fig. 2a shows the
triangular based channel geometry; the base, B, and the height,
h, of the channel are 22.4 and 11.6 mm, respectively, the angle
with horizontal, , being 45 . Fig. 2b shows a closer view of
the metal sheet on which one observes the small scale wall
texture. The latter can be represented as a sinusoidal-shaped
wall as shown in Fig. 2c, the periodic length, , and amplitude,
A, being, respectively, 2.8 and 0.3 mm. Solving the gasliquid
ow on such a large range of scales is not possible so far
for computing resources reasons. However, it is shown in the
present study that the combination of three types of calculation
makes simulations for a whole column possible.

Fig. 1. (a) Sketch of an industrial gas treating column, (b) picture of structured
packing elements as installed within the column and (c) sketch of the smallest
periodic element within the structured packing.

7198

L. Raynal, A. Royon-Lebeaud / Chemical Engineering Science 62 (2007) 7196 7204

Fig. 2. Pictures of the Mellapak 250.Y (a) at channel scales, (b) close view and (c) details of wall texture cut-view.

Fig. 3. Sketch of the present calculation strategy. The type of calculations,


the input and output used at each step and the corresponding geometry and
characteristic length scales are given.

The calculation strategy is illustrated by the dawning shown


in Fig. 3. On the latter sketch, the type of simulation, the
input and the output of the simulations and the associated

geometry are given. The rst type of calculation consists


of two-dimensional gasliquid simulations using the VOF
model as described in Raynal and Harter (2001) and used for
gas/liquid ow simulations in packings by Raynal et al. (2004).
In the latter study, the wall was considered to be smooth, and
calculations were made for very large liquid owrates corresponding to a co-current ow application. Present calculations
consider the gasliquid ow at the scale of the corrugation
which enables to take into account the wall texture and its
inuence on the liquid ow. From such calculations, one can
essentially determine the liquid hold-up within the packing
and the velocity at gasliquid interface (see Fig. 3, top). The
second type of calculations consists of three-dimensional gas
ow simulation in the smallest periodic element representative
of the real structured packing geometry (see Fig. 3, middle).
Those calculations give relationships between pressure drop
and gas supercial velocity. The presence of liquid is indirectly
taken into account by rst converting the supercial gas velocity into an interstitial gas velocity taking the liquid hold-up
into account, and second by choosing a moving boundary condition at walls via imposing the velocity corresponding to the
liquid velocity at interface at walls. These two informations,
liquid hold-up and liquid velocity at gasliquid interface, are
those obtained at the rst step. Last, simulations are carried out
at a very large scale considering the packed bed as a porous
media (see Fig. 3 bottom), the latter being characterised by
pressure drop coefcients obtained at step two. From such
calculation at large scale one is able to study the gasliquid
distributors/packings interactions for achieving optimum
column design. Each steps are described in details in the
following sections.

L. Raynal, A. Royon-Lebeaud / Chemical Engineering Science 62 (2007) 7196 7204

7199

Fig. 4. Views of the two-dimensional computational domain used for the simulations at small scale: (a) case of smooth walls and (b) case of walls with
texture, details.

2.1. Calculations at small scale


In the case of a uniformly wetted smooth vertical wall, and
assuming that the liquid ows as a laminar and fully developed
lm with no gas interaction, the thickness of the liquid lm, e,
is given by


3qL L 1/3
,
(1)
e=
g
where qL is the liquid ow rate per unit width of wetted surface,
or specic ow rate, L the liquid kinematic viscosity and g the
acceleration of gravity. The specic ow rate is given by the
ratio of the liquid load, QL , to the geometric area, aG . Since
the work of Bravo et al. (1985), this model has been commonly
used to deduce the liquid holdup, hL , from the liquid thickness,
e, using
hL = e aG .

