Академический Документы
Профессиональный Документы
Культура Документы
Highly Cited
SPE Peer-Reviewed Papers
January 2018
A Collection of R
ecent Highly Cited
SPE Peer-Reviewed Papers January 2018
SPE JOURNAL
Shale Gas-in-Place Calculations Part I: New Pore-Scale Considerations
R. J. Ambrose, R. C. Hartman, M. Diaz-Campos, I. Y. Akkutlu, and C. H. Sondergeld. 2012. SPE J 17 (1): 219–229.
SPE-131772-PA. http://doi.org/10.2118/131772-PA.
Simultaneous Multifracture Treatments: Fully Coupled Fluid Flow and Fracture Mechanics
for Horizontal Wells
K. Wu and J. E. Olson. 2015. SPE J 20 (2): 337–346. SPE-167626-PA. http://doi.org/10.2118/167626-PA.
Non-Clay Non-Clay
Grain Volume Grain Volume
Volume Volume
Volume
Connected Pore Connected Pore
Volume
Total Void
Total Void
Free Gas
Volume
Volume
Free Oil, Gas, and Water Free Oil, Gas, and Water
Vv0 must be corrected at the beginning and at the end of the pressure If the changes in void volume are not taken into account, the
step, as described in Eqs. 9 and 10 (Menon 1968): isotherm will be in error and will not be usable in engineering
calculations. An example of isotherm data with and without void-
ˆ
n1 M volume corrections is shown in Fig. 2. The aforementioned void-
Vv1 = Vv 0 − , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (9) volume considerations that are necessary to accurately determine
s adsorbed-gas volumes have significant implications on the “live,”
in-situ shale pore volume available for free-gas storage. There-
ˆ
n2 M fore, the effective-porosity/gas-saturation product (Sge) derived
Vv 2 = Vv 0 − . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (10)
s from a total pore volume that is determined under static labora-
tory conditions does not reflect live-reservoir conditions. The
reservoir total pore volume is not only consumed by water and
Accordingly, over the course of the isotherm analysis, the void
oil but also by the adsorbed gas. It is for this reason we propose
volume is further reduced for each subsequent pressure step. In
that the calculated free-gas pore volume requires a correction for
practice, it is often more convenient to determine a so-called
the adsorbed-gas fraction present under reservoir-temperature and
Gibbs isotherm in terms of number of moles of adsorbed gas, as
-pressure conditions.
in Eq. 11:
Additional evidence to support the need for the correction lies
⎛ pr 1 pr 2 ⎞ ⎛ ps1 ps 2 ⎞ in visual investigation of the complex rock fabric typical of fine-
n2′ = n1′ + Vr ⎜ − + Vv 0 ⎜ − . grained, organic-rich shales. It has recently been reported that
⎝ zr 1 RTr 1 zr 2 RTr 2 ⎠⎟ ⎝ z s1 RTs1 z s 2 RTs 2 ⎠⎟ much of the gas-storage capacity within shale is predominantly
. . . . . . . . . . . . . . . . . . . . . . . (11) associated with the organic fraction of the rock matrix (Loucks
et al. 2009; Wang and Reed 2009; Sondergeld et al. 2010).
The Gibbs isotherm can then be converted to volumes using an Therefore, the free-gas volumes measured by pycnometry using
equation of state and can be adjusted for the void volume using a nonadsorbent gas such as helium must be corrected for the
the Gibbs correction factor f /s: presence of adsorbed gas within the organics. Fig. 3 shows a 2D
FIB/SEM gas-shale image supporting the preceding observations.
Ga' In the image, the matrix is represented by the gray color, with the
Ga = . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (12)
dark- and light-gray regions being the organic (kerogen) and the
1− f inorganic constituents (e.g., clay, silica, and feldspar), respectively,
s
whereas the pores are shown in black. Clearly, most of the pores
are significantly small, typically less than 100 nm, and are almost
exclusively found within the kerogen.
Using advanced imaging technologies, we have created 3D
shale segmentations showing kerogen and pore networks that
contain up to 600 such FIB/SEM images. A similar investigation
has previously been conducted by Tomutsa et al. (2007) for the sub-
micron petrophysical analysis of chalk. A typical kerogen network
of our study is shown in Fig. 4a. The network consists of large
interconnected pockets of kerogen, which make up an estimated
7.7% (volume) of the segment. Fig. 4b shows the superimposed
kerogen (bounded with yellow) and pore (on red) networks of the
same shale segment. As can be seen again from these images,
nearly all of the porosity within the system is associated with the
kerogen network. The image-computed porosity for this segmented
volume is 2.5%. A 3D network overlay showing the comparison
of the kerogen/pore networks and the gray-scale image is shown
in Fig. 3c. Orange denotes kerogen, whereas dark orange denotes
pores. These volumes are produced through gray-scale threshold-
Fig. 3—2D FIB/SEM image showing porosity and kerogen within ing. The setting of thresholds is subjective, and care should be
shale. Black depicts pores; dark gray is kerogen; light gray is taken, although the medium-sized pores are best resolved (Curtis
matrix (clay and silica). et al. 2010).
Fig. 4—3D FIB/SEM segmentations of 300 separate 2D SEM scans. Sample size is 4 m high, 5 m wide, and 2.5 m deep. The
kerogen network is shown in (a); yellow outlines the 3D kerogen network. In (b), yellow outlines the 3D kerogen network, while
red outlines porosity (in this sample, all porosity is found within the kerogen). (c) depicts an overlay of porosity and kerogen 3D
network; orange is kerogen, and dark orange depicts porosity within kerogen. The large orange mass is further in depth than
what is seen on the surface in (a) and (b).
Method for Shale Gas-in-Place Calculations The volume occupied by the sorbed gas must be accounted for after
In this paper, a new petrophysical model is proposed that alters the correction for water saturation. This is because of the porosity
the previous concept of effective porosity discussed earlier. Fig. 1 basis of the water-saturation measurement. Appendix A shows the
shows a comparison between the new model and the old model. derivation of the fundamental equation (Eq. 13). The porosity-
The new model emphasizes two distinct conceptual changes with reduction effects can be shown to vary. In deep reservoirs where
respect to the previous model. First, there exists a dependency of there is not much sorbed-gas content, the correction to the free gas
the connected pore space on the organics (Wang and Reed 2009; in place can be less than 5%. In other reservoirs with low porosity
Loucks et al. 2009; Sondergeld et al. 2010). Second, there is a and a high sorbed-gas content, the reduction could be up to 40%
dependency on the free pore space by the inclusion of a sorbed of the free gas in place or greater. Appendix B shows an example
phase. Fig. 5 shows a simple diagram of the current methodology calculation and the impact on the calculated gas in place.
used to determine gas in place vs. the proposed methodology. For
simplicity, the water and oil volumes are not considered in the Sorbed-Phase Density. In order to calculate the volume occupied
diagram. The simple illustration taken in context with the new by the sorbed phase, the density of adsorbed gas in the organic
information from the FIB/SEM images and segmentations visu- pores (s) must be known a priori. Measurement of the sorbed-
ally shows the errors that result by assuming that the sorbed gas phase density is not a trivial matter, however; the local density for
takes up no volume. methane is expected to vary across the pore and to be different
In order to properly account for the total and free gas in place, from its average bulk density because of the added interactions
the volume occupied by the adsorbed gas must be determined and between methane and the organic pore walls. Furthermore, in gas
subtracted from the free-gas calculation. We therefore propose that shales where the reservoir temperature is significantly greater than
the standard calculations used to calculate free-gas storage capacity the critical temperature of the natural gas, it is difficult to study
(i.e., Eq. 4) be modified as follows: phase transitions and determine if the adsorbate is in the form of
a liquid or vapor.
There have been several suggestions, which have been docu-
32.0368 ⎡ (1 − Sw ) 1.318 × 10 −6 M
ˆ ⎛ p ⎞⎤
Gf = ⎢ − ⎜ GsL ⎥ mented in physical-chemistry literature, to determine the density
Bg ⎣ b s ⎝ p + pL ⎟⎠ ⎦ of the adsorbed phase on solid surfaces. Dubinin (1960) suggested
that the adsorbate density is related to the van der Waals co-volume
. . . . . . . . . . . . . . . . . . . . . . . (13) constant b. Independently, Haydel and Kobayashi (1967) used an
experimental method and found the density values for methane
Old Methodology New Methodology and propane to be nearly equal to the van der Waals co-volume
constant. Later, it was argued that the sorbed-phase density is
Void Space Void Space equivalent to the liquid density (Menon 1968) and to the critical
Measured by Measured by density (Tsai et al. 1985) of the sorbed fluid. Ozawa et al. (1976)
Porosity Porosity considered the adsorbed phase as a superheated liquid with a den-
Measurement Measurement sity dependent upon the thermal expansion of the liquid. Recently
+ Ming et al. (2003) compared all of the previously stated methods
+ Sorbed Mass to a Langmuir-Freundlich adsorption model and found that there
Sorbed Mass Measured by exists a temperature dependence in the sorbed-phase density, but
Measured by Adsorption the value approaches those proposed by Dubinin (1960). These
Adsorption Experiment studies, although fundamentally important to our understanding of
Experiment -
gas adsorption in shale, do not show a clear and accurate path to
Free Gas prediction of the adsorbed-phase density of shale gas and, hence,
Volume Taken to estimation of shale gas in place. All the previous studies were
=Total GIP up by Sorbed
conducted at low pressure and temperature values (at most, 104oF
Gas
=Total GIP and 1,000 psi) or using an inorganic wall (silica gel, as in the
case of Haydel and Kobayashi). In this research, density values
Fig. 5—Comparison of the old and new methodologies in pre- in the first layer within the range 0.28–0.3 g/cc were found using
dicting shale gas-in-place; for simplicity, oil and water volumes molecular modeling and simulation. It is found that the density
are not shown. is sensitive to changes in temperature, pressure, and pore size.
Typically, at 3,043 psi and 176oF, the adsorbed-phase density of MD Simulation of Methane Adsorption in Organic Slit-Pores.
methane is 0.35 g/cc for a pore width of 2.3 nm when the Langmuir For the molecular-level investigation, methane is considered at
single-layer theoretical model is used. some supercritical conditions under thermodynamic equilibrium in
In this study, we used a numerical molecular-modeling approach 3D periodic orthorhombic pore geometry consisting of upper and
to determine the adsorbed-phase density from the first principles lower pore walls made of graphene (carbon) layers (see Fig. 6).
of Newtonian mechanics. Molecular modeling and simulation is a For comparison, we take into account three slit-like pores with a
form of computer simulation that enables us to study thermody- pore width equal to H = 3.9, 2.3, and 1.1 nm. Pore width is an
namic and transport properties of many-particle systems, in which important length scale of the study, and it is defined as the dis-
particles (atoms and molecules that make up the natural gas and tance between the innermost graphene planes. The pores maintain
pore walls) with initially known and instantaneously predicted posi- fixed dimension in the y-coordinate: Ly = 3.93 nm; however, the
tions and momenta are allowed to interact for a period of time, giv- dimension is changed in the x-coordinate such that both pores host
ing a view of the motion of the particles as trajectories in space and approximately the same number of methane molecules (400–450)
time. Numerical integration of Newton’s equations of motion make during the simulations. Hence, the dimensions in this direction are
up the core of the simulation; therefore, special algorithms (e.g., equal to Lx = 4.26 nm and to Lx = 7.67 nm for the large and small
Verlet, Leap Frog, or Beeman algorithms) have been developed and pores, respectively. While the effect of pore pressure is investi-
are commonly used during a molecular simulation study. Readers gated, the number of particles is increased from 400 to 600 within a
who are interested in details of molecular simulation are encour- 2.3-nm pore. Typically, the approximate parallel computation time
aged to consult textbooks by Frenkel and Smit (2002) and Allen was around 1,500–3,000 seconds, and 100 to 144 cores have been
and Tildesley (2002) for deterministic and stochastic treatment used during the simulations.
of the numerical-integration process. Molecular simulations have A united-atom carbon-centered Lennard-Jones potential (based
made enormous strides in recent years and are gradually becoming on the optimized potentials for liquid simulations, OPLS-UA
a commonly used tool in science and engineering (de Pablo and force field) is used as the model of a methane molecule; Table 2
Escobedo 2002). Today, molecular simulations are being widely shows the energy and distance parameters used for fluid/fluid
used to construct virtual experiments in cases where controlled and solid/solid interactions (Nath et al. 1998). The methane/solid
laboratory measurements are difficult, if not impossible, to perform. and methane/methane interactions are of the Lennard-Jones type
There exists an exhaustive literature studying the equilibrium ther- (Frenkel and Smit 2002). A Lorentz-Berthelot mixing rule [ij =
modynamics of fluids using molecular simulation involving phase (ii + jj)/2 and εij = (εiiεjj)1/2] is set up in order to describe solid/
change of bulk fluids (Harris and Yung 1995), characterization of fluid (i.e., pore-wall methane) interactions. Here, ij and εij are
porous materials using gas adsorption (Aukett et al. 1992; Sweat- the Lennard-Jones parameters accounting for interactions between
man and Quirke 2001), and multicomponent gas separation (Ghoufi a molecular site of methane species and a carbon atom of the
et al. 2009). In our case, two sets of molecular-dynamics (MD) organic wall. Lennard-Jones interactions were cut off at 4.1ii for
simulations are used to estimate and analyze the adsorbed-phase the 3.9-nm pore width, and at 3ii for the 1.1- and 2.3-nm pore
density under typical reservoir pressure and temperature conditions: width, respectively. The van der Waals interactions were cut off at
(a) runs involving bulk-phase (i.e., in the absence of pore walls) 3ii for all three systems.
methane-density measurements using a fixed number of methane The MD simulation package, DL-POLY (version 2.20), was
molecules at fixed pressure and temperature and (b) runs involving used to perform the simulations (Todorov and Smith 2008). Meth-
measurements of density of methane at the same temperature but ane bulk-density computations are made considering isobaric and
confined to a pore with organic (graphite) walls. We will consider isothermic conditions with a constant number of atoms, constant
deviations in the methane-density profile along the pore width of pressure, and constant temperature (NPT) at three constant tem-
the second set of runs relative to the uniform density value of the peratures (176, 212, and 266°F) using the Nosé- Hoover thermostat
bulk as an impact of the pore-wall effects. and barostat ensemble (Frenkel and Smit 2002). The relaxation
time used for the Nosé-Hoover thermostat is optimized to 15
pico-seconds (ps). In the case of simulations involving methane in
TABLE 2—LENNARD-JONES POTENTIAL PARAMETERS carbon slit-pores, initially the fluid system was equilibrated at a
FOR METHANE AND CARBON constant temperature, with a canonical ensemble [constant number
of atoms, constant volume, and constant temperature (i.e., NVT)].
Atom (nm) /kB (K) The equilibrium is assumed to be achieved if no drift in time was
Carbon 0.34 28 observed in the time-independent quantity, such as the total energy
of the system. The relaxation time used for the Nosé-Hoover ther-
Methane 0.373 147.9
mostat is optimized to 13, 15, and 20 ps for the pore widths of 1.1,
32 0.0
0.20
ρCH4 (g/cc)
0
0.0 3.8 7.6 11.4 15.2 19.
0.163
24
0.15 0.133 0.126 0.124
16
0.10
7.24E+00
8 5.18E+00 0.05
0 0.00
0.0 3.8 7.6 11.4 15.2 19.0 0.0 3.8 7.6 11.4 15.2 19.0
Pore Half Length (Å) Pore Half Length (Å)
Fig. 7—Number-density (left) and discrete-density (right) profile for methane at 176°F (80°C) in a 3.93 nm pore. Density values
are estimated at each 0.2-Å interval for the continuous-density profile. Discrete density corresponds to molecular-layer density
for methane across the pore. The estimated pore pressure at the center of the pores is 3,043 psi. The inset graph in the upper
right corner is the equivalent density using a Langmuir single-layer adsorption model.
2.3, and 3.9 nm, respectively. The Nosé-Hoover algorithm needs to gas density of methane 0.124 g/cc at the center of the pore. The
be used to control the system temperature while the real canonical pore pressure is 3,043 psia, which is a quantity predicted using the
ensemble computations are carried out for the density computa- free-gas molecules and the volume at the center of the pore. The
tions (Frenkel and Smit 2002). The average quantities within observed density profile shows that the assumption of Langmuir
1 nanosecond (ns) intervals after the equilibrium is reached are adsorption theory with monolayer is reasonable to describe the
taken for the analysis. Thermodynamic equilibrium is considered equilibrium adsorption dynamics of methane in the organic pores.
to be reached when the fluctuations in the density profile along the The inset graph on the upper right shows the equivalent Langmuir
time are uniform. Typically, the total run time for a simulation with monolayer plot where the pore half-length has been divided by the
the pore was 1.5 ns and the timestep used was 0.003 ps. adsorbed and free gas. The adsorbed-layer density is estimated as
At the end of each simulation run, number density Number (num- the total of the discrete densities associated with adsorption:
ber of molecules per volume) for methane across the pore space
was computed at every z = 0.2 Å interval (for the continuous s = 0.281 + (0.163–0.124) + (0.133–0.124)
density profile) and at every Lz = 3.8 Å (for the discrete density + (0.126–0.124) = 0.33g/cm3.
profile) volume segment in the z-direction. The number density
for each volume segment is estimated by summing the number of Pore-Size Effects on Methane Adsorption. In essence, super-
molecules counted in each interval within the volume segment and critical methane in small organic pores is structured as a result
dividing the total by the number of intervals. Then, the number of pore-wall effects, showing layers of graded density across the
density is converted to the local mass density of methane using its pore. Depending on the pore size, a bulk-fluid region may exist at
molecular weight MCH4 and Avogadro’s number as follows: the central portion of the pore, where the influence of molecular
interactions with the pore walls is very small. In pores with sizes up
Number M CH4 to 50 nm, a combination of molecule/molecule and molecule/wall
CH4 = .
interactions dictates thermodynamic states of the gas and its mass
6.02252 × 10 23
transport in the pore (Krishna 2009). On the other hand, within a
As explained earlier, it is expected that the local mass density small pore (see Fig. 8), methane molecules are always under the
for methane across the pore will be different from its mean bulk influence of the force field exerted by the walls; consequently, no
density, owing to changing levels of interactions between the bulk-fluid region can be observed in the pore, and no pore pressure
methane/methane and methane/carbon bodies. The purpose of our can be measured. Therefore, behavior of the adsorbed molecules
numerical investigation was to predict a precise density profile should be considered rather than the motion of free-gas molecules.
across the pore as an indication of the presence of these interac- Furthermore, the adsorbed-phase density at the first layer is signifi-
tions and to determine an average adsorbed-gas-density value, a cantly large. The smaller the pore, the larger the adsorbed-phase
macroscopic quantity necessary for the gas-in-place calculations density is. In the limit, Langmuir theory is not suitable for the
using Eq. 13. description of physical interactions between the gas and solid.
Fig. 7 shows the density profile for methane confined to the Effect of Temperature on Methane Adsorption. The effect
3.92-nm pore at 176°F. It is clear that the predicted methane density of temperature on methane density in the large pore is shown in
is not uniform across the pores; its value is significantly greater Fig. 9. The estimated average adsorbed-methane density in the
near the wall, where adsorption takes place, and it decreases with first layer is around 0.3 g/cm3, although it decreases with tem-
damped oscillations as the distance from the pore wall increases. perature (for example, to 0.259 g/cm3 at 266°F). These values
The oscillations are caused by the presence of adsorption in show variations caused by changing levels of kinetic energy at
the pores and involve structured distribution of molecules (i.e., the microscopic scale.
molecular layers). The number of molecules is largest in the first Table 3 shows that the predicted free-gas density values at the
layer near the wall, indicating physical adsorption. The molecules center of the pore correspond to the bulk methane density values
in the second layer are still under the influence of pore walls, predicted independently using the National Institute of Standards
although intermolecular interactions among the methane molecules and Technology’s (NIST) thermophysical properties of hydro-
begin to dominate, not allowing locally high methane densities. In carbon mixtures database, SUPERTRAPP. The comparisons also
this layer, the density of methane is slightly larger than the bulk show that our numerical simulations are accurate.
50 0.35 0.2
0.124
0.1
42.80 0.30 0.292
43.01 0.0
40
0.25 0.0 3.8 7.6
ρCH4 (g/cc)
ρ* CH4 (1000Å–3 )
30 0.20 0.171
0.15 0.135
20
0.10
10 7.65 7.72
0.05
4-5 (bulk)
0 0.00
0.0 3.8 7.6 11.4 15.2 19.0 22.8 0.0 3.8 7.6 11.4
Pore Length (Å) Pore Half Length (Å)
Pore Width = 1.14nm
50 4.57E+01 0.35
0.307
4.69E+01 0.30
40
ρ* CH4 (1000Å–3 )
20 0.15
1.57E+01
0.10
10
0.05
4-5(bulk)
0 0.00
0.0 3.8 7.6 11.4 0.0 3.8
Pore Length (Å) Pore Half Length (Å)
Fig. 8—Number-density (left) and discrete-density (right) profiles for methane at 176°F (80°C) in pore widths of 2.31 nm (upper)
and 1.41 nm (lower). Density values are estimated at each 0.2-Å interval for the continuous-density profile. Discrete density cor-
responds to molecular-layer density for methane across the pore. Estimated pore pressure at the center of the pores is 3,043
psi. The inset graph in the upper right corner is the equivalent density using a Langmuir single-layer adsorption model.
Conclusions adjustments. This change in gas in place persists even when the
In this paper, we addressed the following issues in regard to the sorbed density values reported by other researchers, such as the
volume available for free gas in organic shales: liquid-methane density of 0.4223 g/cc, corresponding to 1 atm
• The sorbed phase follows Langmuir theory for most of the pores and –161°C (Mavor et al. 2004), and of 0.374 g/cc (Haydel and
and takes up a one-molecule-thick portion of a pore, although Kobayashi et al. 1967). These are clearly shown in Table 4.
there is a damped oscillation density profile with an increased In conclusion, a robust method that matches the local phys-
density in the second layer. For a 100-nm pore, the volume is ics is presented to determine an estimate of the gas in place in
fairly insignificant; however, for pores on the order of 1 nm, it organic-rich gas shale. Future work includes re-evaluation of the
is quite large. concepts and approaches that are presented and discussed in the
• The current industry standard disregards the volume consumed presence of multicomponent gases with varying pore sizes. It is
by the sorbed phase, thus inadvertently overestimating the pore expected that there will be added uncertainties in the gas-in-place
volume available for free-gas storage. We showed that with Eq. estimation because of (1) multicomponent nature of gas and its
13, a more correct volume is calculated. Through MD simulation adsorbed phase and (2) volume fraction of organic nanopores that
and Langmuir theory, we showed that the density for adsorbed are currently inaccessible for visual observation and measurement.
methane typically equals 0.34 g/cm3. This value is based on The former requires computer-simulation studies on the adsorbed-
numerical work using simple flat organic pore-wall surfaces. In phase thickness and density using binary, ternary, and quaternary
small pores, the adsorbed-layer thickness will be affected by the mixtures of gases with components such as ethane, propane, and
roughness and curvature of the solid surface, and the density will carbon dioxide. The latter uncertainty, on the other hand, can be
increase because of increased surface area of the wall. Further reduced by measuring the absolute area of the organic surfaces
analysis using more-complex pore-wall structures is necessary. using isotherm data. In addition, a new study on transport proper-
• In our calculation of the gas in place, we predict a dramatic ties of natural gases in nanoporous organic materials is necessary
change in the estimated gas-in-place values because of volume in light of the new pore-scale findings.
266°F
0.20
ρCH4 (g/cc)
24
0.15
16
0.10
8
0.05
0 0.00
0.0 3.8 7.6 11.4 15.2 19.0 0.0 3.8 7.6 11.4 15.2 19.0
Pore Half Length (Å) Pore Half Length (Å)
Fig. 9—Number-density (left) and discrete-density (right) profiles across half-length of a 3.9-nm-width slit-pore as a function of
temperature.
TABLE 4—EFFECT OF SORBED PHASE DENSITY ON FREE AND TOTAL GIP CALCULATIONS
Adsorbed-Phase Density Free Storage Capacity Total Storage Capacity
3
(g/cm ) (SCF/ton) (SCF/ton)
Given that it is in scf, we can convert scf into a mass with the Using the new calculation method for gas in place, the free, sorbed,
ideal-gas law at standard temperature and pressure: and total storage capacities for Shale A are, respectively,
φ 0.06 0.06
Sw 0.35 0.35
So 0.0 0.0
Bg 0.0046 0.0046
M̂ 16 lb/lb-mol 16 lb/lb-mol
GsL 50 scf/ton 120 scf/ton
p 4000 psia 4000 psia
T
o o
180 F 180 F
pL 1150 psia 1800 psia
3 3
b 2.5 g/cm 2.5 g/cm
3 3
s
0.34 g/cm 0.34 g/cm
32.0368 ⎡ (1 − Sw ) 1.318 × 10 −6 M
ˆ ⎛ p ⎞⎤ Ga = 120
4000
= 82.8 scf/ton . . . . . . . . . . . . . . . . (B-11)
Gf = ⎢ − ⎜ GsL ⎥
Bg ⎣ b s ⎝ p + pL ⎟⎠ ⎦ 4000 + 1800
9 Multiphase
Region
8 Phase boundary
7 Clear solution
6 Two clear phases
% NaCl
5 Precipitation
4 One-Phase
Region Cloudy solution
3
Cloudy after nine
2 months
0
1:1 4:1 9:1
IOS N67
N67:IOS (w/w)
Fig. 1—Effect of added NaCl on phase behavior of 3 wt% solutions of N67/IOS mixtures containing 1 wt% Na2CO3 (Liu et al.
2008).
as alkaline/surfactant processes (Nelson et al. 1984; Peru and ponent of surfactant retention. Phase trapping of surfactant can be
Lorenz 1990; Surkalo 1990; Baviere et al. 1995). more significant and will be discussed later.
The primary role of the alkali in an alkaline/surfactant process Alkaline preflush had been advocated for both sequestering
is to reduce adsorption of the surfactant during displacement divalent ions and reducing sulfonate adsorption (Holm and Rob-
through the formation and sequestering divalent ions. An addi- ertson 1981). In subsequent work, alkali has been injected with the
tional benefit of alkali is that the soap is formed in situ from the surfactant. Adsorption of anionic surfactants on Berea sandstone
naphthenic acid in the crude oil (Johnson 1976). As indicated pre- was reduced several-fold with addition of sodium carbonate for
viously, the generation of soap allows the surfactant to be injected petroleum sulfonate (Bae and Petrick 1977) or with addition of
at lower salinities than if used alone, which further reduces adsorp- sodium silicate or hydroxide for alcohol ethoxysulfate (Nelson
tion and facilitates incorporation of polymer in the surfactant slug. et al. 1984). The reduction of adsorption on Berea sandstone with
Also, alkali can alter formation wettability to reach either more sodium bicarbonate was 68% in a dynamic experiment (Peru and
water-wet or more oil-wet states. In fractured oil-wet reservoirs, Lorenz 1990). Static and dynamic adsorption of anionic surfactants
the combined effect of alkali and surfactant in making the matrix on calcite and dolomite was decreased by an order of magnitude
preferentially water-wet is essential for an effective process. These with addition of sodium carbonate, but insignificantly with sodium
benefits of alkali will occur only where alkali is present. Thus, it hydroxide (see Figs. 2 through 4) (Hirasaki and Zhang 2004;
is important to determine “alkali consumption,” which controls the Seethepalli et al. 2004; Zhang et al. 2006; Liu et al. 2008; Tabata-
rate of propagation of alkali through the formation. bal et al. 1993). (The TC Blend of Figs. 2 through 4 is an earlier
blend of isotridecyl 4PO sulfate and C12 3EO sulfate. Research
Reduced Surfactant Adsorption. The discussion here will be on it was discontinued because its optimal salinity was too high
limited to anionic surfactants (Wessen and Harwell 2000). The for the application of interest.)
primary mechanism for the adsorption of anionic surfactants on
sandstone- and carbonate-formation material is the ionic attrac- Divalent-Ion Sequestration. The phase behavior of anionic sur-
tion between positively charged mineral sites and the negative factant systems is much more sensitive to a change in divalent ions
surfactant anion (Tabatabal et al. 1993; Zhang and Somasundaran (e.g., Ca2+ and Mg2+) compared to monovalent ions (e.g., Na+),
2006). Thus, the role of the alkali is to be a “potential-determining especially at low surfactant concentrations (Nelson 1981). This is
ion” to reverse the charge on positively charged mineral sites. The problematic in sandstones because of ion exchange between the
potential-determining ions for oxide minerals are the hydronium clay, brine, and surfactant micelles (Hill et al. 1977; Pope et al.
and hydroxide ions. The pH at which the charge reverses is the 1978; Hirasaki 1982). This exchange can result in the phase behav-
“isoelectric point” if measured by electrophoresis (zeta potential) ior becoming overoptimum, with resulting large surfactant reten-
and is the “point-of-zero-charge” if determined by titration. The tion (Glover et al. 1979, Gupta 1982). Alkali anions (e.g., carbon-
values are tabulated for most common minerals (Lyklema 1995). ate, silicate, and phosphate) that have low solubility product with
Silica is negatively charged at reservoir conditions and exhibits divalent cations will sequester divalent cations to low concentra-
negligible adsorption of anionic surfactants. Clays (at neutral pH) tions (Holm and Robertson 1981). Hydroxide is not as effective for
have negative charge at the faces and positive charge at the edges. sequestration of calcium because the solubility product of calcium
The clay edges are alumina-like and thus are expected to reverse hydroxide is not very low. Sodium metaborate has recently been
their charge at a pH of approximately 9. Carbonate formations and introduced as an alkali that may sequester divalent ions (Flaaten
sandstone-cementing material can be calcite or dolomite. These et al. 2009; Zhang et al. 2008). A common problem with alkali
latter minerals also have an isoelectric point of approximately pH injection is that softened water is needed to avoid scaling.
9, but carbonate ions, as well as the calcium and magnesium ions,
are more significant potential-determining ions. The zeta potential Generation of Soap. The original concept of alkali flooding was
of calcite is negative even at neutral pH in the presence of 0.1 N the reduction of oil/water IFT by in-situ generation of soap, which
carbonate/bicarbonate ions (Hirasaki and Zhang 2004). If a for- is an anionic surfactant, sodium naphthenate (Jennings 1975).
mation contains iron minerals, the oxidation/reduction conditions Ultralow IFT usually required injection of relatively fresh water
influence whether the surface iron sites are Fe3+ or Fe2+. Adsorption with a low concentration of alkali because optimal salinity (total
of anionic surfactant for one sandstone was found to be lower by electrolyte concentration) of the in-situ-generated soap is usually
more than a factor of two for reducing rather than for oxidizing low (e.g., <1% electrolyte). If the alkali concentration is too low,
conditions (Wang 1993). Surfactant adsorption is only one com- alkali consumption reactions may result in a large retardation of
2.0
Adsorption Density,
10–3 mmol/m2
1.5
1.0
0.5
0.0
0.0 0.5 1.0 1.5 2.0 2.5
Fig. 2—Static adsorption of TC blend surfactant on dolomite sand. BET surface area of the calcite: 17.8 m2/g (Zhang et al.
2006).
the alkali displacement front. The concept of alkaline/surfactant Alkali Consumption. The ASP process should be designed such
flooding is to inject a surfactant with the alkaline solution such that that displacement fronts of the alkali, surfactant, and polymer
mixture of the in-situ-generated soap and injected surfactant has travel together. The mechanisms responsible for the retardation of
an optimal salinity that is tailored to the reservoir fluids (Nelson the alkali front include silica dissolution, clay dissolution with zeo-
et al. 1984; Surkalo 1990). lite precipitation, anhydrite or gypsum dissolution with calcite (or
The common method used to determine the amount of naph- calcium hydroxide or silicate) precipitation, dolomite dissolution
thenic acid in crude oil is the total acid number (TAN), determined with calcium and magnesium silicate precipitation, hydrogen-ion
by nonaqueous titration with a base (Fan and Buckley 2007). If exchange, divalent-ion exchange with precipitation, and mixing
sodium naphthenate is to act as a surfactant, it should partition with divalent ions in formation water with precipitation (Ehrlich
into the aqueous phase at low electrolyte concentrations and be and Wygal 1977; Holm and Robertson 1981; Southwick 1985;
measurable by hyamine titration for anionic surfactants. It was Cheng 1986; Novosad and Novosad 1984; Jensen and Radke 1988;
found that the sodium naphthenate determined by extraction into Mohammadi et al. 2009). Naphthenic acids in crude oil also react
the aqueous phase and measured by hyamine titration is less than with alkali and thus contribute to consumption, but the amount
the TAN value (Liu et al. 2010). It is hypothesized that the TAN is usually small compared to the mentioned inorganic mineral
includes components that are too lipophilic to be extracted to the reactions. Silica dissolution can be controlled by using a buffered
aqueous phase and/or too hydrophilic to be detected by hyamine system such as sodium carbonate or silicate rather than hydroxide
titration. (Southwick 1985). Clay dissolution is strongly dependent on the
1.0
Expr. 1
0.9 v=1.2 feet/day
beta=0.34±0.03 Expr. 2
Cl−
0.8 Φ=0.335
Dimensionless Concentration
0.7 Expr. 2
Expr. 1
v=12 feet/day
0.6 beta=0.22±0.03
Φ=0.338
0.5
Calculated
0.4 from dolomite
isotherm
0.3
beta=0.04
0.2
0.1
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Injected Volume, PV
Fig. 3—Dynamic adsorption of 0.2% TC blend surfactant without Na2CO3 on dolomite sand (Zhang et al. 2006).
