Computers and Geotechnics 91 (2017) 5870

Contents lists available at ScienceDirect

Computers and Geotechnics

journal homepage: www.elsevier.com/locate/compgeo

Research paper

Modeling hydraulic fracture propagation using cohesive zone model

equipped with frictional contact capability
Y. Li a,b, J.G. Deng a,b, W. Liu a,b,, Y. Feng c
a
State Key Laboratory of Petroleum Resource & Engineering, China University of Petroleum (Beijing), Beijing 102249, China
b
College of Petroleum Engineering, China University of Petroleum (Beijing), Beijing 102249, China
c
Department of Petroleum and Geosystems Engineering, University of Texas at Austin, Austin, TX 78712, USA

a r t i c l e i n f o a b s t r a c t

Article history: We present a new pore pressure cohesive element for modeling the propagation of hydraulically induced
Received 18 January 2017 fracture. The Park-Paulino-Roesler cohesive zone model has been employed to characterize the fracturing
Received in revised form 1 July 2017 behavior. Coulombs frictional contact model has been incorporated into the element to model the pos-
Accepted 2 July 2017
sible shear reactivation of pre-existing natural fractures. The developed element has been validated
through a series of single-element tests and an available analytical solution. Furthermore, intersection
behaviors between the hydraulic fracture and the natural fracture under various conditions have been
Keywords:
predicted using the present element, which shows good agreement with experimental results.
Hydraulic fracturing
Cohesive zone model
2017 Elsevier Ltd. All rights reserved.
Natural fracture
Frictional contact

1. Introduction
Hydraulic fracturing has been widely used in the petroleum
industry to extract hydrocarbons from reservoirs with ultra-low
permeability which were previously economically unviable. Dur-
ing hydraulic stimulation, pressurized fluid is injected into the rock
formations to create fractures and/or reactivate pre-existing natu-
ral fractures [1]. Other applications of hydraulic fracturing include
the measurement of in situ stress [2,3], preconditioning ore bodies
for caving [4], and stimulation of geothermal reservoirs [5,6]. With
the rapid development of unconventional resources, the initiation
and propagation of hydraulic fractures in tight rocks have been
extensively investigated from an analytical perspective [2,711]
and a numerical point of view [1217]. However, modeling of
hydraulic fracturing remains a challenging task due to the involved
complicated coupled processes [13,14]: (i) rock deformation
caused by the fluid pressure on the fracture faces, (ii) flow of frac-
turing fluid within the fracture, (iii) fracture propagation, and (iv)
leak-off of fluid from the fracture into the formation. In addition,
friction between fracture surfaces also plays a significant role in
shear reactivation of natural fractures. Generally, the rock defor-
mation can be described by the theory of elasticity which relates
the fracture width with the fluid pressure in the fracture. The fluid
Corresponding author at: College of Petroleum Engineering, China University of
Petroleum (Beijing), Beijing 102249, China.
E-mail address: liuwei@cup.edu.cn (W. Liu).
flow in the fracture can be modeled by the cubic law which relates
the flow rate in the fracture to the fracture width and the fluid
pressure gradient. Usually, linear elastic fracture mechanics (LEFM)
theory (i.e. fracture propagation occurs if the stress intensity factor
equals to the fracture toughness) is used as a criterion for fracture
propagation.
Analytical solutions of fluid-driven fracturing played an impor-
tant role in the design of fracturing treatments in the early time.
KGD [18,19] and PKN [20,21] models are the two well-known frac-
turing models, which can predict the behaviors of plane strain bi-
wing fractures with constant height. In recent years, the scaling
law and asymptotic framework [7,8,22,23] have been used to
understand the different propagation regimes of hydraulic frac-
tures. Analytical methods mentioned above are widely used and
adequate for engineering purposes in the early time. However,
their applications are limited to simple situations such as constant
injection rate, simple fracture geometries, and linearly elastic and
homogeneous medium. For hydraulic fracturing treatments in
unconventional reservoirs featuring low-permeability and pre-
existing natural fractures, complex fracture networks may be
created which cannot be adequately described by the simple ana-
lytical methods.
In order to investigate the hydraulic fracturing in unconven-
tional reservoirs with complex fracture geometry and boundary
conditions, numerical methods are a viable alternative since the
complex practical conditions can be conveniently considered in a
numerical model. Various numerical methods have been proposed