(2)

For industrial conditions, the liquid load varies from approximately 10 to 50 m3 m2 h1 . This corresponds to liquid lm
thickness varying from 0.2 to 0.4 mm. Since these values are
of the same order of magnitude as the amplitude of the surface
texture at walls (A = 0.3 mm), the latter has to be considered.
The same model gives the following expression for the liquid
velocity at the interface, or effective liquid velocity as called
by Bravo et al. (1985), UL,eff :


2 1/3
3
3 gq L
,
(3)
UL,eff = 2 UL = 2
3L
where UL is the averaged liquid velocity of the liquid lm given
by the ratio of the specic liquid ow rate, qL , to the liquid lm
thickness. The objectives of the rst step calculations are to
provide the liquid hold-up and the liquid velocity at the liquid
lm interface. Both the informations are compared to the onedimensional laminar falling lm model and are further used in

the second step of the present methodology for the calculations


at meso-scale.
The rst step calculations thus consists in simulating the
gasliquidwall interaction at the liquid lm scale. It is done
with the VOF approach. This two-phase ow model allows
to capture the interface between two non-interpenetrating uids. It has been successfully used by Raynal et al. (2004) in
the case of co-current ow at very large liquid loads owing
along smooth walls in a two-dimensional geometry corresponding to a high geometric area packing. Valluri et al.
(2005) used a similar technique to simulate liquid lms owing down over two-dimensional structured wavy surfaces.
While these authors discuss the impact of the sinusoidal-like
wavy deformations amplitude and frequency on the effective
area, they do not use their results for estimating the liquid
hold-up.
Present calculations aim at determining the inuence of the
wall structure on the liquid hold-up for the Mellapak.250Y geometry as shown in Fig. 2. Simulations have been carried out
with the Fluent.6.2 commercial code. The VOF approach has
been used in the steady calculations approach, with the implicit
scheme for the interface reconstruction and the HRIC module for the VOF solver. Two geometries were used. One with
smooth walls and the other with texture on walls which consists of a sinusoidal-like structure. The whole computational
domain is similar to the one used by Raynal et al. (2004), and
only close views are shown in the present article. Fig. 4a shows
the central periodic element within which the liquid hold-up
is determined for the case of smooth walls. Fig. 4b shows a
closer view near the wall in the case of wall with texture.
The grid is adapted to the computed case in such a way that
at least 810 grid points are within the liquid lm. Most of
the grid is made of structured mesh, the number of cells being about 10,000 cells. The liquid hold-up is determined via
the calculation of the liquid fraction in the central periodic
element.

7200

L. Raynal, A. Royon-Lebeaud / Chemical Engineering Science 62 (2007) 7196 7204

2.2. Calculations at meso-scale

2.3. Calculations at large scale

The second step is done in a similar way as performed by


Petre et al. (2003). It consists of calculations in a computational domain corresponding to the smallest periodic geometry
characteristic of structured packings as shown in Fig. 1c. The
computational domain corresponds to the volume between the
two opposite smooth metal sheets. Flow is considered periodic
in the z and y direction, a lineic pressure drop, DP/L, being
imposed along the z direction. Calculations are run in steady
mode until the surface averaged gas velocity z component is
constant. This requirement is much more demanding than just
the standard 103 requirement on residuals values. Convergence in terms of mass ux often requires about 1000 iterations
and residuals values are less than 105 . The computational domain contains 150,000 tetrahedral cells. To study the viscous
ow model inuence, laminar and standard k. turbulent models have been used. VOF simulations in such a complex threedimensional geometry are considered to be too demanding in
terms of CPU resources and simulations with gas ow only are
considered. However, boundary conditions have been adapted
in two ways in order to take indirectly the liquid inuence into
account. First, since part of the volume should be occupied by
the liquid, the velocity obtained from the calculations, UCFD ,
has to be corrected by the liquid hold-up. Indeed a given value
of pressure drop would be reached at a lower gas velocity with
liquid ow than without any liquid ow. To determine the gas
capacity F -factor, F, for comparison with literature data, the
following expression is thus used:

