You are on page 1of 10

SPE 35531

Society of PetrolewmEngineers

Prediction of Relative Permeability and Capillary Pressure from a Pore Model


L. Scandellari

Ndsen, Stavanger

lCopyrbghl 1!336, Soctety of Pelroleum

University,

P.E. Oren and S. Bakke, Statoil Research

Engrieers

Th!s papel was prepared for presenlatlw

al Ihe European 3-D Reservoir

Iorma!lon

contafned

m an abslracl

Modellmg Conference

by the SPE Program Commmee

submitted

by the author(s)

folhwmg

Contents

rewew of In-

of the paper as pre-

sented, have not been rewewed by the Soctely of Pelroleum Eng,neers and are sublecf to
correction by the author(s) The ma ferlal, as presented, does nol necessarily reflect any Psi!on
of the Society of Pe!roleum

Engineers

or

Its

members

subject to publ,cahon cevlew by Ed!tonal Commmee

Papeis presented

m]ss)on to copy IS reslocled to an abstract of not more Ihan 3M words


copied The abslracl should contain conspicuous
per was presen!ed

al SPE mee!lngs are

of the Scmely of Pelroleum

acknowledgment

Engfneers

Illuslral[ons

Per-

may nol ba

01 where and by whom the Pa

Write L,brar, an, SPE, P O 8333.936 Rchardson

TX 75083.3836

and A. Henriquez,

Statoii as.

with the measurements. This prohibits assigning unique relative


permeability functions to different architectural units in the reservoir (i.e., channels, crevasse splays, wash-over fans, etc.) and
limits the ability of reservoir simulators to accurately predict oil
recovery.
Relative permeabilities and capillary pressure are averaged
transport properties which represent the physical processes occurring on the pore-scale. On the pore-scale, the displacement of
one fluid by another is controlled by interracial tension, viscous
forces, rock-fluid interactions, and the geometry of the pore
space. In principle, it should therefore be possible to determine
relative perrneabilities and capillary pressure by appropriately
averaging the equations describing the physical processes occurring on the microscopic or pore-scale. This approach requires a
detailed understanding of the displacement mechanisms on the
pore-scale and a complete description of the morphology of the
pore space The procedure has successfully been applied to twoand three-phase flow in simple or idealised porous media using
pore-scale physics identified in micromodel experiments with
the morphology of the pore space represented by a topologically
equivalent numerical network .
lhe difficulty in constructing a realistic three-dimensional
(3-D) representation of the complex pore structure of real porous
rocks has limited the above approach to idealised porous media.
Although advanced techniques such as serial- sectioning and
micro-CT $ are available, information about the pore structure of
porous rocks is usually obtained from image analysis of 2-D thin
section images of rocks and from mercury injection capillary
pressure curves. Thin sections provide aerial information which
is relevant to porosity measurements whilst mercury injection
data provide information about the volume of pores which may
be invaded through pore throats within specified size ranges.
These data are insufficient to provide a complete 3-D description
of the architecture and geometry of the pore space and do not allow construction of 3-D pore networks which accurately represent the complex pore space of a given porous rock,
In the present work, stochastic modelling of the main
sandstone-forming
geological processes (i.e., sedimentation,
compaction, and diagenesis) are combined with 3-D image analysis techniques to generate a realistic and fully characterised 3-D
representation of the pore network for a Berea sandstone. The

held In Stavanger, Norway, 1617 APrIl, 1996


This paper was selected for presenlallon

Centre,

USA, lax

01-214-9529435

Abstract

A pore-scale model consisting of a network of pore bodies interconnected by pore throats is used to calculate drainage relative
permeabilities and capillary pressure for a strongly water wet Berea sandstone core, The architecture and geometry of the pore
network which is used in the model is constructed from thin section analysis and numerical modelling of the main sandstoneforming geological processes, i.e., grain sedimentation, compaction, and diagenesis The effect of different pore network descriptors on relative permeability at low capillary numbers has
been simulated. The results show [hat pore shapes strongly influence wetting phase relative permeability, particularly at low saturations where film flow is important. Simulated relative
permeabilities are found to be in good agreement with those predicted from an empirically derived correlation.
introduction

Macroscopic multiphase flow in porous media is usually described in terms of Darcys law and measured or empirically derived saturation dependent relationships for phase relative
permeabilities and capillary pressure. Accurate and consistent acquisition and interpretation of such data are essential for almost
all reservoir engineering calculations and determine to a large extent how reservoir management can optimise oi I production and
recovery.
Relative permeability measurements, either by steady state or
unsteady state methods, are time consuming, expensive, and often difficult to inte~ret. As a result, too few measurements are
usually performed and numerous uncertainties may be associated