http://dx.doi.org/10.1016/j.compgeo.2017.07.001
0266-352X/ 2017 Elsevier Ltd. All rights reserved.
 Y. Li et al. / Computers and Geotechnics 91 (2017) 5870
for the simulation of hydraulic fracturing. These methods can be
roughly classified into two categories: (1) continuum-based meth-
ods and (2) discontinuum-based methods.
A representative in the continuum-based methods is the dis-
placement discontinuity method (DDM). Kresse [24], Zhang
[25,26] and Wu [27,28] used DDM to simulate the propagation of
hydraulic fracture and its interaction with pre-existing natural
fractures. McClure [29] used DDM to investigate the diagnostic
fracture injection tests with complex fracture networks. Later,
DDM was extended to three dimensions by Wu [30] and McClure
[31]. In the context of finite element methods, Carrier [32], Chen
[14], and Li [33] used pore pressure cohesive element to model
the fluid-driven fracture. Dahi-Talegahani [16] and Chen [34] sim-
ulated hydraulic fracturing using extended finite element method
(XFEM). The above mentioned DDM and XFEM can simulate the
propagation of hydraulic fracture with arbitrary paths. In addition
to the conventional continuum approaches summarized above,
some new continuum approaches have been introduced in recent
year. Wang [35] used the RFPA software to investigate the shear
stimulation in naturally fractured reservoirs. Miehe [36], Wick
[37] and Liu [38] used a Phase-Field model to simulate the fluid-
driven fractures and the interaction between multiple fractures.
Ouchi [39] developed a Peridynamics model to simulate the prop-
agation of hydraulic fractures in heterogeneous, and naturally frac-
tured reservoirs. However, until now, large-scale simulation is still
a problem for these new methods.
For typical discontinuum-based methods, continuum domain is
discretized into a series of separate blocks or particles and fracture
propagation is characterized by the separation of these blocks or
particles. The discontinuum-based approach is represented by
the universal distinct element code (UDEC) and 3 dimensional dis-
tinct element code (3DEC). Nagel [40] used the UDEC software to
model the shear reactivation of pre-existing natural fractures
around the hydraulic fracture. The results show that changes of
stress field due to hydraulic fracture propagation have a dramatic
influence on the shear reactivations of the pre-existing natural
fractures, which, in turn, significantly affect the growth of the
hydraulic fracture. Farzin [41] investigated the initiation and prop-
agation of hydraulically induced fractures by 3DEC. Another major
type of discontinuum-based method is discontinuous deformation
analysis (DDA). Choo [42] and Morgan [43] present a new hydrau-
lic fracturing model based on DDA. In addition to the continuum-
and discontinuum-based methods, hybrid method, such as the
finite-discrete element method (FDEM) is also an important
method for hydraulic fracturing simulation [4446].
In general, the continuum-based methods have some difficul-
ties in modeling the intersection between hydraulic fracture and
natural fracture while most of the discontinuum methods need a
small time increment due to the explicit integration scheme.
Therefore, we try to develop a new model, which is similar to the
discontinuum-based methods but uses an implicit integration
scheme to obtain a relatively larger time increment.
In this paper, we developed a 2D, fully coupled finite element
model to simulate the propagation of hydraulic fractures in tight
rocks. Two types of elements are involved in the finite element
mesh, i.e. bulk elements and pore-pressure cohesive zone (PPCZ)
elements. Zero-thickness PPCZ elements are inserted into any
two neighboring triangular elements. The propagation, branching,
merging, and intersection of fractures can be captured by the PPCZ
elements and the deformation of surrounding matrix is model by
bulk elements. For the reason that the bulk element has been
extensively implemented in many commercial or open source
FEM codes, we restrict our attention to the development and
implementation of the PPCZ element in this study. The mechanical
behaviors of the PPCZ element are characterized by Park-Paulino-
Roesler (PPR) cohesive zone model proposed by Park [47].
59
Interfacial friction plays an important role in the shear reactivation
of pre-existing natural fractures, which is one of the main purposes
of fracturing treatment in unconventional reservoirs [29,35,48].
However, the original PPR model can be only used to simulate
the fracturing behaviors of frictionless material. To simulate the
shear reactivation of pre-existing natural fractures, Coulombs fric-
tion law is also incorporated into the cohesive element.
In Section 2, governing equations of the PPCZ element are pro-
vided. Detailed finite element formulations for this new element
are discussed in Section 3. In Section 4, our model is validated
through a series of single-element tests and an available analytical
solution. Furthermore, the intersection behaviors between the
hydraulic fracture and the natural fracture have been predicted
by the PPCZ element and compared to the existing experimental
results. In Section 5, some advantages and limitations of this new
method are discussed briefly. Finally, some conclusions are given
in Section 6.
2. Governing equations
2.1. Fluid flow in the fracture
The fluid flow in the fracture (Fig. 1) is controlled by the mass
conservation equation:
@ qw @ qq
0 1
@t @s
where w is the fracture width; q is the fluid density; q is the volume
flow rate; and s is the coordinate along the fracture.
For a Newtonian fluid, the flow inside the fracture can be
described by the cubic law derived from the Poiseuille equation
between two parallel plates:
ws; t3 @p
qs; t  2
12l @s
where l is the fluid viscosity.
Assuming the fluid is incompressible, combining Eqs. (1) and
(2), the following lubrication equation can be obtained:


@w @ w3 @p
 0 3
@t @s 12l @s
The above equations can be solved with appropriate initial and
boundary conditions. Usually, the conditions at the crack mouth
and tip of the fracture can be expressed as:
qs 0; t q0 t 4
ws l; t 0; qs l; t 0 5
where l is the fracture length. In fact, when the first condition in Eq.
(5) is satisfied, the second one is fulfilled naturally.
2.2. Fracture propagation
Fracture propagation is assumed to be governed by a cohesive
zone model. This model was first proposed by Barenblatt [49] to
overcome some limitations of the LEFM. It avoids the stress singu-
larity at the fracture tip and characterizes the nonlinear fracture
process zone ahead of the tip. Generally, this process zone may
experience three stages: elastic, softening, and complete failure.
The cohesive zone model has been successfully applied to simulate
fracture propagation in various materials from brittle to ductile,
such as rocks, cement, and metals [14,15,32,34,5057].
The behaviors of a cohesive element are governed by a traction-
separation law. In this study, the PPR potential-based cohesive law
60 Y. Li et al. / Computers and Geotechnics 91 (2017) 5870

n
h/t /n i b
Ct /t /t /n 14
n
If the modes I and II energies are the same, the energy constants
are expressed as
 a m
Cn /n 15
m

n
b
Ct 16
n
The normal and tangential final crack opening width can be
expressed as

/n a  a m1
dn akn 1  kn a1 1 kn 1 17
rmax m m


n1
/t b b
Fig. 1. Fluid flow in a hydraulic fracture. dt bkt 1  kt b1 1 kt 1 18
smax n n
Then the propagation of fracture can be described by the cohe-
is used to characterize the fracture behaviors. The potential func- sive zone model presented above. Note that, in the PPR potential
tion of the cohesive law is defined as follows [47]: cohesive zone model, only /n , /t , rmax , smax , a, b, kn and kt are inde-
pendent variables. Refer to [47,58,59] for more details about this
wDn ; Dt min/n ; /t model.


 
Dn a m Dn m
Cn 1  h/n  /t i
dn a dn 2.3. Interface friction
"
b
n #
jDt j n jDt j
 Ct 1  h/t  /n i 6 Shear reactivation of pre-existing natural fractures plays an
dt b dt important role in creating a complex fracture network, which is
where w is the potential function of the cohesive zone model; /n , /t crucial for the successful development of unconventional reser-
are the fracture energies in the normal and tangential direction, voirs. The friction between the fracture surfaces should be consid-
respectively; Cn , Ct are the energy constants in the PPR model; Dn ered in fracture modeling. Although many types of cohesive zone
and Dt are the separations in normal and tangential direction, models are able to model fracture initiation and propagation, most
respectively; dn , dt are the normal and tangential final openings; of them neglect the friction between the sliding surfaces. The cohe-
a, b are the shape function parameters in PPR model; the symbol sive tractions obtained from the potential-based models represent
monotonic separation phenomena. If the frictional sliding along
h i is theMacauley bracket, i.e.
 the fracture surfaces is considered, an additional constitutive rela-
0; x 6 0 tionship should be introduced into the PPR potential-based cohe-
hxi 7
x; x > 0 sive zone model.
In this study, a Coulomb friction model is used for characteriz-
the exponents m and n are associated with the initial slope and can
ing the static and sliding frictions between the fracture surfaces, as
be expressed as:
shown in Fig. 2. Cohesive elements will slide if the shear stress
aa  1k2n bb  1k2t exceeds the sliding friction. To simplify the problem, it is assumed
m ; n 8 that the friction coefficient remains the same for static and sliding
1  ak2n 1  bk2t
frictions. Thus, for the sliding elements, the friction traction T f can
where kn and kt are initial slope indicators and determined by the be given as:
ratio of the critical fracture opening to the final fracture opening.
The partial derivatives of the PPR potential function lead to the
normal and tangential cohesive tractions as: Tf
@wDn ; Dt
T n Dn ; Dt 9 f Tn
@ Dn