(4)
F = G UCFD (1 hL ),

Once liquid hold-up, gasliquid interactions and pressure


drop characteristics are known at small scale, they are merged
together and it enables the determination of pressure drop characteristics of an equivalent porous media which is used in this
third step consisting of CFD calculations at large scale. This last
step consists in considering a computational domain which corresponds to the whole absorber at industrial scale as sketched
in Fig. 1a. The packed bed is considered as successive layers
of porous volumes. Layers of packings, made of parallel metal
sheets, are considered as an anisotropic porous media with two
equal pressure loss coefcients, Kz = K(x or y) , in two directions, and one pressure loss coefcient in the third direction,
K(y or x) , which is innite to model the inuence of the presence of perpendicular plates. In practice, it has been considered
that the third pressure loss coefcients would be 1000 times
larger than the two others. Since successive layers are turned
by 90 between each other when installed in columns (see
Fig. 1b), Kx and Ky pressure loss coefcients are switched between each other for odd or even layers, one being equal to Kz
the other being equal to 1000 Kz . The height of these layers
is 210 mm corresponding to the height of industrial Mellapak
250.Y packing elements.
Inlet and outlet effects may be considered via the use of
different distributors. The inuence of the liquid distributor is
quite well known and many articles have shown the importance
of distributor design in particular in terms of number of drip
points per unit surface (see Bonilla, 1993; Billet, 1995). In the
present study, we assume that the liquid distributor is ideal. The
liquid distributor is thus modelled as a simple pressure jump
surface of zero thickness and it is assumed that the liquid is
homogeneously distributed. This means that there is no radial
distribution of the pressure loss coefcients Kz . For gas inlet,
the distributor is a straight tube with a single open orice at the
centre of the column oriented downward as sketched in Fig. 1a.
The number of cells is about 400,000 cells for a column of 1 m
in diameter containing 10 packing layers.

where the liquid hold-up is the result of the calculations run at


rst step of the present strategy.
The second adaptation consists in modifying the boundary
conditions at walls. First, it is commonly accepted that, below
the loading point, the liquid hold-up in packed columns is not
affected by the gas ow (Stichlmair et al., 1989; Suess and
Spiegel, 1992). Second, as discussed by Sidi-Boumedine and
Raynal (2005) or Alix and Raynal (2006), one observes from tomographic measurements across structured packings that there
is almost no axial evolution of the liquid ow distribution. This
means that the main liquid velocity component is in the vertical direction. From these two observations, it is considered that
the presence of liquid can be modelled by imposing a moving
wall condition as boundary conditions at walls in the vertical
downward direction. The velocity that is imposed is given by
the value of the liquid velocity at interface obtained at rst step
of present methodology.
From such calculations, one can determine the relationship
between the pressure drop coefcient in the vertical direction,
Kz , and the gas and liquid ow characteristics.

P
1
= KZ (QG , QL , L ).
2
L
1/2G UG

3.1. Results at small scale


Hold-up in case of a corrugated smooth wall is evaluated on
one periodic element. In the case of the rough wall, hold-up is
evaluated as the same way as for the one-dimensional model
(see Section 2.1). A mean thickness, eCFD , is rst calculated
on a periodic element. It is then corrected to take into account
the inclination of the plate by
ecorrect = eCFD (sin )1/3 ,
e = eCFD (sin )1/3 .

1
P
= K(QG , QL , L ),
K=
2
L
1/2G UG
KZ =

3. Results and discussion

(5)

(6)

The liquid hold up hL is then determined by Eq. (2).


Hold-ups obtained in the case of smooth corrugated wall and
in the case of wall with texture are compared with the value
of the one-dimensional model and with experimental values

L. Raynal, A. Royon-Lebeaud / Chemical Engineering Science 62 (2007) 7196 7204

7201

Fig. 6. Velocity elds within the liquid lm obtained with the VOF simulations
for a liquid viscosity of 2.5 cP. (a) QL = 54 m3 m2 h1 i.e., ReL = 96 and
(b) QL = 161 m3 m2 h1 i.e., ReL = 288.

Fig. 5. Ratio of liquid hold-up to the liquid hold-up given by the


one-dimensional model versus liquid load. Comparison between experimental
and CFD results for two liquid viscosities.

of Alix and Raynal (2006) and of Spiegel and Meier (1992)


