Comp. Part. Mech.

(2016) 3:229–248
DOI 10.1007/s40571-015-0081-4

Complementary hydro-mechanical coupled finite/discrete element

and microseismic modelling to predict hydraulic fracture
propagation in tight shale reservoirs
Matthew Profit1 · Martin Dutko1 · Jianguo Yu1 ·
Sarah Cole1 · Doug Angus2 · Alan Baird3

Received: 27 March 2015 / Revised: 9 October 2015 / Accepted: 13 October 2015 / Published online: 2 November 2015
© OWZ 2015

Abstract This paper presents a novel approach to predict pared against field microseismicity. A number of hydraulic
the propagation of hydraulic fractures in tight shale reser- fracture numerical examples are presented to illustrate the
voirs. Many hydraulic fracture modelling schemes assume new technology.
that the fracture direction is pre-seeded in the problem
domain discretisation. This is a severe limitation as the Keywords Hydraulic fracture · Finite/discrete element
reservoir often contains large numbers of pre-existing frac- method · Coupled geomechanical · Microseismicity
tures that strongly influence the direction of the propagating
fracture. To circumvent these shortcomings, a new fracture
modelling treatment is proposed where the introduction of 1 Introduction
discrete fracture surfaces is based on new and dynamically
updated geometrical entities rather than the topology of the Hydraulic fracturing is an engineering process often used by
underlying spatial discretisation. Hydraulic fracturing is an the petroleum industry to extract hydrocarbons from very low
inherently coupled engineering problem with interactions porosity reservoir rock which would otherwise be econom-
between fluid flow and fracturing when the stress state of the ically unviable [1–6]. The use of numerical models in the
reservoir rock attains a failure criterion. This work follows a design stage would be an advantageous tool to optimise the
staggered hydro-mechanical coupled finite/discrete element hydrocarbon recovery for a given tight shale reservoir. Stan-
approach to capture the key interplay between fluid pressure dard practice in the industry often assumes equal spacing
and fracture growth. In field practice, the fracture growth between stimulation points [2], and for reservoirs with local
is hidden from the design engineer and microseismicity is variations in stress and material properties, this may be sub-
often used to infer hydraulic fracture lengths and directions. optimal, leading to significantly reduced ultimate recovery
Microseismic output can also be computed from changes of of hydrocarbons.
the effective stress in the geomechanical model and com- A key design variable in hydraulic fracturing is the stimu-
lated reservoir volume (SRV) [7], and its value is a complex
B Matthew Profit function of, amongst others, shale material properties, pump
[email protected] rates of fracking fluids in both clean and proppant laden
Martin Dutko states, the initial stress state and the density of natural
[email protected] fractures in the reservoir. Given the number of interacting
Doug Angus mechanisms, it is no surprise that empirical formulations pro-
[email protected] vide only a limited guide on the relationship between injected
Alan Baird fluid and the resulting SRV.
[email protected] The petroleum industry has focused its efforts in many
1
directions to provide quantitative information between key
Rockfield Software Limited, Swansea SA1 8AS, UK
design variables such as the type and rate of fluid injected and
2 University of Leeds, Leeds, UK the final SRV. These can be broadly broken into four main cat-
3 University of Bristol, Bristol, UK egories: empirical, analytical, semi-analytical and numerical