@wDn ; Dt
T t Dn ; Dt 10 Kt
@ Dt
t (e) t( p) 1
The normal and shear cohesive strengths of the PPR model can
be expressed as:
t
rmax T n Dn ; Dt max 11

smax T t Dn ; Dt max 12
When the modes I and II fracture energies are different, the
f Tn
energy constants Cn and Ct are given by
h/n /t i  a m
Cn /n /n /t 13 Fig. 2. Coulomb friction constitutive model.
m
 Y. Li et al. / Computers and Geotechnics 91 (2017) 5870 61

T f lf jT n j 19 Using this iteration process, the tangential friction traction at

where lf is the friction coefficient and jT n j is the effective normal
stress acting on the surfaces. For the elements that are not
sliding, the magnitude of the friction stress T f is in the range from
zero to lf jT n j depending on the external shear stress acting on the
surfaces.
The tangential separation of the cohesive element is needed to
be calculated with the friction stress at the same time. Generally, it
cannot be expressed explicitly and thus an iterative technique is
required. The implementation of this technique is analogous to
the Mohr-Coulomb plasticity model. At the beginning of the
i 1th time step, the total tangential separation increment is
assumed as elastic strain, and then the trial tangential friction
traction can be calculated as:
stri1 si K t Ddti1
@T t Dn ; Dt 0
Kt
@ Dt
where si is the tangential friction traction at the end of the previous
time step; stri1 is the trial tangential friction traction; Ddt is the
total tangential separation increment; K t is the artificial tangential
stiffness.
The frictional slip criterion governing the frictional behavior
between surfaces can be defined as:
f stri1  lf jT n j
If f 6 0, there is no plastic strain in the i 1th time step and
the tangential friction traction st is calculated from the Eq. (20)
directly.
If f > 0, plastic strain exists in the i 1th time step. The
associative flow rule [60] is adopted to compute the plastic
strain, i.e. frictional sliding, as
@f
Ddtp Dk Dk
@s
the end of the i 1th time step can be obtained.
3. Formulations and numerical implementation
3.1. Pore pressure cohesive element
The cohesive zone model can be easily implemented into the
classic finite element framework. To model hydraulic fracture
propagation, a new zero-thickness PPCZ element based on PPR
cohesive zone model is developed in this paper. As shown in
Fig. 3, a PPCZ element contains 6 nodes. Each node of 14 has
two displacement degrees of freedom, and nodes 5 and 6 only have
the pore pressure degree of freedom. The fluid pressure and flow
rate in the fracture are discretized on the mid-plane nodes 5 and
20 6. The fracture width is calculated from the displacement differ-
ence between the upper and lower surfaces. Displacement and
fluid pressure are assumed varying linearly over the cohesive ele-
21 ment. Hence, the separations of element Dx and fluid pressure
pf x at any position can be determined from the nodal values
using linear interpolation functions.
i1 Since the fracture widths are calculated using the local nodal
displacements, the global coordinate X of a cohesive element is
first transformed to a local coordinate x:
x RX 28
where R is the coordinate rotation matrix, and is given as:
22  
cos h sin h
R 29
 sin h cos h
where h is the angle between the local and global coordinate
system.

is transformed to the local
The global nodal displacement u
~:
nodal displacement u
~ Tu
u
30
23
where u
u1 ; u2 ; . . . ; u7 ; u8 ; T is the displacement transformation
T

where Dk is the consistency factor. matrix, defined as:

The total increment of tangential separation Ddt can be split 2 3
R 0 0 0
into two parts, i.e. an elastic part Ddte and a plastic part Ddtp : 60 R 0 07
6 7
T6 7 31
Ddt Ddte Ddtp 24 40 0 R 05
Substituting Eq. (23) into Eq. (24) gives the elastic strain as:
0 0 0 R

i1 i1 According to the local nodal displacements, the local nodal sep-

Ddte Dd t  Dki1 25
aration can be obtained as:
Then, the following return mapping algorithm (Newton itera- ~ Lu
D ~ 32
tions) is used for calculating the plastic part mentioned above
and the real friction stress si1 : where L is the local displacement-separation relation matrix,
defined as:
(1) The Taylor expansion of f 0 can be written as:

@f
f dDk    0 26
@ Dk
Neglecting the high-order terms result in:

dDk f =dt 27
i
where dt @@fDk ji K t .
(2) The initial value for Dki is 0;
i i
(3) Dki1 Dki dDki Dki  f =dt
i1
(4) stri1
s i
K t Dut  Dki1
i i1
(5) If f > tol, then Dk Dk and return to (3)
(6) If f 6 tol, then si1 stri1 Fig. 3. A typical configuration of PPCZ element.
62 Y. Li et al. / Computers and Geotechnics 91 (2017) 5870

2 3
1 0 0 0 0 0 1 0 The weak form of the equation governing the behavior of fluid
6 7 flow in the PPCZ element can be obtained from Eqs. (3)(5) as
6 0 1 0 0 0 0 0 17
L6
6
7
7 33 follows:
4 0 0 1 0 1 0 0 05 Z 


@w @ w3 @pf w3 @pf
0 0 0 1 0 1 0 0 dpf  dx  dpf q0 j
Cc @t @x 12l @x 12l @x x0
Then, the separation fields can be determined as:

w3 @pf
dpf 0 j 0 43
e
Dx NTw D 34 12l @x xle

where Nw is the interpolation function, defined as: Eq. (43) can be rewritten to:
2 le x 3 Z Z
@w w3 @pf @dpf
0 dpf dx dx dpf q0 jx0 44
6
le
7 Cc @t Cc 12l @x @x
6 0 le x 7
Nw 6
6 x
le 7
7 35 The discrete form of Eq. (44) is given as:
4 le 0 5

Z 
Z 
0 x w3
le Np NT1 Bc dx Du
Dt Mp MTp dx p DtNp q0 jx0
Cc Cc 12l
where le is the length of fracture, x 2 0; le  is the location of the |{z}
I2
point of interest in the local coordinate.
The relationship between local separation field and global nodal 45
displacement can be obtained by substituting Eqs. (30) and (32)
u
u

, u

 is the nodal displacement vector at the end
where Du
into Eq. (34):
of the previous time step.


Dx Bc u 36 Combining Eqs. (42) and (45), the stiffness K and internal force
f int of this new PPCZ element can be obtained:
where Bc is the global nodal displacement-separation relation
" @I1 @I1
#
matrix, Bc NTw LT.