345

PREDICTION OF RELATIVE PERMEABILITY AND CAPILLARY PRESSURE FROM A PORE MODEL

random X,Y position. Upon hitting one of the grains in the sandbed, it rolls down the steepest local gradient until it reaches a stable position (i.e. local minimum). Figure 1 shows an example of a
sandpack generated from simulation of a low energy sedimentation process. The process results in a relatively loose or opeo
packing of grains and the porosity of the resulting sandpack is
high (38Yo).
Low energy sand sedimentation is, however, rare. Normally,
the sedimentation of sand is classified as a high energy geological
event which is influenced by lateral forces such as streams,
waves, and gravity (dipping bed). These lateral forces will move
individual grains around until they settle in a position where the
effect of the transporting forces is at a minimum (to avoid erosion). This process is implemented by simply placing each new
grain at the lowest available position (global minimum) in the existing sandbed. Figure 2 shows an example of a sandpack generated from simulation of a high energy sedimentation process. The
same grain-size distribution as in Figure 1 was used for the modelling. Compared with Figure 1, the process results in a denser
packing of grains and the porosity of the sandpack is reduced
(35%).
In general, fine sand or silt deposition is modelled as low energy sedimentation whilst sand and coarse sand deposition is
modelled as high energy sedimentation.

pore network is used as input to an invasion percolation based


network flow simulator to compute drainage capillary pressure
and relative penneabilities. The main purpose of the present paper is to illustrate how a detailed description of a rocks microstructure may be combined with network flow simulations to predict capillary pressure and relative permeabilities tlom the underlying pore-scale physics. This offers the possibility
of
overcoming the serious experimental difficulties encountered in
measuring two- and three-phase relative permeabiiities and of
providing realistic estimates of transport properties for core material which is not suited for laboratory testing.
Sandstone

Pore Network

SPE 35531

Modelling

Sandstones are the end products of a series of complex geological


processes which, schematically described, starts with erosion of a
(quartz bearing) rock, followed by grain transport via air, water,
or ice, and, finally, sedimentation of grains in a basin. The deposited grains may be reworked by one or several cycles of erosion,
transport, and sedimentation. The sandstone generation process is
completed by compaction and various diagenetic processes. lnforrnation regarding the last sedimentation cycle and subsequent
compaction and diagenesis may be interpreted from thin section
analysis. Although there are still unveiled sides of these processes, they are relatively well described and understood. In the
following, we give a brief description of how numerical modelIing of these processes maybe combined with 3-D image analysis
techniques to generate a realistic 3-D pore network for a given
rock, A complete description of the methodology is given
elsewhere*

Compaction: Compaction during burial in response to vertical


stress causes a decrease in the porosity of sandstones. Compaction is an intricate mechanical process which involves grain rearrangements, fracturing, and pressure solution. Currently, the
effects of compaction are modelled as a linear process in a manner similar to that described by Bryant et af.7. The z-co-ordinate
of each grain centre is shifted vertically downwards by an amount
proportional to the original z-co-ordinate

Sedimentation: The modelling of the sedimentation process


commences with the generation of a grain-size distribution determined from image analysis of thin sections from a given rock. At
present, all the detrital grains are treated as spherical quartz
grains. In modelling sedimentation, sand grains are picked randomly from the grain-size distribution. This mimics the stochastic
nature of the geological processes which determines the deposition ofa sand grain onto a sandbed.
The exact dynamics and geometry which determine the deposition of a sand grain onto an existing sandbed are complex and
can only be solved by numerical means at considerable computational expense. To numerically simplifi the problem, the radius
of each new grain is reduced to zero prior to deposition whilst the
radii of all the grains in the sandbed are increased by an amount
equal to the radius of the new grain. This reduces each new grain
to a point and the sandbed to a solid surface which dynamically
and geometrically are easier to describe. After the new grainpoint has been deposited, all grain radii are reset to their original
values and the above procedure is repeated for the next grain.
The exact location in the sandbed where each new grain is deposited depends on whether the sedimentation process occurs in a
low or high energy environment. For low energy sedimentation, it
is assumed that the horizontal velocity of the grain is zero and
that no lateral forces affect the sandbed. In this case, each new
randomly selected grain is dropped onto the sandbed from a

z= 2.(1 -A) ................................................................. (1)


where Z is the new z-co-ordinate, Z,, is the original z-co-ordinate,
and L is the compaction factor which varies between 0.0 and 1.().
Compaction results in grains intersecting. The material corresponding to the grain intersection is assumed to either be dissolved and washed out of the system or it may be deposited as
quartz cement overgrowth (see below). The effect of different
compaction factors is illustrated in Figure 3 and it shows that
compaction significantly reduces porosity.
Diagenesis: The alterations to the pore space which results from
diagenetic processes are complex and depend on how quickly
minerals are transported to and from the sandstone deposit, the
rate of crystal growth, dissolution of minerals and cement overgrowth. At present, only a simplified subset of known diagenetic
processes have been implemented: quartz cement overgrowth and
clay coating of the free surface.
Quartz overgrowth is modelled in a manner similar to that of
Bryant et d.7 where the radii of all the sand grains are increased
uniformly. The amount of quartz cement overgrowth may be
346