Dimensionless Concentration
v=1.2 feet/day
0.7
0.6
0.5
Experimental Data for NaCl
0.4
Fig. 4—Dynamic adsorption of 0.2% TC blend/0.3-M Na2CO3 on dolomite sand (Zhang et al. 2006).
pH and type of clay, and is kinetically limited (Sydansk 1982). be preferentially water-wet when the salinity is below the optimal
Thus, acidic clay such as kaolinite, as well as high temperature, salinity (Winsor I). When the system is overoptimum (Winsor
will increase the importance of this mechanism. II), macroemulsions tend to be oil-external. An oil-external mac-
A limitation of the application of sodium carbonate in carbonate roemulsion will trap water and have a low oil and water relative
formations is that if anhydrite or gypsum is present, it will dissolve permeability, similar to what one expects with oil-wet porous
and precipitate as calcite (Hirasaki et al. 2005; Liu 2007). This is media. The optimal salinity for a conventional alkali flooding
detrimental for dolomite formations because they may have origi- system is dependent on the in-situ-generated sodium naphthenate
nated from evaporite deposits where gypsum is usually present. An soap, and is usually below approximately 1% electrolyte strength.
alternative alkali is sodium metaborate (Zhang et al. 2008; Flaaten Because salinity of reservoir brine typically exceeds this value, a
et al. 2009). However, longer-term experiments and equilibrium conventional alkali flood often generates overoptimum and oil-wet
calculations indicate that this metaborate will also precipitate. conditions. We show later in this review that this behavior can
be avoided by injecting the alkali and surfactant in the Winsor I
Alkaline/Surfactant Processes: Wettability region. After mixing with the fluids in the reservoir of interest, it
Alteration will pass through the Winsor III, low-IFT region. Even a high-
Wettability is the next most important factor in waterflood recov- salinity sandstone formation that is initially oil-wet may be altered
ery after geology (Morrow 1990). The recovery efficiency of a to preferentially water-wet by injecting alkali with a hydrophilic
flooding process is a function of the displacement efficiency and surfactant in the Winsor I region.
sweep efficiency. These efficiencies are a function of the residual-
oil saturation (waterflood and chemical flood) and mobility ratio, Carbonate Formations. Wettability alteration has received more
respectively. The residual-oil saturation to waterflooding is a attention recently for carbonate formations compared to sand-
function of wettability, with the lowest value at intermediate wet- stones because carbonate formations are much more likely to be
tability (Jadhunandan and Morrow 1995). The mobility ratio is a preferentially oil-wet (Treiber and Owens 1972). Also, carbonate
function of the ratio of water relative permeability to oil relative formations are more likely to be fractured and will depend on
permeability at their respective endpoints or at a specific satura- spontaneous imbibition or buoyancy for displacement of oil from
tion. The mobility ratio or relative permeability ratio becomes the matrix to the fracture.
progressively larger as the wettability changes from water-wet to Wettability-alteration tests on plates of calcite, marble, lime-
oil-wet (Anderson 1987b). When a formation is strongly oil-wet, stone, and dolomite with different surfactants and sodium carbon-
it can have both a high waterflood residual-oil saturation and ate have been used to identify many systems that are altered to
unfavorable mobility ratio. In addition, an oil-wet formation will preferentially water-wet with low anionic-surfactant concentra-
have capillary resistance to imbibition of water (Anderson 1987a). tions (Hirasaki and Zhang 2004; Seethepalli et al. 2004; Zhang
Formation wettability can be altered by pH (Wagner and Leach et al. 2006; Adibhatla and Mohanty 2008; Gupta et al. 2009).
1959; Ehrlich et al. 1974; Takamura and Chow 1985; Buckley Sodium carbonate has an important role because the carbonate ion
et al. 1989, Dubey and Doe 1993), surfactants that adsorb on the is a potential-determining ion for calcite and dolomite (Hirasaki
minerals (Somasundaran and Zhang 2006) or remove adsorbed and Zhang 2004).
naphthenic acids (Standnes and Austad 2000), and acids or bases
(Cuiec 1977). These processes are now incorporated into chemi- Spontaneous Imbibition. Spontaneous imbibition is the process
cal-flood simulators (Anderson et al. 2006; Delshad et al. 2006; by which a wetting fluid is drawn into a porous medium by capil-
Adibhatla and Mohanty 2007, 2008). lary action (Morrow and Mason 2001). The presence of surfactant
in some cases lowers the IFT, and thus the capillary pressure, to
Sandstone Formations. Wettability alteration to more water-wet negligible values. Spontaneous displacement can still occur in this
or more oil-wet conditions was proposed as one of the mechanisms case by buoyancy or gravity drainage (Schechter et al. 1994).
of caustic flooding (Wagner and Leach 1959; Ehrlich et al. 1974; The research group of Austad has investigated spontane-
Johnson 1976). Our current understanding of microemulsion phase ous imbibition into chalk-formation material with enhancement
behavior and wettability is that the system wettability is likely to by cationic and nonionic surfactants and/or sulfate ions present in
0.09
0.08
0.06
0.05
0.04
0.03
0.02
0.01
0
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018
(a) Distance From Center, m --->
0.09
0.08
0.07
Distance From Bottom, m --->
0.06
0.05
0.04
0.03
0.02
0.01
0
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018
(b) Distance From Center, m --->
Fig. 5—Velocity field in a cylindrical core immersed in surfactant solution at 10.89 days: (a) aqueous phase, (b) oil phase (Gupta
and Mohanty 2010).
spinning drop for IFT measurements. The IFT estimated from the the displacement process. Thus, an alkaline/surfactant system
Huh correlation shows this crude oil to also have a wider, ultralow should be considered as a pseudo-two-surfactant system featuring
IFT region of salinities, especially in the underoptimum region the injected surfactant and the soap. The two surfactants will likely
(Liu et al. 2010). have different optimal salinities. Thus, a mixing rule is needed to
model how the optimal salinity changes with surfactant and soap
Alkaline/Surfactant Processes: Phase concentrations.
Behavior of Soap/Surfactant
Alkali saponifies the naphthenic acid in crude oil in situ to generate WOR and Surfactant Concentration. Optimal salinity was
sodium naphthenate, a soap that helps to generate low IFT during observed to be a function of surfactant concentration and WOR
Excess
Oil
Colloidal
Dispersion
Lower-Phase
Microemulsion
1.E+01
Without Na2CO3
With 1% Na2CO3
1.E+00
1.E-01
IFT, mN/m
1.E-02
1.E-03
1.E-04
0 1 2 3 4 5 6
Salinity, % NaCl
Fig. 8—Measured IFT of system with and without Na2CO3 (from Liu et al. 2008).
100 1.E-02
IFT, mN/m
10 1.E-03
Vo/Vs
1 1.E-04
2 2.5 3 3.5 4 2.0 2.5 3.0 3.5 4.0
NaCl, % NaCl, %
Fig. 9—Measured solubilization ratios of salinity scan (from Fig. 10—Comparison of the IFT from the solubilization param-
Liu et al. 2008). eter and spinning-drop measurements (from Liu et al. 2008).
14 14
WOR=1 (TC blend)
The optimal-salinity mixing rule is used in UTCHEM for simu- IFT Measurements. The dependence of the optimal salinity on
lation of alkaline/surfactant processes (Mohammadi et al. 2009). the soap/surfactant ratio can be used to explain the difference of
In addition, the researchers found that the optimum solubilization minimum-equilibrium IFT that may be observed with different
ratio follows a linear mixing rule: surfactant concentrations and WOR. This also explains why the
Opt vs Soap Fraction (theory) Opt vs Soap Fraction (exp) Opt vs Soap Fraction (theory) Opt vs Soap Fraction (exp)
10 10
Optiman Salinity
Optimal Salinity
1
1
0.1
0.1
0 0.2 0.4 0.6 0.8 1
0 0.2 0.4 0.6 0.8 1
Xsoap Xsoap
((a)
a) (b)
Fig. 13—Relationship of optimal salinity and soap mole fraction by different acid-number methods for NI Blend and Yates oil
(Liu et al. 2010). (a) Nonaqueous phase titration; acid number = 0.75 mg KOH/g; (b) soap extraction by NaOH; acid number =
0.44 mg KOH/g.
Two-Phases - II (+)
Two-Phases - II (−)
Three-Phases - III
1982; Hirasaki and Lawson 1986). The problem of divalent ions
0.3 is avoided by use of an alkali such as sodium carbonate or sodium
Residual Oil Saturation, PV
soap fraction at the waterflood residual-oil saturation. The range 2010). The lower salinity of the drive compensates for the lower
of salinity for greater-than-90% oil recovery is a function of the surfactant concentration such that the region of optimal salinity
soap fraction. The salinity for maximum oil recovery decreases again propagates with a near-unit velocity. In addition, injection
from optimal salinity of the surfactant to that of soap as the soap of the surfactant slug and polymer drive with a salinity that is less
fraction increases (straight line in the figure). The range of salini- that the optimal salinity of the surfactant alone makes it possible
ties for potentially high oil recovery is substantial, especially in to inject the surfactant slug with polymer without separation of the
the underoptimum region below the optimal line. surfactant and polymer into separate phases (Gupta 1982; Liu et al.
If the dispersion is increased to a representative field-scale 2008). Also, the salinity gradient avoids the large surfactant reten-
value [Pe = 50, Lake (1989)] with constant salinity and a 0.2-PV tion from microemulsion trapping by the polymer drive (Glover
slug, the region of greater-than-90% recovery all but disappears. et al. 1979; Hirasaki et al. 1983).
Dilution by mixing at the front and back of the surfactant slug low-
ers the surfactant concentration more than the soap concentration, Foam Mobility Control
and the propagation velocity of the soap/surfactant ratio for optimal Foam is usually considered as a means of mobility control for
salinity is greatly retarded. However, if the system is operated with gas-injection processes such as steam foam or CO2 foam. Foam
a salinity gradient, high oil recovery is again possible (Liu et al. mobility control for surfactant flooding is a natural progression
4.0
70% 50%
3.0 30%
90%
1.0
Optimum
70%
Curve
50%
0.5
30%
Fig. 16—Recovery factor with small slug (0.2 PV) and low dispersion (Pe = 500) (Liu et al. 2010).
because the system already has surfactant present (Lawson and surfactant/polymer flooding (Kang et al. 2010). It has also been
Reisberg 1980). Moreover, at high temperatures, foam may be used as mobility control for surfactant aquifer remediation (Hira-
favored because polymer degradation is a concern (Srivastava and saki et al. 1997, 2000). Nonionic surfactants have been evaluated
Nguyen 2010). In fact, foam was used for mobility control for for mobility control of CO2 EOR (Adkins et al. 2010a, b). They
alkaline/surfactant flooding in China (Zhang et al. 2000; Wang can be injected dissolved in the CO2 phase and have less adsorp-
et al. 2001). Recently, it has been used to improve sweep in tion on carbonate formations compared with anionic surfactants.
Fig. 17—Displacement profiles for the displacement of MY residual crude oil by ASPF in 40-darcy sandpack (Li et al. 2010).
NIP Air NIP Air NIP Air IOS Air IOS Air IOS Air IOS Air IOS Air IOS
1 ft/day 20 ft/day
Fig. 18—Profiles of the displacement of 266-cp Crude B with ASP and ASPF (Li et al. 2010).
If, in addition, the oil/water IFT can be reduced to ultralow values, Apparently, the viscous oil was being transported as an oil-in-water
a low-tension CO2 EOR process may be applicable for reservoirs emulsion with much less resistance than that of the crude oil.
with pressures below the minimum miscibility pressure.
Sweep of Layered Sands. Fig. 19 compares sweep in two cases
ASP Foam. The reduction of surfactant adsorption with alkali of a 19:1 permeability contrast layered sandpack initially filled
may result in the polymer being the most expensive chemical in with water dyed green. The sandpack is nearly completely swept
the ASP process. Experiments in 1D sandpacks have shown that with 1.0 TPV of surfactant alternating with gas (SAG) while the
an ASP process with the polymer drive replaced by a foam drive low-permeability layer is only one-quarter swept with water only
is equally efficient. Fig. 17 is an experiment in which the ASP (Li et al. 2010). The sweep efficiency is compared in Fig. 20 as
slug is alternated with equal-sized slugs of gas. The foam drive a function of the PV of liquid injected for SAG, water alternating
consists of slugs of the better-foaming surfactant component gas, and waterflooding.
(without polymer) alternated with equal-sized slugs of gas (Li
et al. 2010). Practically all of the 19-cp oil was recovered after 1.2 Potential for Fractured Formations. The improvement of sweep
TPV injected, but with only 0.6 PV of liquid injected. Experiments in layered sands suggests that foam may be helpful in the sweep
with different sands indicated that foam reduced mobility more of a system of fractures. Yan et al. (2006) showed that pregener-
in higher-permeability media, making it particularly attractive in ated foam does improve the sweep of parallel plates with different
layered systems. apertures to simulate heterogeneous, parallel fractures. Also, the
ASP foam was used to recover a 266-cp, 4.8-mg KOH/g TAN increased pressure gradient caused by foam flow in the fractures
crude oil (Fig. 18). What was remarkable is that the apparent increases the driving force for displacement of oil from the matrix
viscosity of the displacement process was only 80 cp or less. (Haugen et al. 2010; Abbasi-Asl et al. 2010). Farajzadeh et al.
0.0 TPV
0.2 TPV
0.4 TPV
0.6 TPV
0.8 TPV
1.0 TPV
Fig. 19—Comparison of SAG with waterflood in 19:1 permeability ratio sandpack (Li et al. 2010).
Sweep Efficiency
0.8 SAG fg=3/4, 4psi
0.6 SAG fg=2/3, 2psi
Waterflood
SAG fg=1/3, 6psi
0.4
SAG fg=1/2, 4psi
0.2 WAG fg=4/5, 4psi
WAG fg=3/4, 4psi
0.0
0 0.5 1 1.5 2 2.5 3 WAG fg=2/3, 4psi
WAG fg=1/2, 4psi
PVs of Liquid Injected
Water fg=0, 4psi
Fig. 20—Sweep in 19:1 permeability contrast sandpack with SAG, WAG, and waterflood (Li et al. 2010).
(2010) describe how the viscous pressure gradient caused by foam 6. The soap generated in situ by the alkali causes a middle layer
flow in fractures can accelerate the production of oil that would to form and coexist with the lower-phase microemulsion, which
otherwise be produced from the matrix by gravity drainage. results in ultralow IFT over a wide range of salinity.
7. Anionic surfactants and sodium carbonate can alter wettability
Field Pilots for either sandstone or carbonate formations. Spontaneous oil
A pilot of the alkaline/surfactant process is described by Falls et al. displacement can occur by gravity drainage.
(1994). This pilot was tested without polymer, with the intention 8. Foam can be used as the drive of the alkaline/surfactant process
of a subsequent test with polymer. Nevertheless, the interpretation in place of the polymer drive.
of induction logs suggested 100% displacement efficiency in the 9. Foam can efficiently sweep layered and fractured systems.
region swept by the injected fluids.
A refinement of the alkaline/surfactant process with recent Acknowledgments
understanding about the IOS is described by Buijse et al. (2010). The authors acknowledge the financial support by DOE grant
The paper also discusses the importance of the crude-oil composi- DE-FC26-03NT15406 and the Rice University Consortium on
tion. The process was tested in the field with a single well chemi- Processes in Porous Media. The information and insight we gained
cal-tracer test. Stoll et al. (2010) describe pilot tests in Oman. from our long collaboration with Gary Pope and Kishore Mohanty
Gao and Gao (2010) summarized the pilots in Daqing oil are also acknowledged.
field.
References
Conclusions Abbasi-Asl, Y., Pope, G.A., and Delshad, M. 2010. Mechanistic Modeling
The technology of surfactant flooding has advanced to overcome of Chemical Transport in Naturally Fractured Oil Reservoirs. Paper
many of the past causes of failures and to reduce the amount of sur- SPE 129661 presented at the SPE Improved Oil Recovery Symposium,
factant required. These developments are summarized as follows: Tulsa, 24–28 April. doi: 10.2118/129661-MS.
1. Surfactant adsorption can be significantly reduced in sandstone Abe, M., Schechter, D., Schechter, R.S., Wade, W.H., Weerasooriya, U.,
and carbonate formations by injection of an alkali such as and Yiv, S. 1986. Microemulsion Formation with Branched Tail Poly-
sodium carbonate. The alkali also sequesters divalent ions. The oxyethelene Sulfonate Surfactants. Journal of Colloid and Interface
reduced adsorption permits lower surfactant concentrations. Science 114 (2): 342–356. doi: 10.1016/0021-9797(86)90420-0.
2. A wide selection of surfactant structures is now available to meet Adams, W.T. and Schievelbein, V.H. 1987. Surfactant Flooding Carbon-
requirements for specific applications. ate Reservoirs. SPE Res Eng 2 (4): 619–626. SPE-12686-PA. doi:
a. Branched alcohol alkoxylate sulfates and sulfonates are toler- 10.2118/12686-PA.
ant of divalent ions. Ethoxylation increases optimal salinity; Adibhatia, B. and Mohanty, K.K. 2007. Simulation of Surfactant-Aided
propoxylation decreases optimal salinity. In both cases, Gravity Drainage in Fractured Carbonates. Paper SPE 106161 pre-
EO or PO, the optimal salinity decreases with increasing sented at the SPE Reservoir Simulation Symposium, Houston, 26–28
temperature. February. doi: 10.2118/106161-MS.
b. Alkyloxylated glycidyl ether sulfonate is more expensive Adibhatla, B. and Mohanty, K.K. 2008. Oil Recovery From Fractured
than sulfate but is stable at elevated temperatures. Carbonates by Surfactant-Aided Gravity Drainage: Laboratory Experi-
c. IOSs are low-cost, double-tailed surfactants. ments and Mechanistic Simulations. SPE Res Eval & Eng 11 (1):
3. Aqueous solutions of a blend of N67-7PO sulfate and IOS1518 119–130. SPE-99773-PA. doi: 10.2118/99773-PA.
with alkali have a larger single-phase region extending to higher Adkins, S.S., Chen, X., Chan, I., Torino, E., Nguyen, Q.P., Sanders, A.,
salinities and calcium-ion concentrations than either alone. and Johnston, K.P. 2010a. Morphology and Stability of CO2-in-Water
This blend, without alcohol, can form a single phase for injec- Foams with Nonionic Hydrocarbon Surfactants. Langmuir 26 (8):
tion with polymer but can form microemulsions with crude oil 5335–5348. doi: 10.1021/la903663v.
without forming a gel. Adkins, S.S., Chen, X., Nguyen, Q.P., Sanders, A.W., and Johnston, K.P.
4. Soap generated in situ by the alkali is a cosurfactant that can 2010b. Effect of branching on the interfacial properties of nonionic
change the phase behavior of the injected surfactant solution hydrocarbon surfactants at the air-water and carbon dioxide-water
from lower- to middle- to upper-phase microemulsions. It is interfaces. Journal of Colloid and Interface Science 346 (2): 455–463.
lower phase when injected, middle phase at the displacement doi: 10.1016/j.jcis.2009.12.059.
front, and upper phase ahead of the displacement front. Anderson, G.A., Delshad, M., King, C.B., Mohammadi, H., and Pope, G.A.
5. Injection of the surfactant and polymer at salinity that is under- 2006. Optimization of Chemical Flooding in a Mixed-Wet Dolomite
optimum with respect to the injected surfactant avoids surfac- Reservoir. Paper SPE 100082 presented at the SPE/DOE Symposium on
tant/polymer phase separation and microemulsion trapping. Improved Oil Recovery, Tulsa, 22–26 April. doi: 10.2118/100082-MS.
(a) (b)
Fig. 1—(a) 2D backscatter FIB/SEM image of a gas-shale sample. (b) 3D FIB/SEM shale segmentation. The shale segmentation
size is 4 µm high, 5 µm wide, and 4 µm deep.
Slippage
0.30
Phase
0.25 Hopping
transition layers
ρCH4, g/cm3
0.20
Free gas
0.15
0.10
0.05
0.00
0.0 3.8 7.6 11.4 15.2 Direction of external molecular flux
Pore Half Length, Å
(a) (b)
J K,x
(c) J S,x
Fig. 2—(a) Molecular layer density for methane at 176°F (80°C) across the half-length of a 3.74-nm organic slit pore. Results
obtained using equilibrium molecular dynamics simulation carried out in the canonical ensemble. The estimated pore pressure
is approximately 2,000 psi. (b) Schematic representation of gas slippage and hopping mechanisms of molecular transport. (c)
Schematic of gas mass transport in organic nanopores. The arrows represent diffusive mass fluxes of the free-gas phase (thick
arrow) and the adsorbed phase (thin arrow) in the x direction. JK,x and JS,x are the contributions of free-gas pore diffusion and
adsorbed-phase surface diffusion, respectively.
for the free-gas molecules. Density structure of gas molecules in the gas mass transport in the organics should be in the transitional
small organic pores and its effect on shale-gas in-place estima- slippage-flow regime. This means that the assumption of Darcy
tions have been studied recently by our group; see Ambrose et al. flow is invalid in the organic pores in shale. Therefore, a new
(2010). approach is necessary for modeling the gas mass transport, which
• The mean free path of gas molecules (i.e., average distance will incorporate non-Darcy transport of the free-gas molecules.
gas molecules travel between two successive collisions) becomes • Adsorbed-phase transport could be another important mass-
comparable to the available pore space. Consequently, the free-gas transfer mechanism. Fig. 2c shows the free-gas and adsorbed-phase
molecules interact vigorously with the molecules in phase transi- components of total mass flux in the organic micro- and mesopore.
tion and with those that make up the adsorption layer and pore wall, Under the influence of an external flux, some of the molecules in
compared with the level of interaction the free-gas molecules have the adsorbed-phase and those in phase transition can overcome the
among themselves at the central portion of the pore. When thermo- local interactions with the wall and develop a hopping mechanism.
dynamic equilibrium is perturbed because of an externally applied (Much like drifting of a tumbleweed or tree leaves on the ground
molecular flux, for example, these amplified interactions may lead under the influence of a mild wind.) This transport mechanism, also
to kinetic effect that causes streaming of the gas molecules—in known as surface diffusion, is not important if it takes place in a
particular, those in phase transition. This is the occurrence of the large pore with an almost infinite number of free-gas molecules;
so-called slippage effect (or molecular streaming) in the organic but, if the discussion is on nanopores with a large surface area, it
pore. It can cause the collapse of laminar-flow (parabolic) -veloc- may easily contribute to the total mass flux and, hence, enhance
ity profile with zero velocity at the wall. Consequently, the free- the kinetic effect by the pore walls. Let us reconsider the layers of
gas mass transport would show deviations from Darcy’s law and molecules indicating nonuniform structure (damped oscillation) in
would be represented better with molecular (pore) diffusion. A methane density in a 3.6-nm organic pore in Fig. 2. It is not dif-
useful dimensionless parameter in the literature to determine the ficult to imagine that the molecular layers corresponding to free gas
changes in flow regime of a gas system is the Knudsen number would not develop and all of the existing molecules would be either
(NKn). This number is the ratio of the molecular mean free path to adsorbed or under the influence of pore walls if the pore size were
the pore length scale. When NKn > 0.1, the kinetic effect becomes less then 2×1.08 = 2.16 nm. Then, the only transport mechanism
dominant by the wall in the phase-transition region. Further dis- during the gas release from or uptake by the pore would be because
cussion on the flow regimes can be found in Roy et al. (2003). of surface diffusion. We note that, although they both take place in
In Fig. 3, we consider methane existence in organic pore volume the phase-transition region, surface diffusion is a different transport
with characteristic pore length varying between 2 and 10 nm and mechanism than the slippage; at a given pressure and temperature,
estimate the NKn for varying pore pressures. The values inside the surface diffusion is driven by gradients in adsorbed-phase density,
ellipse (typical initial shale-gas-reservoir conditions) indicate that whereas slippage is driven by the pore size.
N Kn
1.00
0.10
0.01
10 100 1,000 10,000
Pressure, psia
Fig. 3—Knudsen number vs. pore pressure for pore-throat radii calculated from the Langmuir isotherm at reservoir temperature
of 188°F.
The additional moles of gas ns stored in the sample provide the
Valve 1 increase in storability as a function of the change in pressure.
Reservoir When this process is iterated over multiple steps, it provides the
Volume 1 storability as a function of pore pressure. This is the basic informa-
tion needed to characterize the reservoir and is an essential input
into simulations that model the way gas will be produced from the
reservoir or be sequestered in it. It should be noted that, because
there can be hysteresis in the storability curves, both the increasing
Valve 2 and decreasing storability cases may need to be measured.
Gas To separate the storage in the rock system into the adsorbed/
Source absorbed component and the component stored as free gas in the
pore space, a model of the storage mechanisms needs to be used.
Volume 3 = Volume 2 The moles of gas nsp stored in the sample pore space Vp in terms
+ Sample Volume of the sample pore pressure Ps (Ps equals P2 or Pf) satisfies the
gas law. Therefore,
Va =
(
Va max P PL ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (13)
Organic material
Inorganic
1 + P PL
⎡ Pf PL P2 PL ⎤
Vpa = −Va max ⎢ −
( P2 PL ) ⎥⎥⎦
. . . . . . . . . . . . . . (14)
⎢⎣ 1 + ( Pf PL ) 1 +
Fracture
Fracture
Inorganic r matrix
matrix o
g Adsorption Viscous
Viscous
flow e + flow
n diffusion
(a) (b)
Fig. 6—Conceptual multicontinuum models for gas flow and transport in organic-rich shale during sequestration.
is distributed continuously in space and holds the fundamental ments of permeability and diffusivity of shale are difficult to
porous-medium conditions specified by Bear and Bachmat (1991). implement because of very low mass fluxes and the extremely
Multicontinuum models can be identified by the number of long time needed to reach the steady-state condition. Therefore,
components characterizing the hydraulic behavior of the system transient methods such as pressure-pulse-decay experiments have
and the type of coupling and mass-exchange terms between the been used for the measurements. This method is much faster in
components. The interaction between the components, defined by comparison to the conventional methods and can be used to mea-
finite mass-exchange terms, affects the overall transport behavior sure permeability valuse as low as 10−9 md (Brace et al. 1968; Ning
of the system significantly, while a single-continuum approach 1992; Finsterle and Persoff 1997; Jones 1997). In this method,
neglects this important effect. There are three different types of a small perturbation is introduced to a shale/gas system under
coupling between the components: parallel, series, and selective thermodynamic equilibrium by rapidly changing the downstream
coupling. In parallel coupling, all specified continua are coupled (or upstream) pressure, typically by an approximate 10% of the
with one another directly, while, in the series coupling, they are equilibrium pressure. Consequently, small pressure gradients are
connected in the order of hydraulic conductivity (Lee and Tan generated across the sample, leading to transport of gas across the
1987). In organic-rich-shale-gas reservoirs, multicontinuum could core plug. During the gas permeation, decline in the upstream or
be defined, at the least, by two matrix continua (i.e., organic and downstream pressure is measured to extract transport properties of
inorganic, along with a fracture continuum). the shale using an analytical or numerical inversion method.
In the Effective Organic Pore Size subsection, we argued that
the gas mass transport in the organics is transitional slippage with Results
adsorbed-phase surface-diffusion effects (i.e., non-Darcy). Model- Table 4 shows results from the models that fit the measured stor-
ing of such transitional flow regimes in the presence of adsorption age data for helium, CO2, and methane with two organic-rich shale
is not straightforward, however. Here, following the work by Fathi samples, Sample 21 and Sample 23. These samples are selected
and Akkutlu (2009, in press), we investigate gas-transport behavior because TOC and mineralogy were known. Because, in the initial
in the kerogen by experimentally quantifying the pore and surface measurement stage, the CO2 passes through a pressure where it
diffusion coefficients DK and DS , respectively. Pore diffusion is the can form a liquid phase, there is considerably more uncertainty in
mechanism of transport for the free-gas molecules occupying the the model fits to the measured storage data. Note the estimated Vp0
available pore space, whereas surface diffusion is the mechanism values. The values are quite different for Sample 23, giving 5.38,
of transport for the adsorbed-phase near the internal surfaces of the 3.95, and 2.89% porosity for the gases, respectively. The model
kerogen. During the production, gas in the kerogen pore network fits to the experimental data are not unique. Assuming that all
is eventually released to the relatively larger pores, voids, micro- reasonable fits to the data have the properties of this fit, the varia-
fractures, and cracks in the inorganic matrix or directly into the tions indicate that methane molecules with spherical molecular
nonuniform fracture network shown in Fig. 5b. In its new environ- geometry cannot have access to some of the micropores that CO2
ment, the mechanism of its transport is primarily viscous (Darcy) molecules with linear molecular geometry have access to, and the
flow and, because of limited surface area, the adsorbed amount is CO2 molecules cannot have access to some of the micropores that
considered negligible. In the case of CO2 sequestration, the same the small helium molecules can access. In essence, the organic
arguments are valid, although in the opposite transport direction material of Sample 23 acts as a molecular sieve for fluids used
(i.e., from the fractures deep into the inorganic shale matrix and during measurements; see Fig. 7. Although this effect creates
into the kerogen). Fig. 6 shows two conceptual models of shale-gas uncertainties during the storage-capacity measurements for Sample
flow and transport that are considered in this study for methane 23, in general, it points to an important CO2-sequestration quality
production and CO2 sequestration. In Fig. 6a, parallel coupling is for the shale-gas formations. It suggests that a large fraction of
shown where organic and inorganic continua are coupled directly the injected CO2 can be stored in small pores in which the other
with fractures. In Fig. 6b, all three continua are coupled in series naturally occurring gases, such as methane and nitrogen, cannot
(i.e., fracture → inorganic → organic). We used these conceptual be stored. Compared to methane, for example, CO2 can access an
models during the development of a mathematical formulation additional 100 × (0.4771−0.345)/0.345 = 38.3% of the organic pore
(a group of coupled partial-differential equations) describing the volume when injected into Sample 23 under the same conditions.
mechanisms of gas storage and transport. The formulation is used to The molecular sieving becomes a secondary effect in the case of
simulate gas-permeation experiments using organic-rich shale core Sample 21, however. In this case, methane and CO2 have access
plugs. Later, the developed simulators are used in the experimental to the same pore space, whereas helium can penetrate into an
investigation of multiscale characterization of transport phenomena additional 7.7% organic-micropore volume.
in organic-rich shales. Further discussion on the modeling and As the consequence of changes in the estimated values of the
simulation of shale-gas flow can be found in Fathi (2010). pore volume Vp0, the predicted gas-sorption parameters in Table 4
are quite different. Effect on the Langmuir adsorption isotherm is
Measurements of Transport Coefficients—Unsteady Gas-Per- shown in Fig. 8. Note that, because the pore-volume estimation for
meation Experiments. Steady-state methods for the measure- Sample 21 is consistent, the predicted methane and CO2 isotherms
Sample 21 S a m p le 2 3
appear distinctly in Fig. 8a. The isotherms for the other sample at any given pore pressure. For both samples, the total CO2-gas-
are shown in Fig. 8b. Indeed, methane isotherms show that the storage capacity is 4 times greater than the total methane-gas-stor-
gas stored as an adsorbed phase is significantly larger when the age capacity at 2,500 psia. Such a large contrast in methane and
pore volume obtained using helium is included in the analysis. The CO2 storage is important for sequestration, although it cannot be
contrast is even larger in the case of CO2. These results indicate explained by the molecular-sieving arguments related to the organ-
the importance of fluid type used for the adsorption analysis. In the ics of certain types of samples. It can be explained only by changes
rest of the paper, we show only the storage and transport results in the fluid’s internal energy with the changes in the fluid type.
for a particular fluid using a sample pore volume measured with Indeed, the molecular-interaction energies (both among the gas
that fluid. molecules and between the organic pore wall and gas molecules)
Modeled total gas storage is given in Figs. 9 and 10. The are different, which leads to significantly enhanced adsorption in
adsorbed-gas and free-gas amounts stored are shown in Fig. 9a the case of CO2. Molecular-level investigation further comparing
for Sample 21 and in Fig. 9b for Sample 23. Clearly, most of the adsorption of fluids and their mixtures in kerogen in the small
injected CO2 is stored in an adsorbed state. For example, at 2,500 organic pores under reservoir conditions is currently necessary.
psia and for Sample 21, the ratio of adsorbed to free CO2 is approx- Results indicate that adsorption is the dominant mechanism
imately 13:1. For Sample 23, that ratio drops to 13:5, although the during the CO2 storage in organic-rich shale. This may also be the
total amount stored is nearly doubled. Hence, the measurements case for methane, although the dominance is not as amplified and
reveal that the samples have larger adsorption capacity for CO2 distinct as in the case of CO2; see Fig. 10. At 2,500 psia, the ratio
0 0.0000 0 0.0000
0 1,000 2,000 3,000 4,000 5,000 0 1,000 2,000 3,000 4,000 5,000
Pressure, psia
(a) (b) Pressure, psia
Fig. 8—Estimated Langmuir isotherms for Sample 21 (a) and Sample 23 (b). The parameters used are given in Table 4. The mark-
ers show the five experimental pressure data points used during the prediction of the gas-storage parameters.
