Beruflich Dokumente
Kultur Dokumente
Abstract
The permeability estimation of reservoir rocks using numerical methods has always been a
challenge for petrophysicist. Several models have been developed for its computation of this parameter.
As the permeability is entirely controlled by the geometry, the possibility arising of estimating the
permeability from quantifiable attributes of the space has always has been a approach.
In the present paper, a model is discussed which gives accurate prediction of the permeability
based on two dimensional SEM image of a core sample of sedimentary rock. The inputs required for the
model are the areas and perimeters measurements from the images of space. The individual
conductances are estimated using hydraulic radius approximation. Before using the data obtained
from images, stereological corrections are used to convert
geometries and various hydraulic
corrections are used to account for converging-diverging flow paths. Kirkpatrick's medium
approximation is finally used to find the value of the hydraulic conductances of the individual pores.
The method has been applied to six data sets of SEM images which include Berea sandstone,
consolidated sandstones and carbonate samples. The laboratory determined air permeabilities of these
samples ranged from 0.5-400mD. The permeability values predicted by this method are within a factor
of two of the values.
This method requires least data manipulation and computation and is more accurate than
conventional methods such as the Kozeny-Carman equation. The method holds promise of permeability
predictions on irregular rock samples like drill cuttings which cannot be in standard lab measurements.
Another possible future is to use down-hole borehole imaging technology to provide an image with the
appropriate resolution, thereby allowing in-situ permeability estimation, without the need for core
samples.
Keywords Permeability, Scanning Electron Microscope, Pore Network Modeling, Porosity,
Conductance, Sandstones and Carbonates
Introduction
Being able to relate the transport properties of rocks to their internal pore structure has long been
of great interest to hydrologists, earth scientists and petroleum engineers (Jurgawczynski, 2007).
Permeability is arguably the most important petro-physical property of a reservoir rock and the ability to
predict its value without time consuming and expensive laboratory measurement would obviously be of
great practical value (Liang et al, 1997). Scanning Electron Microscope (SEM), images the sample
SPE144434
surface by scanning it with high energy beam of electrons, which interact with the atoms producing
signals that contain information about the samples surface topography, composition and other
properties such as electrical conductivity.
Petro-physical properties measurements, on samples of porous rocks, are classically performed in the
context of Core Analysis (porosity, permeability, formation factor). On the other hand, qualitative and
quantitative informations are obtained from the study of thin sections and SEM analysis of the same
materials (Adler et al, 1991).
The purpose of this paper is to present an attempt to establish a link between these two aspects, and to
try to solve a long standing problem :to deduce the transport properties of a 3-dimensional sample (plug)
from the corresponding geometrical structures of the pore space, which can be observed and
characterized in 2 dimensions(thin sections).
This problem is difficult for two main reasons:
It is hard to describe quantitatively the 3-D geometry of the medium in a realistic manner,
without introduction of artificial models and parameters (Adler, 1989).
The partial differential equations corresponding to these transport phenomena are not easy to
solve, except numerically.
Empirical permeability models such as that of Kozeny-Carman equation predict values of the
permeability using knowledge only of porosity and a mean pore diameter or mean grain size. Although
simple to implement, the Kozeny-Carman equation is often found to be insufficiently accurate for
reservoir characterization purposes. It is also known that this model, although fairly accurate for
unconsolidated sands, tends to become unreliable for consolidated sandstones. The Kozeny-Carman
approach therefore requires some means of estimating the specific surface, which can be problematic
(Berryman and Blair, 1986). The Katz-Thomson equation (1986) can yield accurate estimations of the
permeability, using the porosity and the electrical formation factor. However, the requirement of having
a measured value of the electrical formation factor is clearly a disadvantage of this method.
At the other end of the spectrum are models that attempt to reconstruct the complete 3-D microstructure of the rock, which can then be used either to compute the properties directly using the NavierStokes equations (Adler et al., 1990), or as a starting point for the development of network models
(Blunt, 2001). These network models can then be used to compute various flow properties, such as
absolute or relative permeability (Blunt et al., 2002), as well as to gain insight into the relationship
between pore geometry and petro-physical properties (Arns et al., 2004). Spanne et al. (1994), and later
Ferrol and Rothman (1995), used X-ray microtomography to reconstruct the pore structure of a
Fontainebleau sandstone, from which the permeability was calculated numerically, in the latter case
using the lattice-Boltzmann method. Such approaches are capable of good accuracy, but at the expense
of extensive data collection and computation.