SPE 35531

L. SCANDELLARI NiLSEN, P.E, OREN, S. BAKKE, A. HENRIQUEZ

assumed to be equal to the material corresponding to the intersecting grains during compaction or it maybe specified by acement thickness input parameter. The latter allows for migration
of SiOz to and from the sandpack.
Clay coating is modelled by randomly precipitating clay minerals (clay voxels) on the free quartz, quartz cement, or existing
clay surface. The amount of clay which is precipitated is determined from thin section analysis. In thin sections, clay is sometimes observed to be concentrated to certain areas whilst other
areas are almost clay free. This type of behaviour is implemented
via a clustering routine which increases the likelihood of new
clay voxels to be precipitated on already existing clay surfaces.

G/=$

...................................................................... (2)

where A is the area of the cross-section and P is the perimeter


length. Tbe area and perimeter of every pore throat are determined from geometrical analysis and the corresponding shape
factors are computed and recorded. The value of the shape factor
ranges from zero for a slit shaped triangle to 0.048 for and equilateral triangle. For a square pore, the shape factor acquires the
value 0,062. For every pore throat, the minimum radius, r,, which
can be inscribed in the pore space connecting two pore bodies is
identified. This is the relevant radius for computing drainage
threshold capillary pressures and throat conductivities (see discussion below).
Similar to pore throats, pore bodies are modelled as having triangular cross-sections with dimensionless shape factor Gmand inscribed radius r. (rn>r,), The radius rn is defined as the maximum
radius which can be locally inscribed in the pore space. This is
the relevant radius for computing pore volumes and pore body
threshold capillary pressures during imbibition. The pore body to
pore throat aspect ratio (i.e. m/r,) strongly influences hysteresis
in capilla~ pressure and relative permeability and is also measured and recorded.
The above description of the pore network allows all the experimentally observed features of two-phase (and three-phase)
flow to be model led. These include the distribution of pore throat
and body threshold pressures, the simultaneous flow of wetting
and non-wetting fluids in pores, and the dependence of film flow
on the prevailing capillary pressure, A schematic ball-and-stick
representation of a numerical pore network is shown in Figure 7,

and Geometry: Figure 4 shows an


isosurface representation of the grain matrix for the grainpack
shown in Figure 2. Quartz is depicted as medium grey, quartz cement as light grey whilst clay minerals are dark grey. Despite the
necessarily approximate treatment of the complex geological
processes which govern the format ion of sandstones, the model is
remarkably similar to 3-D representations of real sandstones produced by more involved and time consuming methods such as serial sectionings and micro-CT
The grain matrix of a sandstone defines a spatial network
which is complementary to its pore space network. Figure 5
shows an isosurface representation of the complimentary pore
space network for the sandstone model shown in Figure 4. The
architecture and geometry of the pore space are very complex
and far tlom regular.
The skeleton of the pore space may be regarded as the graph,
or the reformational retract, of the pore network and it forms the
basis for analysing the architecture and geometry of the pore
space. The network skeleton corresponding to the pore space network shown in Figure 5 is depicted in Figure 6. 3-D image analysis techniques are used to extract the network skeletond. The
spatial lines in the skeleton correspond to the centreline of the
pore throats. The points where two or more pore throats meet define the pore bodies in the network. The connectivity or coordination number for each pore body is determined directly
horn the pore network skeleton.
To simulate fluid flow in the pore network, it is necessary to
replace the highly irregular geometry of the pore throats and pore
bodies by an equivalent idealised geometry which leads to
mathematically tractable problems but retains the essential features of the pore space which are relevant to fluid flow. Triangles
are an example of an idealised pore shape which allow the basic
features of capillary dominated flow to be modelled. They have
angular corners which can retain wetting fluid and thus allow
both wetting and non-wetting fluids to flow simultaneously
through the same pore. Pore throats are therefore modelled as
long narrow channels of triangular cross-section. The length I of
each throat is defined as the distance between the centres of its
two connecting pore bodies and is determined from the pore network skeleton. The exact shape of the throat cross-section is defined in terms of a dimensionless shape factor G, defined as
Analysis of the Architecture