500 500
Adsorbed gas, CO2 Adsorbed gas, CO2 0.0016
450 0.0016 450 Adsorbed gas, CO4
Adsorbed gas, CO4 0.0014
0.0014 Free gas, CO2
400 Free gas, CO2 400
Fig. 9—Estimated adsorbed- and free-gas amounts for Sample 21 (a) and Sample 23 (b). Solid lines are with CO2, and the dashed
lines are with methane.
Adsorbed Gas Storage/Total Gas Storage
1.5 1.5
CO2 CO2
1 1
CH 4
CH 4
0.5 0.5
0 0
25 1,025 2,025 3,025 4,025 5,025 25 1,025 2,025 3,025 4,025 5,025
Pressure, psia Pressure, psia
(a) (b)
Fig. 10—Estimated ratio of adsorbed gas to total gas stored in Sample 21 (a) and in Sample 23 (b).
–0.5
1,600
Confining pressure 5,000 psi –1
1,590
–1.5
Dark gray curve: experimental data
1,580 Light gray curve: mean of realizations –2
ln (ΔPD)
1,570 Black curve: 112 realizations
–2.5
1,560 –3
1,550 –3.5
1,540 –4 μgLcg
kg=|slope|
1,530 –4.5 6.805×10–5f1A(1/V1+1/V2)
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
×104 –5
Time, seconds
(a) (b)
Fig. 11—(a) Example history-matching results for Sample 21 using upstream pressure decline. 112 realizations are used to obtain
the optimum values of the transport coefficients shown in the first row in Table 5. (b) Permeability for Sample 21 is obtained
using the pulse decay shown on the left and using Jones’ straight-line method (Jones 1997).
of adsorbed CO2 to total CO2 gas storage is approximately 0.93 nature of multiscale gas transport in shale are discussed for the
for Sample 21; on the other hand, the ratio for methane is approxi- sequestration. Most importantly, we found that it is not possible
mately 0.54 for the same sample. In the case of Sample 23, at the to obtain converging (optimum) results using unipore shale-matrix
same pressure, the ratios are 0.72 and 0.46, respectively. models for the forward simulation; the matrix model used for the
The rapid change in free-gas-CO2 measurements near 950 psia history matching should be a dual-porosity (organic/inorganic)
in Figs. 9 and 10 is caused by dramatic changes in the z factor of system. History-matching results at various confining-stress condi-
the gas near the critical fluid state. tions and using a multiscale transport model are shown in Table 5
The estimated coefficient of isothermal pore-volume-com- for Samples 21 and 23. Note that the time required to reach the new
pressibility values in Table 4 are nearly the same for the two sam- equilibrium is significantly less as the pore pressure is increased.
ples, taking values on the order of from 5.0×10−6 to 7.0×10−6 psi−1. This indicates that the transport paths are sensitive to effective
During our gas-storage measurements and calculations, we found stress. The observation is supported further by the increase in the
that the results would have been less than 5% different when the estimated inorganic permeability and organic diffusivity values.
pore-compressibility effect had been ignored. Thus, the primary We note that permeability estimates are significantly less than
correction necessary for the organic pore volume is the adsorption those obtained using conventional measurements, such as Jones’
effect. This behavior is quite different than in coals where most of method (see also Fig. 11b for details). This is possibly because of
the matrix is organic and the pore-volume-compressibility values the explicit introduction of ultralow-permeability organics into the
could be one or two orders of magnitude larger; hence, its effect description of shale-gas transport in this work.
on the pore volume measurements in coal becomes comparable to It was argued in the Interpretation of Organic Pore Diffusion
the sorption effects (Koleowo 2010). subsection that the pore-diffusion coefficient for a single capil-
Finally, the estimated adsorbed-phase density values in Table lary should typically be two orders of magnitude larger than the
4 for methane and CO2 are in agreement with those reported in the experimentally obtained apparent free-gas transport coefficient
literature (Kurniawan et al. 2006; Ambrose et al. 2010), indicating DK. Here, using Eq. 4, we experimentally predict tortuosity val-
that the gas-storage models are reasonable and that the density for ues within the range of 4.0 to 6.0 for the kerogen. The estimated
the adsorbed-phase is close to the liquid density of the chemical values are consistent with the literature, showing high tortuosity
species used during the experiment. values for microporous materials and suggesting that the mea-
A typical upstream pressure-decline curve of the pulse-decay sured diffusion coefficient indeed represents the coefficient for
gas-permeation experiments is shown in Fig. 11. In this case, pore diffusion of free gas. More interestingly, the coefficient for
methane initially in equilibrium with shale Sample 21 is allowed adsorbed-phase transport DS is predicted to be typically three
to permeate. The curve typically yields two distinct behaviors orders of magnitude larger than that for the free gas. This is a
having completely different characteristic times. At early times, somewhat curious behavior, which suggests that the surface diffu-
the curve shows a rapid decline in pressure (fast transport), and, sion could be the dominant mechanism of transport in the organ-
later, it slows down, gradually decreasing the upstream pressure ics. Typically, the adsorbed-phase concentration and its gradient
(slow transport). These distinct portions of the decline curve in the organics are smaller than those values associated with the
include information related to Darcy flow in the inorganic matrix free gas; however, when multiplied with an extremely large sur-
consisting of large voids, and pore/surface diffusion in the organic face diffusivity, the product (i.e., mass flux of the adsorbed phase)
pore network. In addition, the total time required for the shale/gas becomes a quantity that significantly contributes to the total mass
system to reach the new thermodynamic equilibrium is related flux in the kerogen. The importance of surface diffusion to the
to gas storability for the shale and can be used to obtain basic gas transport in heterogeneous coals and shales has been discussed
information such as organic-/inorganic-pore-volume ratio. Having recently by Fathi (2010) and Fathi and Akkutlu (in press) on the
the pressure-decline curve, we used a simulation-based history- basis of theoretical arguments in the spectral (Fourier-Laplace)
matching algorithm based on the randomized maximum likelihood domain. Here, the experimental results support the observation,
method and estimated the optimized transport properties of the suggesting that the sequestered gases will also be transported as
shale samples. The optimization requires different realizations of part of the adsorbed phase in the organics. Fig. 12 shows that the
the parameter space, typically on the order of hundreds. Details estimated mass flux from surface diffusion of the adsorbed gas
of the automated nonlinear history-matching process are described molecules is comparable to and could be even significantly larger
elsewhere (Oliver et al. 2008). Here, salient results related to the than the mass flux from pore diffusion of the free-gas molecules.
0 1.73
5 2
5,000 1,623 83 44,485 70.4 3.9 1.33 15.8 53.87 4.45 3.50 10 1.10 10
5 2
5,000 2,588 84 18,063 102.2 3.9 1.24 20.4 64.16 4.39 4.00 10 4.20 10
5 2
5,000 2,974 70 12,060 122.9 3.9 1.22 28.7 67.56 4.04 4.90 10 8.80 10
0 2.85
5 3
5,000 1,772 78 72,060 12.65 3 2.4 2.6 43.58 6.14 3.50 10 1.55 10
5 3
5,000 25,008 74 54,480 16.31 3 2.32 3.52 47.09 5.10 5.30 10 3.50 10
5 3
5,000 3,066 73 32,520 17.7 3 2.27 5.24 48.23 4.10 8.20 10 6.35 10
* In the tables, porosity represents the void volume available for free gas only; hence, porosity is corrected for the pore com pressibility and methane-adsorption effects;
TOC is measured independently; k0 represents the ratio of kerogen pore volume to the total pore volume; Dk is the molecular diffusivity of free gas in kerogen.
Further investigation using different shale samples is currently and permeability, and it could be dominated by the adsorbed-
necessary. phase transport taking place in the organics. It is a difficult task
to experimentally differentiate slippage of free-gas molecules
Conclusions and surface-diffusion effects taking place in the phase-transition
In this paper, we have developed methodologies that allow us to region near the organic nanopore walls, although strong depen-
investigate in detail CO2 sequestration in organic-rich gas-shale dence on the adsorbed-phase gradient indicates the significance of
samples from Forth Worth basin. It is found that the pore-volume surface diffusion in the organics. Although the study is important
estimation is a crucial step for the sequestration considerations for our understanding of gas storage and transport in organic-rich
in gas shale. This volume may not be important for the storage shale, gas shales in North America may differ in terms of organic
of free gas, however, because up to 97% of the uptaken gas is content, pore structure of the organic material, and temperature.
stored in adsorbed state inside the organic pores, depending on the Further investigation involving samples from different shale
shale-gas-reservoir pressure and temperature. It is also found that plays is necessary for statistical inference and generalization of
the gas transport takes place in the presence of dynamic porosity the results.
Sample 21 Sample 23
2.5×10–9 2.5×10–9
Free gas diffusive mass flux Free gas diffusive mass flux
JK,x and JS,x(mol-cm/cm3/s)
Adsorbed gas diffusive mass flux Adsorbed gas diffusive mass flux
2×10–9 2×10–9
Darcy flux
Darcy flux
1.5×10–9 1.5×10–9
Direction of
gas permeation
1×10–9 1×10–9
5×10–10 5×10–10
0 0
0 1 2 3 4 5 0 1 2 3
Distance, cm Distance, cm
(a) (b)
Fig. 12—Predicted mass fluxes across the shale Sample 21 at initial equilibrium pore pressure of 2,974 psia and Sample 23 at
3,066 psia during the pressure-pulse-decay measurements. The mass fluxes are computed using forward simulation with the
optimized transport parameters shown in Table 5. Only the flux values at the halftime to reach the new equilibrium pressure
during the pulse-decay measurement are shown.
Summary Layne (1991), Mack et al. (1992), and Elbel et al. (1992), in which
Successfully creating multiple hydraulic fractures in horizontal fracture propagation was modeled according to the Geerstma and
wells is critical for unconventional gas production economically. DeKlerk geometry, the Perkins and Kern geometry, and a pseudo-
Optimizing the stimulation of these wells will require models that 3D geometry, respectively, with fluid flow in multiple fractures and
can account for the simultaneous propagation of multiple, poten- the wellbore, but with no mechanical interaction between multiple
tially nonplanar, fractures. fractures nor independent pressure drop in each fracture. Olson
In this paper, a novel fracture-propagation model (FPM) is (2008) used a pseudo-3D approach for vertical but laterally nonpla-
described that can simulate multiple-hydraulic-fracture propagation nar fractures, assuming zero-viscosity injection fluid to reduce
from a horizontal wellbore. The model couples fracture deformation computation time for multifracture problems. Roussel and Sharma
with fluid flow in the fractures and the horizontal wellbore. The dis- (2011) used a 3D finite-difference numerical model to calculate
placement discontinuity method (DDM) is used to represent the stress distributions around uniform-pressure fractures, accounting
mechanics of the fractures and their opening, including interaction for propagation by assuming fractures would follow principal-stress
effects between closely spaced fractures. Fluid flow in the fractures trajectories. Weng et al. (2011) and Wu et al. (2012) describe a
is determined by the lubrication theory. Frictional pressure drop in model for simulating multiple hydraulic fractures in naturally frac-
the wellbore and perforation zones is taken into account by applying tured reservoirs combining standard pseudo-3D equations for frac-
Kirchoff’s first and second laws. The fluid-flow rates and pressure ture-width calculations coupled with the displacement discontinuity
compatibility are maintained between the wellbore and the multiple approach of Olson (2004) to account for stress-interaction effects
fractures with Newton’s numerical method. The model generates between fractures. Meyer and Bazan (2011) presented a discrete-
physically realistic multiple-fracture geometries and nonplanar-frac- fracture-network model to simulate planar-fracture growth in the
ture trajectories that are consistent with physical-laboratory results three principal planes, with the Olson (2004) analytical solution to
and inferences drawn from microseismic diagnostic interpretations. describe the interaction between induced fractures. Wong and Xu
One can use the simulation results of the FPM for sensitivity (2013) modeled multiple-fracture propagation in horizontal wells
analysis of in-situ and fracture treatment parameters for shale-gas with mechanical interaction through the boundary integral formula-
stimulation design. They provide a physics-based complex frac- tion. Castonguay et al. (2013) implemented the weakly singular,
ture network that one can import into reservoir-simulation models symmetric Galerkin boundary-element method to calculate fracture
for production analysis. Furthermore, the results from the model geometry, in which flow in the fracture is modeled as power-law
can highlight conditions under which restricted width occurs that fluid flow in the arbitrary curved channels. Nagel et al. (2011) pro-
could lead to proppant screenout. posed a discrete-element model, in which the discrete-fracture net-
work consists of planar, finite fractures in three dimensions that
bound intervening rigid or deformable reservoir blocks. Dahi Tale-
Introduction ghani and Olson (2011) presented a coupled fluid-flow and rock-
Multiple hydraulic fracturing in combination with horizontal dril- mechanics model that was based on the extended finite-element
ling significantly improves well productivity in shale reservoirs. method to simulate the fracture geometry of plane-strain problems.
Limited-entry fracturing techniques are widely used to create Keshavarzi and Mohammadi (2012) also applied the extended fi-
simultaneously propagating multiple fractures (Daneshy 2011). In a nite-element method to analyze the interaction between hydraulic
single fracturing stage, there can be three to six perforation clusters. and natural fractures under plane-strain and quasistatic conditions.
In comparison with modeling single fracturing propagation, multi- The model presented in this paper follows the single nonplanar-
ple-fracture-propagation modeling needs to consider mechanical fracture model of Olson and Wu (2012). The model incorporates
fracture interaction, which is a dynamic process of subsurface geo- fracture deformation and non-Newtonian fluid flow in the wellbore
mechanical stress changes induced by fracture growth. The signifi- and fractures. Fracture widths and stress-intensity factors are calcu-
cant consequences of multifracture treatments are stress-shadow lated with the modified, pseudo-3D DDM approach of Olson
effects that lead to complex fracture geometry and fracture-width (2004), accounting for all mechanical interactions between every
restriction. Fracture geometry can be further complicated by pre- fracture element at every timestep. The flow rate in each fracture
existing natural fractures in shale-gas reservoirs (Gale et al. 2007; satisfies mass balance and pressure equilibrium between the hori-
Olson 2008; Chong et al. 2010), which is not the focus of this work zontal wellbore and hydraulic fractures. The Newton-Raphson nu-
but which will be discussed in a future publication. To optimize merical method is applied to solve nonlinear fluid-flow equations.
multiple-fracture treatments, characterization of the fracture geom- The Picard iterative method is used to fully couple fracture
etry is very necessary. However, the complex-fracture geometry mechanics and fluid flow. We validate our model against two ana-
cannot be measured directly and completely by any fracture diag- lytical solutions and published numerical results. A field case study
nostic techniques. Therefore, modeling remains an important tool is performed to predict multiple-fracture geometry and to calculate
for engineers to predict the fracture geometry. stress distribution with the purpose of demonstrating the feasibility
Recent efforts to model multifracture treatments have taken var- of the model.
ious forms. Multicrack models were described by Siriwardane and
Fracture Mechanics
Copyright V
C 2015 Society of Petroleum Engineers
Fracture mechanics determines fracture displacements and propa-
This paper (SPE 167626) was accepted for presentation at the SPE Annual Technical gation direction. The 2D DDM (Crouch 1976) with a 3D correc-
Conference and Exhibition, New Orleans, 30 September–2 October 2013, and revised for
publication. Original manuscript received for review 1 November 2013. Revised manuscript tion factor (Olson 2004) is used to calculate normal and shear
received for review 8 March 2014. Paper peer approved 17 April 2014. displacements considering mechanical interaction. The fracture-
y y
Dx Dy
x z
σzz
2a
σxx y
h
+
S x
–
S
x
Fig. 2—Sketch of a uniformly loaded, isolated, vertical fracture
Fig. 1—Illustration of a 2D boundary with crack-like geometry. with a finite height and an infinite length.
propagation direction is determined by the maximum-circumfer- Because hydraulic fractures are best modeled accounting for
ential-stress theory (Erdogan and Sih 1963), a criterion based on their 3D geometry (finite height and length), we use a 3D correc-
the stress-intensity factors at the crack tips. tion factor Gij derived by Olson (2004),
Fluid Flow pcf,i is pressure loss in the horizontal wellbore given by (Valko
Multiple-fracture propagation is driven by an incompressible and and Economides 1995)
non-Newtonian fluid. Fluid flow consists of two parts—fluid flow X
i
0
in the fracture and fluid flow in the horizontal wellbore between pcf ;i ¼ Ccf ðxj xj1 ÞQwj n ;
injection points. j¼1
2ðX
j1Þ
Fluid Flow in a Single Fracture. Fluid flow through a rock frac- Qwj ¼ QT Qk ;
ture is modeled by the Navier-Stokes equations of fluid mechanics k¼1 : . . . . . . . . . .ð11Þ
(Schlichting 1968). When assuming a uniform pressure gradient Qwj ¼ QT at j ¼ 1;
in the channel between two parallel and smooth fracture surfaces,
one can simplify the Navier-Stokes equations (Valko and Econo- and n0
mides 1995): 0 0 1 þ 3n0 0
Ccf ¼ 23n þ2 pn k0 Dð3n þ1Þ :
0 n0
@p 0 1 þ 2n0 n n0 ð2n0 þ1Þ n0
¼ 2n þ1 k0 h w Q : . . . . . . . . . . ð7Þ
@s n0 The pressure drop between the wellbore intervals is a function of
distance between two neighboring fractures and flow rate Qwj in the
From Eq. 7, we can obtain the relationship between pressure drop wellbore intervals. The longer the distance between two neighboring
and fluid properties, fracture geometry, and flow rate. Pressure fractures, the larger the pressure drop. This is a rough approach to esti-
drop increases with increasing flow rate Q and decreases with mate frictional pressure drop in the wellbore without considering
increasing fracture width w. In our model, to enhance conver- flowing conditions. Siriwardane and Layne (1991) gave a more-accu-
gence stability with little cost in accuracy, the flow-rate gradient rate approach to calculate the pressure loss in the wellbore by taking
along the fracture is ignored. Spence and Sharp (1985) and Zim- into account friction factors under turbulent and laminar conditions.
merman et al. (1991) discussed the validation of the lubrication
theory for modeling fluid flow inside the fractures. According to Material Balance. One must satisfy material balance in the
their analysis, fracture aperture cannot change too abruptly and whole fluid-flow system. The total volume of fluid pumped during
should be small in comparison to the other dimensions of the frac- injection time is equal to the fluid volume preserved in the multi-
tures, and fluid flow should be laminar. ple fractures and the volume leaking into the surrounding porous
medium. The global volume balance is
Fluid Flow in Horizontal Wellbore. A numerical model for
multilayer-fracturing treatments in vertical wells was presented ðt i ðtÞ
Lð i ðtÞ ð
Lð t
X
N X
N
by Siriwardane and Layne (1991), Mack et al. (1992), and Elbel QT ðtÞdt ¼ hwds þ qL ðs; tÞdtds: . . . . ð12Þ
et al. (1992). The model described the conservation of volume 1 1
0 0 0 0
and the continuity of pressure by use of a set of constraints, such
as hydrostatic-pressure change, perforation frictional pressure The leakoff volume is described by the Carter leakoff model
drop, and frictional pressure loss in each casing interval. Fluid (Carter 1957), which is expressed as
flow in the horizontal wellbore is similar to fluid flow in the verti-
cal wellbore with multiple layers. The only difference is that we 2hCL
qL ðs; tÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : . . . . . . . . . . . . . . . . . . . . . . . ð13Þ
assume no hydrostatic-pressure change in the horizontal wellbore. t sðsÞ
Fluid flow in the wellbore is analogous to the flow of electric
current through an electrical circuit network, by applying Kirch- The model assumes that leakoff is 1D in the direction perpen-
off’s first and second laws to obtain the flow-rate distribution of dicular to the fracture plane.
multiple fractures (Siriwardane and Layne1991; Mack et al. 1992;
Elbel et al. 1992). Our model assumes that multiple-transverse Iterative Algorithm
fractures are started perpendicular to the horizontal wellbore at
The fracture mechanics and fluid flow are coupled numerically
perforation-cluster locations. The total injection rate QT is divided
through an iterative algorithm. The elastic DDM solution deter-
into injection rates of each fracture Qi. The flow-rate distribution
is a dynamic process, and the diversion of the fluid into various mines the displacement discontinuities (opening and shearing) for
perforation clusters is a function of wellbore friction, perforation each fracture element because of its internal pressure and resolved
friction, and the fluid pressure within the fractures (Fig. 3). external stresses, whereas the internal-pressure distribution is
By ignoring wellbore-storage effects, total injection rate related to fracture opening by the fluid flow. The mechanical solu-
tion is a linear set of equations, whereas the fluid flow is nonlin-
should be the sum of the injection rates of all fractures,
ear. Fluid flow is discretized by the finite-difference method on a
X
N fixed grid of boundary elements. Newton’s numerical method is
QT ¼ Qi : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð8Þ used to solve the system of nonlinear equations. Picard coupling
i¼1 (Li 1991; Adachi et al. 2007) is used to couple the fracture
mechanics and fluid flow at every updated timestep. If the conver-
Kirchoff’s second law describes the continuity of pressure, in gence criterion is not satisfied, the net pressure is modified at each
which the pressure in the wellbore heel is equal to the sum of iteration step as
wellbore frictional pressure drop, perforation frictional pressure
drop, and pressure in the first element of the fracture as (Elbel pkþ1 ¼ ð1 ar Þpk1 þ ar pk : . . . . . . . . . . . . . . . . . . . ð14Þ
et al. 1992)
X
i Displacement discontinuities are recalculated with the modi-
P0 ¼ pw;i þ ppf ;i þ pcf ; j : . . . . . . . . . . . . . . . . . . . ð9Þ fied pressure from the previous iteration step. The convergence
j¼1 criterion is checked again after computing the new net pressure
on the basis of the displacements. ar is a retardation coefficient
pw,i is pressure in the first element of fracture, which is calculated ranging from 0 to 0.5 that dampens variations to avoid stress
by Eq. 7; ppf,i is the perforation frictional pressure loss that is pro- jumps and extreme changes in pressure. If the convergence crite-
portional to the square of flow rate, estimated by (Elbel et al. rion cannot be satisfied, decreasing ar is an approach to improve
1992) convergence. Smaller ar can cause convergence to be too slow,
qs whereas a large ar can result in diverging oscillations. We found
ppf ;i ¼ Qi 2 : . . . . . . . . . . . . . . . . . . . ð10Þ that ar ¼ 0.2 provides a reasonably stable convergence. The flow
0:323qw n2p;i dp;i
4
chart in Fig. 4 describes the structure of the program.
Model Validation pared with the PKN model. At the earlier time of propagation,
Comparison With Analytical Solutions. Our FPM extends from KGD gives more-accurate results than PKN because fracture length
a nonplanar single-fracture model to a multiple-fracture model. is shorter than fracture height. As fracture length grows, KGD grad-
The single-fracture, nonplanar model developed by Olson and Wu ually loses its accuracy, and PKN can give more-accurate results.
(2012) was validated against planar pseudo-3D models from the The other input parameters are listed in Table 2.
Gas Research Institute Staged Field Experiment No. 3 case (War- A comparison of results made for fracture length (Fig. 5), frac-
pinski et al. 1994). In this paper, we compare our model to the ture width at the wellbore (Fig. 6), and net injection pressure
end-member analytical solutions of Khristianovich-Geertsma- (Fig. 7) is presented. As shown, our model can give a very good
Deklerk (KGD) and Perkins-Kern-Nordgren (PKN) (Table 1). agreement with the KGD in the earlier time and tends to match
The KGD model presumes plane strain in the horizontal plane and PKN analytical solutions as the fracture grows longer, which
a rectangular vertical cross section (no width variation over the lends credence to our model results and analysis.
height). The PKN model assumes plane strain in a vertical plane
normal to the fracture-propagation direction, and the vertical cross Comparison With Unconventional Fracture Model. For multi-
section of the fracture is elliptical in shape. To compare with the ple-fracture propagation, fracture interaction plays an important
KGD and PKN solutions, we simulate a case with a fracture height role in fracture geometry. For nonplanar propagation validation,
of 200 ft. The fracture geometry calculated by KGD is much fatter we refer for comparison to the nonplanar model of Weng et al.
and shorter with very low net pressure that climbs with time, com- (2011) and Wu et al. (2012) for the configurations of a single
Initialization
A new timestep
Displacement
Fracture mechanics
Stress shadow
Pressure
Fluid flow
Flow rate
No Converge
Yes
Mass balance
Finished
Postprocessing
TABLE 1—ANALYTICAL SOLUTION OF KGD AND PKN (VALKO AND ECONOMIDES 1995)
1200
TABLE 2—INPUT PARAMETERS
FPM KGD PKN
1000
Fracture half-length, ft
l 100 cp
Injection Rate 20 bbl/min 800
Young’s Modulus 6,530,000 psi
Poisson’s Ratio 0.2 600
Height 200 ft
400
Minimum Horizontal Stress 4,450 psi
200
0.4 0
Average width at the wellbore, in.
0.15
1200
0.1 FPM KGD PKN
Net injection pressure, psi
0.05 900
0
0 10 20 30 40 50 60 70
Injection time, min 600
horizontal wellbore with two initial fractures (Fig. 8a) and two
horizontal wellbores with one fracture each (Fig. 8b). We simu- 0
0 10 20 30 40 50 60 70
lated these two configurations under isotropic and anisotropic in-
situ stress conditions (see Table 3 for input parameters). The dif- Injection time, min
ferential stress for the anisotropic-stress reservoir is 130 psi, with
the maximum horizontal stress oriented in the y-direction. Fig. 7—Comparison of fracture pressure at the wellbore
between the KGD and PKN models and our model.
Two fractures, started parallel with each other in a horizontal
wellbore for isotropic and anisotropic far-field stresses, are illus- trated in Fig. 9. The two fractures propagate away from each other
because of the mechanical fracture interaction (i.e., stress-shadow
effects). An opening fracture exerts additional stresses on sur-
rounding rock and adjacent fractures, which can result in a local
change in the direction of maximum horizontal stress and a devia-
tion of the fracture path from a planar geometry. In addition, the
fractures have larger curvature under the isotropic-stress condition
than that under the anisotropic condition. This is because the ani-
sotropic far-field stresses tend to suppress fracture curving (Olson
and Pollard 1989). The stress shadow and the anisotropic far-field
stresses compete with each other. To quantify the competition,
Olson and Dahi Taleghani (2009) defined a parameter called the
relative net pressure Rn,
ðPfrac Shmin Þ
Rn ¼ ; . . . . . . . . . . . . . . . . . . . . . . . ð15Þ
ðShmax Shmin Þ
(a) (b) where the numerator is the net pressure and the denominator is hori-
zontal differential stress. Stress-shadow effects strongly depend on
Fig. 8—Two initiation fractures in a horizontal wellbore (a) and the net pressure within the fractures (Pollard and Segall 1987; War-
two horizontal wellbores (b); blue lines represent horizontal pinski and Branagan 1989). Larger net pressure will lead to stronger
wellbores, and red lines represent initial fractures. stress-shadow effects. When Rn is much greater than unity, the net
–850 –850
FPM FPM
Wu et al. 2012 Wu et al. 2012
–900 –900
–950 –950
–1000 –1000
y, ft
y, ft
–1050 –1050
–1100 –1100
–1150 –1150
–1200 –1200
–60 –10 40 90 –60 –10 40 90
x, ft x, ft
Fig. 9—Comparison of propagation paths for two initially parallel fractures in isotropic and anisotropic stress fields: left figure
refers to isotropic stress field; right figure refers to anisotropic stress field.
–850
–850 FPM
FPM
Wu et al. 2012
Wu et al. 2012
–900 –900
–950 –950
–1000 –1000
y, ft
y, ft
–1050 –1050
–1100 –1100
–1150 –1150
–1200 –1200
–60 –10 40 90 –60 –10 40 90
x, ft x, ft
Fig. 10—Comparison of propagation paths for two initially offset fractures in isotropic and anisotropic stress fields: left figure
refers to isotropic stress field; right figure refers to anisotropic stress field.
fractures are shorter, narrower, and more nonplanar than the exte- rxx is much larger than that for ryy, because fractures open against
rior fractures. the rock and directly squeeze the rock in the x-direction, whereas
One can further understand the resultant fracture paths and y-direction effects are reduced by Poisson’s ratio effects. Thus, at
widths by examining the stress distributions induced by the frac- some specific points the change in rxx is greater than the change
tures. Before fracturing, the minimum horizontal stress of rxx is ori- in ryy, such that rxx might become larger than ryy. This flip in the
ented parallel to the x-direction, and the maximum horizontal stress orientation of the compressive stress explains why interior frac-
is the ryy. Fractures grow along the y-direction because that path pro- tures change their propagation directions to seek the path with
vides the least resistance. As the fractures grow, geomechanical least resistance.
stresses around fractures are dynamically changed. In the Figs. 14 One also can use the induced stress field to predict the poten-
and 15, red colors indicate an increase of compressional stress, and tial for microseismic activity. Microseismic events are induced by
blue colors represent a decrease of compression. Nearby and ahead changes in stresses and pore pressure caused by hydraulic fractur-
of the fracture tips, compressional stresses of rxx and ryy decrease ing (Warpinski et al. 2001; Vulgamore et al. 2007), and are often
because of induced tensile stresses in the near-tip field. In contrast, used to estimate fracture geometry. The stress changes generated
rock behind the tips is compressed because of the fracture opening. by opening fractures might lead to the shear failure of pre-existing
In comparing the change in rxx and ryy at the end of injection fractures. Fig. 16 shows the change of the maximum shear stress
(Figs. 14 and 15), one can note that the area of stress change for at the end of fracturing. As shown in the figure, areas representing
0 psi of the shear stress change indicate no change from the initial
state. Red colors mean an increase in the maximum shear stress,
TABLE 4—INPUT PARAMETERS FOR CASE STUDY whereas blue colors indicate a decrease. The initial maximum
shear stress is 50 psi because the differential stress (SHmax–Shmin)
Fracture Spacing 50 ft is 100 psi. A larger maximum shear stress suggests the greater
Layer Height 100 ft likelihood to induce shear failure. Fig. 16 illustrates that the large
Minimum Horizontal Stress 4,450 psi increase of the shear-stress primarily occurs around fracture tips
Maximum Horizontal Stress 4,550 psi and the region between two fractures. Accordingly, in these areas,
Fluid Leakoff coefficient 0.00001 ft/min0.5
there is a possibility to induce shear failure of pre-existing fac-
tures and generate microseismic activity. Warpinski et al. (2001)
Injection Rate 60 bbl/min
reported that considerable shear stress near fracture tips developed
n0 0.7 by hydraulic fractures provides a primary mechanism for microse-
k0 0.02 lbf-secn/ft2 isms to accurately determine the length and height of fractures,
Young’s Modulus 6.53106 psi and pore-pressure change because of fluid leakoff is an important
Poisson’s Ratio 0.2 factor in microseismic development behind the tip. Our analysis,
Number of Perforations 60 however, did not take into account fluid-leakoff effects.
Density of Slurry 1.2 g/cm3
Density of Water 1.0 g/cm3
Conclusions
Diameter of Perforations 5 mm
The FPM presented in this paper is able to simulate simultaneous
Diameter of Wellbore 0.1 m
multiple-fracture propagation. The model integrates fracture
400
300
200
100
y, ft
–100
–200
–300
–400
–500
–200 –100 0 100
x, ft
(a) Fracture trajectory (b) Fracture width distribution
Fig. 11—Fracture trajectory and aperture distribution after fracturing (map view), twofold exaggeration of x-axis scale.
interaction between multiple fractures and fluid flow in the single • Shear stress is generated during fracturing and might cause shear
fracture and the horizontal wellbore. Mass balance and pressure slippage of pre-existing fractures and develop microseismic ac-
equilibrium between multiple fractures and the horizontal well- tivity to help determine the geometry of multiple fractures.
bore are satisfied. From our model, we can conclude: • A map of the maximum shear stress change can assess the
• Simultaneous propagating fractures extend, attracting or repel- expected area of the shear failure of pre-existing fractures
ling each other as a result of stress-shadow effects, resulting in induced by hydraulic fracturing.
the direction and magnitude change of maximum and minimum
horizontal stresses.
