You are on page 1of 10

European Congress on Computational Methods in Applied Sciences and Engineering

ECCOMAS 2000
Barcelona, 11-14 September 2000
ECCOMAS

A PORE NETWORK MODEL FOR DRYING PROCESSES


IN POROUS MEDIA
A.G. Yiotis*,#, A.K. Stubos*, and A.G. Boudouvis#
*

National Center For Scientific Research Demokritos


15310 Agia Paraskevi, Athens, Greece
e-mail: giotis@chemeng.ntua.gr
#

Department of Chemical Engineering


National Technical University of Athens
15780 Zografos, Athens, Greece
e-mail: boudouvi@chemeng.ntua.gr

Key words: Drying, Porous media, Pore-network model, Capillarity, Viscous forces.
Abstract. A numerical simulator for the drying of capillary porous media under isothermal
conditions is presented. The simulator is based on the discrete model approach and accounts
for most of the mechanisms that have been identified to play an important role during the
drying process. These include mass transfer by diffusion in the gas phase, viscous flow in the
gas and liquid phase, evaporation at the moving interface and capillary effects at the pore
throats. It has been shown in past works that viscous effects may have non-negligible
contribution on the phase distribution patterns occurring during the drying process and also
upon the overall drying time of the media. The main purpose of this work is to elucidate
through a series of numerical simulations the relative significance of capillary and viscous
forces on phase distribution patterns and evaporation rates.

A.G. Yiotis, A.K. Stubos, and A.G. Boudouvis

INTRODUCTION

Drying of porous media is a subject of significant scientific and technological interest.


Industrial applications include drying of coatings, food, paper, textile, wood, ceramics,
building and granular materials and many more. In a similar context, the problem is of direct
relevance to the recovery of volatile hydrocarbons from oil reservoirs by gas injection.
In the past, efforts to model the drying process in porous materials have been based on a
continuum description of the media and the estimation of phenomenological and empirical
parameters that correlated drying times to pressure and concentration gradients1,2. The
drawback of that approach is the need for a series of experimental data, in order to estimate
phenomenological parameters, such as permeability of the sample, which are applicable to
only a short range of materials. This approach does not require deep understanding of the
physics and mechanisms involved in the process at the microscale.
More recently, numerical simulators have been developed based on the discrete porenetwork model. The porous medium is represented by a network of pores connected through
necks of varying sizes. This approach, although more computational costly than that of
continuum description, offers better understanding of the physics involved and the effects of
the microstructure3-10.
We consider a porous medium initially fully saturated with a volatile wetting fluid.
Through an open side of the porous material, an inlet gas is injected. The fluid evaporates at
the interface formed between gas and liquid phases, and the gas phase gradually penetrates the
porous structure replacing the liquid. Experiments performed by previous researchers on porenetwork micromodels,4-7 packed beads11, and realistic core samples12 have revealed that,
provided that gravity is negligible as it is the case in this work, capillary forces and viscous
forces control the receding of the gas-liquid interface.
Capillary forces create an interfacial pressure difference across an interface between the gas
and the liquid phase. A typical interfacial pressure difference is
Pint

(1)

where is the surface tension and r is a typical dimension at the pore scale. It is evident from
equation (1) that capillary forces are stronger as r gets smaller.
The effect of viscous pressure drop in both the liquid and the gas phases is significant to
the evolution of the drying process. In the gas phase, pressure gradients favor mass transfer by
advection and accelerate drying. In the liquid phase, pressure gradients have been reported to
stabilize the front through the so-called capillary pumping1,12.
Neglecting gravity, three main dimensionless groups govern the process13: (i) a diffusionbased capillary number Ca, expressing the ratio of viscous to capillary forces, based on a
diffusion-driven velocity and defined as
Ca = D" C "e /""

(2)

A.G. Yiotis, A.K. Stubos, and A.G. Boudouvis

where " and " is the viscosity and density of the liquid phase respectively, C "e is the
equilibrium concentration, " is the distance between pore centers and D is the diffusion
coefficient in the gas phase, (ii) a Peclet number, Pe, expressing the ratio of advection to
diffusion in the gas phase, based on the linear gas flow velocity in the fracture;
"