Fluid Flow in the Network


Fluid Distribution: To effectively model fluid flow on the pore-

scale, it is necessary to know the distribution of the fluids within


the pore space. The local equilibrium distribution of two immiscible fluids in a porous medium is determined by nettability and
capillary pressure. For the present discussion, it is assumed that
the wetting fluid perfectly wets the pore solid. The fluid distribution is therefore governed by capillary pressure alone.
When a non-wetting fluid is located in a pore body which is
connected to a pore throat filled with wetting fluid, capillary
forces prevent the non-wetting fluid from spontaneously entering
the pore throat. The non-wetting fluid can only enter the pore
throat when the pressure in the non-wetting fluid exceeds the
pressure in the wetting fluid by a value equal to the threshold
capillary pressure. For a triangular pore throat, the threshold capillary pressure, P, , is given by

lc=acose

1+2~
,,

..... .. .. .. ...... . .. ....... ........ .. .. .. .....

(3)

where o is the interracial tension, 6 is the contact angle, and r, is


the minimum inscribed radius of the pore throat.
Figure 8 shows the pore-scale fluid distribution of oil and water in a strongly water wet micromodel. Water is clearly the wetting phase. [t preferentially occupies the smaller pore throats and
347

SPE 35531

PREDICTION OF RELATIVE PERMEABILITY AND CAPILLARY PRESSURE FROM A PORE MODEL

dius of the cylinder whose volume is equal to the volume of the


fluid in the pore, i.e.,

forms continuous wetting films which cover the glass surface


throughout the entire porous network. The hydraulic continuity of
the wetting phase, clearly observable in glass micromodels, has
been confirmed in studies using actual porous rocks.
Oil is the non-wetting phase and preferentially occupies the
larger pore bodies and some of the interconnecting pore throats.
We note that some wetting fluid is always retained in the comers
of non-wetting fluid tilled pores. As a result, both wetting and
non-wetting fluid may flow simultaneously through non-wetting
fluid tilled pores. This ensures mobility of the wetting fluid down
to very low saturations.

where A, is the cross-sectional area open to flow of fluid i in the


pore (throat or body), For pores which are occupied by both fluids simultaneously, the cross-sectional areas open to flow for individual fluids are determined by the pore geometry and the
prevailing capillary pressure. For triangular pores the areas are
readily calculated from elementary geometry and are given by

Drainage: We want to simulate drainage displacements at low capillary numbers. At low capillary numbers, the
displacement is sufficiently slow for viscous losses associated
with the movement of an interface through a pore to be negligible, The condition for an interface to invade a pore throat, displace wetting fluid tlom the throat and reach the connecting pore
body, is then determined almost entirely by capillary pressure.
This forms the basis for the invasion percolation model for simulating drainage22.
Initially the network is fully saturated with wetting fluid. Nonwetting fluid is injected uniformly along the inlet side of the network whilst wetting fluid escapes through the outlet on the opposite side. For a displacement to proceed, it must be possible for
non-wetting fluid to flow to the displacement site and wetting
fluid to flow away from the displacement site to the outlet. injected non-wetting fluid is always continuous and flows from the
inlet to the displacement site through interconnected pore bodies
and throats filled with non-wetting fluid (bulk flow). For strongly
wetting conditions, the wetting fluid is hydraulically connected
throughout and flows away from the displacement site to the outlet either by bulk or film flow, or a combination of both.
At every stage of the displacement, a number of pore throats
are available for invasion by the non-wetting fluid. The pore
throat which is actually invaded is the one having the lowest
threshold capillary pressure (Equation (3)), The displacement is
thus driven by increasing capillary pressure and the simulation records the fluid distribution in the network as a fimction of capillary pressure.
Simulating

An=&[l-(fi)2,1-4XG)] . (5)

()

d,,=+

2
(1

-41@

. .(6)

where An and A. are the areas open to flow for the non-wetting
and wetting fluids, respectively, r is the inscribed pore radius,
and G is the dimensionless shape factor, The conductivity of fluid
i in a pore, g,, is given by Poiseuilles equation and may be written as
R;A,

g{ =
gp,/

. . . .. ...... .... ...... ... ..

. ..... . .