• Under the same situation, fractures closer to the injection point Nomenclature
can obtain more fluid and propagate longer and wider. a ¼ half-length of the element, ft
• Nonplanar-geometry reduced propagation occurs for interior CL ¼ fluid-loss coefficient, ft/min0.5
fractures, causing a significant restriction of fracture aperture D ¼ diameter of wellbore, in.
and reduction of length. Ds ¼ shear displacement of the element, in.
• Compression beside the fractures increases, but is decreased at Dn ¼ normal displacement of the element, in.
fracture tips. At some points around fractures, the increment of dij ¼ distance between the center of elements i and j, ft
minimum horizontal stress is much larger than that of maxi- dp,i ¼ diameter of perforation, in.
mum horizontal stress, resulting in direction change of maxi- E ¼ Young’s modulus, psi
mum horizontal stress and nonplanar fracture geometry. E0 ¼ plane-strain modulus, psi
800
0.25 Frac. 1 Frac. 2 Frac. 3
Frac. 1 Frac. 2 700 Frac. 4 Frac. 5 Frac. 6
Frac. 3 Frac. 4
0.02 Frac. 5 Frac. 6 600
Net pressure, psi
Fracture width, in.
500
0.15
400
300
0.01
200
0.05 100
0
0 0 100 200 300 400 500
0 100 200 300 400 500
Fracture length, ft
Fracture length, ft
Fig. 13—Net pressure in each fracture changing along fracture
Fig. 12—Fracture width vs. fracture length. length.
y, ft
0 0 0 0
–100 –100
–100 –100
–200 –200
–300 –200 –300 –200
–400 –400
–500 –300 –500 –300
Fig. 14—Stress change of normal stress rxx (map view); twofold Fig. 15—Stress change of normal stress ryy (map view); twofold
exaggeration of x-axis scale. exaggeration of x-axis scale.
Elbel, J.L., Piggott, A.R., and Mack, M.G. 1992. Numerical Modeling of Schlichting, H. 1968. Boundary-Layer Theory, sixth edition. New York:
Multilayer Fracture Treatments. Presented at the SPE Permian Basin McGraw-Hill.
Oil and Gas Recovery Conference, Midland, Texas, 18–20 March. Sheibani, Farrokh and Olson, Jon. 2013. Stress Intensity Factor Determi-
SPE-23982-MS. http://dx.doi.org/10.2118/23982-MS. nation for Three-Dimensional Crack Using the Displacement Disconti-
Erdogan, F. and Sih, G.C. 1963. On the Crack Extension in Plates Under nuity Method With Applications to Hydraulic Fracture Height Growth
Plane Loading and Transverse Shear. J. Basic Eng. 519–527. and Non-Planar Propagation Paths. In Effective and Sustainable Hy-
Gale, Julia F.W., Reed, Robert M., and Holder, Jon. 2007. Natural Frac- draulic Fracturing, ed. A.P. Bunger, J. McLennan, and R. Jeffrey,
tures in the Barnett Shale and Their Importance for Hydraulic Fracture Chap. 37. Rijeka, Croatia: InTech. http://dx.doi.org/10.5772/56308.
Treatments. AAPG Bull. 91 (4): 603–622. http://dx.doi.org/10.1306/ Shou, K.J. 1993. A High Order Three-Dimensional Displacement Discon-
11010606061. tinuity Method With Application to Bonded Half-Space Problems. PhD
Keshavarzi, R. and Mohammadi, S. 2012. A New Approach for Nu- thesis, University of Minnesota, Minnesota (August 1993).
merical Modeling of Hydraulic Fracture Propagation in Naturally Siriwardane, H.J. and Layne, A.W. 1991. Improved Model for Predicting
Fractured Reservoirs. Presented at the SPE/EAGE European Un- Multiple Hydraulic Fracture Propagation From a Horizontal Well. Pre-
conventional Resources Conference and Exhibition, Vienna, Aus- sented at the SPE Eastern Regional Meeting, Lexington, Kentucky,
tria, 20–22 March. SPE-152509-MS. http://dx.doi.org/10.2118/ 22–25 October. SPE-23448-MS. http://dx.doi.org/10.2118/23448-MS.
152509-MS. Spence, D.A. and Sharp, P.W. 1985. Self-Similar Solution for Elastohy-
Li, Yi. 1991. On Initiation and Propagation of Fractures From Deviated drodynamic Cavity Flow. Proc. of the Royal Society of London, Series
Wellbores. Dissertation, The University of Texas at Austin (May A (400), pp. 289–313.
1991). Valko, P. and Economides, M.J. 1995. Hydraulic Fracture Mechanics.
Mack, Mark G., Elbel, Jacques L., and Piggott, Andrew R. 1992. Numeri- New York: John Wiley & Sons.
cal Representation of Multilayer Fracturing. Rock Mechanics, ed. Till- Vulgamore, T., Clawson, T., Pope, C. et al. 2007. Applying Hydraulic
erson and Wawersik. Balkema, Rotterdam. ISBN 90 5410 0451. Fracture Diagnostics to Optimize Stimulations in the Woodford Shale.
Meyer, B.R. and Bazan, L.W. 2011. A Discrete Fracture Network Model Presented at the SPE Annual Technical Conference and Exhibition,
for Hydraulically Induced Fractures: Theory, Parametric and Case Anaheim, California, 11–14 November. SPE-110029-MS. http://
Studies. Presented at the SPE Hydraulic Fracturing Technology Con- dx.doi.org/10.2118/110029-MS.
ference and Exhibition, The Woodlands, Texas, 24–26 January. SPE- Warpinski, N.R. and Branagan, P.T. 1989. Altered-Stress Fracturing. J Pet
140514-MS. http://dx.doi.org/10.2118/140514-MS. Technol 41 (9): 990–997. SPE-17533-PA. http://dx.doi.org/10.2118/
Nagel, N., Gil, I., Sanchez-Nagel, M. et al. 2011. Simulating Hydraulic 17533-PA.
Fracturing in Real Fractured Rock—Overcoming the Limits of Pseudo Warpinski, N.R., Moschovidis, Z.A., Parker, C.D. et al. 1994. Comparison
3D Models. Presented at the SPE Hydraulic Fracturing Technology Study of Hydraulic Fracturing Models—Test Case: GRI Staged Field
Conference and Exhibition, The Woodlands, Texas, 24–26 January. Experiment No. 3. SPE Prod & Fac 9 (1): 7–16. SPE-28158-PA.
SPE-140480-MS. http://dx.doi.org/10.2118/140480-MS. Warpinski, N.R., Wolhart, S.L., and Wright, C.A. 2001. Analysis and Pre-
Olson, J.E. 1990. Fracture Mechanics Analysis of Joints and Veins. Dis- diction of Microseismicity Induced by Hydraulic Fracturing. Presented
sertation, Stanford University, Stanford, California (December 1990). at the SPE Annual Technical Conference and Exhibition, New Orle-
Olson, J.E. 2004. Predicting Fracture Swarms—The Influence of Subcriti- ans, Louisiana, 30 September–3 October. SPE J. 9 (1): 24–33. SPE-
cal Crack Growth and the Crack-Tip Process Zone on Joint Spacing in 87673-PA. http://dx.doi.org/10.2118/87673-PA.
Rock. In The Initiation, Propagation, and Arrest of Joints and Other Weng, X., Kresse, O., Cohen, C. et al. 2011. Modeling of Hydraulic Frac-
Fractures, ed. J.W. Cosgrove and T. Engelder, 73–87. Geological So- ture Network Propagation in a Naturally Fractured Formation. Pre-
ciety of London Special Publication 231. sented at the SPE Hydraulic Fracturing Technical Conference and
Olson, J.E. 2007. Fracture Aperture, Length and Pattern Geometry Devel- Exhibition, The Woodlands, Texas, 24–26 January. SPE-140253-MS.
opment Under Biaxial Loading: A Numerical Study With Applications http://dx.doi.org/10.2118/140253-MS.
to Natural, Cross-Jointed Systems. In Fracture-Like Damage and Wong, Sau-Wai and Xu, Guanshui. 2013. Interaction of Multiple Hydraulic
Localisation, ed. G. Couples and H. Lewis, Vol. 289, 123-142. Geo- Fractures in Horizontal Wells. Presented at the SPE Middle East Uncon-
logical Society of London, Special Publications. ventional Gas Conference and Exhibition, Muscat, Oman, 28–30 Janu-
Olson, J.E. 2008. Multi-Fracture Propagation Modeling: Applications to ary. SPE-163982-MS. http://dx.doi.org/10.2118/163982-MS.
Hydraulic Fracturing in Shales and Tight Gas Sands. Presented at the Wu, R., Kresse, O., Weng, X. et al. 2012. Modeling of Interaction of Hy-
42nd U.S. Symposium on Rock Mechanics, San Francisco, California, draulic Fractures in Complex Fracture Networks. Presented at the SPE
29 June–2 July. Hydraulic Fracture Technology Conference, The Woodlands, Texas,
Olson, J.E. and Wu, K. 2012. Sequential Versus Simultaneous Multi-zone 6–8 February. SPE-152052-MS. http://dx.doi.org/10.2118/152052-MS.
Fracturing in Horizontal Wells: Insights From a Non-planar, Multi- Zimmerman, R.W., Kumar, S., and Bodvarsson, G.S. 1991. Lubrication
frac Numerical Model. Presented at the SPE Hydraulic Fracturing Theory Analysis of the Permeability of Rough-Walled Fractures. Inter-
Technology Conference, The Woodlands, Texas, 6–8 February. SPE- national J. Rock Mechanics Mine Sci. & Geomechanics. Abstr. 28 (4):
152602-MS. http://dx.doi.org/10.2118/152602-MS. 325–331.
Olson, J.E. and Dahi Taleghani, A. 2009. Modeling Simultaneous Growth
of Multiple Hydraulic Fractures and Their Interaction With Natural Kan Wu is a PhD candidate in the Department of Petroleum and
Fractures. Presented at the SPE Hydraulic Fracturing Technology Con- Geosystems Engineering at the University of Texas at Austin (UT).
ference, The Woodlands, Texas, 19–21 January. SPE-119739-MS. Her research is focused on developing a fracture-propagation
http://dx.doi.org/10.2118/119739-MS. model to simulate multiple fracture propagation in naturally
Olson, J.E. and Pollard, D.D. 1989. Inferring Paleostresses From Natural fractured reservoirs. Wu holds MS and BS degrees in petroleum
Fracture Patterns: A New Method. Geology 17 (4): 345–348. http:// engineering from the China University of Petroleum. She is a
member of SPE. Wu won the First Place in the PhD Division at the
dx.doi.org/10.1130/0091-7613.
2013 SPE Gulf Coast Region Paper Contest and Second Place in
Pollard, D.D. and Segall, P. 1987. Theoretical Displacements and Stresses the PhD Division at the 2013 SPE Local Paper Contest.
Near Fractures in Rock: With Applications to Faults, Joints, Veins,
Dikes and Solution Surfaces. In Fracture Mechanics of Rock, ed. B.K. Jon E. Olson is an associate professor in petroleum and geosys-
Atkinson, 277-350. London: Academic Press. tems engineering at UT. He is co-principal investigator of the
Fracture Research and Application Consortium at UT, and spe-
Roussel, Nicolas P. and Sharma, Mukul M. 2011. Strategies to Minimize
cializes in rock-mechanics applications to structural geology,
Frac Spacing and Stimulate Natural Fractures in Horizontal Comple- hydraulic fracturing, and reservoir characterization. Before join-
tions. Presented at the SPE Annual Technical Conference and Exhibi- ing UT in 1995, Olson spent 6 years as a research engineer at
tion. Denver, Colorado, 30 October–2 November. SPE-146104-MS. Mobil Research and Development Corporation. He holds a
http://dx.doi.org/10.2118/146104-MS. PhD degree in geomechanics from Stanford University.
and has been mostly ignored in practice. However, in unconven- gaseous and aqueous. For simplicity, the gas and water compo-
tional shale formations with nanosized pores or nanosized micro- nents are assumed to be present only in their associated phases
fractures, such geomechanics effects can be relatively large and and adsorbed gas is within the solid phase of rock. Each fluid
may have a significant impact on both fracture and matrix perme- phase flows in response to pressure and gravitational and capillary
ability, which has to be considered, in general. Wang et al. (2009) forces according to the multiphase extension of Darcy’s law or
show that permeability in the Marcellus shale is pressure-depend- several extended non-Darcy-flow laws, discussed next. In an iso-
ent and decreases with an increase in the confining pressure (or thermal system containing two mass components, subject to mul-
total stress). The effect of confining pressure on permeability is tiphase flow and adsorption, two mass-balance equations are
caused by a reduction of porosity. Bustin et al. (2008) report the needed to fully describe the system, as described in an arbitrary
effect of stress (confining pressure) in Barnett, Muskwa, Ohio, flow region of a porous or fractured domain for flow of phase b
and Woodford shales and show that the degree of permeability (b ¼ g for gas and b ¼ w for water),
reduction with confining pressure is significantly higher in shales
than in consolidated sandstone or carbonate. @
ð/Sb qb þ mb Þ ¼ r ðqb vb Þ þ qb : . . . . . . . . . . . . ð1Þ
This paper presents a generalized mathematical model and nu- @t
merical approach for unconventional-gas-reservoir simulation.
where / is the effective porosity of the porous or fractured media;
We present a unified framework model that is able to incorporate
Sb is the saturation of fluid b; qb is the density of fluid b; vb is the
many known mechanisms and processes for two-phase gas flow in
volumetric velocity vector of fluid b, determined by Darcy’s law
shale gas or tight gas formations with a continuum modeling
or non-Darcy-flow models, discussed next; t is time; mb is the
approach. The numerical scheme is based on generalized flow
adsorption or desorption mass term for the gas component per
models with unstructured grids. We discuss the numerical imple-
unit volume of formation; and qb is the sink/source term of phase
mentation of Klinkenberg effects, non-Darcy flow, gas adsorption,
(component) b per unit volume of formation.
and geomechanics effects into the mathematical model. In the nu-
merical-modeling examples, we apply geomechanics-coupled per-
meability to fractures, and apply the Klinkenberg effect to matrix Incorporation of Gas Adsorption and Desorption. The amount
media. This is because the Klinkenberg effect is apparent with of adsorbed gas in a given shale gas formation is generally
extremely low permeability or tiny pores, whereas fractures are described with the Langmuir’s isotherm (e.g., Moridis et al. 2010;
very sensitive to the change in effective stress. Results of our Mengal and Wattenbarger 2011; Silin and Kneafsey 2011; EIA
model-verification effort are also presented. We demonstrate 2011; Wu et al. 2012; Wu and Wang 2012) (i.e., it is correlated to
model application to quantify hydraulic fractures and transient reservoir gas pressure). To incorporate the gas adsorption or de-
flow behavior in shale gas reservoirs. sorption mass term in the mass-conservation equation, the amount
One of the critical issues in shale-gas-reservoir simulation is of adsorbed gas is determined according to the Langmuir’s iso-
how to handle fracture flow and fracture/matrix interaction. This therm as a function of reservoir pressure. As the pressure
is because the gas flow and production rely on fractures in these decreases with continuous gas production through production
reservoirs. Cipolla et al. (2009) built a methodology on modeling wells in reservoirs, more adsorbed gas is released from the solid
complex fracture geometry and heterogeneity from the microseis- to the free-gas phase in the pressure-lowering region, contributing
mic data. In this paper, we present a hybrid fracture-modeling to the total gas flow or production. In our model, the mass of
approach, defined as a combination of explicit fracture modeling adsorbed gas in unit formation volume is described (Leahy-Dios
and multicontinuum, multiple-interacting-continua (MINC) (Pruess et al. 2011; Silin and Kneafsey 2011; Wu et al. 2012) as
and Narasimhan 1985), and single-porosity modeling approaches,
which seems the best option for modeling a shale gas reservoir mg ¼ qR qg VE ; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð2Þ
with both hydraulic fractures and natural fractures. This is because
hydraulic fractures, which have to be dealt with for shale gas pro- where mg is adsorbed gas mass in unit formation volume; qR is
duction, are better handled by the explicit fracture method but can- rock bulk density; qg is gas density at standard condition; and VE
not be modeled, in general, by a dual-continuum model. On the is the adsorption isotherm function or gas content in scf/ton
other hand, naturally fractured reservoirs are better modeled by a (or standard gas volume adsorbed per unit rock mass). If the
dual-continuum approach, such MINC, for extremely low-perme- adsorbed-gas terms can be represented by the Langmuir isotherm
ability matrix in shale gas formations, which cannot be modeled by (Langmuir 1916), the dependency of adsorbed-gas volume on
an explicit fracture model. Specifically, we demonstrate how to use pressure at constant temperature is given as
the hybrid modeling approach to describe different types and scales
of fractures from the explicit modeling of hydraulic fractures and P
fracture network in the SRV to distributed natural fractures, micro- VE ¼ VL ; . . . . . . . . . . . . . . . . . . . . . . . . . . . ð3Þ
P þ PL
fractures, and tight matrix.
where VL is the Langmuir’s volume in scf/ton; P is reservoir gas
pressure; and PL is Langmuir’s pressure, the pressure at which
Flow-Governing Equations 50% of the gas is desorbed. In general, Langmuir’s volume VL is
In most cases of gas production from shale gas formations, a two- a function of the organic richness (or TOC) and thermal maturity
phase (gas/liquid) -flow model or a multiphase-flow model is con- of the shale.
sidered to be sufficient for simulation studies. This is because Note that Eq. 3 is valid only for the case when the Langmuir
what we are most concerned with in shale-gas-reservoir simula- model is applicable. In general, VE in Eq. 2 can be determined
tion is the modeling of gas flow from reservoir to well. However, from any correlation of gas adsorption as a function of reservoir
in addition to the gas phase, liquid-phase flow is often occurring gas pressure, which may be defined by a table lookup from labora-
simultaneously with gas flow; it needs to be considered when two tory studies for a given unconventional reservoir.
cases exist—mobile in-situ connate water and an abundance of In the literature, the most commonly used empirical model
aqueous hydraulic-fracturing fluids, which are sucked into the for- describing sorption onto organic carbon in shales is analogous to
mations surrounding the wells. Therefore, in this paper, we pri- that used in coalbed methane and follows the Langmuir isotherm
marily discuss the two-phase (gas and liquid) -flow model and (Gao et al. 1994; Moridis et al. 2010), such as Eq. 2. This adsorp-
formulation and treat single-phase gas flow as a special case of tion-modeling approach is based on the assumption that an instan-
the two-phase flow for the simulation studies of unconventional taneous equilibrium exists between the sorbed and the free gas
gas reservoirs. (i.e., there is no transient-time lag between pressure changes and
A multiphase system of gas and water (or liquid) in a porous the corresponding sorption/desorption responses; i.e., the equilib-
or fractured unconventional reservoir is assumed to be similar to rium model of the Langmuir sorption is assumed to be valid,
what is described in a black-oil model, composed of two phases: which provides a good approximation in shale gas modeling).
60 60
Permeability, µD
Permeability, µD
40 40
20 20
0 0
0 1000 2000 3000 0 0.005 0.01 0.015
Pore Pressure, psia Reciprocal of Pore Pressure, 1/psia
Fig. 2—Contribution of the Klinkenberg effect to the apparent matrix permeability (Ozkan et al. 2010).
microscale or pore-scale models have considered the beta factor We analyze the Klinkenberg effect with three different matrix
as a function of gas pressure or the Knudsen number. In applica- permeabilities—1:0 103 md, 1:0 105 md; and 1:0 107
tion, the Klinkenberg effect should be modeled with the labora- md; as shown in Fig. 2 and Fig. 4. We can see that the contribu-
tory-determined beta factor either as a constant or as a pressure- tion of the Klinkenberg effect is more significant at low pressures
dependent function or simply treating the apparent gas permeabil- and for lower values of permeability. This estimation also pro-
ity as a function of pressure from a table lookup to include the vides reliable values of the beta factor for analyzing the Klinken-
Klinkenberg effect or the Knudsen diffusion. An example relation berg effect.
between permeability and pressure, as shown in Fig. 2, can be
directly used for the reservoirs concerned, if a site-specific study The Incorporation of NonDarcy Gas Flow. In addition to mul-
provides such correlations or plots. tiphase Darcy flow, non-Darcy flow may also occur between and
After a comparison of Fig. 1 and Fig. 3, the Klinkenberg effect among the continua, such as along fractures, in unconventional
seems to have less impact than that of geomechanics, and they are gas reservoirs. The flow velocity, vß, for the non-Darcy flow of
going in the opposite directions. Geomechanics has an effect each fluid may be described with the multiphase extension of the
mainly on the microfractures and stimulation fractures, whereas Forchheimer equation (e.g., Wu 2002),
the Klinkenberg effect is primarily on the matrix media with
nanosized pores or fractures. lb
ðrUb Þ ¼ vb þ bb qb vb jvb j; . . . . . . . . . . . . . ð14Þ
A dynamic gas-slippage theory was proposed on the basis of k krb
the assumption that gas travels under the influence of a concentra-
tion field (random molecular flow) and a pressure field (macro- where ßß is presented the effective non-Darcy-flow coefficient
scopic flow) (Ertekin et al. 1986). According to this theory, the with a unit m–1 for fluid b under multiphase-flow conditions. The
Klinkenberg factor is not a constant anymore, but a pressure-de- correlation proposed by Evans and Civan is used to determine the
pendent value. They gave the expression of the Klinkenberg fac- non-Darcy-flow beta factor in the Forchheimer equation (Evans
tor as and Civan 1994) in our simulation examples, such as
where cg is gas compressibility and D is diffusivity coefficient. where the unit of k is md and the unit of b is ft1 . This correlation
In Eq. 12, the correlation to compute the diffusivity constant is for b matched with more than 180 data points, including those for
given by Ertekin et al. (1986): propped fractures (correlation coefficient ¼ 0.974).
1.0E–2
3200
1.0E–3 km = 1E–7 md
km = 1E–3 md
1600
Permeability, md
1.0E–4
b factore, psi
1.0E–5 800
km = 1E–5 md km = 1E–5 md
1.0E–6 400
1.0E–7 200
km = 1E–7 md
1.0E–8 km = 1E–3 md
100
300.0 600.0 1200.0 2400.0 4800.0
300.0 600.0 1200.0 2400.0 4800.0
Pressure, psi
Pressure, psi
Fig. 3—Effect of pore pressure on gas permeability in the Mar-
cellus shale, with a confining pressure of 3,000 psi (Soeder Fig. 4—Estimations of Klinkenberg beta factor for three perme-
1988; Wang et al. 2009). ability values.
other nonlinear flow processes, such as adsorption and non-Darcy In this numerical approach, we apply the upstream weighting
flow. In general, the flow model needs to be solved with a numeri- method to the mobility term and the harmonic mean method to
cal approach. This work follows the methodology for reservoir the transmissivity term to guarantee the convergence and accu-
simulation (i.e., the use of numerical approaches to simulate gas racy of the calculation. The flow-potential term in Eq. 17 is
and water flow), following three steps: (1) spatial discretization of defined as
mass-conservation equations; (2) time discretization; and (3) iter-
ative approaches to solve the resulting nonlinear, discrete alge- Ub j ¼ Pbi qb;ijþ1=2 gZi ; . . . . . . . . . . . . . . . . . . . . . ð20Þ
braic equations.
where Zi is the depth to the center of block i from a reference
Discrete Equations. The component mass-balance equations datum.
(Eq. 1) are discretized in space with a control-volume or inte-
grated finite-difference concept (Pruess et at. 1999). The control- Handling the Klinkenberg Effect. To include the Klinkenberg
volume approach provides a general spatial discretization scheme effect on gas flow, the absolute permeability to gas phase in Eq.
that can represent a 1D, 2D, or 3D domain with a set of discrete 19 should be evaluated with Eq. 11 as a function of gas-phase
meshes. Each mesh has a certain control volume for a proper pressure.
averaging or interpolation of flow and transport properties or ther-
modynamic variables. Time discretization is carried out with a Handling the Non-Darcy Flow. Under the non-Darcy-flow con-
backward, first-order, fully implicit finite-difference scheme. The dition of Eq. 14, the flow term (flowbij ) in Eq. 17 along the con-
discrete nonlinear equations for components of gas and water at
nection (i, j), between elements i and j, is numerically defined as
gridblock or node i can be written in a general form:
(Wu 2002)
h i 8
b;n Vi 2 !2 31=2 9
ð/qSÞb;nþ1
i þ mb;nþ1
i ð/qSÞb;n
i mi >
< >
=
X Dt Aij 1 1
flowb;ij ¼ þ4 c ij ðUb j Ub i Þ5 ;
¼ flowb;nþ1
ij þ Qb;nþ1
i 2ðkbb Þijþ1=2 >
: kb kb >
;
j 2 gi
Microseismic
cloud
Dual-porosity, or dual-permeability, or
Well
MINC
1.2E+7 8.E+05
Analytical c_β = 0
Analytical b = 7.6E6 Analytical c_β = 1.0E–3
Numerical b = 7.6E6 7.E+05 Analytical c_β = 1.0E–4
1.0E+7 Numerical b = 0 Numerical c_β = 0
Numerical c_β =1.0E–3
Analytical b = 0 6.E+05 Numerical c_β =1.0E–4
Pressure, Pa
8.0E+6
Pressure, Pa
5.E+05
6.0E+6 4.E+05
4.0E+6 3.E+05
2.E+05
2.0E+6
1.E+05
0.0E+0 0.E+00
0 2 4 6 8 10 0 2 4 6 8 10
Horizontal Distance, m Horizontal Distance, m
Fig. 7—Analytical and numerical results for linear flow with the Fig. 8—Analytical and numerical results for linear non-Darcy
Klinkenberg effect. flow.
and gas saturation. The rest of the dependent variables—such as steady and transient analytical solutions are derived or used for
relative permeability, capillary pressures, viscosity and densities, considering these flow mechanisms. The problem concerns steady-
adsorption term, and nonselected pressure and saturation—are state and transient gas flow across a 1D reservoir. The system con-
treated as secondary variables, which are calculated from selected tains steady-/transient-state gas flow at an isothermal condition, and
primary variables. a constant gas mass injection/production rate is imposed at one side
In terms of the primary variables, the residual equation, Eq. of the rock or well. The other boundary of the rock/reservoir is kept
23, at a node i is regarded as a function of the primary variables at at constant pressure. Eventually, the system will reach steady state,
not only node i, but also at all its direct neighboring nodes j. The if the production is maintained for a long period of time. A compar-
Newton iteration scheme gives rise to ison of the pressure profiles along the rock block from the simula-
tion and the analytical solution is shown in Figs. 7 and 8,
X @Rb;nþ1 ðxm;p Þ indicating that our simulated pressure distribution is in excellent
i
ðdxm;pþ1 Þ ¼ Rb;nþ1
i ðxm;p Þ; . . . . . . ð24Þ agreement with the analytical solutions for all the problems of 1D
m
@xm
linear flow with the Klinkenberg or non-Darcy-flow effect.
where xm is the primary variable m with m ¼ 1 and 2, respectively, Details about the analytical solution derivation considering the
at node i and all its direct neighbors; p is the iteration level; and Klinkenberg and non-Darcy-flow effect are included in our previ-
i ¼ 1, 2, 3, …, N. The primary variables in Eq. 23 need to be ous work (Wu et al. 2012), and we will show their verification
updated after each iteration, results only in this section for the 1D linear-flow steady-flow situa-
tion. Comparisons between the analytical and numerical solutions
xm;pþ1 ¼ xm;p þ dxm;pþ1 : . . . . . . . . . . . . . . . . . . . . . ð25Þ for the radial-flow and transient-flow cases are also presented in
our former work. Constant coefficients for the Klinkenberg effect
The Newton iteration process continues until the residuals and correlation (Eq. 13) for the non-Darcy-flow coefficient are used
Rb;nþ1
i or changes in the primary variables dxm;pþ1 over iteration with comparison results shown in Figs. 7 and 8.
are reduced below preset convergence tolerances.
Numerical methods are generally used to construct the Jaco- Verification for Flow With Adsorption. For the gas flow with
bian matrix for Eq. 24, as outlined in Forsyth et al. (1995). At adsorption, the approximate analytical solution is given in Appen-
each Newton iteration, Eq. 24 represents a system of (2 N) line- dix A. The parameters used for this comparison study are porosity
arized algebraic equations with sparse matrices, which are solved U ¼ 0.15; permeability k ¼ 100 md; formation temperature
by a linear equation solver. T ¼ 25 C; gas viscosity l ¼ 1:64 102 cp; initial pressure
Pi ¼ 105 Pa; and thickness of the radial system is 1 m. The
Numerical-Model Verification well-boundary condition is a constant gas/mass-injection rate:
To examine the accuracy of our simulator formulation in simulat- Q ¼ 1:0 104 kg=s.
ing porous-medium gas flow with the Klinkenberg, non-Darcy- Fig. 9 presents the comparisons of the pressure profile at 1.67
flow, gas-adsorption, and geomechanics effects, several relevant days from the numerical and analytical solutions. Two situations,
200000 200000
P_analytical
P_analytical
180000 180000 P_numerical
P_numerical
Pressure, Pa
Pressure, Pa
160000 160000
140000 140000
120000 120000
VL = 0 VL = 50
t = 1.67 days t = 1.67 days
100000 100000
0 0.5 1 1.5 2 2.5 0 1 2 3
Radius, m Radius, m
Fig. 9—Comparison of gas-pressure profiles considering gas adsorption in a radial system at 1.67 days, calculated with the numer-
ical and analytical solutions.
3.0E+7
TABLE 1—PARAMETERS FOR CHECKING INTEGRAL
SOLUTION FOR FLOW WITH GEOMECHANICS EFFECT 2.5E+7
Pressure, Pa
2.0E+7
7
Initial pressure Pi ¼ 10 Pa
Initial porosity Ui ¼ 0:20 1.5E+7
3 Constant K Pressure (N)
Initial fluid density qw ¼ 975:9 kg=m
1.0E+7 Constant K Pressure (A)
Cross-sectional area a ¼ 1.0 m2 unconstant K Pressure (A)
Formation thickness h ¼ 1.0 m 5.0E+6 unconstant K Pressure (N)
Fluid viscosity l ¼ 0:35132 103 Pa s
Fluid compressibility Cf ¼ 4:556 1010 Pa1 0.0E+0
Rock compressibility Cr ¼ 5:0 109 Pa1 0 5000 10000
Initial permeability k0 ¼ 9:860 1013 m2 Time, seconds
Water-injection rate qm ¼ 0:01 kg=s
Fig. 10—Comparison of injection pressures calculated from in-
Hydraulic radius rw ¼ 0:1 m
tegral and numerical solutions for linear flow in a permeability-
Exponential index c ¼ 2.22 dependent medium with constant and nonconstant permeabil-
ity function.
8000 105 md, and the initial reservoir pressure is 3,800 psi. With this
Cumulative Production, MMscf
Pg Pg
2.4E+07 2.4E+07
2.2E+07 2.2E+07
2E+07 2E+07
1.8E+07 1.8E+07
1.6E+07 1.6E+07
1.4E+07 1.4E+07
1.2E+07 1.2E+07
1E+07 1E+07
8E+06 8E+06
6E+06 6E+06
4E+06 4E+06
Fig. 14—Pressure distribution at 1 year (left) and 20 years (right) of Fracture Model #2 (unit: Pa).
6000 3000
4000 2000
3000
1500
case without Darcy flow
2000
Klinkengberg 1000
case with non-Darcy flow
1000 Klinkenberg
500
0
0 20 40 60 80 100 0
Time, years 0 2 4 6 8
Time, years
Fig. 15—Gas-cumulative-production behavior with the Klinken-
berg effect. Fig. 16—Gas-cumulative-production behavior with non-Darcy
flow in the first 6 years.