in Fig. 5. First, one observes that the one-dimensional model
and the CFD calculations with corrugated smooth wall are in
close agreement and strongly underestimates the experimental
results. This is observed for liquid supercial velocity ranging
from 1 to 250 m3 m2 h1 and for viscosity ranging from 1 to
2.5 cP. This shows that the liquid lm/wall interaction is not
properly modellised. Second, it is observed that the simulations
taking into account the small scale roughness of the wall give a
much better agreement with the experimental data even if some
discrepancy is still observed.
Liquid ow over a one-dimensional periodic surface has been
largely investigated by experimental and numerical studies (see
e.g., Zhao and Cerro, 1992; Pozrikidis, 1988 for creeping ow,
respectively). Such studies show that the relevant parameters
are the liquid Reynolds number based on the liquid ow rate
per unit width, ReL = 4qL /, the ratio of the amplitude to the
wavelength of the roughness, the ratio of the liquid thickness
of the corresponding one-dimensional model to the amplitude
of the roughness and the Bond number which is the ratio of
gravity effects by surface tension effects. In the present study,
only the liquid Reynolds has been varied; it ranges from 40 to
1000. Depending on the liquid Reynolds number, one observes
that recirculation zones form in the cavities of the roughness,
as shown in Fig. 6, and that the recirculation zone grows as the
Reynolds liquid increase. The presence of such recirculation
zones explains why the liquid hold-up is greater in the case of
rough walls than in the case of smooth walls. The liquid hold
up can therefore be considered as a sum of a static hold up and
a dynamic hold up. The static hold up given by the volume ratio
occupied by the recirculation is limited for geometric reason
to 3.2% which represents the whole volume of the cavities
for the Sulzer M252.Y packing. This value is far from being
negligible when compared to experimental values which ranges
approximately from 4% to 8% as measured by Spiegel and
Meier (1992). The present calculations clearly show that the
traditional smooth wall approach cannot apply and that texture
at walls has to be taken into account.

Fig. 7. Wet lineic pressure drop, DP/L, versus F -factor.

The second important parameter that has to be determined


by these calculations at small scale is the velocity at gasliquid
interface. The present CFD calculations show that the velocity
at the interface is not affected by the presence of recirculating
zones. In the case of a rough wall, one actually observes that the
liquid velocity at the interface is equal to the interfacial velocity
calculated by the one-dimensional model of an innite smooth
wall with a discrepancy less than 2%. It is thus reasonable to
take the value of the one-dimensional model for the velocity as
given in Eq. (3) for the calculations at larger scale.
3.2. Results at meso-scale
Calculations at meso-scales have been run rst without any
consideration of the liquid phase, that is for dry packing, and
second with the inuence of the liquid being taken into account. The results obtained for the wet pressure drop per unit
length versus the F -factor are shown in Fig. 7. Results for dry
pressure drop are rst discussed. The closed symbols correspond to the different models which have been used for the gas
ow, laminar or standard k. turbulent model. Present results
are compared with the experimental data obtained by Spiegel
and Meier (1992). One rst observes that, while one would expect a power law of 1 for laminar ow and a power law of 2

7202

L. Raynal, A. Royon-Lebeaud / Chemical Engineering Science 62 (2007) 7196 7204

for turbulent ow, both models give results with power laws
in between these two values, 1.4 and 1.8, respectively. This
may be explained by the fact that strong changes in geometry
that occur at a similar scale as the hydraulic diameter induce
a complex organisation of the ow differing signicantly from
common fully developed ow. Second, one observes that the
best agreement with experimental data is obtained with simulation run assuming laminar ow. The straight line in Fig. 7
indeed corresponds to the best t as suggested by Spiegel and
Meier (1992) for their own experimental data. The agreement
with present CFD calculations assuming laminar is very good
while turbulent ow models give much higher pressure drops.
This agreement with laminar ow model is obtained while gas
Reynolds numbers cover a wide range from 400 up to 2 104
where the gas ow Reynolds number, ReG is given by
ReG =

G VSG / sin()4/aG
,
(1 hL )G

(7)

where 4/aG corresponds to the hydraulic diameter being four


times the hydraulic radius, the ratio of the wetted perimeter to
the surface area and , the packing porosity (here,  = 0.95).
With such high Reynolds numbers in a such complex geometry,
one would expect a better agreement with turbulent models.
It is thus believed that the turbulent models are not adapted
for such a complex three-dimensional ow; the laminar model,
which gives satisfactory results, is thus recommended.
Results for wet or irrigated pressure drops are also shown
in Fig. 7. Since moderate effects are observed, only one liquid load is considered. The latter corresponds to the maximum
value one would expect for CO2 absorption process, that is
50 m3 m2 h1 . As obtained from calculations at rst step, such
liquid load corresponds to a velocity at gas/liquid interface, and,
in the present strategy, to a wall velocity, Uw , of 0.5 m s1 .
From Fig. 7, one observes that imposing a liquid velocity at
walls and taking into account the volume fraction occupied by
the liquid, lead to two modications on the pressure drop curve
compared to that obtained without any liquid. First, for low gas
velocity, or low F -factor, less than approximately 0.7, one observes an important increase in the pressure drop. Second, at
higher F -factor values, the curves at Uw = 0.5 m s1 closely
follow the curve obtained for dry packing, the offset between
the two curves being due to the correction of the F -factor, taking into account the volume fraction occupied by the liquid as
shown in Eq. (4). The present results are compared to values obtained with the Sulpak software developed by Sulzer and based
on the work of Spiegel and Meier (1992). It is seen that the
presence of liquid is well reproduced for low F -factor values
and a close agreement for intermediate F -factor values. However, when gas velocity becomes too large, the discrepancy between experimental and CFD values becomes important. This
difference happens when the gasliquid interaction increases,
inducing an increase in experimental pressure drop values. This
happens above loading point and just below the point corresponding to 80% of ooding. The point corresponding to 80%
of ooding is shown as a point with larger size. Note that the
choice for an industrial column diameter would be such that
the ooding rate is less than 80%, which means pressure drops