. . . (7)

where y, is the fluid viscosity and I is the length of the pore


(throat or body). The overall resistance to flow of fluid i between
the centres of two neighboring pore bodies / and J, g,,,,, is the
sum of the two pore body resistances and the pore throat resistance, i.e.,
1

.~++++
gli

................................ ................... (8)

where g,, is the conductivity of the pore throat or link and g,, and
g,,,are the two pore body conductivities.
The flow rate of fluid i between the centres of two pore bodies
or nodes, Q,,,,, is related to the pressures in the nodes by
Poiseuilles equation which maybe written as

Relative Permeability:
To compute relative permeabilities at a
given saturation (or capillary pressure), we need to define the hydraulic conductivity of the fluids in the network. As discussed
above, both wetting and non-wetting fluids may be flowing simultaneously in the same pore, The simultaneous flow of two fluids through a pore (throat or body) is complicated by the pore
geometry and the fact that the flow of individual fluids are coupled by momentum transfer across fluid-fluid interfaces. In the
present treatment we make the assumption that the flow is suftlciently slow for the interaction between fluids to be negligible.
Furthermore, we make the simplification that the resistance to
flow of fluid i in a pore may be characterised in terms of an
equivalent radius R,. The equivalent radius is defined as the ra-

fi2LlJ= gLIJ(p,,I - P,,J)

.......... ...... .... ... . . . . . (9)

where P,, and P,,, are the nodal pressures. Since the fluids are incompressible, flux conservation requires that

z, Q,,/J= o ......... .......................................... ......... (lo)


where J runs over all pore throats connected to pore body I.
Equations (9- 10), together with the appropriate boundary conditions, form a complete solution to the steady flow of an incompressible fluid in the pore network. The equations were solved
using successive over-relaxation.

348

SPE 35531

L. SCAN DELLARI NILS EN, P.E, 13REN, S. BAKKE, A. HENRIQUEZ

an equivalent 3-D numerical network which the flow simulations


were performed on, Pore throat radii were assigned according to
a log normal distribution where the maximum and average pore
throat radii were determined from analysis of the generated pore
network.
The permeability of the numerical network was approximately
800 mt) which is SIightly higher than the experimentally determined one (714 mD). In Figure 9, simulated relative permeabilities are compared with those predicted using Brooks and
Coreys: empirical correlation. Compared with the empirical
predictions, the simulator predicts a higher wetting phase relative
permeability and a lower non-wetting phase relative permeability. t.Unfortunately, no experimentally determined relative permeabilities were available and it is unclear if the empirically derived
or simulated relative permeabilities most closely resemble the acmal ones. Simulations were therefore performed to examine the
impact of different pore structure descriptors on the computed
relative permeabiiities,

The absolute permeability K of the pore network is calculated


from Darcys law}by imposing a unit pressure drop across the network and computing the resulting single phase flow rate

K=.Qp

.............................................................(n)

where Q is the fluid flux per unit area,


Relative permeabilities are computed in a similar manner. The
fluid distribution at a specified saturation (or capillary pressure)
is established by the network flowI simulator. A unit pressure
drop is applied to each of the phases and the individual phase
flow rates are computed. Relative permeability of fluid i is then
given by
k.,

= -~

.. . ...

,. ...,,, (]~)

Since the wetting fluid is continuous throughout the entire network (either by bulk or films), Equations (9-1 O) must be applied
at every node in the network when solving for the wetting phase
relative permeability. Unlike the uetting phase, the non-wetting
phase only occupies part of the network at any given saturation
and Equations (9-1 O) only need to be solved for nodes which are
occupied by non-wetting fluid.

Shape Factor: At low wetting phase saturations, the continuity


and mobility of the wetting phase are mainly controlled by the
amount of wetting fluid retained in the comers of pores with angular cross-section. The pore shape factor, together with capillary pressure, determine the amount of wetting fluid retained in
[he corners of non-wetting fluid filled pores. The hydraulic conductivity of the wetting phase is therefore strongly influenced by
the magnitude of the shape factor. This is illustrated in Figure 10
\vhich shows the effect of different shape factors on wetting
phase relative permeability. The hydraulic conductivity of the
wetting phase is drastically reduced when the shape factor increases, particularly at low wetting phase saturation where film
flow is important.
The non-wetting fluid occupies the bulk pore space of interconnected non-wetting fluid tilled pore bodies and throats. Bulk
conductivity is dominated by the narrowest constriction along the
pore space. The hydraulic conductivity of the non-wetting phase
is therefore mainly determined by the minimum inscribed pore
throat radii along the flow path. As a result, the shape factor has
little effect on non-wetting phase relative permeability.