@u @p
Using Eq. (36), the width of the fracture can be expressed as: K @I2 @I2
46
@u
@p
 T
0
wx NT1 Dx Dx 37  
1 I1
f int 47
I2
The fluid pressure is considered as traction acting on the inter-
nal surfaces of failure PPCZ element, which can be given by:

pf x NTp p 38 4. Benchmarks and results

where p p1 ; p2 T is the vector of fluid pressure; Np is the fluid It is essential to test the mechanical and hydraulic behaviors of
pressure interpolation vector, defined as: the new PPCZ element before using it. In this section, the compu-
" # tational implementation of the element is first verified using
le x
Np le
39 single-element tests. Later, the accuracy of the element is validated
x
le against analytical solutions and laboratory experiments. In the
single-element tests, pure mode-I, pure mode-II, and mixed
The partial derivative of Np with respect to the local coordinate mode-II/friction tests are performed. Next, the simulation results
x is given as: of the KGD problem are compared with the asymptotic solutions
@Np provided by Detourney [8]. Finally, simulation results of interac-
Mp 40 tion between hydraulic facture and natural fracture are compared
@x
with the results of laboratory experiments provided by Gu [61].

3.2. Finite element formulation

4.1. Singe-element tests
The weak form of the equation governing the mechanical
The geometries of the single-element tests are shown in Fig. 4. A
behavior of the PPCZ element is obtained from the principle of
PPCZ element is assembled with a bulk element (1 m  1 m). The
virtual work. The total virtual work done by the internal forces
(summation of cohesive and friction traction Tc , and fluid
pressure p) on the internal element surfaces Cc is equal to the
work done by the external forces Text on external element sur-
faces C:
Z Z Z
dD  Tc dS  dwxpf xdS du  Text dS 41
Cc Cc C

where dD, dwx, and du are the virtual separation, virtual width,
and virtual displacement, respectively.
The first-order backward difference scheme is used for time
integration, and the discrete form of Eq. (41) can be given as:

Z 
Z  Z
BTc Tc dS  BTc N1 NTp dS p NTw Text dS 42
Cc Cc C
|{z} Fig. 4. Geometries of (a) pure tensile test, (b) pure shear test, and (c) mixed mode-
I1 II/friction test.
 Y. Li et al. / Computers and Geotechnics 91 (2017) 5870 63

Table 1
Input parameters of the PPR cohesive zone model for the single-element tests.

/n (J/m2) /t (J/m2) rmax (MPa) smax (MPa) a b kn kt lf

400 800 4 8 4 4 0.1 0.1 0.6

two lower nodes of the PPCZ element are fixed. Additionally, the
pore pressures of the mid-plane nodes are fixed at zero because
this test focus on the mechanical behavior of the element and
the hydraulic behavior of the element will be validated later. The
Youngs modulus and Poissons ratio of the bulk element are
10,000 GPa and 0, respectively. The relevant parameters of the
cohesive element are shown in Table 1.
For the pure mode-I test, the bulk element is stretched at the
top with a displacement of 0.3 mm. The stress-displacement rela-
tionship is plotted in Fig. 5(a). The stress initially increases to the
tensile strength of the element, and then decreases to zero. For
the pure mode-II test, a displacement of 0.3 mm is applied to the
right side of the bulk element. Similar to the pure mode-I test, as
shown in Fig. 5(b), the shear stress increases to the shear strength
of the element, and then decreases to zero. Next, a mixed mode-II/
friction test is performed to examine the frictional behavior of the
PPCZ element. First, a normal stress of 5 MPa is applied on the top
of the bulk element; next, the bulk element is stretched to the right
by 0.3 mm; and then it is displaced to the left by 0.6 mm; finally,
the element is stretched back to the initial position. Fig. 5(c) shows
the changes of friction stress and total tangential stress with shear
displacement during the entire simulation. The friction stress
equals to the product of the normal stress and the friction coeffi-
cient, and the total tangential stress equals to the summation of
the cohesive strength and friction stress. These results indicate that
the computational implementation of the PPR cohesive zone model
is correct.
4.2. KGD problem
In this section, the PPCZ element is validated by comparing its
results with available analytical solutions for hydraulic fractures
under viscosity-dominated and toughness-dominated conditions.
A bi-wing fracture is created by injecting an incompressible New-
tonian fluid into rock (Fig. 6). The rock is assumed to be imperme-
able and under a plane strain condition. The fluid injection rate is
Q 0 . Fracture half-length lt, fracture aperture wx; t, and net
pressure of the fracture px; t will be determined, where x is the

Fig. 6. Geometry of the KGD problem.

Table 2
Input parameters for the KGD problem.

Youngs modulus, GPa E 10

Poissons ratio v 0.1
Mode-I fracture energy, J/m2 /n 10 (viscosity, Km 0:304.)
1000 (toughness, Km 5:409.)
2
Mode-II fracture energy, J/m 0
Tensile cohesive strength, MPa rmax 0.04 (viscosity)
8 (toughness)
Shear cohesive strength, MPa smax 1.5
Shape function parameter a 3
Shape function parameter b 3
Initial slope indicator kn 0.001 (viscosity)
0.05 (toughness)
Initial slope indicator kt 0.01
Friction coefficient lf 0.6
Injection rate, m2/s Q0 0.001
Viscosity, Pas l 0.01 (viscosity)
0.001 (toughness)
Fig. 5. Computational results of (a) pure mode-I test, (b) pure mode-II test, and (c)
Minimum principal stress, MPa rh 2
Injection time, s t 10
mixed mode-II/friction test.
64 Y. Li et al. / Computers and Geotechnics 91 (2017) 5870

distance along the fracture and t is injection time. The analytical where n x=lt is the scaled coordinate 0 6 n 6 1; et is a small
solutions of this problem are provided by Detourney [8] as func- dimensionless parameter; Lt is length scale; c, X, and P are
tions of the injection rate Q 0 and three material parameters E0 , dimensionless fracture length, aperture, and net pressure,
K 0 , and l0 , respectively.

1=2 In viscosity-dominated propagation regime, the aforemen-
E 2 tioned dimensionless parameters can be expressed as follows:
E0 ; K 0
4 K IC ; l0 12l; 48
1  v2 p !1=6
l0 1=3 E0 Q 30
where K IC is the fracture toughness, E is the Young modulus, v is the et 0 ; Lt t 2=3 50
Poisson ratio, and l is the fluid viscosity.
Et l0
The solutions provided by Detourney [8] can be expressed as:
cm0 0:616 51
lt Ltct; wx; t etltXn; t; px; t etE0 Pn; t
49
m0 n A0 1  n2
X
2=3 1
A1 1  n2
5=8

" q p#
10 m 1 1  1  n2
B 4 1  n 2n ln 2 2
p 52
1 1  n2
2m



1 1 2 1 1 10 1 7 1
Initial P1
m0 B ; A02 F 1  ; 1; ; n2 A1 2 F 1  ;1; ; n2
3p 2 3 6 2 7 6 2
crack
B1 2  pjnj
Q0 53
2 80 m 3.2 m 1
where A0 31=2 ; A1 0:156; B1 0:0663; B is the Euler beta
function; and 2 F 1 is a hyper-geometric function.
Cohesive In toughness-dominated propagation regime, the above param-
element eters are defined as:
!1=3
2=3
K 04 E0 Q 0 t
et ; Lt 54
E04 Q 0 t K0
(a) (b)
2
ck0 55
Fig. 7. Geometry and mesh for the numerical model of KGD problem. p2=3