In the present work, we have developed a model for predicting permeability from two-dimensional
SEM images of the pore space, without requiring any computationally intensive procedures of the
aforementioned three-dimensional models. But, unlike simple models such as Kozeny-Carman, we do
make use of rock-specific pore geometry information, obtained from two-dimensional pore images. The
hydraulic conductivities of the individual pores are estimated from their areas and perimeters using the
hydraulic radius approximation (Lock, 2001). Stereological correction factors are applied to determine
the true cross-sectional shapes from the images, and to determine the true number density of pores per
unit area. A constriction factor accounts for the variation of the cross-sectional area along the tube
length. The pores are assumed to be arranged in a cubic lattice, after which the effective-medium theory
of Kirkpatrick (1973) is used to estimate the effective conductance of the pores. Finally, the
permeability is estimated from the effective pore conductance and the number density of pores.
Because of the more complex pore structure found in carbonates, the methodology set up by Lock
(2001) for sandstones had to be entirely reviewed. The areal, thresholding process is adjusted to account
for the wide pore size distributions encountered in carbonates. Finally, in order to improve the
SPE 144434
Ar 2 A 2 A
A Rh 2
A3
=
C (circle) =
=
=
8
8
8
2 2 .......... (1)
Where the hydraulic radius is defined by Rh=2*A/. According to the hydraulic radius approximation,
we can use equation (1) for the conductance of any cylindrical pore, even if its cross section is not
circular.
Koplik et al, (1984) solved the governing laminar flow equation numerically for several pore shapes
found from SEM Microscopy of Massilon sandstone and found that the hydraulic radius approximation
was usually +-30% of the exact conductance. Sisavath et al (2001a) found similar results for pores in
Massilon and Berea sandstone based on comparison with boundary element calculations. Furthermore,
they found that this approximation does not systematically under/over predict the conductance so that
the errors (partially) cancel out. When applied to a network of pores of different cross sections, these
fact in mind and in the spirit of avoiding computationally intensive procedures such as a boundary
elements or numerical conformal mapping (Sisavath et al, 2001a). We use the hydraulic radius
approximation to estimate the flow conductance of the individual pores.
SPE144434
larger than the actual values for the pore cross-sections. For example, consider a cylindrical pore of
radius r. In general, the plane of the image will intersect this pore at some arbitrary angle relative to
the pore axis, and so the pore will appear as an ellipse with a semi-minor axis of r, but a semi-major axis
of r/cos. Hence, the area and perimeter of the image would consequently be larger than the actual area
and perimeter of the pore, and so the estimated hydraulic conductance will be greater than the actual
value. An approximate stereological correction factor that converts the measured values of the
hydraulic conductance into actual values can be found by averaging the overestimation in the
conductance C over all possible angles, assuming that the pores are randomly oriented with respect to
the plane of the image. The result (Lock et al, 2002) of this calculation is that the pore conductances
estimated from the image must be multiplied by 0.375 to arrive at the true conductance of the pore.
Stereological correction for number density
Consideration must also be made of the overestimation in the areal number density of pores that
occurs as a consequence of taking an arbitrary two-dimensional slice that probably does not lie in a
plane perpendicular to a lattice direction. If we again consider the idealization of pore microstructure by
a hypothetical cubic lattice, then a slice taken perpendicular to a given lattice direction will only
intersect those pores that lie along that direction. If, however, the slicing plane is not normal to the
lattice direction, it will also intersect some pores that are orthogonal to that first lattice direction.
Evaluation of this effect (Lock et al, 2002) leads to the conclusion that the apparent number density of
pores (i.e., pores per unit area of image) must be divided by 1.47 to yield the actual density of pores that
are oriented in a given lattice direction. An analogous calculation for a rock containing pores having a
random distribution of orientations gives an identical result.
NETWORK OF PORES
Network models have been used to study the transport properties of porous media. The pore-space
model that we use is that of a network of conductors, connected at volume-less nodes. (At the very low
Reynolds numbers encountered in most subsurface flow situations, hydraulic energy losses at
junctions, which are important in pipe flow at high Reynolds numbers, are negligible). If the
values of all of the conductances were known exactly, and the topology of the network was also
known, evaluation of the overall macroscopic conductance of the network would require the solution of
the large system of linear equations. These equations arise by applying the equation Q = CP/L to
each tube, and invoking the fact that, in order to conserve mass, the sum of the fluxes into each node
must be zero.this procedure has been carried out for idealized networks, generally two-dimensional,
and can yield much insight into the behavior of network models (Koplik, 1981; David et al., 1990).