Results

The method was tested using a 700 mD Berea sandstone core


plug, Berea sandstone is a Mississippian terrestrial sandstone and
previous work has indicated that the Berea sandstone sequence
was deposited as a constructive, fluvial-dominated epeiric (shallow inland) sea deltaic complex. A thin section of the core plug
was prepared and analysed at four different areas in order to obtain the necessary input data for the modelling of the pore space
network.
The grain-size distribution was analysed using Contextvisions
programme MicroGOP 2000 whilst the porosity and clay fraction
were determined using a Jeol JSM-840 Scanning Electron Microscope, The quartz cement overgrowth was determined visually
and an average value of 5 pm was estimated. In general, the
variation was between 0-20 ~m, depending on whether the grains
were in contact with each other or not. Mostly quartz and feldspar minerals were identified, but siderite (carbonate cement) and
mica minerals were also seen. Kaolinite (authogenic clay mineral) was seen in areas where feldspar minerals had dissolved due
to diagenesis. The data from the thin section analysis were used
to generate a 3-D representation of the Berea pore network.
The network flow simulator used in the present study was
originally developed for a regular cubic grid. At the time of the
study, the network simulator had not been modified to account
for irregular grids and it was therefore not possible to perform
the flow simulations directly on the generated pore network (this
has, however, later been implemented).
To overcome this limitation, the generated pore network was
analysed and average values for the most important pore structure descriptors were determined (i.e. co-ordination number, aspect ratio, and shape factors). These data were used to generate

Number: Networks with different average coordination numbers (i.e. the average number of pore throats connected to a pore body) were generated by randomly removing
pore throats in the network. The effect of different co-ordination
numbers on simulated relative permeabilities is illustrated in Figure 1I. The figure shows that the co-ordination number has only
a small effect on wetting phase relative permeability, For
strongly wetted systems, the wetting phase is hydraulically connected throughout and displaced wetting fluid can always escape
to the outlet either by bulk or film flow.
The connectivity of the non-wetting phase, however, is dependent on the co-ordination number. For a given pore size distribution, increasing the co-ordination number decreases the
breakthrough capillary pressure and moves the relative permeability curve to the right (see Figure 11).
Co-ordination

349

PREDICTION OF RELATIVE PERMEABILITY AND CAPILLARY PRESSURE FROM A PORE MODEL

caused by non-wetting-wetting
into the angular pore corners.

Aspect Ratio: The pore body to pore throat aspect ratio has little

effect on drainage relative perrneabilities. For low capillary number displacements, the sequence of pore invasion and therefore
the pore-scale fluid distribution are mainly determined by throat
threshold capillary pressures. This is, however, not the case for
imbibition where the aspect ratio plays an important role in the
competition between piston like advance in pore bodies and
snap-off in throats. The aspect ratio is probably the most important single feature determining hysteresis in capillary pressure
and relative permeability,

SPE 35531

fluid interfaces penetrating further

Conclusions
1. A new method for generating realistic 3-D pore networks

which are representative of different sandstones has been presented The method is based on stochastic modelling of the three
main sandstone-forming geological processes; (i) grain sedimentation, (ii) compaction, and (iii) diagenesis. Input data for the
modelling are obtained from image analysis of thin sections.
2. Generated pore networks are used as input to a network
flow simulator to predict capillary pressure and relative perrneabilities fi-om the underlying pore-scale physics.
3. In the network simulator, pore throats and bodies are modelied as having triangular cross-section. This models in a simple
way many of the basic features relevant to capillary dominated
flow, including the simultaneous flow of wetting and non-wetting
fluids in the same pore.
4. The pore shape has a strong influence on wetting phase relative permeability, especially at low wetting phase saturations
where film flow is important.
5. Simulated relative permeabilities for a Berea sandstone are
shown to be in fair agreement with those predicted from an empirical model.

Based on mercury inIrreducible Wetting Phase Saturation:


jection capillary pressure data, the irreducible wetting phase saturation was determined to be 0.053. To account for the presence of
irreducible wetting phase saturation in invasion percolation based
models, some rules regarding the continuity of the wetting phase
(high capillary
pressures)
must be
at low saturations
implemented.
At high capillary pressures, the area available for film flow of
wetting fluid in the comers of non-wetting fluid filled pores is
very small, In this case, the viscous pressure drop associated with
film flow is large and the time required to drain a pore by film
flow can be exceedingly large. The capillary pressure may then
increase during the course of the displacement to values which
exceed the threshold pressure for the next invasion event, For
these conditions, a second invasion may commence prior to the
completion of the first invasion and the displacement process is
no longer described by classical invasion percolation.
To account for irreducible wetting phase saturation, a primary
drainage process was first performed to a specified capillary pressure f<m~,. Above Pcmm,no further displacement occurs. The location of the irreducible wetting phase saturation in the network is
then identified and the conductivity of pore throats filled with
wetting fluid is set to a very small value. This ensures that the irreducible wetting fluid does not participate in the flow and is
therefore not displaced.
Following the establishment of irreducible wetting phase saturation, a second drainage process is simulated and relative perrneabilities are calculated, Figure 12 shows a comparison between
the empirically derived relative permeabilities and the simulated
ones using the modified network model. Clearly, the presence of
in-educible wetting phase saturation and increased shape factor
values (0.04) have decreased the wetting phase relative permeability. As a result, the simulated relative permeabilities are much
closer to those predicted by the empirical model (compare Figures 9 and 12).
Figure 13 compares simulated capillary pressures with those
determined horn mercury injection. In general, simulated capillary pressures are lower than the experimental ones, strongly suggesting that pore throat sizes in the network have been
overestimated. The rapid increase in simulated capillary pressures close to irreducible wetting phase saturation is mainly