6000 1.0E+5
Cumulative Production, MMscf
5000 1.0E+4
1.0E+3
4000
Velocity, ft/D
1.0E+2
3000
1.0E+1
Darcy flow
2000
non-Darcy flow Darcy Velocity
1.0E+0
1000 non-Darcy Velocity
1.0E–1
0
1.0E–2
0 20 40 60 80 100
1.0E–6 1.0E–4 1.0E–2 1.0E+0
Time, years
Pressure Gradient, psi/ft
Fig. 17—Gas cumulative-production behavior with non-Darcy Fig. 18—Darcy and non-Darcy velocities with pressure
flow in 100 years. gradient.
keeps increasing from 1:0 103 psi/ft, the difference will input of the figure data. As shown in Fig. 20, geomechanics/flow
become larger. Fig. 19 shows the calculated average gas flow rate coupling has a large impact on formation permeability, especially
inside hydraulic fractures with production time in the case that for the natural-fracture system. Consider the Muska formation, for
does not consider the non-Darcy flow. For the first 6 years, flow example, when the effective stress increases from 1,600 to 4,800
velocities locate in the range in which the difference between the psia and permeability decreases to 1/20 of its original value. With
Darcy flow and the non-Darcy flow is obvious. After that, flow the gas production, reservoir effective stress increases as pore
velocities move to the area in which the difference is negligible. pressure decreases, leading to the large reduction of cumulative
Fig. 20 shows the simulated-well cumulative production vs. gas production.
time with and without geomechanics effect. The relationship used Figs. 21 and 22 present the results for adsorption analysis with
for describing the effective stress and permeability of the uncon- the numerical model. On the basis of the data in Table 2, we cal-
ventional reservoir is shown in Fig. 1, by use of a table-lookup culate the total gas mass as free gas in the micropores and
1.0E+5 6000
Cumulative Production, MMscf
1.0E+3
4000
1.0E+2
3000
1.0E+1
2000 case without
1.0E+0 geomechanics
1000 case with
1.0E–1 geomechanics
1.0E–2 0
0 10 20 30 0 10 20 30 40 50 60 70 80 90 100
Time, years Time, years
Fig. 19—Calculated gas-flow velocity with time in hydraulic Fig. 20—Gas-cumulative-production behaviors with geo-
fractures. mechanics.
6000 6000
0 0
0 20 40 60 80 100 0 20 40 60 80 100
Time, years Time, years
Fig. 21—Gas-cumulative-production behaviors with adsorption. Fig. 22—Gas-production analysis from free gas and adsorbed
gas.
adsorbed gas at initial condition. The proportion of gas stored in presented at the SPE Gas Production Conference, Fort Worth, Texas,
the pore space is approximately 77%, whereas that stored as 16–18 November. http://dx.doi.org/10.2118/119892-MS.
adsorption is 23%. Then, we compare the cumulative gas produc- Cipolla, C.L. 2009. Modeling Production and Evaluating Fracture Per-
tion with and without considering adsorption. Simulation results formance in Unconventional Gas Reservoirs. J. Pet Tech 61 (9): 84–90
(Fig. 21) show that the estimated gas production will increase (Distinguished Author Series). http://dx.doi.org/10.2118/118536-PA.
with considering adsorption. This difference will become more Cipolla, C.L., Lolon, E.P., Erdle, J.C. et al. 2010. Reservoir Modeling in
and more evident. For the situation considering gas adsorption/de- Shale-Gas Reservoirs. Res Eval & Eng. 13 (4): 638–653. http://
sorption, gas production from the desorption is approximately dx.doi.org/10.2118/125530-PA.
13%, and the produced portion of the free gas consists of 87%, as Cipolla, C.L., Warpinski, N.R., Mayerhofer, M.J. et al. 2009. The Rela-
shown in Fig. 22. tionship Between Fracture Complexity, Reservoir Properties, and
Fracture Treatment Design. Paper SPE 115769 presented at the SPE
ATC&E, Denver, Colorado, 21–24 September. http://dx.doi.org/
Summary and Conclusions 10.2118/115769-MS.
This paper discusses a generalized-framework mathematical Davies, J.P. and Davies, D.K. 1999. Stress-Dependent Permeability Char-
model for modeling gas production from unconventional gas res- acterization and Modeling. Paper SPE 56813 presented at the SPE
ervoirs. The model formulation incorporates known nonlinear ATCE, Houston, Texas, 3–6 October. http://dx.doi.org/10.2118/
flow processes, associated with gas production from low-perme- 56813-MS.
ability unconventional reservoirs, including the Klinkenberg, EIA (US Energy Information Administration). 2011. World Shale Gas
non-Darcy-flow, and nonlinear-adsorption effects. The model for- Resources: An Initial Assessment of 14 Regions Outside the United
mulation and numerical scheme are based on a generalized two- States (April).
phase (gas/liquid) -flow model with unstructured grids. Specifi- Ertekin, T., King, G.R., and Schwerer, F.C. 1986. Dynamic Gas Slippage:
cally, a hybrid modeling approach is presented by combining dis- A Unique Dual-Mechanism Approach to the Flow of Gas in Tight For-
crete fracture, multidomain, and multicontinuum concepts for mations. SPE Form Eval 1 (1): 43–52. http://dx.doi.org/10.2118/
handling hydraulic fractures and a fracture network in SRV, dis- 12045-PA.
tributed natural fractures, microfractures as well as porous matrix. Evans, R.D., and Civan, F. 1994. Characterization of Non-Darcy Multi-
We have verified the numerical models against analytical solu- phase Flow in Petroleum Bearing Formation. Final Report for US
tions for the Klinkenberg, non-Darcy-flow, and nonlinear-adsorp- Department of Energy Assistant Secretary for Fossil Energy.
tion effects. Forsyth, P.A., Wu, Y.S., and Pruess, K. 1995. Robust Numerical Methods
As application examples, we present modeling studies with for Saturated-Unsaturated Flow With Dry Initial Conditions in Hetero-
three fracture models for gas production from a 10-stage hydrau- geneous Media. Advance in Water Resources 18: 25–38.
lic-fractured horizontal well, incorporating the Klinkenberg, non-
Freeman, C.M., Moridis, G.J., Ilk, D. et al. 2010. A Numerical Study of
Darcy-flow, and nonlinear-adsorption effects. The model results
Transport and Storage Effects for Tight Gas and Shale Gas Reservoir
show that there is a large impact of various fracture models on
Systems. Paper SPE 131583 presented at the SPE International Oil and
gas-production rates as well as cumulative production.
Gas Conference and Exhibition, Beijing, China, 8–10 June. http://
dx.doi.org/10.2118/131583-MS.
Acknowledgments Freeman, C.M., Moridis, G.J., and Blasingame, T.A. 2009a. A Numerical
Study of Microscale Flow Behavior in Tight Gas and Shale Gas Reser-
This work was supported in part by EMG Research Center and voir Systems. Proceedings of the 2010 TOUGH Symposium, Berkeley,
UNGI of the Petroleum Engineering Department at Colorado California, 14–16 September.
School of Mines; by Foundation CMG; by RIPED of PetroChina
Freeman, C.M., Moridis, G.J., Ilk, D. et al. 2009b. A Numerical Study of
Company; and by IFPEN.
Tight Gas and Shale Gas Reservoir Systems. Paper SPE 124961 pre-
sented at the SPE Annual Technical Conference and Exhibition, New
Orleans, Louisiana, 4–9 October. http://dx.doi.org/10.2118/124961-
References MS.
Blasingame, T.A. 2008. The Characteristic Flow Behavior of Low-Perme- Gao, C., Lee, J.W., Spivey, J.P. et al. 1994. Modeling Multilayer Gas Res-
ability Reservoir Systems. Paper SPE 114168 presented at the SPE ervoirs Including Sorption Effects. Paper SPE 29173 presented at the
Unconventional Reservoirs Conference, Keystone, Colorado, 10–12 SPE Eastern Regional Conference and Exhibition, Charleston, West
February. http://dx.doi.org/10.2118-MS. Virginia, 8–10 November. http://dx.doi.org/10.2118/29173-MS.
Bustin, R.M., Bustin, A.M.M., Cui, X. et al. 2008. Impact of Shale Proper- Kelkar M. and Atiq, M. 2010. Upgrading Method for Tight Gas Reser-
ties on Pore Structure and Storage Characteristics. Paper SPE 119892 voirs. Paper SPE 133301 presented at the SPE Annual Technical
Conference and Exhibition, Florence, Italy, 19–22. http://dx.doi.org/ Wu, Y.S. 2002. Numerical Simulation of Single-Phase and Multiphase
10.2118/133301-MS. Non-Darcy Flow in Porous and Fractured Reservoirs. Transport in Po-
Klinkenberg, L.J. 1941. The Permeability of Porous Media to Liquids and rous Media 49: 209–240.
Gases. In API Drilling and Production Practice, 200–213. Wu, Y.S. and Pruess, K. 2000. Integral Solutions for Transient Fluid Flow
Langmuir, I. 1916. The Constitution and Fundamental Properties of Solids Through a Porous Medium With Pressure-Dependent Permeability.
and Liquids. J. Am. Chem. Soc. 38 (11): 2221–2295. International J. Rock Mechanics and Mining Sci. 37: 51–61.
Wu, Y.S., Pruess, K., and Persoff, P. 1998. Gas Flow in Porous Media
Leahy-Dios, A., Das, M., Agarwal, A. et al. 2011. Modeling of Transport
With Klinkenberg Effects. Transport in Porous Media 32: 117–137.
Phenomena and Multicomponent Sorption for Shale Gas and Coalbed
Wu, Y.S., Rutqvist, J., Karasaki, K. et al. 2008. A Mathematic Model for
Methane in an Unstructured Grid Simulator. Paper SPE 147352 pre-
Rock Deformation Effect of Flow in Porous and Fractured Reservoirs.
sented at the SPE Annual Technical Conference, Denver, Colorado, 30
Paper ARMA-08-142 presented at the 42nd US Rock Mechanics Sym-
Octomber–2 November. http://dx.doi.org/10.2118/147352-MS.
posium and 2nd US–Canada Rock Mechanics Symposium, San Fran-
Leverett, M.C. 1941. Capillary Behavior in Porous Media. Trans AIME cisco, California, 29 June–2 July.
142: 341–358. Wu, Y.S. and Wang, C. 2012. Transient Pressure Testing Analysis of Gas
Li, J., Du, C.M., and Zhang, X. 2011. Critical Evaluation of Shale Gas Wells in Unconventional Reservoirs. Paper SPE-SAS-312 presented at
Reservoir Simulation Approaches: Single-Porosity and Dual-Porosity the SPE Annual Technical Symposium and Exhibition (ATS&E), Kho-
Modeling. Paper presented at the SPE Middle East Unconventional bar, Saudi Arabia, 8–11 April.
Gas Conference and Exhibition, Muscal, Oman, 31 January–2 Wu, Y.S., Wang, C., Li, J. et al. 2012. Transient Gas Flow in Unconventional
February. Gas Reservoirs. Paper SPE 154448 presented at the EAGE Annual Con-
Mengal, S.A. and Wattenbarger, R.A. 2011. Accounting for Adsorbed Gas ference and & Exhibition Incorporating SPE Europec, Copenhagen, Den-
in Shale Gas Reservoirs. Paper SPE 141085 presented at the SPE Mid- mark, 4–7 June. http://dx.doi.org/10.2118/154448-MS.
dle East Oil and Gas Show and Conference, Manama, Bahrain, 25–28
September. http://dx.doi.org/10.2118/141085-MS. Appendix A
Moridis, G.J., Blasingame, T.A., and Freeman, C.M. 2010. Analysis of Here, we derive the analytical solution for gas flow with adsorp-
Mechanisms of Flow in Fractured Tight-Gas and Shale-Gas Reser- tion/desorption. If the system is isothermal, the ideal-gas law
voirs. Paper SPE 139250 presented at the SPE Latin American and applies, and the gravity effect is negligible, then gas flow in porous
Caribbean Petroleum Engineering Conference, Lima, Peru, 1–3 media with adsorption is described by the following equations:
December. http://dx.doi.org/10.2118/139250-MS.
Ozkan, E., Raghavan, R., and Apaydin, O.G. 2010. Modeling of Fluid
@ð/q þ mg Þ
r ðqvÞ ¼ ; . . . . . . . . . . . . . . . . . . ðA-1Þ
Transfer From Shale Matrix to Fracture Network. Paper SPE 134830 @t
presented at the SPE ATC&E, Florence, Italy, 19–22 September. where q is the gas density; v is the gas-flow velocity; / is the po-
http://dx.doi.org/10.2118/134830-MS. rous-media porosity; mg is the adsorbed gas mass in a unit forma-
Pruess, K. 1983. GMINC—A Mesh Generator for Flow Simulations in tion volume at a given pressure; and t is the time.
Fractured Reservoirs, Report LBL-15227, Berkeley, California: Law- According to the ideal-gas law,
rence Berkeley National Laboratory.
Pruess, K. and Narasimhan, T.N. 1985. A Practical Method for Modeling PV ¼ nRT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-2Þ
Fluid and Heat Flow in Fractured Porous Media. Soc. Pet. Eng. J. 25:
14–26. and
Pruess, K., Oldenburg, C., and Moridis, G. 1999. TOUGH2 User’s Guide, M
q¼ P ¼ bP; . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-3Þ
Version 2.0, Report LBNL-43134, Berkeley, California: Lawrence RT
Berkeley National Laboratory.
where M is gas molecular weight; R is the universal gas constant;
Rubin, B. 2010. Accurate Simulation of Non-Darcy Flow in Stimulated M
Fractured Shale Reservoirs. Paper SPE 132093 presented at the SPE b is a coefficient, for simplicity, defined as b ¼ ; and T is the
Western Regional Meeting, Anaheim, California, 27–29 May. http://
RT
system temperature.
dx.doi.org/10.2118/132093-MS. From Darcy’s law and the Langmuir isotherm (Eqs. 2 and 3),
Rutqvist, J.Y.,Wu, S., Tsang, C.F. et al. 2002. A Modeling Approach for
Analysis of Coupled Multiphase Fluid Flow, Heat Transfer, and Defor- k
v ¼ rP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ðA-4Þ
mation in Fractured Porous Rock. International J. Rock Mechanics l
and Mining Sci. 39: 429–442.
and
Silin, D. and Kneafsey, T. 2011. Gas Shale: From Nanometer-Scale Obser-
vations to Well Modeling. Paper CSUG/SPE 149489 presented at the P P
mg ¼ qR qg VE ¼ qR qg VL ¼ Va ; . . . . . ðA-5Þ
Canadian Unconventional Resources Conference, Calgary, Alberta, P þ PL P þ PL
Canada, 15–17 November. http://dx.doi.org/10.2118/149489-MS.
where qR is rock bulk density; qg is gas density at standard condi-
Soeder, D.J. 1988. Porosity and Permeability of Eastern Devonian Gas
tion; VE is the adsorption isotherm function for gas content; VL is
Shale. SPE Form Eval 3 (1): 116–124. http://dx.doi.org/10.2118/
the Langmuir’s volume in scf/ton; and PL is Langmuir’s pressure.
15213-PA.
a is a coefficient, for simplicity, defined as a ¼ qR qg VL .
Wang, F.P., Reed, R.M., Jackson, J.A. et al. 2009. Pore Networks and By substituting Eqs. A-4 and A-5 into Eq. A-1, we obtain
Fluid Flow in Gas Shales. Paper SPE 124253 presented at the SPE An-
nual Technical Conference and Exhibition, New Orleans, Louisiana, P
@
4–7 October. http://dx.doi.org/10.2118/124253-MS. k @P P þ PL
r b PrP ¼ /b þa : . . . . . . . .ðA-6Þ
Winterfeld, P.H. and Wu, Y.S. 2011. Parallel Simulation of CO2 Seques- l @t @t
tration With Rock Deformation in Saline Aquifers. Paper SPE 141514
prepared for presentation at the SPE Reservoir Simulation Symposium, In radial coordinates,
The Woodlands, Texas, 21–23 February. http://dx.doi.org/10.2118/ P
2
@
141514-MS. 1@ @P 2/l @P 2al P þ PL @P
r ¼ þ ; . . . . ðA-7Þ
Winterfeld, P.H. and Wu, Y.S. 2012. A Novel Fully Coupled Geomechan- r @r @r k @t bk @P @t
ical Model for CO2 Sequestration in Fractured and Porous Brine Aqui-
fers. Paper presented at the XIX International Conference on Water " #
1@ @P2 2/l 2al PL @P
Resources CMWR 2012, University of Illinois at Urbana-Champaign, r ¼ þ ; . . . . . . ðA-8Þ
r @r @r k bk ðP þ PL Þ2 @t
Illinois, 17–22 June.
Summary many shale reservoirs was made possible through improved stimu-
The exploitation of unconventional gas reservoirs has become an lation techniques and horizontal drilling.
ever increasing component of the North American gas supply. The economic viability of many unconventional gas develop-
The economic viability of many unconventional gas developments ments hinges on effective stimulation of extremely low-permeability
hinges on effective stimulation of extremely low-permeability rock rock, typically 10 to 100 nanodarcies (10−6 md). In most cases,
by creating very complex fracture networks that connect huge economic production is possible only if a very complex, highly
reservoir surface area to the wellbore. In addition, gas desorption nonlinear fracture network can be created that effectively connects
may be a significant component of overall gas recovery in many a huge reservoir surface area to the wellbore. Many conventional
shale-gas reservoirs. The widespread application of microseismic hydraulic-fracture treatments use high-viscosity fluids to reduce
(MS) mapping has significantly improved our understanding of fracture complexity and promote planar fractures, allowing the
hydraulic fracture growth in unconventional gas reservoirs (pri- placement of high concentrations of large proppant. However,
marily shale) and has led to better stimulation designs. However, stimulation treatments in shale reservoirs use large volumes of
the overall effectiveness of stimulation treatments is difficult to low-viscosity fluid (water) to promote fracture complexity and
determine from MS mapping because the location of proppant place very low concentrations of small proppant—an approach
and the distribution of conductivity in the fracture network cannot that is completely different from conventional hydraulic fracturing.
be measured (and are critical parameters that control well perfor- Hydraulic-fracture treatment stages in horizontal wells can require
mance). Therefore, it is important to develop reservoir-modeling over a million gallons of water and up to 1 million pounds of prop-
approaches that properly characterize fluid flow in and the prop- pant and are pumped at rates of 75–150 bbl/min. Because of the
erties of a complex fracture network, tight matrix, and primary complexity of fracture growth in many shale-gas reservoirs, it is not
hydraulic fracture (if present) to evaluate well performance and possible to accurately predict the distribution of proppant within
understand critical parameters that affect gas recovery. the network and the conductivity of the network fractures is dif-
This paper illustrates the impact of gas desorption on produc- ficult to predict. Previous work has characterized proppant distribu-
tion profile and ultimate gas recovery in shale reservoirs, showing tion within complex fracture networks using two limiting cases:
that in some shale-gas reservoirs desorption may be a minor com- either the proppant is evenly distributed throughout the complex
ponent of gas recovery. In addition, the paper details the impact of network or the proppant is confined to a primary or main fracture
changing closure stress distribution in the fracture network on well (Cipolla et al. 2008a). This previous work showed that for prop-
productivity and gas recovery. In shale-gas reservoirs with lower pant volumes that are typical for many shale-gas hydraulic-fracture
Young’s modulus rock, stress-dependent network-fracture conduc- treatments, if the proppant is evenly distributed throughout the
tivity may reduce ultimate gas recovery significantly. The paper complex fracture network, proppant concentrations (lbm/ft2) are
includes an example that contrasts the application of numerical likely to be too low to materially affect network-fracture conductiv-
reservoir simulation and advanced decline-curve analyses to illus- ity and the network fractures will essentially behave as unpropped
trate issues associated with conventional production-data-analysis fractures. If the proppant is confined to a primary or main hydraulic
techniques when applied to unconventional reservoirs. fracture and is not transported into the network fractures, then the
Selected examples from the Barnett shale are included that relative conductivity of the primary fracture should be high, but the
incorporate MS fracture mapping and production data to illustrate fracture network will be unpropped. Therefore, unpropped-fracture
the application of production modeling to evaluate well perfor- conductivity may play an important role in shale-gas stimulation
mance in unconventional gas reservoirs. This paper highlights performance. Network-fracture conductivity in many shale-gas
production modeling and analysis techniques that aid in evaluat- reservoirs is estimated by modeling well performance. Combining
ing stimulation and completion strategies in unconventional gas detailed reservoir modeling with MS-mapping data that define the
reservoirs. size of the fracture network and laboratory cores tests that provide
estimates of matrix permeability and propped- and unpropped-
Introduction fracture conductivity can result in a much better understanding of
stimulation performance.
Gas shales are organic-rich shale formations and are apparently
Gas flow from ultralow-permeability rock through the complex
both the source rock and the reservoir. The gas is stored in the
fracture network must be modeled to evaluate stimulation designs
limited pore space of these rocks and a sizable fraction of the gas
and completion strategies properly. Therefore, the complex frac-
in place may be adsorbed on the organic material. The natural-
ture network and primary hydraulic fracture (if present) must be
gas resource potential for gas shales in the USA is estimated to
discretely characterized in our reservoir-simulation models. MS
be from 500 to 1,000 Tcf (Arthur et al. 2008). Typical shale-gas
fracture mapping provides a measurement of the overall reservoir
reservoirs exhibit a net thickness of 50 to 600 ft, porosity of 2 to
volume that is stimulated [referred to as stimulated reservoir vol-
8%, and total organic carbon of 1 to 14% and are found at depths
ume (SRV)], and special core analysis can provide measurements
ranging from 1,000 to 13,000 ft. The success of the Barnett shale
of matrix permeability. With these two parameters, reservoir-simu-
has illustrated that gas can be economically produced from rock
lation models can be used to estimate the spacing of the network
that was previously thought to be source rock and/or caprock, not
fractures (complexity) and their conductivity. Simulated produc-
reservoir rock. This revelation has led to the development of many
tion profiles can then be compared to actual well performance to
other shale-gas reservoirs, including the Woodford, Fayetteville,
evaluate the effectiveness of stimulation designs and completion
Marcellus, and the Haynesville (Fig. 1). In addition to increasing
strategies and diagnose the relative conductivity of the primary
natural-gas prices (until recently), the economic development of
fracture (if present). This paper builds upon and extends previous
shale-gas reservoir-modeling work (Mayerhofer et al. 2006, 2010;
Copyright © 2010 Society of Petroleum Engineers
Warpinski et al. 2008; Cipolla et al. 2008a, 2009b), focusing on
This paper (SPE 125530) was accepted for presentation at the SPE Eastern Regional the impact of gas desorption and stress-sensitive network-fracture
Meeting, Charleston, West Virginia, USA, 23–25 September 2009 and revised for publication.
Original manuscript received for review 21 July 2009. Revised manuscript received
conductivity on well performance. In addition, this work discusses
for review 29 December 2009. Paper peer approved 1 March 2010. various shale-gas reservoir-simulation approaches, detailing an
approach that appears best suited for most shale-gas reservoirs. stimulation fluid from the treatment well, confirming that large
The highlights of the previous work are summarized to provide volumes of water are transversing the complex fracture network.
background to support the work presented in this paper. The volume of reservoir rock that is contacted by the stimulation
treatment (SRV), as evidenced by the MS activity, has a direct
Background impact on well performance—with larger SRVs resulting in better
The initial MS-mapping work in the Barnett shale revealed that wells (Mayerhofer et al. 2010). This early MS-mapping work led
fracture growth can be surprisingly complex in some environments. to a much better understanding of fracture growth in the Barnett
Fig. 2 shows the MS-mapping data for a typical vertical well in shale and was instrumental in many improvements in stimulation
the Barnett shale (Fisher et al. 2002). Stimulation treatments in and completion strategies, including horizontal drilling.
vertical wells consist of approximately 750,000 gal of slickwater Fig. 3 shows the MS-mapping data for a horizontal Barnett
and 150,000 lbm of proppant pumped at concentrations of 0.1 to shale well completed with an uncemented liner (five sets of per-
0.25 ppa and rates of 50–70 bbl/min. Early work showed that very forations) and stimulated using a single-stage hydraulic-fracturing
large fracture networks were created in the Barnett shale, typically treatment consisting of 4.2 million gal of slickwater and 853,000
in excess of 2,500 ft long and 1,000 ft wide, covering the Barnett lbm of proppant pumped at 130 bbl/min [Fisher et al. 2004]. In this
reservoir thickness (≈300 ft). In Fig. 2, the fracture network is case, the single-stage hydraulic-fracture treatment covered most of
more than 3,000 ft long and approximately 1,000 ft wide, extend- the 2,400 ft long lateral, but there are areas with little MS activity
ing more than 70 acres and contacting 900 million ft3 of reservoir that may not be stimulated effectively. The stimulation treatment
rock. Fig. 2 also shows that five offset wells were “killed” by appears to have effectively covered an area of approximately
1500 2000.00
1000 1200.00
800.00
500 Fairway “Length”
“Length”
400.00
Observation Well Fairway“Width”
Fairway “Width”
0.00
South-North (ft)
0
South-North (ft)
–400.00
–500 –800.00
Treatment Well
–1200.00
–1000
–1600.00
–2000.00
–1500
–2400.00
–2000 –2800.00
–4000.00
–3000
–2800
–2400
–2000
–1600
–1200
–800
–400
400
800
1200
1600
2000
2400
2800
Fig. 2—MS-mapping data from a vertical Barnett shale well Fig. 3—MS-mapping data for a horizontal Barnett shale well,
(Fisher et al. 2002). uncemented liner, single-stage treatment (Fisher et al. 2004).
1 year
5 years
15 years
Fig. 7—Pressure distribution after 1, 5, and 15 years for the Barnett shale example (with desorbed gas).
conductivity of 2 md-ft is used in the simulations in Figs. 8 and 9. result in similar production profiles, but the modeling may indicate
Fig. 9 shows that reducing the FBHP from 1,000 to 500 psi results that Well A has a more complex fracture network compared with
in increases in cumulative gas production of up to 10% after 30 Well B. Non-unique solutions to production history matching
years for the case with 10−4 md matrix permeability (Fig. 8) and are common in shale-gas reservoirs because matrix permeability,
12.5% after 30 years of production for the case with 10−5 md matrix network-fracture spacing, and the conductivity of the network
permeability (Fig. 9). Comparing the results with desorbed-gas and primary fractures are not known exactly. However, laboratory
production (solid lines) to those without desorbed-gas production core tests can be performed to estimate matrix permeability and
(dashed lines) in Figs. 8 and 9 shows that reducing the FBHP from unpropped-fracture conductivity, constraining the solution and
1,000 to 500 psi does not significantly increase the production of allowing significant insights into well performance.
desorbed gas. In addition, reviewing the production profiles for
the first 5 years shows that reducing the FBHP from 1,000 to 500 Barnett Shale: Impact of Stress-Dependent Network-Fracture
psi has very little impact on the early-time production profile, Permeability. The impact of increasing closure stress on propped-
especially in the first 3 years. However, the impact of reducing fracture conductivity has been studied for decades, but little atten-
the FBHP on gas production is greater for more-complex fracture tion has been given to the conductivity of unpropped hydraulically
networks (i.e., smaller blocks). created fractures until recently. There is a likelihood that large
Comparing the production profile from Well A to the simulation portions of the fracture networks created during shale-gas stimula-
results in Fig. 9 shows that Well A follows the production profile tion treatments may be unpropped or ineffectively propped (Fredd
for a matrix permeability of 10−5 md and a network-fracture spacing et al. 2001; Cipolla et al. 2008a). Therefore, unpropped- or partially-
of 100 ft. This is not a unique solution, because other combina- propped-fracture conductivity may play a significant role in shale-
tions of matrix permeability and network-fracture spacing may gas production. The impact of primary- and network-fracture
Fig. 8—Impact of BHFP for the desorbed-gas case (k = 10−4 Fig. 9—Impact of BHFP for the desorbed-gas case (k = 10−5
md, Dx = 600 ft). md, Dx = 100 ft).
for matrix permeability of 10−4 and 10−5 md, respectively. The stress-dependent network-fracture conductivity does not appear
simulations assume a primary- or main- fracture spacing of 600 ft to be significant, mostly affecting the mid- to late-life production
and a network-fracture spacing of 100 ft (k=10−4 md) and 50 ft behavior, and could reduce gas recovery by approximately 10%.
(k=10−5 md). The cumulative production for a constant and uni- However, if the relationship between closure stress and unpropped-
form 2-md-ft network-fracture conductivity and is shown for fracture conductivity is more severe than that shown in Fig. 10
comparison. The results show that more-severe stress-dependent (E = 5×106 psi curve), then ultimate gas recovery could be affected
fracture conductivity can reduce gas recovery significantly. When materially.
Young’s modulus is 4×106 psi, stress-dependent fracture conduc-
tivity reduces gas recovery by 11 to 14% in comparison with the Marcellus Shale: Impact of Stress-Dependent Network-Fracture
constant-conductivity case. For lower-modulus rock (2×106 psi), Permeability. The impact of gas desorption and stress-dependent
gas recovery decreases by 40 to 43% compared to the E=5×106 psi fracture conductivity was also evaluated for typical Marcellus shale
case. At lower modulus, the fracture conductivity is reduced sig- reservoir properties, although the Marcellus is more variable in
nificantly as drawdown in the fracture network increases with time, depth and thickness and much earlier in its development than the
resulting in poor drainage of the tight matrix rock and significantly core area of Barnett—making it more difficult to define average
lower ultimate gas recovery. However, initial well performance is or generic properties. In addition, actual production data were not
not markedly different for the various cases, and the severity of available to “calibrate” the Marcellus simulations. Table 2 lists the
stress-dependent fracture conductivity may not be evident during generic reservoir properties used for the Marcellus shale. The area
the first 1–2 years of production, requiring 3–5 years of produc- of the reservoir model is the same as that used in the Barnett shale,
tion history to better define the possible effects of stress-dependent 3,400×2,400 ft (≈190 acres), and a cased-and-cemented horizontal
network-fracture conductivity. completion with multiple fracture-treatment stages is assumed. The
The production profile for Well A follows the trend for a matrix SRV for the Marcellus simulations is 680×106 ft3 (100 ft of pay).
permeability of 10−5 md and a fracture-network spacing of 50 ft Fig. 15 shows a graph of gas content (scf/ton) vs. pressure
(lowest curve in Fig. 14) very closely, while also following the for the gas in the matrix porosity and the adsorbed gas used for
trend for a matrix permeability of 10−4 and network-fracture spac- the Marcellus shale (Faraj and Duggan 2008). The free gas in the
ing of 100 ft (lowest curve in Fig. 13). This illustrates some of the matrix porosity is shown as a nearly straight line (circles), while
issues with potentially non-unique solutions when history matching the adsorbed gas is represented by a curved line (squares). In this
shale-gas production profiles—because fracture conductivity, matrix case, 46.5% of the total gas is adsorbed, a percentage similar to
permeability, and network-fracture spacing are many times not well that in the Barnett shale. At higher pressures, more gas is stored
known. However, reservoir modeling can provide important insights in the matrix porosity and the gas porosity will contribute more
into production mechanisms and stimulation performance when to production than gas desorption because of the difference in the
values of matrix permeability and unpropped-fracture conductivity curve shapes (straight line vs. curved line).
can be bounded using laboratory data. It may be possible to estimate The Marcellus simulations include the effect of stress depend-
network-fracture spacing and network structure using MS-mapping ent network-fracture conductivity. Assuming uniform and constant
data, further constraining the history match.
Summary of Barnett Shale-Gas Desorption and Stress-Sensi-
tive-Permeability Simulations. The simulation study using typical TABLE 2—GENERIC MARCELLUS SHALE PROPERTIES
Barnett shale reservoir and rock properties indicated that gas
Depth (ft) 5,000
desorption and stress-dependent fracture conductivity probably
have a minor, but not necessarily insignificant, impact on well Reservoir pressure (psi) 2,750
performance. The impact of gas desorption is primarily in the Pore pressure gradient (psi/ft) 0.55
later life of the well when pressures in the tight matrix blocks may Closure pressure (psi) 3,500
become low enough to produce meaningful amounts of gas, result- Closure pressure gradient (psi/ft) 0.70
ing in an increase in 30-year gas recovery of 5–15%. However, as
Net pay (ft) 100
network-block size increases and fracture conductivity and matrix
permeability decrease, the ability to produce adsorbed gas becomes Porosity (% ) 8
increasingly difficult. While decreasing FBHP from 1,000 to 500 Matrix permeability (md) 1 E 4; 5 E 4
psi does not significantly increase the production of adsorbed Water saturation (%) 25
gas, it will improve 30-year gas recovery by approximately 10%
Reservoir temperature (°F) 105
(mostly because of the additional production of free gas). Because
of the relatively high modulus of the Barnett shale, the impact of FBHP (psi) 500
Conductivity (mD-ft)
1
2 mD-ft
0.02 mD-ft
0.1
0.01
Initial closure stress Closure stress at 500 psi FBHP
0.001
0 500 1000 1500 2000 2500 3000 3500 4000
Stress (psi)
Fig. 17—Impact of desorbed-gas production in the Marcellus Fig. 18—Impact of desorbed-gas production in the Marcellus
shale (k = 10−4 md). shale (k = 5×10−4 md).
1 year
5 years
15 years
Fig. 19—Pressure distribution after 1, 5, and 15 years for the generic Marcellus shale example (with desorbed gas).