123
230 Comp. Part. Mech. (2016) 3:229–248

methods. Empirical methods often apply very simple math- does not require topology change. However, fine meshes are
ematical modelling schemes such as curve fitting methods often required near fractured regions. Even then the fracture
based on experiences and post-appraisal data from previous smears local stress fields leading to spurious values. In addi-
reservoir stimulations [8]. Although these techniques pro- tion, the inclusion of a proppant transport model is difficult
vide ‘rule of thumb’ guides to the design engineer, they lack when undertaking a purely continuum FEM approach.
robustness in terms of honouring physical first principles, The DFN geomechanical and continuum flow models are
and so they lack good feedback information once data trends often used within the petroleum industry [14]. They permit
drift from previous responses. Analytical solutions are a step the use of different discretisations for the geomechanical and
in the right direction in that they offer a sound theoretical flow fields (e.g., finite element for the geomechanical field
basis for the observed responses [9]. So, for example, the and finite volume for the flow field). Standard dual porosity
fracture aperture width will be a function of fluid pressure reservoir simulators are used to capture matrix and fracture
based on Sneddon’s elasticity formulation [8]. Avoiding the fluid flow. The hydraulic fracture process, however, strongly
introduction of any non-elastic effects means that the final couples the two main governing fields, and this is poorly
equations can be readily solved without resorting to numeri- represented by these classes of models.
cal techniques such as the finite difference and finite element The DDM uses the elastic boundary element method to
methods (FEMs). Therefore, although these techniques pro- capture the relationship between fracture surface pressure
vide a feedback mechanism, the underlying assumptions, like and the resulting aperture [15–17]. In these models only
elastic behaviour of the rock body or steady-state flow in the the fractures are discretised, and this reduces the size of the
fracture region, might be too restrictive when dealing with a problem which is an attractive benefit in 3D. Furthermore,
more general case such as a tight shale reservoir with numer- in contrast to continuum FEM models, it is possible to cap-
ous sets of natural fractures. The semi-analytical solutions ture the transport of proppant inside the propagating fracture.
build on the analytical solutions by offering a numerical However, these models set restrictions on the material char-
approach on specific parts of the governing equations. An acterisation which are generally limited to homogenous and
example here would be a hybrid technique which uses an isotropic elastic. With these models, it is difficult to accom-
analytical solution on the fracture width versus fluid pres- modate complex 3D geometries and stress fields. In addition,
sure relationship, but solves numerically the transient fluid heterogeneity is only defined via the existence of natural frac-
flow in the fracture region [9]. Numerical methods provide tures.
the most robust schemes by allowing non-linear behaviour to The discrete element approach is able to mimic bond
be represented in both the evolution of mechanical stresses breakage, which occurs during strain softening, via a cohe-
in the rock, plus the fracture fluid flow. This benefit comes sive element [18,20]. It is, however, often assumed that an
at the expense of increased run-time and the necessity of element edge pre-seeds the fracture direction, negating the
an experienced numericist who understands the underly- possibility of truly complex fractures forming which are com-
ing assumptions in the model. With increasing CPU power, monly observed in largely heterogeneous reservoirs.
numerical techniques such as the FEM and discrete ele- XFEM contrasts sharply with the DEM by using a purely
ment methods (DEMs) are now seen as attractive modelling continuum approach and captures crack deformation via dis-
approaches even when complex fracturing is one of the key continuous fields, namely the partition of unity functions [4].
mechanisms in the design process. The method has been coupled to fracture fluid flow models to
In broad terms, the main numerical schemes appropriate simulate the propagation of hydraulic fractures [19]. To pre-
for this class of problem include the FEM [10–12] in its pure dict crack growth, these special functions must track through
continuum form [13], the discrete fracture network (DFN) the mesh, and this can lead to book-keeping difficulties in
geomechanical and continuum flow models [14], the dis- complex 3D fracturing.
placement discontinuity method (DDM) geomechanical and Kolditz and co-workers have implemented object-oriented
flow models [15–17], the combined FEM/DEM [18] tech- coupled thermal-hydro-mechanical (THM) FEM techniques
nique plus the more recent extended FEM (XFEM) [4,19]. to investigate a range of geomechanical problems, including
The pure continuum FEM models rely on a combined cou- the flow of water and heat in fractured porous media [21,22].
pled damage and tensile constitutive model (e.g., a unified The investigations are limited to pre-defined DFNs. The same
Mohr–Coulomb with Rankine cap constitutive model) with author has a monograph on bench-case solutions for THM
a single porosity model to capture porous flow. The frac- processes in fractured porous media [23].
ture propagation is captured via the volume increase through This paper describes the governing equations and imple-
the material model, which is able to represent the pressure mentation of a combined FEM/DEM coupled hydro-
drops observed during fracture [13]. These models readily mechanical method which is able to simulate hydraulic frac-
allow the implementation of sophisticated constitutive mod- ture in tight shale reservoirs. A novel approach based on new
els, and due to their simplistic form, the fracture propagation dynamically updated geometrical identities is undertaken to