An alternative to an exact network calculation is the effective medium approach, in which each
conductor in the network is replaced by a conductor having some effective value Ceff. One popular
method of estimating Ceff is the procedure of Kirkpatrick (1973), who used the criterion that if each
conductor within a certain region is replaced by a conductor with conductance Ceff., the resulting
perturbations within this region should average out to zero. This consideration leads to the
following equation that implicitly defines Ceff :
N
C eff C i
= 0 . (2)
i =1 ( z / 2 1)C eff + C i
Where the co-ordination number z is the number of pores that meet at each node and the summation
is taken over each individual pore in the network. Kirkpatrick (1973) derived equation (2) under the
assumption that the co-ordination number of each node was the same, but it was later shown (Koplik,
1982) that it can be used for topologically irregular networks by defining z to be the mean co-ordination
number of all the nodes in the network. In the present study we idealize the network as being a cubic
lattice, and set z = 6. Other researchers have estimated the mean co-ordination number for sandstones to
be in the range of 3-8.
SPE 144434
The effective-medium approximation becomes exact in the two limiting cases z=2 and z which
corresponds to series and parallel network. For z=2 equation (2) reduces to the harmonic mean. Ceff=N /
(1/Ci), whereas z, it yields the arithmetic mean, Ceff=Ci / N. For 2<z<, the accuracy of the
effective medium approximation decreases as the conductance distribution becomes broader. Using twodimensional square, triangular, and hexagonal networks, Koplik (1981)and David et al.(1990) found that
the effective-medium approximation has errors of less than 10% for narrow peak-like distributions of
conductances. Lock et al. (2004) compared the predictions of the Kirkpatrick equation to exact solutions
of the network equations, for conductance distributions inferred from actual sandstone pore images, and
found errors of less than 3%.
The last step involves deriving the permeability of the macroscopic rock from the effective
conductance of the individual pores. Imagine a plane that slices the lattice perpendicular to one of the
lattice directions, containing N pores in a region of cross-sectional area A. The total flowrate through
this region is given by
. (3)
Darcys law, on the other hand, expresses the flow-rate as
.. (4)
Equating (3) and (4) then gives
. (5)
SPE144434
SPE 144434
SPE144434
SPE 144434
LAB
porosity
23.84
22.58
7.21
9.74
28.9
31.9
23.58
19
17.86
16.61
16.28
16.64
MODEL
porosity
26.10
21.90
7.80
9.70
26.5
29
24
20
19
16
16.5
16.2
LAB
permeability
118.17
17.88
2.16
4.67
18.33
17.12
12.56
38.55
6.56
6.65
10.71
11.73
MODEL
Permeability
119
11.54
3.27
4.96
20
18.07
13.1
37.9
7.26
7.08
10.98
12.01
10
SPE144434
Figure 5: Crossplot of predicted and measured permeabilities for both the sandstone and carbonate
data set.
Figure 6: Crossplot of predicted and measured porosity values for both the sandstone and carbonate
data set.
After the generation of the SEM Image, Image analysis is done with the help of Image processing
software, ImageJ, which analysis the thresholded image and total number of pores in it and gives the area
and perimeter of each individual pores. Area and perimeter values are then used to determine the hydraulic
conductance of each individual pore and then, after applying stereological corrections and constriction
factor, kirkpatrik effective medium theory is used to determine the hydraulic conductance of the entire
network. With the help of overall conductance value, permeability of the sample is determined.
With the above described methodology the results of permeability and porosity predicted and
determined by laboratory is very close within about a percentage factor of two (Figure 5 and Figure 6).
Sample C, the carbonate, appeared not to have been carefully handled during the laboratory procedures,
resulting in the formation of fractures within the rock, which very likely caused an increase in the
laboratory permeability. Carbonate rocks are much more heterogeneous than sandstones, and so it is
quite possible that the small area of our image did not capture the full pore size distribution that existed
in the core. In addition, there is also the ongoing challenge of addressing the impact of vugs and their
SPE 144434
11
12
SPE144434
18. Koplik, J., On the effective medium theory of random linear networks, J. Phys. C, vol.14, pp. 48211981, 1981.
19. David, C., Gueguen, Y. and Pampoukis, G.: Effective medium theory and network theory applied
to the transport properties of rock, Journal of Geophysical research (1990) 95: 6993-7005.
20. Koplik, J., Creeping flow in two-dimensional networks, J. Fluid Mech., vol. 119, pp.219-247, 1982.
21. Lock, P. A., Jing, X. D., Zimmerman, R. W., Comparison of methods for upscaling permeability
from the pore scale to the core scale, J. Hydr. Res., vol. 42, pp. 3-8, 2004.
Nomenclatures:
Q
Volumetric flow rate
C
Flow conductance
P
Pressure drop
Fluid viscosity
L
length of the individual pore
r
Radius
Perimeter
A
Area
Rh
Hydraulic radius
Ceff
Effective Conductance
z
Coordination Number
Porosity
k
Permeability