Effect of Stress-Dependent Fracture Conductivity. The impact 10% additional gas production in 30 years, while having little effect
of stress-dependent conductivity is shown in Fig. 20 for a matrix on the initial production profile. Therefore, desorbed-gas produc-
permeability of 5×10−4 md, primary-fracture spacing of 100 ft, tion is probably a minor component in the economic development
and a network-fracture spacing of 100 ft. All simulations include of the Marcellus shale. In contrast to the Barnett shale, the lower
gas desorption. The cumulative production for a constant 2-md-ft Young’s modulus of the Marcellus could mean that stress-depen-
uniform-conductivity network is shown by the top curve, indicating dent network-fracture conductivity will play a significant role in
a 30-year gas recovery of 8,342 MMscf. The bottom curve (dashed well performance and could substantially reduce initial production
line) shows the production for a uniform-conductivity network rates and ultimate gas recovery. However, the detrimental effects
with stress-dependent fracture conductivity (initial conductivity of stress-dependent network-fracture conductivity may be offset
of 2 md-ft, declining to 0.02 md-ft when the pressure in a net- by stimulation designs that result in high-relative-conductivity
work fracture reaches 500 psi, Fig. 16). The 30-year gas recovery primary fractures.
is reduced significantly when stress-dependent network-fracture
conductivity is modeled, decreasing by almost 60%. Vertical-Well Case Study
The detrimental effects of stress-dependent network-fracture The example well is a vertical well located in the core area of the
permeability can be offset, to some degree, by designing stimulation Barnett shale; it was fracture treated and mapped in 2001 (May-
treatments that create a high-relative-conductivity primary fracture. erhofer et al. 2006). MS mapping clearly illustrates that very large
The presence of a high-relative-conductivity primary fracture fracture networks are generated, with the predominant hydraulic-
significantly reduces network-fracture conductivity requirements fracture azimuth in the northeasterly direction and a secondary
(Cipolla et al. 2008a, 2009a). The impact of a primary fracture
with an initial conductivity of 50 md-ft is also shown in Fig. 20,
middle curve with higher initial production and 7200×106 ft3 cumu-
lative gas production. It should be emphasized that the conductiv-
ity of the primary fracture is also stress dependent, declining to
0.25 md-ft when the pressure in the primary fracture reaches
500 psi. The cumulative gas production after 30 years is roughly
doubled for the case with a high relative conductivity primary
fracture in comparison with the 2-md-ft uniform-fracture-conduc-
tivity case (with stress-dependent network-fracture conductivity)—
essentially offsetting the detrimental affects of stress-dependent
network-fracture conductivity. It is interesting to note that the case
with a high-relative-conductivity primary fracture (Fig. 20, middle
curve) exhibits much higher initial gas-production rates compared
to the 2-md-ft constant- and uniform-conductivity case (Fig. 20,
top curve), yet 30-year cumulative gas production is approximately
15% lower. These simulations emphasize the importance of under-
standing the stress dependence of network-fracture conductivity in
fracture design and production forecasting.
Summary of Marcellus Shale Gas-Desorption and Stress-
Sensitive-Permeability Simulations. The impact of desorbed-gas Fig. 20—Impact of stress-dependent fracture conductivity, with
production in the generic Marcellus reservoir-simulation cases was and without a high-relative-conductivity primary fracture, Mar-
similar to that for the Barnett shale, possibly adding approximately cellus shale example.
Parameters V a lu e
Fracture-Network Model
Temperature (°F) 180
Gas gravity 0.6
MS mapping
gas rate was used as a constraint (Fig. 22). The total well history
includes 4 years of production. Fig. 23 shows the history match of
flowing tubing and casing pressures vs. time. The model predicts
higher wellhead flowing pressures in late time, which tracks the
surface annulus pressure (open-ended tubing) and represents liquid
loading. The loading condition was confirmed by higher flow rates
after a foam cleanout. Although this match is not entirely unique,
the effective total-fracture-network length (sum of all fracture seg-
West/East (ft) ments open to flow in the two principal frac directions) is on the
order of 6,000 to 7,800 ft, with an average conductivity of 4 md-ft
Fig. 21—Fracture network as determined from MS mapping (matrix permeability was fixed at 0.0001 md or 100 nanodarcies,
is approximated with numerical network model (Mayerhofer which was estimated from core tests). The effective network width
et al. 2006). was modeled to be 180 ft (90-ft fracture spacing in the northeast
direction and four orthogonal stringers) with a network extent of
approximately 2,000 ft in the southwest direction. While the total
component orthogonal to that. One of the remaining questions effective network length (sum of all open segments) must fall
concerns how much of the generated network is actually effective within the 6,000- to 7,800-ft range, the aspect ratio of effective
during production. Fig. 21 shows the MS-mapping results. The network extent to width could vary to some degree. An alternative
total extent of the fracture network is at least 2,700 ft, with apparent match, which would also be within the MS-mapped network area,
asymmetric growth to the southwest. It appears that the northeast was achieved with a wider 360-ft network (180 ft fracture spacing)
wing of the network structure did not develop as effectively, on the and shorter 1,300-ft extent in the southwest direction. Decreasing
basis of created MS events (Note: The observation well was in a the matrix permeability by an order of magnitude (to 10 nanodar-
position that should have detected growth on the northeast side). cies) would result in a similar pressure history match; however, the
The network width is fairly small in this case (approximately 350 effective fracture-network length would increase to approximately
ft), and network height shows adequate coverage of the Lower 16,500 ft and result in closer fracture spacing of 60 ft with 360-ft
Barnett interval. The network structure (bold lines) in Fig. 21 network width and 2,700-ft total extent. The exact position, aspect
represents the effective network modeled in the reservoir simulator ratio of effective network width to length, and fracture spacing
as an approximation of the MS results. This is the final network within the network can be uniquely determined only with the help
structure used to history match well production and is a result of of multiwell production data.
several iterations that explored alternative answers. In short, the Fig. 24 shows the pressure distribution after 1 year of produc-
conclusion is that once matrix permeability is fixed, the resulting tion. Red represents initial reservoir pressure, and dark blue is
total effective network length (sum of all open segments) and approximately the FBHP. Fig. 24 illustrates that nanodarcy matrix
conductivity are fairly unique. permeability will not drain very far beyond the created fracture
Table 3 shows the reservoir input parameters used in the simu- network. Additionally, low fracture conductivities play an impor-
lation. Three parameters were varied to obtain the history match: tant role because they create higher pressure drops in the network
fracture length, fracture density, and fracture conductivity. Actual resulting in less-efficient drainage farther from the well.
Wellhead Pressure (psi)
Time (days)
Fig. 22—History match of flowing tubing and casing pressures, Fig. 23—Gas-flow rate was used as a constraint for production
open-ended tubing (Mayerhofer et al. 2006). modeling (Mayerhofer et al. 2006).
Conclusions
The reservoir-simulation work focused on evaluating the effects of
gas desorption and stress-dependent network permeability in hori-
zontal completions using reservoir properties typical of the Barnett
and Marcellus shales, but the results and modeling approach should
be applicable to many other shale-gas reservoirs. The following
conclusions can be drawn from this work.
1. Gas desorption may not be a significant component of produc-
tion in many moderate-to-deep shale-gas reservoirs, such as the
Barnett, Marcellus, and Haynesville shales. Although adsorbed
gas may constitute 40–50% of the total gas in place, the ability
to produce the adsorbed gas is limited because of the ultratight
matrix rock, relatively high FBHP, and the desorption profile,
which requires relatively low pressures to produce significant
amounts of adsorbed gas.
2. Desorbed gas may account for 5–15% of ultimate gas produc-
tion, but the desorbed gas is produced primarily during the later
life of the well, having little impact on the initial well produc-
tivity. Therefore, the impact of gas desorption on development Fig. 26—Blasingame analysis, Barnett shale (xf = 130 ft).
Hydraulic Fracture
Fig. A-1—Grid design for single-plane vertical-propped-fracture simulation.
fracture treatments. Several options were evaluated to determine The MINC approach for modeling the fracture network was
which “grid design” and “porosity model” approach resulted also evaluated but proved inadequate at early time, as can be seen
in the best agreement with very finely gridded, single-porosity in Fig. A-5, which compares logarithmically and equally spaced
reference models, including the use of the multiple interacting gridding for both MINC (with 4 levels of refinement) and the new
continuum (MINC) and the dual-permeability “dual-continuum” simplified 9×9 LGR grid with 2-ft-wide grids representing the
models, while minimizing model run time. fractures. The MINC approach provided worse results when flow
Our work has determined that the best method for both accu- was allowed from the shale surrounding the stimulated volumes
rately and efficiently modeling transient, non-Darcy gas produc- (not shown).
tion from hydraulically fractured horizontal shale-gas wells is
to (a) use the dual-permeability method to represent all network Modeling Variations in Fracture Intensity in the Stimulated
fractures in both the unstimulated and stimulated volumes and Volume. We extended the new DK-LS-LGR approach to account
to (b) locally refine the grid in the stimulated volumes using the for variable fracture intensity in the stimulated volume by (a)
same logarithmically spaced grid design as was developed for the using a base gridblock size throughout the model whose length is
single-plane vertical-fracture work mentioned previously. This new set equal to the largest fracture spacing observed in a given situ-
approach is henceforth referred to in this paper as the “DK-LS- ation, (b) applying logarithmic-spaced LGR within the simulated
LGR method” (i.e., for dual permeability, logarithmically spaced, volumes, and (c) varying the dual-permeability parameters within
locally refined grid) for modeling hydraulically fractured shale-gas the stimulated volumes to account for variable fracture intensity
wells/reservoirs. (i.e., spacing) and effective permeability. Fig. A-6a depicts the
Fig. A-3 compares the grid designs within the stimulated vol- stimulated-volume region surrounding a horizontal shale-gas well
umes for both the very-fine-grid reference model (a), which uses as is typically interpreted from MS data gathered during hydraulic-
43×43 LGR with 0.001-ft-wide grids representing each fracture fracturing treatments. The base grid length in this example is 200 ft
and the simplified model (b), which uses 9×9 LGR with 2-ft-wide on a side and represents the coarsest fracture spacing observed in
grids representing each fracture. Both the reference grid and the this particular shale formation. Fig. A-6b shows how we apply log-
simplified grid use logarithmic spacing to capture transient flow arithmically spaced, local grid refinement to the stimulated-volume
behavior to and within the fractures. The gas-production rates from region, and also how we modify effective fracture permeability (the
these two grids are shown in Fig. A-4 and confirm that the simpli- green areas) and the fracture spacing (not shown) to represent vari-
fied approach gives an excellent match to the reference model. able-intensity fracturing within the stimulated volumes. Fig. A-7
100.00
xe/xf=3
xe/xf=7
xe/xf=15
10.00
Infinite
IMEX Infinite
PD
IMEX xe/xf-7
IMEX xe/xf=3
0.10
0.01
0.0001 0.001 0.01 0.1 1 10 100 1000 10000
tDxf
(a) (b)
Fig. A-2—FBHP comparison between Fig. A-1 grid design and (a) a very-finely gridded single-porosity reference model and (b)
Gringarten’s infinite-conductivity type-curve analytical solution.
-9,900
-10,099.90
-10,099.90
-10,000
-10,000
Network Fractures
Network Fractures
-10,100.00
-10,100.00
Producer1
-10,100
-10,100
Producer1
-10,200
-10,200
-10,100.10
-10,100.10
-10,300
-10,300
0.00 75.00 150.00 feet
0.00 25.00 50.00 meter
10,099.80 10,099.90 10,100.00 10,100.10 9,800 9,900 10,000 10,100 10,200 10,300 10,400
Fig. A-3—Grid designs for (a) reference grid (43×43 LGR with 0.001-ft-wide grid for each fracture) and (b) simplified grid (9×9
LGR with 2-ft-wide grid for each fracture) representing network of propped fractures in stimulated volume of a hydraulically
fractured shale-gas well.
compares the production from a very-fine-grid single-porosity • Fractures of greater or variable intensity inside the stimulated
reference model and the DK-LS-LGR model (Fig. A-6) for shale volumes are modeled using scaled dual permeability properties
matrix permeability of (a) 0.0001 md and (b) 0.01 md, showing that (matrix and fracture permeability and fracture spacing).
the simplified gridding approach provides acceptable accuracy. • Outside the stimulated volume the grid is not refined.
The new approach to modeling hydraulically fractured shale- The advantages of the approach described in this paper
gas wells can be summarized as follows: for modeling hydraulically fractured shale-gas wells are as
• A dual-permeability approach has been validated for mod- follows:
eling the matrix and natural fractures both inside and outside of the • Dual-permeability behavior is accounted for both inside and
stimulated volumes in hydraulically fractured shale-gas wells. outside the stimulated volumes (important in some shale plays in
• The base fracture network in the stimulated volumes is mod- which unstimulated fracture permeability may be significant).
eled with 9×9 locally refined, logarithmically spaced grids in the • Variations in fracture intensity can be accommodated without
stimulated volumes. having to vary the base gridblock size.
1.87e+6
1.37e+6
8.71e+5
3.71e+5
0 10 20 30 40 50 60
Time (day)
Fig. A-4—Gas-production-rate comparison between Fig. A-3 grid designs for the case of 0.0001-md matrix permeability.
2.50e+6
2.50e+6
2.00e+6
2.00e+6 9x9 LGR 2 ft Bf Other Cells Equally Spaced
9x9 LGR 2 ft Bf Log Scaled
Minc 4 Equally Spaced
1.50e+6 1.50e+6 Minc 4 Log Scaled
1.00e+6 1.00e+6
9x9 LGR 2 ft Bf Other Cells Equally Spaced
9x9 LGR 2 ft Bf Log Scaled
5.00e+5 Minc 4 Equally Spaced 5.00e+5
Minc 4 Log Scaled
0.00e+0 0.00e+0
0 10 20 30 0 1,000 2,000 3,000 4,000
Time (day)
Time (day)
(a) (b)
Fig. A-5—Comparison of MINC dual-porosity approach to 9×9 LGR with 2-ft-wide grid representing each fracture.
Fig. A-6—(a) Typical stimulated volume resulting from a hydraulic-fracture treatment in a shale-gas well, and (b) the DK-LS-LGR
grid design created to model gas-production rate accurately in a numerical reservoir simulator.
New Dual Perm Model (0.0001 mD Shale) New Dual Perm Model (0.01 mD Shale)
Using Variable Fracture Spacing to Represent Using Variable Fracture Spacing to Represent
Different Fracture Intensity Different Fracture Intensity
Representing 66 ft Fracture Spacing in 200 ft DK blocks Representing 66 ft Fracture Spacing in 200 ft DK blocks
4.00e+6 4.00e+6
Single Porosity Run (Explicit Modelling Single Porosity Run (Explicit Modelling
or Variable Fracture Spacing or Variable Fracture Spacing
DK Run (Using Fracture Spacing DK Run (Using Fracture Spacing
3.00e+6 Scaling or 200 ft Blocks) 3.00e+6
Gas Rate SC (ft3/day)
Gas Rate SC (ft3/day)
2.00e+6 2.00e+6
1.00e+6 1.00e+6
0.00e+0 0.00e+0
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Time (day) Time (day)
(a) (b)
Fig. A-7—Gas-production-rate comparison between Fig. A-6 grid design and a very finely gridded single-porosity reference model
for matrix-permeability cases of 0.0001 and 0.01 md.
r wj /D
Fig. 1—Multiple-fractured horizontal well in a conventional tight
formation and the effective (total) fracture concept. CFD,
FD,eff
,eff
of the hydraulic fractures and the long-term performance of a Fig. 2—Effective wellbore radius vs. effective system conduc-
fractured-horizontal-well/reservoir system can be represented by tivity for a multiple-fractured horizontal well in a conventional
that of a single hydraulic fracture. The length of the equivalent tight reservoir [after Raghavan et al. (1997)].
hydraulic fracture is equal to the aggregate length of the hydraulic
fractures, and the conductivity of the equivalent fracture is equal formation is dominated by linear-flow regimes. Our interpretation
to the average of the conductivities of the individual fractures, as is similar to the ideas used for fractured vertical wells in commin-
shown in Fig. 3. The drainage volume around the equivalent frac- gled reservoirs by Camacho-V. (1984), Bennett et al. (1985), and
ture is a parallelepiped with the length of the equivalent fracture Camacho-V. et al. (1987). As shown by the schematic in Fig. 4,
(2xF,eff=2nFxF) and the width equal to the average hydraulic-fracture the trilinear-flow model couples linear flows in three contiguous
spacing (dF). flow regions: the outer reservoir (denoted by the subscript O), the
In the following sections, we will introduce a trilinear-flow inner reservoir between fractures (denoted by the subscript I), and
model based on the premise that the drainage volume is limited the hydraulic fracture (denoted by the subscript F). In this work,
to the reservoir volume between hydraulic fractures. In extremely we assume uniform distribution of identical hydraulic fractures
tight formations, such as shale, an invasive (natural or induced) along the length of the horizontal well. (If fractures are not equally
fracture network is required for the efficient drainage of the matrix. spaced, then the trilinear model is applicable before the interaction
This introduces additional flow characteristics without changing of the hydraulic fractures.) Although it is possible to incorporate
the predominantly linear-convergence of flow from reservoir to the individual hydraulic-fracture properties in the model by an
hydraulic fractures. Because the linear flow performances of approach similar to that of Raghavan et al. (1997), the common
multiple-fractured horizontal wells in conventional and unconven- field practice of creating equally spaced hydraulic fractures of
tional tight formations can be evaluated along the same lines, we similar properties justifies our assumption. Furthermore, unless
will discuss them together. First, we will introduce the analytical there is significant contrast in individual-fracture properties, our
trilinear-flow model used in this work to simulate the performances model should closely approximate the performance of the system
of multiple-fractured horizontal wells. Then we will use it for the when average fracture properties are used.
numerical computations and demonstration examples. To simulate the performances of multiple-fractured horizontal
wells in shale, we allow the inner reservoir to be naturally frac-
Trilinear-Flow Model tured. We use the dual-porosity idealization to represent the natu-
The basis of our trilinear-flow model is the premise that the rally fractured medium. In this paper, we consider transient fluid
productive life of a multiple-fractured horizontal well in a tight transfer from the matrix (m) to natural fractures (f) as idealized by
Kazemi (1969), de Swaan O (1976), and Serra et al. (1983). The
model, however, is general and can be used with the pseudosteady-
fluid-transfer formulation of Warren and Root (1963). Our choice
of the transient dual-porosity model in this paper is based on the
expectation of elongated transient flow periods in tight matrices.
This expectation is supported by the quarter-slope behavior of the
log-log pressure response of pressure-transient tests observed in
the field, which is a result of the intermediate-time behavior of the
transient-fluid-transfer model. We discuss the flow-regime charac-
teristics of transient and pseudosteady dual-porosity models and
their implications for pressure-transient analysis in a companion
paper (Brown et al. 2009). We note only that the pseudosteady-
fluid-transfer model may be applicable in formations with matrix
permeabilities much higher than those observed in common appli-
cations of multiple-fractured horizontal wells.
The derivation and verification of the trilinear model are given
by Brown (2009) and Brown et al. (2009). Final analytical expres-
sions are presented in Appendix A. For our discussions, here we
note some essential definitions. We let qF,sc denote the flow rate
for each hydraulic fracture such that qF,sc = qsc/nF, where qsc is the
total flow rate from the horizontal well and nF is the number of
Fig. 3—Effective-fracture concept for a multiple-fractured hori- hydraulic fractures along the well length. For a naturally fractured
zontal well in an unconventional tight (shale) reservoir. inner zone, we use the intrinsic properties of the matrix and the
y e =d F /2 y e =d F /2
No-Flow Boundary
xe xe
INNER RESERVOIR INNER RESERVOIR
NATURALLY NATURALLY
xF FRACTURED xF FRACTURED
k f, φf, c tf , k f, φf, c tf ,
k m, φm , c tm km, φm,, cctm tm
Line of symmetry
No-Flow Line
HORIZONTAL WELL HORIZONTAL WELL
Conceptual Representation of the Trilinear Model Symmetry Element Considered in the Mathematical Model
1/nF of the Well’s Drainage Area 1/4 of the Fracture’s Drainage Area
Producing q/nF stb/d (oil) or Mscf/d (gas) Producing qF /4 =q/(4nF) stb/d (oil) or Mscf/d (gas)
Fig. 4—Schematic of the trilinear-flow model used for the analytical solution of multiple-fractured-horizontal-well performance.
fractures. Specifically, kIh = kf hf , where kf is the intrinsic perme- convergence in fractures is radial around the wellbore intersection.
ability of the natural fractures, and hft is the total thickness of the However, unless the fracture length (2xF) is comparable to fracture
natural fractures intersecting the height of the hydraulic fracture height, flow within the fracture is predominantly linear, except
hF (in this work, hF is equal to the pay thickness h). If nf is the with the radial-flow convergence around the wellbore that causes
count of natural fractures intersecting the height of the hydraulic a choking effect (Soliman et al. 1990; Larsen and Hegre 1994;
fracture and hf is the aperture of the natural fractures, we have hft = Al-Kobaisi et al. 2006). We account for the effect of flow choking
nf hf . We also introduce the density of natural fractures, f = nf /h, around the intersection of the horizontal well and hydraulic frac-
in units of the number of natural fractures per foot of pay zone ture by incorporating a choking-skin (sc) expression (see Appendix
(f/ft). An estimate of f can be obtained from cores, logs, rate of A, Eq. A-23) given by Mukherjee and Economides (1991). The
penetration, and other sources. choking-skin concept should be valid after the end of radial flow
For the clarity of the discussions presented later, it is useful in the fracture. Early-time radial flow within hydraulic fractures
to comment on the no-flow boundaries and the lines of symmetry does not last long and is masked by the effect of wellbore stor-
considered in the trilinear model. As shown in Fig. 4, for the age. For the purposes of this work, we do not consider the effect
assumptions of the trilinear model, the lines bisecting the width of wellbore storage in our discussions. However, wellbore-storage
of each fracture and the lines bisecting the distance between two effect can be incorporated easily into the trilinear model by using
identical fractures (dF) are no-flow lines of symmetry. Similarly, Eq. A-24 in Appendix A. For the discussion of wellbore-storage
the axis of the horizontal well forms a no-flow line of symmetry. effect on trilinear-model responses, we refer the reader to Brown
Therefore, it is sufficient to model a symmetry element, which is et al. (2009).
one-quarter of a fracture’s drainage area with a production rate of
qF,sc/4=qsc/(4nF). In this model, the effect of the no-flow bound- Effect of the Outer Reservoir
ary at ye = dF/2 corresponds to the effect of interference between In Figs. 5 and 6, we present some results to support the premise
hydraulic fractures. and field observations that the drainage of fractured horizontal
We also consider another no-flow boundary at a distance xe ≥ wells is confined to the inner reservoir between the hydraulic frac-
xF from the axis of the horizontal well. The boundary at xe may tures (also referred to as the stimulated volume). Before discussing
correspond to a no-flow boundary between two wells or to a physi- the results, we must comment on the value of the matrix perme-
cal barrier to flow. Assuming xe = xF limits the drainage area to the ability (km=1,000 md) used in Figs. 5 and 6. When we used matrix
stimulated inner reservoir between hydraulic fractures. If xe > xF permeabilities in the millidarcy range in our simulations, we did
and the outer reservoir has considerably lower permeability com- not observe any influence of the outer reservoir because the outer
pared with the inner reservoir, then the flow across the boundary reservoir is made of the same matrix (without natural fractures).
between the inner and the outer reservoirs is negligible initially; A tighter matrix causes faster pressure drop in the inner zone,
that is, x = xF is a no-flow boundary at early times. The outer res- but the support of the outer zone is also less efficient and takes a
ervoir may contribute to flow when pressure drops significantly to longer period to be felt. Therefore, the cases presented in Figs. 5
start depletion in the inner reservoir. The magnitude and the time and 6 may not correspond to tight, unconventional reservoirs but
of contribution of the outer reservoir depend on the permeability they are useful to make the pertinent observations on the general
and volume contrast between the inner and the outer reservoirs. characteristics of the support of the outer reservoir.
Interplay among the effects of the boundaries just discussed is In Figs. 5 and 6, we consider a dry-gas reservoir and for the
responsible for the changes in flow and production characteristics identification of the conventional flow regimes, we present the
of multiple-fractured horizontal wells in shale reservoirs. results in terms of pseudopressure (Al-Hussainy et al. 1966;
We should also explain our assumption of linear flow in hydrau- Al-Hussainy and Ramey 1966) defined by Eq. A-3 in Appendix A.
lic fractures, which is similar to fractured-vertical-well models. Pseudopressure change is defined as the difference from the initial
This assumption is applicable when the wellbore intercepts the pseudopressure (Δmwf = mi−mwf) and is presented as a function of
hydraulic fracture through its entire thickness. In the case of a time for three different distances to the outer reservoir boundary
fractured horizontal well, the wellbore communicates with the (xe). In Figs. 5 and 6, the case for xe = xF = 250 ft represents the
fracture through a small cylindrical surface piercing through the conditions in which the drainage volume is limited to the inner
fracture surface. As a result of this restricted communication, flow reservoir and the hydraulic-fracture spacing (dF=2ye) is 500 ft. For
this fracture spacing, the effect of fracture interference (that is, the from the outer reservoir is even more insignificant compared with
effect of the no-flow boundary at ye=dF/2) starts at approximately the volume produced from the inner zone). It is important to note
100 hours. The unit-slope behavior of the responses after 1,000 that mwf = 8.69×108 psi2/cp corresponds to pwf = pi−pwf ≈ 4,000
hours is the same as pseudosteady-state flow behavior. This behav- psi for these examples. Because the initial pressure is pi=4,000
ior is a result of the fact that the outer region does not respond psi, the absolute limit of production is reached at approximately
significantly until the inner zone is depleted to a low pressure. The 10.5 years. Therefore, subject to the constraints of the model con-
deviation of the pseudopressures from the xe = xF case (deviation sidered in our work, for the case presented in Fig. 5a, the effect of
from unit-slope behavior) indicates the start of the support of the the outer boundary will not be felt during the life of the well.
outer reservoir. If, for example, the outer region in Figs. 5 and 6 In Fig. 5b, the natural-fracture permeability is increased to
had permeability comparable to that of the inner region, the sup- kf = 20,000 md, which may not be realistic for most natural
port of the outer reservoir could cause a well-defined half-slope fractures. This case, however, may be used to simulate a hydrau-
behavior (unit slope for pseudosteady state in the inner reservoir lic-fracture network (resulting from dense, complex hydraulic-
and half slope for infinite-acting liner flow in the outer reservoir). fracture growth) with a fracture spacing of 1.25 ft (f = 0.8 f/ft).
For the permeability contrast used in Figs. 5 and 6, the value of As indicated by the comparison of the results in Figs. 5a and 5b,
the late-time slope tends toward one-half before becoming unity at higher natural-fracture permeability causes the support of the outer
late times when the outer reservoir goes into depletion under the reservoir to become evident earlier; however, the effect of the outer
effect of its no-flow boundary at xe. reservoir is still outside the practical life span of the well. These
Fig. 5 shows the effect of natural-fracture permeability. In observations support the common practice of small well spacing
Fig. 5a, for a realistic value of natural-fracture permeability (kf = to efficiently drain tight, unconventional reservoirs with fractured
2,000 md), the support of the outer reservoir becomes evi- horizontal wells.
dent after approximately 100 years for a well spacing of 107 ft Fig. 6 demonstrates the effect of natural-fracture density on the
(xe = 5×106 ft). For smaller well spacing, it takes even longer to feel contribution of the outer reservoir. Fig. 6a is the same as Fig. 5b
the effect of the outer boundary (the relative volume of production having a natural-fracture density of f = 0.8 f/ft. In Fig. 6b, the
(a) (b)
A – ρ f = 0.8 f / ft B – ρ f = 2.0 f / ft
xF = 250 ft f xF = 250 ft f
1E+7 ---------------------
dF = 500 ft 1E-8 dF = 500 ft
1E-6 Matrix km (md)
1E-4 1E+6
--------------------- qF,sc = 100 Mscf/d
1E-2
1E-8
1E+6 1E-1
1E-6 Reservoir
Reservoir
1E+5 1E-4 -------------------
-------------------
1E-2 h = 250 ft
h = 250 ft
1E-1 T = 150oF
T = 150oF γ = 0.55
1E+5 γ = 0.55
qF,sc = 100 Mscf/d 1E+4 pi = 4000 psi
pi = 4000 psi
μ = 0.0184 cp
WITH CHOKING SKIN μ = 0.0184 cp i
i
xe = 500 ft
WITHOUT CHOKING SKIN xe = 500 ft
1E+4 1E+3
1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E+6 1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E+6
TIME, t, hr TIME, t, hr
(a) (b)
Fig. 7—Effect of matrix permeability on multiple fractured horizontal-well performance—naturally fractured inner reservoir.
fracture density is increased to f = 2 f/ft. Comparing the results cannot be used. For the particular example in Figs. 7 and 8, the
in Figs. 6a and 6b, higher fracture density does not cause a notice- effective, or bulk, permeability of the natural-fracture system is
able change in the time required to feel the effect of the outer k f =kf hft/h=1.6 md. Flow capacity of the natural fractures will be
reservoir; and although at higher fracture densities the contribution discussed in the next section. Here, we note that the validity of
of the outer reservoir is more significant at late times, the effect the preceding discussion requires a large permeability contrast
becomes evident only at impractically late times. The discussion between the matrix and fractures; that is, the production of the
in this section, therefore, supports the common interpretation that matrix must be by virtue of the natural fractures as in the dual-
fractured horizontal wells do not drain the region beyond the tips porosity idealization (Warren and Root 1963; Kazemi 1969; de
of the hydraulic fractures for practical matrix permeabilities. Swaan O. 1976). Most tight gas reservoirs and shale-gas reservoirs
satisfy this requirement and the preceding conclusions can be
Effect of Matrix Permeability generalized for practical applications.
When the matrix permeability decreases from milli- to micro- To demonstrate the difference between the performances of
and nanodarcy range, the reservoir description is changed from multiple-fractured horizontal wells in conventional tight (e.g.,
conventional to unconventional. For unconventional reservoirs, a tight gas) and unconventional tight (e.g., shale-gas) formations,
dense fracture network is required for a sufficient effective perme- we present two cases in Fig. 9. The unconventional tight reservoir
ability in the system. Here, we will first present a discussion of the has a natural-fracture network with matrix and natural-fracture
effect of matrix permeability when the inner reservoir is naturally permeabilities of km = 10−6 md and kf = 2,000 md, respectively (the
fractured, as in unconventional tight formations. Then, we will corresponding effective fracture-network permeability is 1.6 md).
present a comparison with the case in which there are no natural The conventional reservoir is homogeneous and has a permeability
fractures, as in conventional tight-sand applications. of k=10−1 md. Despite the large contrast between the matrix perme-
Fig. 7 shows the pseudopressure change vs. time (Fig. 7a) abilities of the two systems, Fig. 9a shows similar pseudopressure
and log-derivative of pseudopressure vs. time (Fig. 7b) for matrix responses. In fact, Fig. 9b shows that the productivity of the higher-
permeabilities in the range of 10−8 md ≤ km ≤ 10−1 md. In Fig.