Pe = V f
D

(3)

where Vf is the linear gas velocity through the fracture. and (iii) the viscosity ratio, M, between
liquid and gas viscosities. The latter is typically large and will not be considered in the
sensitivity analysis to follow.
In general, slow drying of a large pore network has been identified as an Invasion
Percolation in a Stabilizing Gradient (IPSG) process. The front is stabilized by very slow
viscous flows in the liquid phase that are the result of capillary pumping, and the steepness of
the concentration gradients in the gas phase1,9,14.
2 FORMULATION
In the present paper, the porous medium (Figure 1A,B) is represented by a 2-D square
lattice of spherical pores connected through cylindrical throats, as shown in Figure 1C. The
pores serve as containers for either of the two phases and it is assumed that they have no
capillary or flow resistance. The throats serve as conductors of the flow and mass transfer and
they act as capillary barriers. This discrete representation of the porous solid can be easily
extended to a 3-D geometry, although that would be computationally costly and would not
offer much to the numerical results.

Figure 1: Schematic of a porous medium and its network representation

A.G. Yiotis, A.K. Stubos, and A.G. Boudouvis

The network has three impermeable sides. Along the fourth side runs a fracture, which is
represented by a 1-D chain of pores and throats. An inlet gas (air) is injected at various flow
rates at the one edge of the fracture, which is represented by the first throat of the 1-D chain.
The network is initially saturated with a single component volatile liquid hydrocarbon
(hexane). The fracture is assumed to contain only gas at the beginning of the drying process.
The liquid evaporates at the interface between the liquid and the gas phases. A twocomponent gas phase, consisting of the inlet gas and the vapors of the liquid, is then formed.
Liquid vapors are transferred by advection and diffusion towards the exit of the fracture. At
the same time, the gas phase replaces the liquid in the pores. The two-component gas phase
exits the network from the last throat of the 1-D chain.
For drying applications, we can further distinguish three types of pore bodies: Those fully
occupied by gas (belonging to the gas phase and denoted by G), those fully occupied by liquid
(belonging to the liquid phase and denoted by L) and those at the gas-liquid interface (in
which a meniscus resides, denoted by I). The latter may be further subdivided in completely
empty (CE) and partly empty (PE) pores. In the simulations, pore body and throat radii were
uniformly distributed in the range 0.37-0.74 mm and 0.16-0.32 mm respectively. The lattice
length " (pore center to pore center) was taken to be 2 mm. The fracture was also represented
as a pore network (here a 1-D chain), with pore and throat radii taken equal to 0.77 mm and
0.275 mm, respectively. We must point out that this fracture representation may not
necessarily apply to a real problem, where larger sizes should be expected, but it was taken
here for convenience of the sensitivity analysis to follow.
Mass transfer of the vapor in the gas phase obeys the convection-diffusion equation
C
+ u C = D 2 C
t

(4)

where C is the vapor concentration, u is the gas-phase velocity and D is the diffusion
coefficient. In G pores ,this is further discretized as
Vi

rij4 ( Pi Pj )
Ci C j

C i
= Drij2
+
C ij

" j
t
8 "
j

(5)

In I pores of CE type, the convective term is not accounted for and equation (4) is discretised
as
Vi

Ci C j

C i
= Drij2

"
t
j

(6)

where Vi is the volume of pore i, Ci is the change in Ci during the elapsed time t, rij is the
radius of the throat connecting pores i and j, is the gas viscosity, P is the pressure and C is
the concentration at the pore. Note that the advection term is upstream weighted, namely
C ij = Ci

if Pi>Pj ,

and

C ij = Cj

if Pj>Pi

A.G. Yiotis, A.K. Stubos, and A.G. Boudouvis

The single-component liquid in the liquid phase and the non-condensable gas in the gas
phase satisfy continuity equations. By expressing them as shown below, we obtain the
pressure fields in G and L sites. Fluxes between adjacent sites of the same type are computed
by Poiseuille-law type flow resistances, where the viscosity is considered constant
Pi Pj
Qij =
"

ij

rij

=0

(7)
(8)

SIMULATION PROCEDURE
The drying simulation procedure in the network can be summarized as follows;