Fig. 8. Comparisons between the numerical and the analytical solution of the KGD problem. (a) Net pressure at the injection point; (b) fracture aperture at the injection point;
(c) net pressure along the fracture length at the end of injection (t = 10 s); (d) fracture aperture along the fracture length at the end of injection (t = 10 s).
 Y. Li et al. / Computers and Geotechnics 91 (2017) 5870 65

k0 n p1=3  1=2 elements to the far field. There are 40,000 linear quadrilateral ele-
X 1  n2 56
2 ments and 200 zero-thickness PPCZ elements in the model. The
same mesh is used for both cases of viscosity-dominated regime
p1=3 and toughness-dominated regime. A total duration of 10 s of injec-
Pk0 57 tion is simulated.
8
Fig. 8 shows the comparisons between the numerical results
Dimensionless toughness K m can be used to estimate the frac-
and the asymptotic solutions. The temporal evolutions of the net
ture propagation regimes (i.e. viscosity-dominated or toughness-
dominated), which is defined as [8]
Table 3

1=2  1=4
2 K IC 1  v 2 E Input parameters for simulating Gus experiments.
Km 4 58
p E 12lQ 0 1  v 2 Rock parameters
Youngs Modulus, GPa E 10
The hydraulic fracture propagation regime is toughness- Poissons ratio v 0.1
dominated when K m is larger than 4.0, and viscosity-dominated Friction coefficient lf 0.615
when K m is smaller than 1.0. Under the plane strain condition, Tensile strength, MPa rmax 4
the relation between fracturing toughness K IC and fracturing Shear strength, MPa smax 20
Tensile fracturing energy, J/m2 /n 100
energy /n is given as: Shear fracturing energy, J/m2 /t 4500
r Shape function parameter a 3
/n E Shape function parameter b 3
K IC 59
1  v2 Initial slope indicator kn 0.1
Initial slope indicator kt 0.1
The input parameters used for simulations are listed in Table 2.
Natural fracture parameters
The dimensionless toughness K m of viscosity-dominated and Friction coefficient lf 0.615
toughness-dominated regimes are about 0.304 and 5.409, Tensile strength, MPa rmax 0.4
respectively. Shear strength, MPa smax 2
The geometry of the numerical model is shown in Fig. 7(a). The Tensile fracturing energy, J/m2 /n 10
Shear fracturing energy, J/m2 /t 450
size of the half-model is 10  80 m. Symmetric displacement
Shape function parameter a 3
boundary condition is imposed to the left edge. The normal dis- Shape function parameter b 3
placements of the right, top, and bottom edges are fixed. To initial- Initial slope indicator kn 0.05
ize the simulation, a very small hydraulic aperture (0.5 mm) is Initial slope indicator kt 0.05
assigned to two PPCZ elements near the injection point at the mid- Fluid parameters
Pump rate, m2/s Q0 3  108
point of left edge to allow fluid injection. Part of the mesh near the Viscosity, Pas l 1.0
injection point is shown in Fig. 7(b). The model is discretized with Injection time, s t 150
finer elements near the injection point and progressively larger

Fig. 9. (a) Geometry model and (b) mesh topology of the numerical simulation model.
66 Y. Li et al. / Computers and Geotechnics 91 (2017) 5870

pressure and fracture aperture at the injection point are shown in
Fig. 8(a) and (b), respectively. The numerical results of time history
plots agree very well with the asymptotic solutions for both the
viscosity-dominated and toughness-dominated conditions. Fig. 8
(c) and (d) plot the net pressure and fracture aperture profiles at
the end of the simulation (t = 10 s). Far from the near tip region,
the profiles match very well with the asymptotic solutions. In
the near tip region, the viscosity-dominated case exhibits a slightly
lower net pressure and aperture than the asymptotic solution. This
is because that the numerical method and the asymptotic solutions