7a, pseudopressure responses are also shown for the idealized
case of no choking-skin effect (sc) within hydraulic fractures. 1E-3
Fig. 7a indicates that the choking skin masks the characteristics
PRODUCTIVITY INDEX, J / nF , Mscf - cp / d - psi2
of the inner reservoir. Note that the productivity index stabilizes xe = 500 ft dF = 500 ft cm = 10-5 psi-1
when boundary-dominated flow is established at late times. Figs. 1E-7
7 and 8 indicate that the well performance improves as the matrix 1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5
permeability increases but that matrix permeabilities higher than or TIME, t, hr
equal to 10−4 md do not contribute to the well performance. This
indicates that when the maximum flow capacity of the natural- Fig. 8—Effect of matrix permeability on the productivity of a mul-
fracture system is reached, additional productivity of the matrix tiple-fractured horizontal well—naturally fractured reservoir.
mwf, psi2/cp
kF = 105 md kf = 2000 md km = 10-6 md kf = 2000 md km = 10-6 md kF = 105 md
wF = 10-2 ft hf = 10-3 ft φm = 0.05 hf = 10-3 ft φm = 0.05 wF = 10-2 ft
φ = 0.38 φ = 0.45 c = 10-5 psi-1 φ = 0.45 c = 10-5 psi-1 φ F = 0.38
1E+8 F f m f m
cF = 10-5 psi-1 cf = 10-5 psi-1 cf = 10-5 psi-1 cF = 10-5 psi-1
ρ = 0.8 f / ft ρ = 0.8 f / ft
PSEUDOPRESSURE CHANGE,
xF = 250 ft f f xF = 250 ft
dF = 500 ft dF = 500 ft
1E+7 1E-4
qF,sc = 100 Mscf/d
Reservoir
Reservoir
-------------------
-------------------
h = 250 ft
h = 250 ft
1E+6 T = 150oF
T = 150oF γ = 0.55
γ = 0.55 qF,sc = 100 Mscf/d
pi = 4000 psi
pi = 4000 psi
μ = 0.0184 cp NATURALLY FRACTURED
NATURALLY FRACTURED μ = 0.0184 cp i
i
xe = 500 ft HOMOGENOUS, k = 10-1 md
HOMOGENOUS, k = 10-1 md xe = 500 ft
1E+5 1E-5
1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E+6 1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5
TIME, t, hr TIME, t, hr
(a) (b)
Fig. 9—Effect of matrix permeability on the change in pseudopressure (a) and productivity (b) of a multiple-fractured horizontal
well—homogeneous and naturally fractured inner reservoirs.
permeability conventional, homogeneous reservoir is less than that A higher density of natural fractures, however, increases the flow
of the unconventional tight reservoir with natural fractures. This capacities of both the natural fractures and the matrix because the
example underlines the importance of natural fractures in uncon- increase in surface area for fluid transfer leads to a more significant
ventional tight formations, such as shale. increase in productivity. From a practical viewpoint, this is a desir-
able feature because increasing the permeability of natural fractures
Effect of Natural-Fracture Permeability and may not be a viable option while increasing the density of natural
Density fractures may be accomplished by designing complex hydraulic
In the preceding section, we emphasized the importance of the fractures in brittle formations (Cramer 2008; Cipolla 2009). It can
flow capacity of natural fractures for production from tight for- also be deduced that creating a complex hydraulic-fracture network
mations. Figs. 10 and 11 indicate that a higher natural-fracture of high permeability (but relatively low density) is not a good option
permeability may not significantly improve the well performance if hydraulic fracturing does not also increase the density of natural
while a higher density of natural fractures may have a significant fractures. This indicates that the experience in brittle shale forma-
effect on productivity. As discussed in the preceding section, this tions, such as the Barnett, may not be extrapolated directly to soft-
is because of the relative flow capacities of the matrix and frac- shale formations such as the Marcellus and the Haynesville.
ture systems. A higher permeability increases the flow capacity
of natural fractures but does not change the flow capacity of the Effect of Hydraulic-Fracture Permeability
matrix because the flow capacity of the matrix is proportional to and Spacing
the surface area. If there is not enough fluid transfer from matrix Determining the optimum number of hydraulic fractures to be
to fracture, a high permeability in the natural fractures does not created along the length of the well is an important practical con-
create additional productivity. cern. A relevant, but more conventional, concern is determining
1E+7 1E-3
Natural Fractures Natural Fracture Matrix Natural Fractures Natural Fracture Matrix
------------------------- Permeability
PRODUCTIVITY INDEX, J / nF , Mscf - cp / psi2
Reservoir
-------------------
h = 250 ft
1E+6 T = 150oF 1E-4
γ = 0.55 Hydraulic Fractures Hydraulic Fractures Reservoir
pi = 4000 psi ---------------------------- ---------------------------- -------------------
μ = 0.0184 cp kF = 105 md kF = 105 md h = 250 ft
i
T = 150oF
xe = 500 ft wF = 10-2 ft wF = 10-2 ft γ = 0.55
φ F = 0.38 φ F = 0.38 pi = 4000 psi
cF = 10-5 psi-1 c = 10-5 psi-1 μ = 0.0184 cp
F i
xF = 250 ft xF = 250 ft xe = 500 ft
qF, sc = 100 Mscf/d
dF = 500 ft dF = 500 ft
1E+5 1E-5
1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4
Fig. 10—Effect of natural-fracture permeability on the change in pseudopressure (a) and productivity (b) of a multiple-fractured
horizontal well.
Fracture Density
Hydraulic Fractures ρ (f / ft)
f
---------------------------- ---------------------
1E+7 kF = 105 md xF = 250 ft 1.2E-0
8.0E-1
wF = 10-2 ft dF = 500 ft
Reservoir 4.0E-1
φ F = 0.38 2.0E-1 Reservoir
-------------------
cF = 10-5 psi-1 h = 250 ft 1E-5 1.2E-1 -------------------
Natural Fractures h = 250 ft
T = 150oF
1E+6 γ = 0.55 ------------------------- T = 150oF
kf = 2000 md γ = 0.55
pi = 4000 psi
μ = 0.0184 cp hf = 10-3 ft pi = 4000 psi
i
φ = 0.45 μ = 0.0184 cp
qF, sc = 100 Mscf/d xe = 500 ft f i
cf = 10-5 psi-1 xe = 500 ft
1E+5 1E-6
1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E+6 1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5
TIME, t, hr TIME, t, hr
(a) (b)
A – PSEUDOPRESSURE VS. TIME B – PRODUCTIVITY INDEX VS. TIME
Fig. 11—Effect of natural-fracture density on the change in pseudopressure (a) and productivity (b) of a multiple-fractured hori-
zontal well.
the optimum conductivity of hydraulic fractures. Fig. 12 is a density. Therefore, it is desirable to obtain higher dimensionless
demonstration of the conventional concept that increasing hydrau- hydraulic-fracture conductivities by creating larger natural-fracture
lic-fracture conductivity, kFwF, leads to an increase in productivity. densities. It must be noted, however, that higher natural-fracture
The incremental benefit, however, decreases as the conductivity densities also increase the effective natural-fracture permeability
increases. From a fundamental viewpoint, higher productivity and, therefore, decrease the dimensionless hydraulic-fracture con-
is achieved by making the dimensionless conductivity, CFD = ductivity. This leads to a complex design problem for hydraulic
kFwF/(kxF), larger. This concept is relatively straightforward for fractures in tight, naturally fractured formations.
the design of hydraulic fractures in conventional tight formations As shown in Fig. 13a, reducing the hydraulic-fracture spacing
(such as tight gas sands) in which higher conductivities can be significantly influences the magnitudes of the intermediate- and
achieved for low reservoir permeability k. A large dimensionless late-time pseudopressures. On the other hand, Fig. 13b shows
conductivity indicates that the hydraulic fracture can transmit all that hydraulic-fracture spacing does not significantly change flow
the fluid provided by the reservoir. characteristics (it does, however, affect the duration of the flow
For unconventional tight formations, dimensionless hydraulic- regimes). It must be noted that the flow rate of each fracture has
fracture conductivity is defined on the basis of the effective fracture been kept constant while changing the hydraulic-fracture spacing
permeability (k = k f ) and may appear to be relatively large; yet, in Fig. 13. If the total flow rate was kept constant, then the indi-
as discussed in the preceding section, flow capacity of the matrix vidual-fracture rates would have to be reduced when the fracture
may not be sufficient to supply the fracture network to its full spacing decreased. This would yield a separation of the pseudo-
flow capacity because of low matrix permeability or low surface pressure curves at early times (with the smallest pseudopressure
area available for fluid transfer because of low natural-fracture change corresponding with the smaller fracture spacing).
1E+9 1E-3
Hydraulic Fractures Natural Fractures Hydraulic Fractures Reservoir
---------------------------- Natural Fractures ------------------------- ---------------------------- -------------------
PRODUCTIVITY INDEX, J / nF , Mscf - cp / psi2
mwf, psi2/cp
------------------------------ ρ = 1.2 f / ft f
f ---------------
1 x 105 km = 10-6 md
8 x 104
1E+7 φm = 0.05 1E-4
5 x 104
3 x 104 cm = 10-5 psi-1
Reservoir
------------------- Fracture Permeability
h = 250 ft kF (md)
1E+6 T = 150oF Matrix
γ = 0.55 ------------------------------ ---------------
pi = 4000 psi 1 x 105 km = 10-6 md
μ = 0.0184 cp 8 x 104 φm = 0.05
qF, sc = 100 Mscf/d
i
5 x 104
xe = 500 ft cm = 10-5 psi-1
3 x 104
1E+5 1E-5
1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E+6 1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5
TIME, t, hr TIME, t, hr
(a) (b)
A – PSEUDOPRESSURE VS. TIME B – PRODUCTIVITY INDEX VS. TIME
Fig. 12—Effect of hydraulic-fracture conductivity on the change in pseudopressure (a) and productivity (b) of a multiple-fractured
horizontal well.
mwf, psi2/cp
--------------------- kf = 2000 md km = 10-6 md --------------------- kf = 2000 md km = 10-6 md
1000 hf = 10-3 ft φm = 0.05 1000 hf = 10-3 ft φ = 0.05
750 1E+8 750 m
1E+8 500 φ = 0.45 cm = 10-5 psi-1 φ f = 0.45 cm = 10-5 psi-1
f 500
250 cf = 10-5 psi-1 250 -5
cf = 10 psi -1
ρ = 0.4 f / ft ρ = 0.4 f / ft
PSEUDOPRESSURE CHANGE,
f f
1E+7
Hydraulic Fractures
Hydraulic Fractures
----------------------------
1E+7 ----------------------------
kF = 105 md
kF = 105 md
wF = 10-2 ft Reservoir 1E+6 wF = 10-2 ft Reservoir
φ = 0.38 ------------------- -------------------
F φ F = 0.38
-5
cF = 10 psi -1 h = 250 ft h = 250 ft
T = 150oF cF = 10-5 psi-1 T = 150oF
1E+6 xF = 250 ft γ = 0.55 γ = 0.55
xF = 250 ft
pi = 4000 psi 1E+5 pi = 4000 psi
μ = 0.0184 cp μ = 0.0184 cp
i i
qf, sc = 100 Mscf/d xe = 500 ft qF, sc = 100 Mscf/d xe = 500 ft
1E+5 1E+4
1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E+6 1E-3 1E-2 1E-1 1E+0 1E+1 1E+2 1E+3 1E+4 1E+5 1E+6
TIME, t, hr TIME, t, hr
(a) (b)
A – PSEUDOPRESSURE VS. TIME B – LOG DERIVATIVE VS. TIME
Fig. 13—Effect of hydraulic-fracture spacing on the pseudopressure responses of a multiple-fractured horizontal well.
The productivity of each fracture is expected to increase when increase. This conclusion supports the field practice of creating
the fracture spacing is decreased. As shown in Fig. 14a, however, complex fractures to increase natural-fracture density and decrease
the incremental gain in productivity with each additional fracture hydraulic-fracture spacing.
decreases. To find the total productivity of the system, the produc-
tivity indices in Fig. 14 must be multiplied by the corresponding Conclusions
number of hydraulic fractures. For example, for a 4,000-ft horizon- In this paper, we investigated the performance characteristics of
tal well, 1,000-ft and 250-ft hydraulic-fracture spacing corresponds fractured horizontal wells and discussed the effects of hydraulic
to two and eight hydraulic fractures, respectively. The total produc- fractures and reservoir properties. The trilinear-flow model pre-
tivity of the well with 1,000-ft fracture spacing is then obtained sented in this work has provided an effective tool for understanding
by multiplying the corresponding individual-fracture productivity and predicting the productivities of fractured horizontal wells.
given in Fig. 14 by two. Similarly, the well productivity for 250- The following conclusions are warranted on the basis of this
ft fracture spacing is obtained by multiplying the corresponding work:
productivity index by eight. This would yield a larger productivity 1. Differences in the performance of fractured horizontal wells in
for the well with eight fractures (250-ft spacing) than for the well conventional and unconventional tight formations are mostly
with two fractures (1,000-ft spacing). determined by the existence of a natural- or induced-fracture
To complete our discussions, in Fig. 14b we present individual- network in the latter.
fracture productivities for two values of natural-fracture density 2. Unlike conventional tight gas reservoirs, relatively high matrix
and two values of hydraulic-fracture spacing. As expected, Fig. 14b permeability and increased hydraulic-fracture conductivity
indicates that increasing natural-fracture density while decreas- are not sufficient to have good productivity in unconventional,
ing hydraulic-fracture spacing provides the greatest productivity shale-gas reservoirs. Existence of a dense natural- or induced-
1E-3 1E-3
Fracture Spacing Natural Fractures Matrix Natural Fractures Hydraulic Fractures Matrix Reservoir
PRODUCTIVITY INDEX, J / nF , Mscf - cp / d - psi2
Fig. 14—Effect of hydraulic-fracture spacing and natural-fracture density on the productivity of multiple-fractured horizontal
wells.
2F s The expression given by Eq. A-23 should be valid after the end of
F = + , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-10) the radial flow within the hydraulic fracture, which corresponds to
CFD FD
a very short period of time at the beginning of flow. For practical
cases, the early-time radial flow within the hydraulic fracture is
F completely masked by the effect of wellbore storage. The effect
FD = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (A-11) of wellbore storage can be incorporated easily by using the fol-
I
lowing relationship:
pwD
F = R tanh ⎡⎣ R ( yeD − wFD 2 ) ⎤⎦ , . . . . . . . . . . . . . . . (A-12) pwD ,storage = mwD ,storage = , . . . . . . . . . . . . . . . . (A-24)
1 + C D s 2 pwD
k(Kn=0)/kads
5
1
101
d (nm)
3 P=28 MPa
P=20 MPa
P=10 MPa
k[Kn(P,T=300 K)]/k(Kn=0)
P=5 MPa
2.5
Fig. 2—SEM of Barnett shale (Ambrose et al. 2010; Wang and 1.5
Reed 2009).
/kin-situ[Kn(P,T=360 K)]
P=5 MPa
2.5 8
2 6
1.5 4
1 2
0.5 0
101
0 d (nm)
101
d (nm) Fig. 6—The ratio of gas conductance klab without adsorbed
layer and with slippage (laboratory condition of 300 K and 5
MPa) to gas conductance kin-situ with adsorbed layer and slip-
Fig. 5—Combined effect of adsorbed layer and slip on hydraulic
page (field condition of 360 K and P as per legend) increases as
conductance of a single cylindrical conduit [ratio of gas con-
pressure increases and as cylindrical throat size decreases.
ductance kin-situ with adsorbed layer and slippage (in-situ con-
These curves are relevant to estimating field permeability from
dition of 360 K and P as per legend) to liquid conductance
laboratory measurements.
k(Kn 5 0) without adsorbed layer and without slippage]. This
behavior is relevant to evolution of permeability during
production. tory condition is notably higher than at the in-situ condition at
larger pressures. This is because of the lack of an adsorbed layer
and the enhancement of slippage at laboratory condition vs. in-situ
computing slip flow within the open cross section of the throat condition. Clearly, the laboratory measurements should be cor-
remaining after implementing the adsorbed layer at the specified rected for any flow modeling at the field condition. Considering the
pressure. The temperature is held constant at 360 K. Fig. 5 shows importance of this issue, we will analyze the laboratory measure-
qualitatively different trends, depending on pressure. At large ments using network modeling and propose a correction in the fol-
pressures, the ratio of gas conductance at in-situ conditions to the lowing paragraphs.
corresponding no-slip conductance increases as the pore size Liquid permeability is also of interest in shale, either for the
increases. At small pressures, the ratio decreases as the pore size problem of water production or for the challenge of “tight oil”
increases. production. To estimate liquid permeability at the field condition
This calculation helps us understand which effect governs the from gas permeability measured at the laboratory condition, we
flow behavior at different stages of production. The purpose of calculate the ratio of gas to liquid conductances of a single cylin-
the comparison between gas and liquid conductances is to analyze drical conduit. As in the preceding, the adsorbed layer at the labo-
the gas-conductance variation vs. a reference value. It compares ratory condition is negligible because of low pressure. Therefore,
the gas conductance (360 K, various P) that is representative of the gas conductance at the laboratory condition is computed from
reservoir condition to liquid conductance, the reference value the liquid conductance by implementing the slippage at the labo-
evaluated in the absence of an adsorbed layer and ignoring slip. ratory pressure and temperature (5 MPa, 300 K). For the liquid
The analysis shows the adsorbed layer dominates the flow behav- conductance at the in-situ condition, we calculate the throat area
ior at high pressure (P ¼ 28 MPa in Fig. 5) regardless of the throat not obstructed by the thickness of the adsorbed layer. The thick-
size. That is, the gas conductance is smaller than the reference ness of the adsorbed layer depends on pressure. We ignore the
conductance for all throat sizes because the adsorbed layer signifi- swelling effect for the in-situ condition, which may reduce the hy-
cantly reduces the cross section for gas flow. The influence is draulic conductance. This may occur because of the presence of
greater for smaller throats. However, the slippage has the primary water in the formation, which is not available at the same amount
effect at small pressures (P ¼ 10 MPa or P ¼ 5 MPa). That is, the at the laboratory condition. The ratio of gas to liquid conductan-
gas conductance is greater than the reference conductance for all ces at laboratory and field conditions is shown in Fig. 7. The
throats, with the effect being greater in narrower throats. The results reveal the gas permeability obtained at the laboratory con-
implication of Fig. 5 is that the adsorbed layer should be taken into dition is significantly larger than the liquid permeability at the in-
account for modeling gas transport during early production, while situ condition. The ratio of permeabilities estimated here is larger
the influence of slippage is dominant during later production. than in the preceding analysis (Fig. 6) because the liquid does not
Finally, for the single-tube study, we compare the gas conduc- slip at laboratory condition. In reality, the ratio of gas to liquid
tances at laboratory and in-situ conditions. This is useful for conductances may be even greater than in Fig. 7 because the hy-
estimating gas permeability in the field from a laboratory measure- draulic conductance of the throat decreases because of possible
ment. The adsorbed layer is presumed negligible at the laboratory swelling clays available in the formation.
condition because the pressure (5 MPa) is low; the curve for the
corresponding pressure in Fig. 3 shows that the correction is less Analysis of Network of Cylindrical Conduits of Distributed
than 25% for all throat sizes. The thickness of the adsorbed layer Sizes. We employ a regular square lattice network (Bennett and
varies at the in-situ condition, depending on the pressure. Note that Myers 1962) to model the gas flow through organic materials in
the slippage occurs in both conditions. We calculate the gas con- shale. We adopt the pore-size distribution from the drainage
ductance at the laboratory condition (T ¼ 300 K) from the liquid experiment shown in Fig. 1. We also assume the width of the net-
conductance, accounting for slippage at laboratory pressure in the work (transverse to the direction of flow) and the throat length are
relevant flow regime (Kn criterion evaluated at 300 K and 5 MPa) 17 and 50 nm, respectively. The model width is chosen so that the
by means of Eq. 2, Eq. 3, or Eq. 9. For the in-situ condition, we network model yields a desired porosity. Here, 17 nm is the value
compute the cross-sectional area open to flow, depending on the that results in the desired porosity of 10%. The small width means
thickness of the adsorbed layer. The gas conductance is computed that the network corresponds to a thin 2D slice through the or-
from the liquid conductance at the field pressure after implement- ganic material oriented along the axes of the circular holes. That
ing the slippage. Fig. 6 shows that the gas conductance at labora- is, we consider flow perpendicular to the plane of the images in
2.5
10
2
5 1.5
1
0
1
10 0.5
d (nm) 5 10 15 20 25 30
P (MPa)
Fig. 7—For a single cylindrical tube, a laboratory measurement
of gas permeability greatly overestimates the permeability to
liquid in the reservoir, as shown by the ratio of gas conduct- Fig. 8—The ratio of gas permeability with slippage and without
ance klab with slippage and without adsorbed layer (laboratory adsorbed layer (laboratory condition of 300 K and 5 MPa) to gas
condition of 300 K and 5 MPa) to liquid conductance kads with- permeability with adsorbed layer and slippage (field condition
out slippage and with adsorbed layer (field condition of 360 K of 360 K and P from the x-axis) obtained from a network of cy-
and P as in legend). Pressure determines the thickness of the lindrical throats having the size distribution shown in Fig. 1.
adsorbed layer.
2.5 1.4
1.3
2
1.2
1.5
1.1
1 1
5 10 15 20 25 30 0 1 2 3 4 5 6 7
P (MPa) P (MPa)
Fig. 9—The ratio of gas permeability at pressures below initial Fig. 10—The ratio of gas permeability without an adsorbed
reservoir pressure (kg2,in-situ) to gas permeability at initial pres- layer and with slippage [klab(P, T5300 K)] to the liquid perme-
sure (P 5 28MPa) (kg1,in-situ) increases as production continues ability without an adsorbed layer and without slippage
and pressure declines accordingly. The ratio is calculated from [k(Kn 5 0)] is obtained from a network of conduits and com-
a network of cylindrical throats having the size distribution pared to the laboratory measurements (Letham 2011). The
shown in Fig. 1. effective stress is constant, and hence, the pore-pressure
change does not affect pore-throat-size distribution. Gas per-
meability increases with lowering pressure because of slip-
page, unlike the liquid permeability, which is constant. The
dependency of rock resistance against gas flow on the pore pres- flowing gas is methane.
sure when the moving fluid is methane. The test with methane
should be run at reservoir temperature and pressures to represent
the in-situ conditions. Our model predicts that the effect of the mates the normalized gas permeability, the maximum difference
adsorbed layer becomes more important as the organic content of with the laboratory measurement being only 6%. This indicates
shale increases. The organic-rich region has more tendancy to that the slippage model provides a good estimate for the effect of
adsorb methane to its pore wall, which must be considered when pressure on gas permeability at moderate pressures.
choosing the core samples.
Data across the range of conditions of Figs. 8 and 9 for rocks
with different organic content are not currently available. How- Conclusions
ever, measurements of CH4 transport reported on a particular type Pore throats of shale are mostly narrower than 10 nm and are
of shale at pressures below 7 MPa (Letham 2011) let us partially inside organic material on which CH4 adsorbs. As a result, the gas
test the effect of slippage. The experimental data were measured permeability of these rocks is affected significantly by the
with methane for a sample with a permeability of 390 nanodarcies adsorbed gas and by the slip of flowing gas on the pore walls (or
at 5 MPa and 300 K. The effective stress was kept constant in the on layers of molecules adsorbed on the walls). The adsorbed layer
laboratory measurements (Letham 2011); that is, the confining does not play an important role in conventional rocks because the
stress was lowered to yield the same effective stress at a lower pore throats are much wider, nor is the gas volume desorbed from
pore pressure. The porosity of the laboratory sample was reported the organic material significant in conventional reservoirs. To bet-
to be 13%. The porosity of our original sample was 10%, which is ter understand gas-flow behavior through shale, we evaluated
slightly different from the laboratory sample. To compensate for these effects in individual conduits of a cylindrical cross section
these differences, we increased the pore-throat sizes of our origi- and in simple networks of such conduits. The effect of the
nal network model, of which permeability adsorbed layer was treated as purely geometric: The cross section
qwas
ffiffiffiffiffiffi 9 nanodarcies
at 5
open to flow was reduced by the thickness of the adsorbed layer,
390
MPa and 300 K, by a factor of 6.58 9 ¼ 6:58 . We also which was assumed to vary linearly with pressure. The effect of
increased the width of the network by a factor of 43.3 (¼ 6.582) to slip was accounted for by applying the model appropriate to the
keep the porosity of the network model unaffected by the change flow regime, according to the Knudsen number (Kn). The latter
in the pore-size distribution. Using the modified model, we com- depends on pressure, temperature, and conduit diameter. For the
pute gas permeabilities for a range of small pressures (between 1 slip flow regime, the first-order slip model was judged more suita-
and 6 MPa) and at 300 K. Since the effective stress was kept con- ble than the dusty gas model (DGM) for the shale-gas application.
stant during laboratory measurements, we do not change the At large pressures such as typical initial shale-gas-reservoir
modified pore-throat-size distribution with pore pressure. (In the pressures, the effect of the adsorbed layer dominates the effect of
field, effective stress increases with reservoir depletion, and the slip on gas-phase permeability. Slip dominates at smaller pres-
effect on pore-throat-size distribution, which is not included in sures typical of those after longer periods of production. Conse-
the model presented here, would diminish the slippage effect in quently, the reservoir matrix permeability is predicted to increase
Fig. 10.) Fig. 10 plots these gas permeabilities, normalized by the significantly throughout the life of a well, by a factor of 4.5, as
nominal liquid permeability at laboratory condition. Similar to ev- production continues and pressure declines. The models predict
ery calculation performed for the laboratory condition through that the typical conditions for laboratory measurements of perme-
this study, the gas flow here is with slippage and without the ability cause those values to overestimate field permeability by as
adsorbed layer, and the liquid flow is without slippage and with- much as a factor of four. The model results are captured in simple
out the adsorbed layer. However, unlike the laboratory condition analytical expressions that allow convenient estimation of these
we defined in which pressure was 5 MPa, we change the pressure effects.
to investigate its influence on gas permeability because of slip- For complete validation of the proposed model, laboratory
page. The liquid permeability for the laboratory data is 390 nano- measurements at elevated pressures (on the order of 28 MPa) and
darcies, estimated by extrapolating the measured gas permeability reservoir temperature made with CH4 are needed. Experimental
to large pressure. Fig. 10 shows that the model slightly underesti- data at such conditions, for which the adsorbed layer is expected
1 2rv
where the nondimensional no-slip velocity is defined as follows: kg 4 þ rv Kn 2 rv
¼ 1
¼ 1 þ 4 Kn : . . . . . . . . . ðA-9Þ
r
2 kl 4
rv
Uno-slip ¼ 1 ¼ 1 x2 ; . . . . . . . . . . . . . . . . ðA-3Þ
R Here kg is the permeability of the conduit to gas phase, and kl is
where R is the radius of the tube, r is the distance from the center the permeability of the same conduit when the no-slip boundary
of the tube, and x is the normalized radial distance from the center condition applies. The latter condition usually holds when a liquid
of the tube, which is equal to unity at the wall. To include the fills the conduit, hence the choice of subscript l. After implement-
effect of molecule/wall collisions, the first-order slip boundary ing rv ¼ 0.9, we obtain the enhancement of apparent permeability,
condition is imposed as (Klinkenberg 1941) called a1 in the text (refer to Eq. 3), to be equal to 5.
2 rv @Us DGM Expression in Terms of Kn. The apparent permeability of
Us j w ¼ Kn ; . . . . . . . . . . . . . . . . . . . . ðA-4Þ
rv @n w the conduit based on the DGM is calculated as
where n is the outward (unit) vector, which is normal to the tube kg DK l b
¼ 1þ n ¼ 1 þ : . . . . . . . . . . . . . . . . . ðA-10Þ
wall, and rv is the tangential momentum-accommodation coeffi- kl kl p p
cient. The tangential momentum accommodation is calculated by
measuring the gas-flow rate at the slip flow regime and is close to The second term on the right-hand side of Eq. A-10 can be simpli-
0.9 (Roy et al. 2003). fied after substituting the Knudsen diffusivity coefficient in terms
The first-order slip boundary condition is applied to derive the of a pertinent parameter, using Eq. 11. We also use the Hagen-
nondimensional velocity profile (see Eq. A-2) (Klinkenberg Poiseuille model for the permeability term, as follows:
1941): rffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffi
d 8RT 8RT
l
@Us @Uno-slip @Ucorrection @Uno-slip b DKn l 3 pm l 32 1 pm
¼ þ ¼ ¼ ¼ ¼
@n w @n @n
w @n w p kl p d2 3 d p
p
¼ ð2xÞjx¼1 ¼ 2: ðA-5Þ 32
rffiffiffiffiffiffiffiffiffi
2RT
The minus sign is because of the direction of the normal vector. l
64 1 pm
The correction term is constant, and, thus, its gradient vanishes. ¼ : ðA-11Þ
Therefore, Eq. A-4 simplifies to (Klinkenberg 1941) 3 d p
2 rv @Us 2 rv Here, we implement the ideal-gas assumption p ¼ q mR T ¼ qRT
K ¼ 2 Kn : . . . . . . . . . . . . . . ðA-6Þ to express the pressure in terms of density, which yields the
rv @n w
n
rv results in terms of Kn:
Thus, we can express the nondimensional velocity profile based rffiffiffiffiffiffiffiffiffi
2RT rffiffiffiffiffiffiffiffiffi
on Eqs. A-2 and A-3 as follows (Klinkenberg 1941): l
b 64 1 pm 64 l 2m
¼ ¼
2 rv p 3 d R 3 qd pRT
U s ¼ 1 x2 þ 2 Kn : . . . . . . . . . . . . . . . . ðA-7Þ q T
rv r mffiffiffiffiffiffiffiffiffi
128 l p
To cast this profile in terms of permeability, we obtain the flow ¼ ¼ 13:58Kn : ðA-12Þ
3p qd 2RT
rate as
40000
35000
25000
20000
15000
10000
5000
0
0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000
Fluid Volume (bbl)
Fig. 1—Relationship of total fracture-network length as a function of total job fluid volume pumped (Fisher et al. 2004).
oil window. Although general guidelines are applicable for SRV dimensions are given by the total fracture-network length (2xf) and
measurements, it is important to evaluate any given SRV measure- width (xn). However, the key property for well production is really
ment in conjunction with the particular reservoir setting. the total sum of all fracture-network segments (linear feet) within
the SRA, which is a strong function of fracture spacing (density).
SRV and Fracture Spacing in Shales Eq. 1 shows the calculation of total fracture length, L ftotal , for
It is important to note that the SRV is just the reservoir volume the entire SRV as a function of fracture-network half-length, xf ,
affected by the stimulation. It does not provide any details of the width, xn, and fracture spacing, Δxs:
effectively producing fracture structure or spacing. Maxwell et al.
(2006) introduced a concept that could eventually be used to charac- 4 x f xn
L ftotal = + 2 x f + x n . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1)
terize fracture density. In this approach, “additional seismic-signal ⌬x s
characteristics allow investigation of the source of the mechanical
deformation resulting in the microseisms. In particular, the seismic 1500
moment, a robust measure of the strength of an earthquake or All microseisms
microearthquake, can be used to quantify the seismic deformation.”
Besides this potential geophysical approach, reservoir modeling
1000 ~2,700 ft
also provides an avenue to evaluate the effectively producing net-
work better. The details of this approach will be described in the
following section. As an introduction, Fig. 6 shows an identical
SRV with different fracture spacings (densities) and the effect of 500
fracture spacing on the gas-recovery factor. The SRV in this graph is
approximately 2,000 × 106 ft3 (for h = 300 ft). In contrast to conven-
tional single-fracture modeling using fracture half-length, the SRA Frac
well
Northing (ft)
Simple fracture
Fracture Complex
Complexfracture
Fracture –500 ~1,200 ft
Possible
–1000
aligned
features
–1500
Observation well
–2000
–1000 –500 0 500 1000 1500
Complex fracture
Fracture Complex fracture
Fracture Easting (ft)
With fissure
with Fissureopening
Opening Network
network
Fig. 3—Microseismic fracture mapping shows complex-net-
Fig. 2—Types of fracture growth (Warpinski et al. 2008). work growth in shales (Warpinski et al. 2008).
Fig. 7 shows a plot of Eq. 1 on a log-log plot for the given SRV
in Fig. 6. The resulting curve forms an approximate straight line
on a log-log plot. As an example, reducing fracture spacing from
Free spacing: 200 ft Free spacing: 100 ft
300 to 50 ft would result in a more than five-fold increase in total
fracture length (from 48,600 to 264,600 ft) and would accelerate
the 3-year cumulative gas recovery by about the same multiplier.
This illustrates the importance of viewing SRV in context with the
Xf
Treatment horizontal well
Xn After 1 year production
Gas Recovery Factor (%)
1.E+05
1.E+04
10 100 1000
Fig. 7—Fracture spacing vs. total fracture length based on Eq. 1 for SRV of 2,000 × 106 ft3.
potential fracture spacing. A dual-porosity approach is not used at sizes (fluid volumes). While the data show some scatter, it clearly
this point because the fracture spacing is still relatively sparse and indicates a trend that large SRV’s will result in better well perfor-
the numerical approach better illustrates the linear flow patterns in mance. A similar trend is still valid when plotting SRV vs. longer
a low-permeability shale system. term 3-year cumulative gas production (Fig. 9). The scatter in the
data is to be expected as a result of variations in fracture spacing
SRV and Well Performance within the SRV and potential differences in effective network size
Previous publications have briefly discussed the relationship and conductivity. While the presented trends in Figs. 8 and 9 show
between actual SRV and well performance. Here we will expand a general relationship between SRV generated from microseismic
on the subject with more details and comparisons with longer-term mapping and production data, they do not provide any deterministic
production data from the Barnett shale. Fig. 8 shows a plot of quantification of how the actual effective network (portion of the
SRV measured from microseismic fracture mapping vs. 6-month network contributing to gas production) is structured to provide the
cumulative gas production from a group of horizontal wells in one specific gas production. The quantification of the effectively pro-
specific county in the Barnett shale. The plot shows that SRVs in ducing fracture network can be performed only with flow simula-
this county range from as low as 120 × 106 ft3 (2,755 acre-ft) to tions, such as those provided by a numerical reservoir simulator.
approximately 1,900 × 106 ft3 (43,618 acre-ft). Small SRVs were Previous work has provided extensive numerical reservoir mod-
generated partly as a result of shorter laterals and smaller treatment eling (Mayerhofer et al. 2006, Warpinski et al. 2008, Cipolla et al.
1000
900
800
6-month Cum Gas (MMscf)
700
600
500
400
300
200
100
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
SRV (Mio ft3)
Fig. 8—SRV trend vs. 6-month cumulative horizontal-well production for one Barnett-shale county.
2000
1000
500
0
0 400 800 1200 1600 2000
Fig. 9—SRV trend vs. 3-year cumulative horizontal-well production for one Barnett-shale county.