1. A random network of pores and throats is created using a uniform size distribution between
the specified limits.
2. The pressure drop along the fracture is calculated using equations (7) and (8).
3. All isolated liquid clusters are located and identified.
4. Flow conductivities at all network throats are calculated. When an interface is pinned
within a throat then zero flow conductivity is assigned to it.
5. The type and number of pores on each interface is identified. When all interface throats are
pinned at a specific liquid cluster, then we set pressure to zero at all pores within that cluster.
6. We find all throats where gas penetration occurs as the local capillary pressure exceeds the
capillary pressure threshold. If there is no such throat at a specific cluster, then the gas
penetrates at the throat that has the lowest capillary threshold along the interface, i.e. the one
with the largest radius.
7. The evaporation rate from every pore across all interfaces is calculated using equation (6).
8. We solve for the pressure field in both liquid and gas phases taking into account the
volumetric rate of evaporating liquid at all interface pores using equations (7) and (8).
9. We locate the pore that will be emptied first at the current flow rates and find the time
required.
10. The time step is selected taking care not to allow the concentration value at a gas pore to
drop bellow zero or exceed the equilibrium concentration.
11. The concentration field is calculated in the gas phase using equation (5) for the current
time step.
This calculation is straightforward based on the values of concentration at the end of the
previous time step.
12. Gas saturation (fraction) is calculated at all pores on the interface for the current flow rates
using the current time step.
13. If at the end of the current time step a pore becomes fully filled with gas at a specific

A.G. Yiotis, A.K. Stubos, and A.G. Boudouvis

liquid cluster, we return to step 3. If, despite the fact that gas saturation at all penetrated
interface pores increases, there is no pore completely filled with gas at the current time step,
then we return to step 6.
The above procedure is repeated until all liquid in the porous media has evaporated.
4

RESULTS

A number of runs were conducted to simulate drying of liquid hexane in a matrix block of
size 50x50 when air is injected through the fracture. The representative test cases selected for
the present study are shown in Table 1.
A characteristic time t* is used to non-dimensionalize time. This corresponds to the overall
time required to fully empty the network at the maximum possible evaporation rate. The
maximum evaporation rate, as shown in the following results, occurs during the first time step
at the matrixfracture boundary.
Pe

Q
(m /s)*10-6
0.45
0.25
0.20
0.10
0.05
0.015
0.005
0.0005
0.0
3

1
2
3
4
5
6
7
8
9

596
331
265
132
66
19
6.7
0.66
0.0

t*
(s)
17983
21340
22742
27645
33325
45611
57819
88752
166132

Table 1. Test cases

We study the phase distribution patterns and drying rates at different gas flow rates Q. The
corresponding physical parameters of the problem are shown in Table 2.
Surface tension
Diffusion coefficient in gas phase
Equilibrium concentration
Liquid phase viscosity
Gas phase viscosity
Liquid phase density
Gas phase density

19*10-3 N/m
6.38*10-6 m2/s
2.66*102 g/m3
2.85*10-4 Pa*s
1.71*10-5 Pa*s
0.650*10-6 g/m3
4.4*10-6 g/m3

Table 2. Physical parameters for the test cases

Runs 1 and 8 are typical of two limiting regimes. The first (run 1 with high Peclet number,
Figures 2,3) is characterized by dominant viscous forces and advective mass transfer in the
gas phase. The second (run 8 with low Peclet number, Figures 4,5) corresponds to dominant

A.G. Yiotis, A.K. Stubos, and A.G. Boudouvis

capillary effects and diffusive mass transfer.


In run 8 capillary forces control the invasion process. As shown in Figure 4, the cluster
takes the pattern of Invasion Percolation (IP), in which the next throat to be invaded by the gas
is that with the smallest capillary threshold (here, the one with the largest radius) among all
perimeter throats of that cluster. Figure 2 shows that due to the dominance of the viscous over
capillary forces in run 1, the phase distribution pattern is mainly influenced by the pressure
gradients in the gas phase.
Regarding the concentration fields, diffusion is the main mechanism of mass transfer in the
gas phase in run 8. This is reflected in the smooth profiles of Figure 5. In contrast, Figure 3
presents the concentration profiles of run 1, which are quite steep close to the interface and
almost flat away from it. This signifies the advective character of mass transfer in the bulk of
the gas phase.

Figure 2: Phase distribution patterns for run 1


(black is liquid, white is gas)

Figure 3: Concentration contours for run 1

Figure 4: Phase distribution patterns for run 8


(black is liquid, white is gas)

Figure 5: Concentration contours for run 8

A.G. Yiotis, A.K. Stubos, and A.G. Boudouvis

Runs 1 and 8 being typical of the two limiting regimes can be interpreted in a relatively
straightforward fashion. Intermediate patterns are more difficult to analyze, as they process
features from both regimes. A typical case in which viscous and capillary forces compete is
run 2, shown in Figures 6 and 7. The invasion pattern belongs to neither of the two limiting
regimes. There is evidence of IPSG in the matrix, similar to run 1, but also a multiple number
of isolated liquid clusters, similar to run 8.