Fig. 10. Fracture geometries and contours of the magnitude of the displacement in direction x (in meters).
 Y. Li et al. / Computers and Geotechnics 91 (2017) 5870
use different propagation criteria. On the basis of these results it
can be concluded that the PPCZ element developed in this study
is able to accurately capture the viscosity-dominated and the
toughness-dominated solutions.
4.3. Interaction between hydraulic fracture (HF) and natural fracture
(NF)
The new element is also used to simulate the experiments of
interaction between HF and NF in sandstone by Gu [61]. In exper-
iments, six Colton sandstone samples with a tensile strength of
588 psi (4.054 MPa) were used. The size of the sandstone blocks
is 11  11  15 inches (28  28  38 cm). The friction coefficient
of the interface is 0.615, and the cohesion of the interface is nearly
zero. Silicon oil with a kinematic viscosity of 1000 cSt (correspond-
ing dynamic viscosity is 1.0 Pas) was injected into the wellbore
drilled in the blocks at a constant rate of 30 mL/min.
To simulate Gus experiment using the PPCZ elements, the typ-
ical geometry and mesh scheme are shown in Fig. 9. The model is
under plane-strain condition with a size of 30 cm  30 cm. PPCZ
elements are inserted into any two neighboring triangular bulk
elements as shown in Fig. 9(b). Constant stress boundaries are
imposed on the outer boundaries: surface pressure rh is imposed
on the left and right surfaces, and surface pressure rH is imposed
on the top and bottom surfaces. Due to some parameters required
for the numerical simulation were not provided explicitly in Gus
article [61], for better comparison between the numerical and
experimental results, the input parameters used in the numerical
model are chosen as reasonable as possible. The input parameters
for the modeling are reported in Table 3.
Six cases are simulated with different stress contrasts and inter-
section angles. Fig. 10 shows the fracture profiles and contours of
Table 4
Comparison between the numerical and experimental results.
Case Angle () rH (MPa)
a 45 8.27
b 45 17.24
c 75 8.27
d 75 17.24
e 90 7.58
f 90 13.79
Fig. 11. (a) Distribution of natural fractures; (b) finite element meshes.
67
the magnitude of displacement in direction x. First, we compare
the simulation results of two cases ((a) and (b)) with same inter-
section angle 45. Case (a) and case (b) have the same rh .
(6.895 MPa) but different rH (8.27 MPa and 17.24 MPa, respec-
tively). Fig. 10(a) and (b) shows the simulation results of case (a)
and (b). The domains are colored with the magnitudes of the dis-
placement in direction x with a color bar that is depicted on the
bottom-left of each plot (in meters). It can be observed that the
HF propagates along the direction of the maximum stress in the
early time and only deflects into the right wing of the NF when
the HF hits the NF. This is because the opening of the vertical HF
tends to push the left-wing of NF close while helps to pull the
right-wing of NF open. However, because of the different stress
contrasts, there are significant differences between the deforma-
tion behavior of NFs in case (a) and (b). It seems that the NF is acti-
vated in pure tension (mode I) in case (a) while is activated in pure
shear (mode II) in case (b). This difference is to be expect because
they have different normal and shear stress acting on the surfaces
of NFs.
Next, according to the experiments, we consider the cases
shown in Fig. 10(c) and (d) with same intersection angle 75. The
stress conditions of case (c) and (d) are same with the case (a)
and (b), respectively. In Fig. 10(c), as with the case (a), the HF
grows vertically until it hits the natural fracture, and then diverts
into the NF. However, the zoomed-in view of the near-
intersection region in case (c) shows that the left-wing of natural
fracture is also activated because the aforementioned push effect
acting on the left-wing of NF decreases as the intersection angle
increases. In Fig. 10(d), the HF grows vertically in the early time,
and then directly crosses the NF without diversion because of the
larger interfacial friction along the natural fracture induced by
larger rH and intersection angle.
rh (MPa) Experimental results Numerical results
6.895 No crossing No crossing
6.895 No crossing No crossing
6.895 No crossing No crossing
6.895 Crossing Crossing
6.895 No crossing No crossing
6.895 Crossing Crossing
68 Y. Li et al. / Computers and Geotechnics 91 (2017) 5870
Fig. 12. Effect of friction coefficient on the evolution of fracture network. (a) friction coefficient = 0.4; (b) friction coefficient = 0.6; (c) friction coefficient = 0.8; (d) friction
coefficient = 1.0.
The last two simulations have been carried out for the cases
(e and f) with an orthogonal intersection angle. According to the
experiments, in Fig. 10(e), we provide the simulation result of case
subjected to a small stress contrast condition (rH 7:58 MPa,
rh 6:895 MPa). The HF propagates vertically initially, and then
the first element of left-wing and right-wing of NF are activated
at time t = 105 s, as shown in the zoomed-view of Fig. 10(e). At
later times, the upper branch of the HF terminates at the intersec-
tion point and only the lower branch of the HF grows as the injec-
tion continues. According to the experiments, rH is increased to
13.79 MPa in case (f). In this case, as shown in Fig. 10(f), the HF
crosses the NF directly as if the NF does not exist.
The comparisons between the numerical and experimental
results are shown in Table 4. The comparisons show there is a good
agreement in the intersection behaviors of the fractures between
the numerical modeling and the experimental observation. These
results suggest that the PPCZ element can reasonably capture the
intersections between HF and NF during the hydraulic fracturing
treatments.
5. Effect of friction coefficient on the evolution of fracture
network
Friction coefficient, which plays a key role in the shear failure of
NF, may differ considerably for NFs with different roughness and/
or fillings. To investigate the effect of friction coefficient of NFs
on the evolution of fracture network during hydraulic fracturing
treatment, we conducted a sensitivity analysis that considered 4
different NF friction coefficients (0.4/0.6/0.8/1.0). The geometry
of the model and the distribution of NFs in the domain are shown
in Fig. 11(a). The angle between NFs and y-axis are 75. The size of
the model is 40  40 m, and injection point is positioned at the
center of the domain. The domain was discretized into triangular
elements, as shown in Fig. 11(b). Zero-displacement and no-flow
conditions were imposed on the outer boundaries. The magnitudes
of minimum and maximum principal stresses are 10 and 14 MPa,
respectively. The minimum and maximum principal stresses are
along the x-axis and y-axis, respectively. The injection rate is
0.0002 m2/s, and the viscosity of fluid is 1.0 mPas. The other input
parameters are assumed to be the same as those listed in Table 3. A
total duration of 200 s of injection is simulated.
Fracture patterns for 4 cases with different NF friction coeffi-
cients are shown in Fig. 12. The bold gray lines represent the
pre-existing NF elements, the blue lines represent the activated
HF elements, and the red1 lines represent the activated NF elements.
In Fig. 12(a), HF propagates vertically in the early time, and then
deflects into NFs, and then turns back to the direction of maximum
1
For interpretation of color in Fig. 12, the reader is referred to the web version of
this article.
 Y. Li et al. / Computers and Geotechnics 91 (2017) 5870
principal stress after the coming out from the NFs. In Fig. 12(b), as
with the case (a), one can clearly see that the type of most intersec-