Fig. 8. Velocity elds and velocity magnitude contours at the exit of the rst
four packing layers; colourmap in m s1 ranging at 5% around 1.48 m s1 .

less than 2 mbar m1 for this precise case. From Fig. 7, one
observes that it is precisely above that value that discrepancy
between experiments and calculations becomes important. This
difference is thus explained by the fact that present methodology is not able to determine strong gasliquid interactions as it
happens above loading. However, it is not of great importance
since this region corresponds to too high pressure drops for the
CO2 capture industrial application.
One can thus conclude that, for the case of CO2 industrial
capture plants which would be designed for liquid load between 10 and 50 m3 m2 h1 and with maximum pressure drop
per unit length of about 2 mbar m1 , the agreement between
present calculations and experiments is satisfactory. The relationship between pressure drop and local values of gas and liquid velocities may thus be considered for use at larger scale.
3.3. Results at large scale
Calculations at large scale have been run for different ow
conditions and for different column diameters. However, only

L. Raynal, A. Royon-Lebeaud / Chemical Engineering Science 62 (2007) 7196 7204

7203

Fig. 9. Velocity magnitude contours at the exit of the rst four packing layers; colourmap in m s1 . Two assembly procedures are compared (a) rst packing
layer plates are normal to the x direction and (b) rst packing layer plates are parallel to the x direction.

one case is presented here, the purpose of the present paper being to demonstrate the interest of the approach. Simulations have been carried out for an equivalent liquid load of
50 m3 m2 h1 and with a gas velocity of 1.47 m s1 which
corresponds to a F-factor of 1.62 Pa1/2 that is almost at 80%
of ooding as previously discussed. Fig. 8 shows the velocity
eld in the y = 0 plane and the velocity magnitude contours in
sections corresponding to exits of the rst four packing layers
of the packed bed. At the exit of the rst layer, the ow is observed to be strongly heterogeneous, and an important portion
of the section corresponds to local velocities more than 5% different from the average velocity. Three layers of packings are
needed to ensure a uniform ow. One also observes that the
velocity contours are oriented along the x direction at the exit
of the rst packing layer, and along the y direction at the exit
of the second packing layer. This is due to the fact that the
equivalent porous media is turned by 90 between each layer
as previously discussed. This type of simulations considering
the structured packing as an anisotropic porous media is found
to be able to reproduce the impossibility for the gas ow to
redistribute itself across the section in the direction normal to
the plates which is characteristic of structured packings. Fig. 9
shows the impact of packing layers orientation. In case (a) the
rst packing layer is oriented such that packing walls are normal to the inlet tube direction; in case (b) the rst packing layer
is oriented such that packing walls are parallel to the inlet tube
direction. The above packing layers are turned by 90 to each
other as usual. The colours correspond to the velocity contours
ranging from 1.35 to 1.7 m s1 . One observes that in the rst
case strong heterogeneities are observed along the x direction,
while they are along the y direction in the second case. More