2008) to show how specific fracture-network properties can affect fracture length within a microseismically mapped SRV of 219 ×
well performance. The work demonstrated that numerical model- 106 ft3 (gray dots in graph). Fracture spacing was modeled to be
ing using a black-oil reservoir simulator can history match well approximately 180 ft in the principal fracture direction, with a uni-
performance adequately. The numerical modeling describes the form fracture conductivity of 4 md-ft and shale-matrix permeability
fracture network as a discrete set of high-permeability fractures in of 100 nanodarcies (0.0001 md). The fracture network is larger on
two orthogonal directions. One direction represents the principal the southwest side of the wellbore, with only a few microseismic
hydraulic-fracture direction as a result of newly created fractures events to the northeast.
or the opening of natural fractures. The orthogonal set represents Fig. 11 shows the simulation setup for a 3,000-ft-long horizon-
the opening of conjugate natural-fracture sets. Besides geochemical tal well, from which an SRV of 5,000 × 106 ft3 with equidistant
shale properties and pore pressure, the key fracture-network proper- fracture spacing of 400 ft was generated. The plot clearly shows
ties that affect well performance include SRV, shale-matrix perme- that the ultimate drainage area is constrained to the created SRV in
ability, fracture spacing, and fracture conductivity. It was discussed supertight shale reservoirs. Using this approach, a series of simula-
in these papers that once the value for shale permeability is fixed, tions was generated with different SRV sizes and fracture spacings
the solutions for total cumulative fracture length (the sum of all to evaluate the effect of these properties on well performance. The
fracture segments in the network) and conductivity become fairly SRV was varied from approximately 800 × 106 ft3 to 6,000 × 106
unique. Fig. 10 shows the results from a vertical-well history match ft3 and equidistant fracture spacing from 50 to 800 ft.
(Mayerhofer et al. 2006), where the network was approximated with To cross check the generic simulations with actual well-production
an effectively producing network having almost 8,000 ft of total data and a more-detailed production history match, the vertical-well
k (matrix) = 0.0001 md
2,450 ft
3,000 ft
Treatment Horizontal Well Treatment Horizontal Well
Fig. 11—Simulation of horizontal-well fracture network (SRV = 5,000 × 106 ft3, frac spacing = 400 ft, lateral length approximately
3,000 ft).
example in Fig. 10 was plotted against the more-generic simulation from 400 to 100 ft would result in a 3.5-fold increase in production.
results. Although the well configuration is different and flowing The larger the SRV, the more significant the fold of increase in
pressures are not exactly the same, the correlation between SRV, production with smaller fracture spacing (because of much longer
well performance, and fracture spacing is approximately consistent, total fracture length with larger SRV). On the other hand, generat-
showing that the 6-month cumulative production of this vertical ing a larger SRV with large 800-ft fracture spacing will not provide
well would fall on the 200-ft fracture-spacing curve (Fig. 12). increased production, which can be explained by the low fracture
Fig. 13 shows a plot of the same generic simulations with actual conductivity (and associated increasing pressure drop with distance
horizontal-well Barnett-shale data throughout the producing areas from the well) in a sparsely fractured network.
in north Texas. The figure shows a wide scatter of well results, After comparing short-term production, the discussion will now
with cumulative gas production ranging from less than 100 to 500 focus on longer-term production (3 to 15 years). Fig. 14 shows a
MMscf for SRVs ranging from 100 × 106 ft3 to 3,000 × 106 ft3. The plot of the SRV from the same generic horizontal-well simulations
data indicate that most of the wells plot at a fracture spacing of 200 vs. 3-year cumulative gas production, fracture spacing, and actual
ft or larger, with many wells at fracture spacings larger than 800 ft, horizontal-well Barnett-shale data. The data indicate a slight shift
indicating a sparse fracture network (smaller total fracture length) of the real data to smaller fracture spacings, which may indicate
within the measured SRV (Fig. 7) and therefore a less effective improved cleanup for the longer-term production. The actual cumu-
fracture network. The plot also shows the significant acceleration lative production ranges from as low as 100 to 2,000 MMscf. The
component of smaller fracture spacings on short-term production. simulations also show that the folds of increase in production as a
For example, at an SRV of 1,000 × 106 ft3, reducing fracture spacing function of smaller fracture spacing become less for longer-term
Vertical Well from Mayerhofer et al. 2006 falls on 200 ft frac spacing curve
Fig. 12—Simulated SRV and actual 6-month vertical well production response (Mayerhofer et al. 2008).
Fig. 13—Simulated SRV and simulated 6-month cumulative production vs. actual Barnett-shale data (entire north Texas area).
recovery, as previously illustrated in Fig. 6 (early-time accelera- fracture spacing. For example, increasing SRV from 1,500 × 106
tion). The plot also clearly shows that significant improvements in ft3 to 3,000 × 106 ft3 with 800-ft fracture spacing will create no
well performance by generating larger SRVs is more significant at meaningful increase in gas production.
low to moderate SRVs (less than 1,000 × 106 ft3). For larger SRVs, Fig. 15 shows the results for 15-year cumulative produc-
the benefits can be achieved only with a simultaneous decrease of tion. In this case, no actual production data is available, because
1600
3-Year Cumulative Gas (MMscf)
1200
800
400
0
0 500 1000 1500 2000 2500 3000
SRV (Million cuft)
Network spacing=800 ft Network spacing=400 ft Network spacing=200 ft
Network spacing=100 ft Network spacing=50 ft Actual data
Fig. 14—Simulated SRV and simulated 3-year cumulative production vs. actual Barnett-shale data (entire north Texas area).
Inputs:
Lateral length, ft: 3000
14000
Reservoir permeability, md: 0.0001 Largest observed SRV per well
Pore pressure, psi 3000
Net pay thickness, ft: 300
15-Year Cumulative Gas (MMscf)
10000
8000
6000
4000
2000
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000
SRV (Million cuft)
Network spacing=800 ft Network spacing=400 ft Network spacing=200 ft
Network spacing=100 ft Network spacing=50 ft
horizontal-well development started in 2002 in the Barnett shale. Applying SRV and Network Azimuth to
The plot shows that for longer-term production, the impact of SRV Well-Placement and -Spacing Strategies
becomes more significant even at larger fracture spacing because Figs. 16 and 17 show conceptual diagrams illustrating how SRV
SRV provides the ultimate drainage area and diffusion has had and fracture-network azimuth can be used for well-placement and
time to access the entire network. In this case, increasing SRV -spacing considerations, respectively. Fig. 16 shows the importance
from 1,500 × 106 ft3 to 3,000 x106 ft3 with 800-ft fracture spacing of lateral orientation with respect to fracture-network azimuth and
will result in a 50% cumulative-production increase from 2,000 to SRV. The graph assumes that the lateral will be drilled in a north-
3,000 MMscf (compared to no meaningful increase after 3 years west/southwest orientation. If fracture direction follows the same
of production). southwest-to-northwest trend, the scenario on the left side of Fig. 16
Fig. 16—Importance of SRV and network azimuth for lateral orientation and well placement.
SRV 2
SRV 2 Well 3
Well 3
SRV 1
SRV 1 Well 2
Well 2
Well 1
Well 1
Fig. 17—Horizontal-well-placement strategy considerations in conjunction with SRV and fracture spacing in shale reservoirs.
will promote longitudinal fractures with a very small SRV. If wells critical to understand how they will impact well results, comple-
are to be drilled in this orientation, more wells with closer spacing tions and well-development strategies.
will be required to drain a given section of reservoir. The fracture 3. Engineering measures to increase SRV and fracture spacing
orientation provided in the center of Fig. 16 provides an oblique ori- include lateral length and orientation, treatment sizes, number
entation of the fracture direction with respect to the lateral and also of stages, perforation clusters, and diversion techniques and/or
results in less-efficient reservoir coverage, but SRV is now larger openhole packer completion systems. More perforation clusters
than in the longitudinal case. The best scenario with largest SRV and stages in cemented-and-cased completions increase the
is illustrated on the right side of Fig. 16, with transverse network likelihood of dense fracturing.
orientation providing the largest SRV per lateral. 4. The key completion strategy is to balance the creation of a large
Keeping these considerations in mind, one could postulate that SRV with maximum possible fracture density. Large SRVs with
well spacing (distance between laterals) should roughly coincide small fracture spacing provide maximum well performance
with the extent of the SRV perpendicular to the wellbore, as illus- and gas-recovery factors. However, the practical and physical
trated in Fig. 17 on the left-hand side (SRV from each lateral is limitations of achieving this goal will require an economic
adjacent to the offset well). This approach may appear to be the optimization process.
optimum configuration. While the areal coverage of SRV is defi-
nitely optimized with this approach, gas-recovery factors within Nomenclature
the SRV can be optimum only with sufficiently close fracture spac- hf = fracture height, L
ing, as illustrated in Fig. 6. If fracture spacing cannot be optimized xf = hydraulic-fracture wing or half-length, L
with large SRVs, it may be more beneficial to drill laterals at closer
xn = hydraulic-fracture network width (from microseismic-
spacing, with the chance of overlapping SRVs (right-hand side in
event pattern), L
Fig. 17). This is the reason for the so-called “simul-fracs” and “zip-
per fracs” in shale reservoirs, where multiple parallel wellbores are ⌬xs = orthogonal fracture spacing, L
either stimulated simultaneously or in short alternating sequence L ftotal = total composite fracture length
from one lateral to the next. The hypothesis in these strategies
is to maximize fracture density between laterals as a result of Acknowledgments
stress diversion from simultaneously growing opposing fractures The authors would like to thank Pinnacle (a Halliburton Service)
(simul-fracs) and/or creating new branches as a result of increased and Carbo Ceramics for supporting the publication of this work.
stresses around newly opened fractures. Real data from Barnett- Some of this work is also based on previous publications of Devon
shale completions may suggest that this approach has merits, but Energy Corporation data, and we thank them for their support of
no specific study has been published on this topic. The challenge those efforts.
lies in understanding the practical and physical limitations of
what is possible in terms of network size and fracture spacing at References
reasonable completion costs (i.e., lateral length, number of stages Albright, J.N. and Pearson, C.F. 1982. Acoustic Emissions as a Tool for
and perforations, and treatment size). Hydraulic Fracture Location: Experience at the Fenton Hill Hot Dry Rock
Site. SPE J. 22 (4): 523–530. SPE-9509-PA. doi: 10.2118/9509-PA.
Cipolla, C.L., Warpinski, N.R., Mayerhofer, M.J., Lolon, E.P., and Vincent,
Conclusions M.C. 2008. The Relationship between Fracture Complexity, Reservoir
1. Low-permeability shale reservoirs require a large fracture Properties, and Fracture Treatment Design. Paper SPE 115769 pre-
network with small fracture spacing and adequate conductivity sented at the SPE Annual Technical Conference and Exhibition, Denver,
to maximize well performance. The size of the SRV, fracture 21–24 September. doi: 10.2118/115769-MS.
spacing, and azimuth are key components for well-placement Fisher, M.K., Heinze, J.R., Harris, C.D., Davidson, B.M., Wright, C.A.,
and -spacing strategies. and Dunn, K.P. 2004. Optimizing Horizontal Completion Techniques
2. Natural features controlling SRV and fracture spacing in shales in the Barnett Shale Using Microseismic Fracture Mapping. Paper SPE
include shale thickness, stress field and magnitude, presence of 90051 presented at the SPE Annual Technical Conference and Exhibi-
pre-existing open or healed natural fractures, fracture barriers, tion, Houston, 26–29 September. doi: 10.2118/90051-MS.
rock brittleness, and geologic features such as faults and karsts. Fisher, M.K., Wright, C.A., Davidson, B.M., Goodwin, A.K., Fielder,
While these parameters are not under direct human control, it is E.O., Buckler, W.S., and Steinsberger, N.P. 2002. Integrating Fracture
n ′−1
∂p 1 q q
= −␣0 2 n′+1 , . . . . . . . . . . . . . . . . . . . . . . . . (1)
∂s w H fl H fl
flow in the fracture network and the elastic deformation of the ∫ Q (t ) d t = ∫ Hw d s + ∫ ∫q L d t d s , . . . . . . . . . . . . . . . . . (3)
fractures, which has similar assumptions and governing equations 0 0 0 0
to those of conventional pseudo-3D fracture models. But instead where Q(t) is pump rate, L(t) is the total length of all fracture
of solving the problem for a single planar fracture, UFM solves branches at time t, and H(s, t) is fracture height. Additionally, at
the equations for the complex fracture network. Fracture-height any junction of fracture branches, the sum of flow rates into the
growth is modeled in the same manner as in conventional pseudo- junction from all branches must be equal to zero. At the wellbore,
3D models. A three-layer proppant-transport model, consisting of the sum of flow rates into all open perforation intervals must be
a proppant bank at the bottom, a slurry layer in the middle, and equal to the prescribed injection rate Q(t); i.e.,
clean fluid at the top, is adopted for simulating proppant transport
in the fracture network. Transport equations are solved for each ∑ q (t ) = Q(t ),
i i = 1,.. N perf , . . . . . . . . . . . . . . . . . . . . . . . . (4)
component of the fluids and proppants pumped. A key difference i
between UFM and the conventional planar-fracture model is being where qi(t) is the injection rate into each perforation cluster (i =
able to simulate the interaction of hydraulic fractures with pre- 1,..Nperf) and is determined as a part of the solution.
existing natural fractures (i.e., determining whether a hydraulic Eqs. 1 and 2 provide the expressions for solving pressure and
fracture propagates through or is arrested by a natural fracture flow distribution in the fracture network for a given fracture width
when they intersect and whether it subsequently propagates along distribution. On the other hand, fracture-width is related to fluid
the natural fracture). The branching of the hydraulic fracture at the pressure through the elasticity equation. To have a computationally
intersection with the natural fracture gives rise to the development efficient model for general engineering design applications, the
of a nonplanar, complex fracture pattern. A crossing model that is same assumption of 2D plane-strain solution for fracture width
extended from the Renshaw and Pollard (1995) interface crossing as in conventional pseudo-3D models is adopted. For a vertical
criterion, applicable to any intersection angle, has been developed fracture in a layered medium with variable minimum horizontal
(Gu and Weng 2010) and validated against the experimental data stress σh(x, y, z) and fluid pressure p, the width profile can be
(Gu et al. 2011), and it is integrated in the UFM. In addition determined from an analytical solution given as (refer to Eq. 9
to the hydraulic-fracture/natural-fracture interactions, the UFM below for detailed form of solution)
also takes into account the interaction among adjacent hydraulic-
fracture branches by computing the stress-shadow effect on each w( x , y, z ) = w( p( x , y), H , z ). . . . . . . . . . . . . . . . . . . . . . . . . . (5)
fracture by the adjacent fractures.
In the limiting case of a fracture in a zone of uniform in-situ
Governing Equations. To simulate the propagation of a complex stress and constant height, the width/pressure relation in Eq. 5
fracture network that consists of many intersecting fractures, the reduces to that of a 2D PKN model (here E' is the plane-strain
equations express the underlying physics of the fracturing process Young’s modulus):
must be satisfied. These include the equation expressing the fluid
flow in the fracture network, the equation expressing the fracture H( p − n )
deformation, and the fracture-propagation criterion. In the most w= . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (6)
2E '
general sense, hydraulic fractures are 3D in nature, and the fracture
planes may not be vertical. This is especially the case when the At fracture tips, the following boundary conditions are
initial natural fractures are not vertical, potentially leading to non- satisfied:
vertical hydraulic fractures when the fracturing fluid opens up the
natural fractures. However, solving a fully 3D nonplanar fracture p = n , W = 0, q = 0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (7)
problem is extremely computation intensive and not suitable for
engineering design models. In the model presented in this paper, Eqs. 1 through 7 represent the complete set of governing
we make the assumption that all natural and hydraulic fractures equations for the problem and must be solved at each timestep to
are vertical. Therefore, the model is applicable only to a formation find fracture opening, fluid pressure, and flow rate in the entire
with vertical or near-vertical natural fractures. Fig. 1 schematically fracture network.
⎡ ⎛ h − 2hi ⎞ ⎤
⎢ z⎜ ⎟⎠ + hi ⎥ where xx and yy are far-field effective stresses (tensile is positive)
− ⎝ h
4 n −1 ⎢(hi − z ) cosh 1
⎥ acting parallel with and perpendicular to a propagating hydraulic
+ ∑
E ′ i =1
( i +1 − i ) ⎢ z − hi ⎥, fracture correspondingly, T0 is rock tensile strength, and kfric is a
⎢ − ⎥ coefficient of friction at the interface. This criterion was validated
⎢ + z (hh − z ) arccos ⎜⎛ h 2 hi ⎞ ⎥
⎝ h ⎠ ⎥⎦ ⎟ by laboratory experiments.
⎢⎣
Natural fractures are often not aligned with the contemporary
. . . . . . . . . . . . . . . . . . . . . . . . (9) principal in-situ-stress directions, in which case the intersection
angle of a hydraulic fracture approaching a natural fracture is
where KIu and KIl are the stress intensity-factors at the upper and between 0 and 90° (angle  in Fig. 2). Therefore, it is important
lower tips, respectively; w(z) is the width at depth z; n and i are to extend the Renshaw and Pollard criterion to nonorthogonal
the in-situ stresses at the top tip and the ith layer, respectively; h intersections.
is the fracture height; and hi represents the height from the lower The Renshaw and Pollard model recently has been extended
tip to the top of the ith layer. to nonorthogonal intersections by Gu and Weng (2010). Because
This model, called the equilibrium height model, is extended the model is based on an analytical solution, it is efficient in com-
further to a nonequilibrium height model, in which the pressure putation and suitable for incorporation in the hydraulic-fracture-
gradient because of the fluid flow in the tip regions in the verti- network model described in this paper.
cal direction is taken into account. This adds an apparent fracture For hydraulic-fracture applications, the ratio of the remote
toughness proportional to the fracture’s top and bottom velocities in-situ stresses (H to h) is larger than unity. The coefficient of
in the equations above. The resulting equations are solved numeri- friction for many rocks is typically between 0.1 and 0.9. Fig. 3
cally to determine the tip velocities. shows the results from the extended criterion for a number of
1
βy 0 0.2 0.4 0.6 0.8 1
fracture x Coefficient of friction
Dsj
∂ ( H − H bank ) wc fl ,k
+
(
∂ q fl c fl ,k )=−f c fl ,k , . . . . . . . . . . (13)
∂t ∂x
Elem j leakoff
where qfl is the flow rate (m3/s) and fleakoff is the flux of fluid through
σsi Elem i
the fracture walls (m2/s) divided by the fracture height above the
bank. The equation is similar in the case of proppant species,
σni
except that there is no proppant leakoff into the matrix. For the
calculation of slurry height, a similar transport equation is used to
calculate the volumetric concentration of the slurry. The numerical
Fig. 4—2D displacement discontinuity method (DDM) approach: implementation of these transport equations uses an explicit
influence of displacement discontinuities from jth element on scheme in time and a total variation diminishing method for the
stress of ith element. advection term (LeVeque 1992).
where vset,k is the settling velocity for proppant type k, prop,k is its
Extend all tips by density, and Dk is its diameter. K' and n' are the averaged flow con-
ΔLi based on velocity sistency index and averaged flow behavior index, respectively, both
weighted by the fluid concentration. A more accurate calculation
of the settling velocity has also been implemented, on the basis of
the correlation from Shiller and Naumann (1933), to calculate the
Apply junction/ drag-force coefficient on a spherical particle.
crossing rules The modeling of erosion of the bank is based on the correlation
from Wang et al. (2003), which describes the minimal height of
clean fluid and slurry at given flow conditions, under which the
proppant does not settle and the bank is eroded.
Solve coupled flow
Numerical Solution. At any given timestep, the created hydraulic-
and elasticity
Eqs. for p, q, w
fracture network is represented by connected fracture elements.
New elements are added as fracture tips advance. Many new
Iterate elements can be added at all propagating fracture tips in a single
timestep, and new element numbers are assigned to these new
elements. As a result, the element numbers are not continuous
Update Δt based from one element to the next. Therefore, a special code structure
on mass balance is established to track the neighbors, and special junction elements
at intersection of multibranches are also introduced.
The combined Poiseuille and continuity equations (Eqs. 1
and 2) are discretized in terms of pressure pi and average width
Apply proppant transport wi defined at the center of each element and flow rate qij defined
at the boundary between Element i and Element j. The resulting
mass-balance equation is written as
Update fracture height
d Volfluid (i, t ) = d Volfrac (i, t ) + d Volleakoff (i, t ), . . . . . . . . . . . . (15)
or, in the form of pressure, width, and flow rate at each element,
Compute stress shadow
h ( k , i ) ⎡ hocc ( k , i )li t − 0 ( k, i ) ⎤
t t
N zones
⎧
+ 2 ∑ ⎨2C L ( k , i ) L ⎢ ⎥
k =1 ⎩ h( k , i ) ⎢ − hocc ( k , i )li
t − dt t − d t
t − 0 ( k , i ) ⎥⎦
old
⎣
In the vertical direction, the model for proppant considers two hL ( k , i ) t ⎫
+ S p (k, i) ⎡ h ( k , i )lit − h t − d t ( k , i )lit − d t ⎤⎦ ⎬ . . . . . . . . . . . (16)
h( k , i ) ⎣
competing mechanisms: the proppant settling into the bank and the
erosion of the existing bank. The settling is calculated by first cal- ⎭
culating the settling velocity for each proppant and then calculating
the increase of the bank height on the basis of both the proppant Eq. 16 includes the leakoff coefficient CL (k, i) for each Zone k
concentration and the height of the slurry. The settling velocity for covered by Fracture Element i, as well as the height of each zone
each proppant was calculated initially on the basis of the extension h (k, i), leakoff height of each zone hL (k, i), the height of the
t
of the Stokes’s law to power fluids given by Daneshy (1978): occupied part of the zone hocc ( k , i ) at position of Element i, and
33400 1100
33300 900
33200
700
33100
500
33000
UFM
32900 Conventional bi-wing 300 UFM
32800 Conventional bi-wing
100
32700
0 2000 4000 6000 8000 –100
0 1000 2000 3000 4000 5000 6000 7000 8000
Time (sec)
Pumping Time (sec)
Fig. 6—Comparison of injection pressure and fracture half-length between UFM and a conventional model for a biwing fracture.
(␣ ⋅ (l + l ) ⋅ ( n′ + 1))
1/ n ′
i j Then (p + δp) is taken as an approximate zero of f, or solution
of f(p) = 0. In this sense, Newton’s method replaces a system of
ij = ( pi − i ) ( ) ( ) ( )
2 n ′+ 2 2 n ′+ 2 2 n ′+ 2 2 n ′+ 2
− p* − i + p* − j − pj − j , nonlinear equations with a system of linear equations that must be
solved repeatedly to converge on an adequate solution. In this way,
pi l j + p j li as it iterates, the solution is updated as
p* = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (17)
li + l j
p kj +1 = p kj + ␦ p kj +1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (20)
where p* is an averaged value of the pressure at the interface
between Elements i and j. Convergence is achieved when some norm associated with
Different forms of flow-rate equation are used and implemented the right-hand side of Eq. 19 (i.e., solution) reaches the specified
in the model for the cases of laminar, turbulent, and Darcy fluid minimum value.
flow. A combination of the above equations leads to a nonlinear After the solution converges, a new timestep starts, and an
system of equations, simplified as f(p) = 0, which is solved in explicit method is used to extend the fracture tips. The algorithm
X-axis
–3200 –2900 –2400 –2000 –1000 –1200 –800 –400 0 400
1400 1400
1200 1200
1000 1000
NF 2
800 800
NF 1
600 600
H perfs
400 400
NF 3 h
200 200
0 Treatment Well 0
Fig. 7—Top view of wellbore, perforation, and natural fractures (NF) (x- and y-coordinate in feet).
X-axis
Width
–2200 –2000 –1800 –1600 –1400 –1200 –1000 –800 –600
0.5
0.45
0.4
0.35
0.3
0.25
0.2 –2000 X-axis
0.15
–1600
0.1 –1200
0.05 –800
0
Figs. 8a and b—Fracture-geometry evolution for Example No. 1 case 1 (x- and y-coordinate in feet).
first finds all the hydraulic-fracture tips that satisfy the propagation A special tip element is implemented, and the pressure at the
criterion (stress-intensity factor is greater than rock toughness). For propagating tip element is prescribed by using asymptotic PKN tip
each propagating tip element, the flow rate into the tip is calculated solution as a function of the element’s length and velocity.
on the basis of the Poiseuille equation (Eq. 1) to find the corre- Fig. 5 shows the general numerical scheme at each timestep.
sponding tip velocity (equal to fluid velocity near the tip):
Model Validation and Examples
vtip = qtip / (h fl w). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (21) Comparison With Conventional Model for a Biwing Fracture.
To check the UFM validity, it is compared with the limiting case
The tip with the highest velocity is extended by a prescribed of biwing fractures. Fig. 6 shows the comparison of UFM results
maximum length lfix. The other tips are extended proportionally to shown as a red curve with a conventional hydraulic-fracture model
their velocities: shown as a blue curve (Gulrajani et al. 1997), for the input param-
eters as given in Table 1. The predicted injection pressure and
ltip = lfix vtip / vtip
max
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (22) fracture length from the two models are identical.
X-axis
Width
–2200 –2000 –1800 –1600 –1400 –1200 –1000 –800 –600
0.5
0.45
0.4
0.35
0.3
0.25
0.2 –2000 X-axis
0.15
–1600
0.1 –1200
0.05 –800
0
Figs. 8c and d—Fracture-geometry evolution for Example No. 1 case 1 (x- and y-coordinate in feet).
Example No. 1: Simple Natural Fractures. A major advantage only a few natural fractures is given for different values of these
of the UFM is to be able to predict the impact of natural fractures parameters and different intersection angles. Fig. 7 shows the top
and in-situ stresses on the created hydraulic-fracture network. The view of the wellbore and a single perforation cluster where the
factors that influence fracture complexity include natural-frac- fracture is initiated and the prescribed natural fractures. The input
ture size, orientation, and distribution; natural-fracture mechanical parameters for the example are given in Table 2.
properties (interfacial friction coefficient, cohesion, and fracture Case 1: Friction Coefficient = 0.5, Stress Anisotropy = 200
toughness); stress anisotropy (defined as difference between maxi- psi, Intersection Angle = 90°. In this example case, the property of
mum and minimum horizontal stresses); and rock tensile strength. NF 3 is chosen so that a hydraulic fracture cannot propagate along
As shown in Fig. 4, whether a hydraulic fracture crosses a natural NF 3 easily (a large toughness is used). This biases the fracture
fracture when they intersect depends largely on the interfacial propagation to the right side of the wellbore, so the fracture geom-
friction coefficient and stress anisotropy. Therefore, these two etry does not become very complex. Fig. 8 shows the simulation
parameters play a critical role in fracture complexity. fracture geometry at different times, containing top views of both
To illustrate the importance of the natural friction coefficient the hydraulic fracture and natural fractures and 3D views with
and stress anisotropy, a simple example involving a rock containing color contour of width for the hydraulic fracture only.
Fig. 8e—Fracture-geometry evolution for Example No. 1 case 1 (x- and y-coordinate in feet).
Even though this is a simple case, it illustrates some impor- and a net pressure of approximately 230 psi, while the width in NF
tant behaviors of complex fracture. As shown in Fig. 8a, a single 1 remains small because of a larger confining stress. The fractures
fracture is initially created at the perforation cluster. Because that branch off the ends of NF 1 also have a small width because
of the low fluid viscosity, the fracture width is relatively small, of low fluid viscosity and net pressure. The fracture branches that
approximately 0.05 in. When the fracture tips reach NF 1 and NF are directly opposing the main fracture have a smaller width com-
3, the friction coefficient and stress anisotropy are such that the pared with their opposite wings as a result of fracture-interference
hydraulic fracture is not able to cross the natural fractures. In the (stress-shadow) effect (Fig. 8d). These fractures then intersect NF 2
ensuing injection, the fracture length is not growing, but its width and NF 3 and are stopped by these natural fractures. As before, the
balloons up to 0.4 in. (Fig. 8b), while the net pressure increases pressure in these fracture branches starts to build up, and the width
from 40 to approximately 230 psi. balloons until the net pressure inside these fractures exceeds the
When the pressure is sufficiently large to exceed the maximum maximum horizontal stress to allow NF 2 to open up (Fig. 8e).
horizontal stress acting on the orthogonal NF 1, NF 1 is hydrauli- This example shows that when a hydraulic-fracture branch
cally opened. The fracture propagates quickly along NF 1 and intersects a natural fracture and cannot cross, the propagating
reaches the ends of the fracture, where the fracture branches off and branch is stopped and some other branches can continue to grow.
propagates in the preferred fracture direction again (Fig. 8c). The When all propagating tips are stopped by natural fractures, the net
orthogonal fracture NF 1 acts as a choke for the fluid in the main pressure inside the fracture network will have to rise to exceed the
fracture connected to the perforation and maintains a large width stress anisotropy. That is, the net pressure in the fracture system
is regulated by stress anisotropy. This can lead to much greater
1000 fracture width than predicted by the conventional biwing-fracture
1 2 model, or even the wiremesh model, in which straight fracture
planes is assumed. The net pressure response typically shows
a rapid rise in early time and then flattens after it exceeds the
maximum horizontal stress (Fig. 9). The high net pressure resulting
from width restriction and large pressure drop through the natural
Net Pressure Fracture 1 (psi)
0.12
0.11
0.1
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0
Fig. 10—Fracture geometry for Example No. 1, Case 2 (large anisotropy, t = 9 minutes, x- and y-coordinate in feet).
fractures from crossing, resulting in complex fracture rather than statistical parameters given in Table 3. Other input parameters are
biwing fracture as in Case 2 for an orthogonal intersection. Because the same as those given in Table 2. Stress anisotropy is assumed to
of the greater normal stress acting on the natural fractures (375 psi), be 200 psi. The pumping schedule is given in Table 4.
the fluid net pressure has to rise above 380 psi before the fluid pres- Fig. 13 shows the top view of the statistically generated natural
sure can open up NF 1. As a result, the fracture width in the main fractures and the predicted hydraulic-fracture network. Fig. 14
fracture is approximately 0.6 in., much wider than in the previous shows the fracture width contour and area proppant concentration
cases because of the greater net pressure. in 3D view. It shows that while a large hydraulic-fracture network
is created, proppant occupies only a small fraction of the fracture
Example No. 2: More-Complex Natural Fractures. In this exam- surface area created. This is mainly because of fast proppant
ple, more-complex natural fractures are generated on the basis of settling as a result of the low-viscosity fluid used, but it is also
X-axis Width
–2000 –1600 –1200 –800
0.7
0.65
0.6
0.55
0.5
0.45
0.4
0.35 –2400
0.3 –2000
0.25 X-axis
0.2 –1600
0.15 –1200
0.1
0.05 –800
0
Fig. 11—Fracture geometry for Example No. 1, Case 3 (large anisotropy, 60° angle, x- and y-coordinate in feet).
Conclusions
The new complex-hydraulic-fracture model presented in this paper
Net Pressure Fracture 1 (psi)
Stage Name Pump Rate (bpm) Fluid Fluid Vol (gal) Proppant Prop Conc. (ppga) Proppant Mass (lbs)
2000
1800
1600
1400
1200
1000
800
600
400
200
–200
–400
–600
–800
Fig. 13—Natural fractures and created-hydraulic-fracture geometry for Example No. 2 (x- and y-coordinate in feet).
References Cipolla, C.L., Weng, X., Mack, M.G., et al. 2011. Integrating Microseismic
Adachi, J., Seibrits, E., Peirce, A., and Desroches, J. 2007. Computer simulation Mapping and Complex Fracture Modeling to Characterize Hydraulic
of hydraulic fractures. Int. J. Rock Mech. Min. Sci. & Geomech. Abstracts Fracture Complexity. Paper SPE 140185 presented at the SPE Hydrau-
44 (5): 739–757. http://dx.doi.org/10.1016/j.ijrmms.2006.11.006. lic Fracturing Technology Conference, The Woodlands, Texas, USA,
Blanton, T.L. 1982. An Experimental Study of Interaction Between Hydrau- 24–26 January. http://dx.doi.org/10.2118/140185-MS.
lically Induced and Pre-Existing Fractures. Paper SPE 10847 presented Cipolla, C.L., Williams, M.J., Weng, X., Mack, M.G., and Maxwell,
at the SPE Unconventional Gas Recovery Symposium, Pittsburgh, S.C. 2010. Hydraulic Fracture Monitoring to Reservoir Simulation:
Pennsylvania, USA, 16–18 May. http://dx.doi.org/10.2118/10847-MS. Maximizing Value. Paper SPE 133877 presented at the SPE Annual
Width Proppant–Areal_Distribution
0.5 5.5
0.45 5
0.4 4.5
0.35 400 ft 4
3.5
0.3
0.25 3
0.2 2.5
0.15 2
1.5
0.1 1
0.05 0.5
0 0
Fig. 14—Fracture width and proppant area concentration for Example No. 2.