tion of HF and NF can be classified as turning crossing-type. Unlike
the case (b), most intersection in case (c) and all intersection in case
(d) are direct crossing type, as shown in Fig. 12(c) and (d). These
results indicate that HF is more likely to directly cross NFs than to
deflect into and propagate along the NFs as the friction coefficient
of NFs increase, which implies that the low friction coefficient of
NF tends to increase complexity of fracture geometry.
6. Discussions
Numerical simulation of the hydraulic fracture propagation is a
challenging task because of multiple coupled physical mecha-
nisms, strong nonlinear behaviors of the coupling between fracture
propagation and fluid flow, and the existences of the natural frac-
tures and other discontinuous structures. As stated in Section 1,
many numerical methods have been developed to reproduce the
process of hydraulic fracture propagation. Each of those methods
has its own strengths and weaknesses. The PPCZ element proposed
in this paper is no exception. In this section, the advantages and
limitations of the PPCZ element compared with other methods
are discussed briefly.
The main advantage of the PPCZ element is the ability to simu-
late the branching and intersection of the fractures as a natural
outcome of the original boundary value problem. Some other
methods, such as XFEM/GFEM and DDM, are very complicated in
dealing with fracture intersection even in 2D. They need some spe-
cial criteria (usually obtained by combining theoretical and exper-
imental approaches) and sophisticated algorithms, which hinder
the application of these methods to simulate the propagation of
hydraulic fractures in naturally fractured reservoir. Furthermore,
although not demonstrated in this paper, heterogeneity and plas-
ticity of the rock matrix can be easily incorporated into the model
if necessary. Secondly, the PPCZ element can be easily imple-
mented into many existing finite element codes (in this study,
PPCZ element is implemented in the commercial FEM software
ABAQUS). Compared with the built-in cohesive element in ABA-
QUS, PPCZ element can take into account the friction between frac-
ture surfaces, which plays a key role in the reactivation of pre-
existing natural fractures. Thirdly, an implicit time integration
scheme is employed to solve the fully coupled equations of rock
deformation, fracture propagation, and fluid flow within the frac-
ture. The implicit time integration scheme is unconditionally
stable and allows significantly larger time step than methods using
the explicit time integration scheme, which is adopted by many
discontinuum-based approaches. Finally, PPCZ method is similar
to the discrete element methods, such as UDEC and DDA, but the
extensive contact search, which is an intensive computation oper-
ation in the discrete element method, is not required.
However, the PPCZ method is sensitive to the topology of the
element meshes due to the assumption that a fracture can only
propagate along the element edges. Thus, some strategies, such
as unstructured meshes and statistically distributed material prop-
erties, have been used to reduce the influence of meshing on frac-
ture trajectories [62,63]. In addition, the poroelastic effect and fluid
leakoff have not been considered in the current model, so it is not
suitable for simulating fracture propagation in highly permeable
rocks.
7. Conclusions
A new PPCZ element based on the finite element method is
developed and verified for modeling hydraulic fracture propaga-
tion. Coulombs friction law is incorporated into the PPCZ element
69
to describe the interfacial friction. The discrete governing equa-
tions for the PPCZ element are derived. The element is imple-
mented into the commercial finite element software ABAQUS/
Standard and validated through a series of single-element tests
and available analytical solutions of the viscosity-dominated and
toughness-dominated KGD problem. Further simulation results
show that the PPCZ element is able to capture the interaction
behaviors of hydraulic fracture and natural fracture and is effective
for simulating the hydraulic fracture propagation in naturally frac-
tured reservoirs. It is recommended to use the PPCZ element to
simulate hydraulic fracture propagation in impermeable or low-
permeable reservoirs because the poroelastic effect and fluid leak-
off have not been taken into account.
Future work could include extending the PPCZ element to 3D
and considering complex fluid flow models, shearing-induced dila-
tion of the fractures, and proppant transport in the fractures.
Acknowledgements
This work is financially supported by National Natural Science
Found (No. 11502304), and National Science and Technology Major
Project (2016ZX05058002-006), and China University of Petro-
leum, Beijing (No. 2462013YJRC023, No. C201601).
References
[1] Weng X, Kresse O, Cohen C-E, Wu R, Gu H. Modeling of hydraulic-fracture-
network propagation in a naturally fractured formation. SPE Prod Oper
2011;26(04):36880.
[2] Lakirouhani A, Detournay E, Bunger A. A reassessment of in-situ stress
determination by hydraulic fracturing. Geophys J Int 2016:ggw132.
[3] Feng Y, Gray K. Discussion on field injectivity tests during drilling. Rock Mech
Rock Eng 2016;16.
[4] Bunger AP, Jeffrey RG, Kear J, Zhang X, Morgan M. Experimental investigation
of the interaction among closely spaced hydraulic fractures. American Rock
Mechanics Association; 2011.
[5] Zimmermann G, Moeck I, Blcher G. Cyclic waterfrac stimulation to develop an
Enhanced Geothermal System (EGS)conceptual design and experimental
results. Geothermics 2010;39(1):5969.
[6] Yoon JS, Zimmermann G, Zang A. Discrete element modeling of cyclic rate fluid
injection at multiple locations in naturally fractured reservoirs. Int J Rock Mech
Min Sci 2015;74:1523.
[7] Detournay E. Mechanics of hydraulic fractures. Annu Rev Fluid Mech
2016;48:31139.
[8] Detournay E. Propagation regimes of fluid-driven fractures in impermeable
rocks. Int J Geomech 2004;4(1):3545.
[9] Kusmierczyk P, Mishuris G, Wrobel M. Remarks on application of different
variables for the PKN model of hydrofracturing: various fluid-flow regimes.
Fracture Phenomena in Nature and Technology. Springer; 2014. p. 185213.
[10] Linkov A. Speed equation and its application for solving ill-posed problems of
hydraulic fracturing. Doklady Physics. Springer; 2011. p. 4368.
[11] Wrobel M, Mishuris G. Hydraulic fracture revisited: particle velocity based
simulation. Int J Eng Sci 2015;94:2358.
[12] Hunsweck MJ, Shen Y, Lew AJ. A finite element approach to the simulation of
hydraulic fractures with lag. Int J Numer Anal Meth Geomech 2013;37
(9):9931015.
[13] Adachi J, Siebrits E, Peirce A, Desroches J. Computer simulation of hydraulic
fractures. Int J Rock Mech Min Sci 2007;44(5):73957.
[14] Chen Z. Finite element modelling of viscosity-dominated hydraulic fractures. J
Petrol Sci Eng 2012;88-89:13644.
[15] Chen Z, Bunger AP, Zhang X, Jeffrey RG. Cohesive zone finite element-based
modeling of hydraulic fractures. Acta Mech Solida Sin 2009;22(5):44352.
[16] Dahi Taleghani A, Olson JE. How natural fractures could affect hydraulic-
fracture geometry. SPE J 2014;19(01):16171.
[17] Peirce A, Bunger A. Interference fracturing: non-uniform distributions of
perforation clusters that promote simultaneous growth of multiple hydraulic
fractures. SPE J 2013;20(02):112.
[18] Khristianovic S, Zheltov Y. Formation of vertical fractures by means of highly
viscous fluids. In: Proc 4th world petroleum congress, Rome. p. 57986.
[19] Geertsma J, De Klerk F. A rapid method of predicting width and extent of
hydraulically induced fractures. J Petrol Technol 1969;21(12). 1,571-1,81.
[20] Nordgren R. Propagation of a vertical hydraulic fracture. Soc Petrol Eng J