importantly, the heterogeneities are observed to be stronger for


the case (a) than for the case (b). From this type of calculations, it thus possible to choose the most appropriate assembly
to avoid local ooding of the column. Or it is possible to test
different gas distributors design (not shown here) to obtain the
more homogeneous ow as possible.
Note that the total pressure drop is 22.2 mbar. This shows that
the pressure drop may be important even for a moderate size
column (4 m in total height). Singular pressure drops at inlet
and outlet are not negligible in such columns where the packing
induces pressure drop as low as 2 mbar m1 . This shows that
such calculations make optimised design of distributing devices
possible. A good design would correspond to cases where the
ow is as homogeneous as possible within the shortest distance
in the packed bed for minimum distributor pressure drop. To
do so, it is of course very important to take the packing into
account as showed in the present study and not the distributor
only as it is commonly done.
4. Conclusions
This study proposes a methodology that enables representative CFD calculations at large scales. It is based on different types of calculations run at three different scales, the
result obtained at one scale being used at the larger scale.
The methodology gives satisfactory results in the case of low
gasliquid interaction, that is below the loading point and for
low and intermediate liquid ows. Since these two conditions
are those that would be encountered in CO2 capture plants, the
present strategy is well adapted for calculations of gasliquid
ow in column containing structured packings. It will allow

7204

L. Raynal, A. Royon-Lebeaud / Chemical Engineering Science 62 (2007) 7196 7204

for the development and design of appropriate packings and


distributors.
References
Alix, P., Raynal, L., 2006. Liquid distribution and liquid hold up in a high
capacity structured packing. Presented at the 8th International Conference
on Green House Gas control Technologies (GHGT 8), Trondheim, 2006,
June, 1922.
Aroonwilas, A., Veawab, A., Tontiwachwuthikul, T., 1999. Behavior of the
mass transfer coefcients of structured packings in CO2 absorbers with
chemical reactions. Industrial and Engineering Chemistry Research 38,
20442050.
Ataki, A., Bart, H.J., 2006. Experimental and CFD simulation study for the
weeting of structured packing element with liquids. Chemical Engineering
and Technology 29 (3), 336347.
Billet, R., 1995. Packed Towers. VCH, Weinheim, Germany.
Bonilla, J.A., 1993. Dont neglect liquid distributors. Chemical Engineering
Progress, 4761.
Bravo, J.L., Rocha, J.A., Fair, J.R., 1985. Mass transfer in gauze packing.
Hydrocarbon Processing, 9195.
Fair, J.R., Bravo, J.L., 1990. Distillation columns containing structured
packing. Chemical Engineering Progress, 1929.
Freguia, S., Rochelle, G.T., 2003. Modeling of CO2 capture by aqueous
monoethalonanmine. A.I.Ch.E. Journal 49 (7), 16761686.
Petre, C.F., Larachi, F., Illiuta, I., Grandjean, B.P.A., 2003. Pressure drop
through structured packings: breakdown into the contributing mechanisms
by CFD modeling. Chemical Engineering Science 58, 163177.
Pozrikidis, C., 1988. The ow of a liquid along a periodic wall. Journal of
Fluid Mechanics 188, 275300.

Raynal, L., Harter, I., 2001. Studies of gasliquid ow through reactor


internals using VOF simulations. Chemical Engineering Science 56,
63856391.
Raynal, L., Boyer, C., Ballaguet, J.-P., 2004. Liquid holdup and pressure
drop determination in structured packing with CFD simulations. Canadian
Journal of Chemical Engineering 82, 871879.
Sidi-Boumedine, R., Raynal, L., 2005. Inuence of the viscosity on the liquid
hold-up in trickle-bed reactors with structured packings. Catalysis Today
105, 673679.
Spiegel, L., Meier, W., 1992. A generalized pressure drop model for structured
packings. IChemE Symposium Series No. 128, B85B94.
Stichlmair, J., Bravo, J.L., Fair, J.R., 1989. General model for prediction of
pressure drop and capacity of countercurrent gas/liquid packed columns.
Gas Separation and Purication 3, 1928.
Suess, P., Spiegel, L., 1992. Hold-up of Mellapak structured packings.
Chemical Engineering and Processing 31, 119124.
Tobiesen, F.A., Svendsen, H.F., 2006. Study of a modied amine based
regeneration unit. Industrial and Engineering Chemistry Research 45,
24892496.
Valluri, P., Matar, O.M., Hewitt, G.F., Mendes, M.A., 2005. Thin lm ow over
structured packings at moderate Reynolds numbers. Chemical Engineering
Science 60, 19651975.
Wehrli, M., Hirschberg, S., Schweizer, R., 2003. Inuence of vapour feed
design on the ow distribution below packings. Chemical Engineering
Research and Design 81 (1), 116121.
Zhao, L., Cerro, L.R., 1992. Experimental characterization of viscous lm
ows over complex surfaces. International Journal of Multiphase Flow 18
(4), 495516.

Вам также может понравиться