Nomenclature

area
G = pore shape factor
g = hydraulic conductance
K = permeability
k,= relative permeability
I = pore throat length
Q = flux per unit area
P = pressure
P, = capilla~ pressure
fJ = interracial tension
y = viscosity
A = compaction factor
A =

Subscripts
1 = node index

J = node index
I = link
n = node
rrw = non-wetting fluid
w = wetting fluid

Acknowledgements

The authors would like to acknowledge Den norske stats o~eselskap as. (Statoil) for granting permission to publish this paper.
One of the authors (L, S.N,) wishes to thank G.D. Sharrna,
Stavanger University, Paul Nadeau and Lars Aasberg, Statoil Geology Centre, for valuable assistance with the thin section

350

SPE 35531

L, SCANDELLARI NILSEN, P.E. OREN, S. BAKKE, A. HENRIQUEZ

18. t{olt, R.M., Fjzr, E., Torszther, O., and Etakke, S., Pefrophysical

analysis. Special thanks are given to Marit Dovle, Statoil Research Centre, for assistance and guidance with the simulations.

laboratory

of one fluid

by another

in a network

wet~ing liquid

Fluid. Mech., 1983, 135, 337-353.


2, Oren, [E., Billiotte, J., and Pinczewski, W.V., h40biliza/ion

on fhe mo-

Proc. 6th.
European 10 R-Symposium, Stavanger, May 2 I-23, 199}, 1,
705-716.
and sprecrd4. @ren, P.E. and Pinczewski, WV,, Effect ofweflability
bll[zatton

of waterflood

residual

oil

by gasflooding,

wrg on the recove~


of waterflood
residual oil by immiscible gasjJood/ng, SPE Formation Evaluation, 1994,9, 149-156.
5. Oren, PC. and Pinczewski, WV., Fluid distribution and pore-scale
displacement
mechamsms
In drainage dominated three-phase /70w,
6.

Transport in Porous Media 1995,20, 105-133.


Jerauld, G.R. and Salter, S.J., The effect ofpore-swucrure
teresis

m relatwe

permeabdtty

and capdla~

pressure

on hyspm-e-level

modelling, Transport in Porous Media, 1990, 5, 103-151.


7. Blunt, M. and King, P., Relative permeabrlilies
from two- and
network modell[ng, Transport in Pothree-dimensional
pore-scale
rous Media. 1991, 6, 407-433.
Blunt, M., King, M., and Scher, II., Srmu/a/mn and (Aemy oftwm
phasejlow
in porous media, Phys. Rev. A, 1992, 46, 7680-7699.
9. Billiotte, J,, De Mocgen, H., and ~ren, P.I;., F~perlmen[a/ micromodeilmg
withdrawal

and numerical
simulation
of gas-waler
lnjec[ioncycles as applied to underground
gas storage, S PE.

Adv. Tech, Ser., 1993, 1, 133-139.

10 Chan, D.Y.C., Hughes, B.D., Paterson, L., and Sirakoff, C., .S~mu-

II

basin and reservoir

evaluation,

Ma-

in irregular

triangular

tubes,

morpholo~,
behavior

Academic

of a perfedy

Journal of Colloid and

Interface Science, 1991, 141, 262-274.


21. Dullien, F,A.L., Francis, S., Kai, Y,, and MacDonald, IF., }{ydrau/ic corr~inui~ of residual wetting phase in porous media, J. Colloid
and Interface Science, 1991, 109, 21O-2I8.
a new
22. Wilkinson, D. and Willemsen, J.F., Invasion percolation
form ofpercolation
fheory, 1983, J, Phys, A., 16,3365-3376,
23. Brooks, R.H. and Corey, A.T., Hydraulic properties of porous medm, }{ydrology Papers, No. 3, Colorado State University, Ft. Collins, CO, 1964,
24. Brooks, R H. and Corey A.T., Properties ofporous media afjjecting
j7uIdf70wI, J, Irrig. Drain. Div., 1966,6,61

ofwaconditions, S111

3,

for

19. Serra j., Image analysis ;nd m~thematical


Press, New York, 1982,
20. Mason, G. and Morrow, N. R., Capillaty

OJ the d~sof capillary ducts, J.

terf700d residual oil by gas injeclion for water-we/


Formation Evaluation, 1992, 7, 70-78,
L3rcn,P E. and Pinczewski, W. V., The effect offilmf70w

measurements

rine and Petroleum Geolom, in mess.