1972;12(04):30614.
[21] Perkins T, Kern L. Widths of hydraulic fractures. J Petrol Technol 1961;13
(09):93749.
[22] Bunger A, Lakirouhani A, Detournay E. Modelling the effect of injection system
compressibility and viscous fluid flow on hydraulic fracture breakdown
70 Y. Li et al. / Computers and Geotechnics 91 (2017) 5870
pressure. In: Rock stress and earthquakes-proceedings of the 5th international
symposium on in-situ rock stress; 2010. p. 5967.
[23] Bunger A, Jeffrey RG, Zhang X. Constraints on simultaneous growth of
hydraulic fractures from multiple perforation clusters in horizontal wells.
SPE J 2014;19(04):60820.
[24] Kresse O, Weng X, Gu H, Wu R. Numerical modeling of hydraulic fractures
interaction in complex naturally fractured formations. Rock Mech Rock Eng
2013;46(3):55568.
[25] Zhang X, Jeffrey RG. Reinitiation or termination of fluid-driven fractures at
frictional bedding interfaces. J Geophys Res: Solid Earth 2008;113(B8).
[26] Zhang X, Jeffrey RG, Thiercelin M. Deflection and propagation of fluid-driven
fractures at frictional bedding interfaces: a numerical investigation. J Struct
Geol 2007;29(3):396410.
[27] Wu K. Simultaneous multi-frac treatments: fully coupled fluid flow and
fracture mechanics for horizontal wells. In: SPE annual technical conference
and exhibition: society of petroleum engineers; 2013.
[28] Wu K, Olson JE. Investigation of the Impact of Fracture Spacing and Fluid
Properties for Interfering Simultaneously or Sequentially Generated Hydraulic
Fractures. SPE Prod Oper 2013;28(04):42736.
[29] McClure M, Horne R. Characterizing hydraulic fracturing with a tendency-for-
shear-stimulation test. SPE Reservoir Eval Eng 2014;17(02):23343.
[30] Wu K, Olson JE. A simplified three-dimensional displacement discontinuity
method for multiple fracture simulations. Int J Fract 2015;193(2):191204.
[31] McClure M, Babazadeh M, Shiozawa S, Huang J. Fully coupled
hydromechanical simulation of hydraulic fracturing in three-dimensional
discrete fracture networks. In: SPE hydraulic fracturing technology
conference: Society of Petroleum Engineers; 2015.
[32] Carrier B, Granet S. Numerical modeling of hydraulic fracture problem in
permeable medium using cohesive zone model. Eng Fract Mech
2012;79:31228.
[33] Li Y, Deng J, Liu W, Yan W, Feng Y, Cao W, et al. Numerical simulation of
limited-entry multi-cluster fracturing in horizontal well. J Petrol Sci Eng
2017;152:44355.
[34] Chen Z. An ABAQUS implementation of the XFEM for hydraulic fracture
problems. In: ISRM international conference for effective and sustainable
hydraulic fracturing: International Society for Rock Mechanics; 2013.
[35] Wang Y, Li X, Zhou R, Tang C. Numerical evaluation of the shear stimulation
effect in naturally fractured formations. Sci China Earth Sci 2015;113.
[36] Miehe C, Mauthe S, Teichtmeister S. Minimization principles for the coupled
problem of DarcyBiot-type fluid transport in porous media linked to phase
field modeling of fracture. J Mech Phys Solids 2015;82:186217.
[37] Wick T, Singh G, Wheeler MF. Fluid-filled fracture propagation with a phase-
field approach and coupling to a reservoir simulator. SPE J 2015.
[38] Liu G, Li Q, Msekh MA, Zuo Z. Abaqus implementation of monolithic and
staggered schemes for quasi-static and dynamic fracture phase-field model.
Comput Mater Sci 2016;121:3547.
[39] Ouchi H, Katiyar A, Foster J, Sharma MM. A peridynamics model for the
propagation of hydraulic fractures in heterogeneous, naturally fractured
reservoirs. In: SPE hydraulic fracturing technology conference. Society of
Petroleum Engineers; 2015.
[40] Nagel N, Sanchez-Nagel M, Zhang F, Garcia X, Lee B. Coupled numerical
evaluations of the geomechanical interactions between a hydraulic fracture
stimulation and a natural fracture system in shale formations. Rock Mech Rock
Eng 2013;46(3):581609.
[41] Hamidi F, Mortazavi A. A new three dimensional approach to numerically
model hydraulic fracturing process. J Petrol Sci Eng 2014;124:45167.
[42] Choo LQ, Zhao Z, Chen H, Tian Q. Hydraulic fracturing modeling using the
discontinuous deformation analysis (DDA) method. Comput Geotech
2016;76:1222.
[43] Morgan WE, Aral MM. An implicitly coupled hydro-geomechanical model for
hydraulic fracture simulation with the discontinuous deformation analysis. Int
J Rock Mech Min Sci 2015;73:8294.
[44] Yan C, Zheng H. FDEM-flow3D: a 3D hydro-mechanical coupled model
considering the pore seepage of rock matrix for simulating three-
dimensional hydraulic fracturing. Comput Geotech 2017;81:21228.
[45] Yan C, Zheng H, Sun G, Ge X. Combined finite-discrete element method for
simulation of hydraulic fracturing. Rock Mech Rock Eng 2015;49(4):122.
[46] Lisjak A, Kaifosh P, He L, Tatone BSA, Mahabadi OK, Grasselli G. A 2D, fully-
coupled, hydro-mechanical, FDEM formulation for modelling fracturing
processes in discontinuous, porous rock masses. Comput Geotech
2017;81:118.
[47] Park K, Paulino GH, Roesler JR. A unified potential-based cohesive model of
mixed-mode fracture. J Mech Phys Solids 2009;57(6):891908.
[48] Weng X, Sesetty V, Kresse O. Investigation of shear-induced permeability in
unconventional reservoirs. In: 49th US rock mechanics/geomechanics
symposium. American Rock Mechanics Association; 2015.
[49] Barenblatt GI. The mathematical theory of equilibrium cracks in brittle
fracture. In: H.L. Dryden TvKGKFHvdD, Howarth L, editors. Advances in applied
mechanics. Elsevier; 1962. p. 55129.
[50] Chen Z, Jeffrey RG, Zhang X. Numerical modeling of three-dimensional T-
shaped hydraulic fractures in coal seams using a cohesive zone finite element
model. Hydraulic Fract J 2015;2(2):2037.
[51] Dahi Taleghani A, Ahmadi M, Wang W, Olson JE. Thermal reactivation of
microfractures and its potential impact on hydraulic-fracture efficiency. SPE J
2014;19(05):76170.
[52] Alfano G, Sacco E. Combining interface damage and friction in a cohesive-zone
model. Int J Numer Meth Eng 2006;68(5):54282.
[53] Bryant EC, Hwang J, Sharma MM. Arbitrary fracture propagation in
heterogeneous poroelastic formations using a finite volume-based cohesive
zone model. In: SPE hydraulic fracturing technology conference. Society of
Petroleum Engineers; 2015.
[54] Do BC, Liu W, Yang QD, Su XY. Improved cohesive stress integration schemes
for cohesive zone elements. Eng Fract Mech 2013;107(01):1428.
[55] Yao Y. Linear elastic and cohesive fracture analysis to model hydraulic fracture
in brittle and ductile rocks. Rock Mech Rock Eng 2012;45(3):37587.
[56] Snozzi L, Molinari JF. A cohesive element model for mixed mode loading with
frictional contact capability. Int J Numer Meth Eng 2013;93(5):51026.
[57] Chen Z, Jeffrey RG, Zhang X, Kear J. Finite-element simulation of a hydraulic
fracture interacting with a natural fracture. SPE J 2016.
[58] Park K, Paulino GH. Computational implementation of the PPR potential-based
cohesive model in ABAQUS: educational perspective. Eng Fract Mech 2012;93
(01):23962.
[59] Spring DW. Failure processes in soft and quasi-brittle materials with
nonhomogeneous microstructures. University of Illinois at Urbana-
Champaign; 2015.
[60] De Borst R, Crisfield MA, Remmers JJ, Verhoosel CV. Nonlinear finite element
analysis of solids and structures. John Wiley & Sons; 2012.
[61] Gu H, Weng X, Lund JB, Mack MG, Ganguly U, Suarez-Rivera R. Hydraulic
fracture crossing natural fracture at nonorthogonal angles: a criterion and its
validation. SPE Prod Oper 2012;27(01):206.
[62] Zhou F, Molinari J-F. Dynamic crack propagation with cohesive elements: a
methodology to address mesh dependency. Int J Numer Meth Eng 2004;59
(1):124.
[63] Tijssens MGA, Sluys LJ, van der Giessen E. Simulation of fracture of
cementitious composites with explicit modeling of microstructural features.
Eng Fract Mech 2001;68(11):124563.