Figure 6: Phase distribution patterns for run 2


(black is liquid, white is gas)

Figure 7: Concentration contours for run 2

Figure 8 shows drying curves for the various simulated cases. As already mentioned, the
time was made dimensionless with the time t* it would take to empty the matrix block under
conditions of the maximum rate. It then follows that the slope of the drying curves is the
dimensionless rate of drying (relative to the maximum drying rate). Note that because the
maximum rate depends on the Peclet number in the fracture this plot emphasizes processes
under convection control and should be interpreted with care, as far as the upscaling of the
process is concerned. The figure shows clearly the existence of a Constant-Rate-Period (CRP),
followed by a period of continuously declining rates. From an analysis of the patterns, we
have found that the CRP lasts roughly until the time when the main liquid cluster has lost
continuity with the fracture. This is in agreement with indirect experimental findings from
tests on chalk samples initially containing liquid pentane and dried by methane injection along
the fracture12. This is evident when considering the low Peclet number run 8 or the IP run 9 in
Fig. 8 where the CRP is very short due to the fast receding of the liquid phase from the
fracture.

A.G. Yiotis, A.K. Stubos, and A.G. Boudouvis

Figure 8: Drying curves for the different test cases

CONCLUSIONS

A numerical simulator for the description of drying processes in capillary porous media is
presented in the present work. The simulator is based on the discrete network model approach
and accounts for all the main mechanisms that have been identified to play an important role
in the process. These include mass transfer by diffusion in the gas phase, viscous flow in the
gas and liquid phase, evaporation at the interface and capillary effects at the pore throats.
Compared to previous studies in the literature, the main novelty of this work is the
introduction of viscous forces at the pore and pore-network level. Indeed, viscous effects may
have a major contribution on the phase distribution patterns occurring during drying as well as
on the obtained drying curves with time. We present two limiting cases regarding the relative
influence of capillary and viscous forces (low and high Peclet number respectively) and
examine the observed changes in the phase distribution patterns and evaporation rates.
Intermediate cases are also shown and discussed. The existence of Disconnected Liquid
Clusters, the type of concentration profiles in the gas phase and the relative extent of constant
rate and continuously declining periods are noted. Whenever possible, reference is made to the
relation of the results to the usually invoked in drying studies Invasion Percolation process.
REFERENCES
[1] Tsimpanogiannis, I. N., Yortsos, Y. C., Poulou, S., Kanellopoulos, N., Stubos, A. K.,
Phys. Rev. E., 59 (1999) 4353.

A.G. Yiotis, A.K. Stubos, and A.G. Boudouvis

[2] Panagiotou, N. M., Stubos, A. K., Bamopoulos, G., Maroulis, Z. B., Drying Tech., 17
(1999) 2107.
[3] Prat, M., Int. J. Multiphase Flow, 21 (1995) 875.
[4] Haghigi, M., Xu. B. , Yortsos, Y.C., J. Coll. Interf. Sci, 166 (1994) 168.
[5] Laurindo, J. B. & Prat, M., Chem. Eng. Sci., 51 (1996) 5171.
[6] Laurindo, J. B. & Prat, M., Chem. Eng. Sci., 53 (1998) 2257.
[7] Li, X. & Yortsos, Y. C., Chem. Eng. Sci., 50 (1995) 1247.
[8] Nowicki, S. C., Davis, H. T., Scriven, L. E., Drying Tech., 10 (1992) 925.
[9] Prat, M., Bouleux, Phys. Rev. E, 60 (1999) 5647.
[10] Satik, C. & Yortsos, Y. C., J. Heat Trans., 118 (1995) 455.
[11] Shaw, T. M, Phys. Rev. Lett., 59 (1987) 1671.
[12] Le Gallo, Y., Le Romancer, J. F., Bourbiaux, B., Fernades, G., paper SPE 38924 (1997).
[13] Yiotis, A. G., Stubos, A. K. , Boudouvis, A. G., Yortsos, Y. C., A 2-D Pore Network
Model of the Drying of Single-Component Liquids in Porous Media (submitted for
publication to AWR special issue)
[14] Xu, B., Yortsos, Y.C., Salin, D., Phys. Rev. E, 57 (1998), 739.

10

You might also like