References
1. Lenorrnand, R., Zarcone, C., and Sarr, A., Mechanism
piacenren(

Iating /low
in porous
media.
Physical Review A, 1988. 38,
4106-4120.
Oren, P, E., Billiotte, J, and Pinczewski, WV., Pore-scale netiork
modelling of waterflood
residual oil recove~
by immiscible
gas
/700ding, paper SPE 27814, presented at the SPE/DOE 9th Sympo-

sium on IOR, Tulsa, OK, 17-20 April, 1994.


12 Fenwick, D.H. and Blunt, M. J., Pore level modelling of three phase
f70w m porous medza, paper presented at the 8th European 10R
Symposium, Vienn% Austria, May 15-17, 1995.
13 Pereira, G.G., Pinczewski, W. V., Chan, D.Y.C., Paterson, L.. and
network model ~or drainage
dominated
Oren, P.E., Pore-scale
lhree-phase
f70w in porous
media, accepted for publication in
Transport In Porous Media.
and perme14 Koplik, J., Lin, C., and Vermette, M., Conductivities
Journal of Applied Phyisics. 1984, 56,
abil~ty from trricrogeometty,
3127-3131.
displacements
in
15 Hazlett, R.D., Simulation of capillan!-dommated
microromograph{c
tmage$ OJ reservoir rocks, Transport in Porous
Medi< 1995,20, 21-35.
16 Bakke, S and t2Jren, PE., 3-D pore-scale modeliing of heterogeneous sandstone reservow rocks and quantitative
analysis of the architecture,
geometty
and spatial continuity
of the pore network,

paper SPE 35479, presented at the European 3-D Reservoir ModelIing Conference, Stavanger, Norway, April 16-17, 1996.
17 Bryant, S., Cade, C., and Mellor, D , Permeability
prediclton firom
geo/oglcal models, AAPG Bulletin, 1993, 77, 1338-1350.

351

PREDICTION OF RELATIVE PERMEABILITY AND CAPILLARY PRESSURE FROM A PORE MODEL

Fig. 1 - Low energy (local minimum)


= 387.).
open sandpack (pOrOSity

sedimentation

results

SPE 35531

Fig. 2 - High energy (global minimum) sedimentation results In a


more dense sandpeck (porosity = 35A). The same grain-size distribution as in Figure 1 was used.

In an

Fig. 4 . Isosurface representation of the sandstone matrix. Quartz


grains are medium grey, quartz cement is light grey, and clay Is clerk

Fig. 3- Modelllng of Ilneer compaction for compaction factors equal


to 0.0, 0.1, 0.2, and 0.3. Decrease irrporosity ia clearly observed.

grey.

352

SPE 35531

L, SCANDELLARI NILSEN, P,E. OREN, S, BAKKE, A. HENRIQUEZ

Fig. 5- Isosurface representation of the complementary


of the sandstone matrix shown in Figure 4.

Fig. 7- Schematic ball-and-stick


pore network.

representation

pore space

Fig. 6- The skeleton of the pore space network shown In Flgura 5.

Fig. 8- Pore-scale fluid distribution of oil and water In a strongly


water wet glasa mlcromodel (glass Is white, 011is dark grey, and water Is light grey).

of a 3-D numerical

353

10

PREDICTION OF RELATIVE PERMEABILITY AND CAPILLARY PRESSURE FROM A PORE MODEL

SPE 35531

1,0 ,

\
!
i
\
\
\
\
\

./
\

.,

._-. .&

.
-.,

,,.
0.0

--

0.2

.-. ._ ._.

0.4
0.6
Wenma Phase Saturabon

0.8

Fig. 9- Simulated relative permeabilities compered with predicted


curves using Brooks end Coreys empiricsl correlation.

10

1 .U

11
\

\
v

Fig. 10 - Effect of shepe factor values on wetting phase relative


permeability.

Cvcidmtml
W.ndrml.
mmh%m

&r

Uminr
mmb+r.

.4
.5

0.6

correlation
mmulaoon

\
\

0.5

$
a

0.0

- ;.2

/
/

1.1

0,5

00

,,

02

0.0

i..

,.
/,.

0.4
0.6
Wellinn Ph8e0 Satumtin

o.e

Werrhw Phase Ssturatlon

Fig. 12- Simulated relstive permeability curvas using the modified


network model compared with predicted curves (Brooks and Corey).

Fig, 11- Effect of co-ordination number on reletive permeability.

,,,
}

ex~rinwntal
slmulallon

,OJO..->-.

;4

;8

;8

,1(

Wetting Phase Salurat Ion

Fig. 13- Simulated capillary pressures compared with experlmantal


(mercury injection) measurements.

354

You might also like