123
Comp. Part. Mech. (2016) 3:229–248 231

simulate complex fracture propagation. Section 2 describes stresses generated in the reservoir rock. A drained response
the key governing equation, along with links to important is therefore assumed during hydraulic stimulation. As a con-
flow concepts such as channel flow and leak-off. Section 3 sequence of this assumption, the porous flow in the rock
develops the governing equations into a Galerkin finite formation (i.e., seepage field governing equations) is only
element discretisation. In addition, the adopted coupling partially included in this paper when it is required in the
scheme, namely implicit–explicit solution and communi- context of equating equilibrium between the total external
cation, is presented along with a novel geometry-based and internal load vectors; more details on the seepage field
fracturing algorithm which includes local remeshing to formulation can be found in [28].
avoid a computationally expensive global remesh. Section 4
describes the link between the geomechanical model and 2.2 Geomechanical equations
microseismicity. Finally, Sect. 5 presents results illustrating
the numerical capability for modelling fracture propaga- The main governing equations are derived assuming:
tion in both intact and naturally fractured reservoirs. In this
section, a validation model is also presented. The new tech- (1) Equilibrium of stresses with an appropriate constitutive
nology ELFEN tight gas reservoir (TGR) is implemented model which is able to mimic both tensile and shear
in the software package ELFEN (Rockfield Software Ltd., failure;
Swansea, UK) [24]. (2) Mass conservation of fluid flow inside the fracture region
with a flow constitutive response able to recover parallel
plate flow theory. This leads to the well-known cubic
2 Governing equations flow rule [29].

2.1 Overall methodology The update of the mechanical stresses satisfies the momen-
tum balance equation, with the assumption that fluid accel-
Based on hydraulic fracture field practice, an ELFEN TGR eration relative to the solid and the convective terms can be
analysis is divided into five key stages [24]: ignored [30]. The mechanical governing equation is given by
[28]
(1) Initiation of model effective stresses, pore pressures and
fracking fluid pressures;  
LT σ  − αm ps + ρB g = 0, (1)
(2) Pad hydraulic fracturing;
(3) Slurry hydraulic fracturing;
where L is the spatial differential operator, σ  is the effective
(4) Flowback and clean-up of the fractured region;
stress tensor, α is the Biot coefficient, m is the identity tensor,
(5) Gas production.
ps is the pore fluid pressure in the rock formation, ρB is the
This paper concentrates on stages 1 and 2 with the remain- wet bulk density and g is the gravity vector.
ing stages left for future publications. Coupling between the The fracture fluid flow governing equation is given by (see
effective stresses in the rock matrix, pore fluid flow in the Fig. 1 for schematic of link between fracture fluid flow and
rock matrix and finally fracture fluid flow is accomplished fracture opening) [28]
via a staggered coupling scheme [25]. The three main sets of  
∂ k fr   d pn
governing equations are [26,27]: ∇ pn − ρ fn g = S fr + α (ėε ) , (2)
∂x μn dt
(1) Equilibrium of the mechanical stress and pore fluid pres-
sure of the rock formation with external loads (structure where k fr is the intrinsic permeability of the fractured region,
field); μn is the viscosity of the fracturing fluid, pn is the fracturing
(2) Porous flow in the rock formation (seepage field); fluid pressure, ρfn is the density of the fracture fluid, S fr is
(3) Fluid flow in the fracture region (network field). the storage coefficient which is effectively a measure of the
compressibility of the fractured region when a fluid is present
The main target application is the hydraulic stimulation of and ėε is the aperture strain rate [26]. Assuming parallel
tight gas shale reservoirs. To fully mimic, the physics for this plate theory, the intrinsic permeability of a fractured region
class of problem requires a complex multi-phase flow simu- is given by [28]
lator due to the interaction between the invading fracture fluid
and in-situ reservoir gas. In this paper, it is assumed that due k fr = e2 /12, (3)
to the high compressibility of the dry gas (e.g., methane)
inside the shale pore space, the effective resistance of this where e is the fracture element aperture. The storage term is
fluid phase is insignificant when compared to the effective given by

123
232 Comp. Part. Mech. (2016) 3:229–248

Fig. 2 Mohr–Coulomb with Rankine tensile corner in principal stress

space

Fig. 1 Schematic of hydraulic stimulation (top) and the modelling

approach (bottom)

   
S fr = (1/e) 1/K nfr + e/K ffr , (4)
Fig. 3 Typical uniaxial continuum damage response
where K nfr
is the fracture normal stiffness and K ffr is the bulk
modulus of the fracturing fluid [28].
The Rankine tensile model is particularly important in
hydraulic fracture modelling as the minimum principal stress
2.3 Mechanical material model
at the fracture tip is tensile which allows for continued frac-
ture propagation. The FEM is a continuum-based theory,
The reservoir rock stresses are governed by three core mate-
and so it is subject to the well-known limitations for strain
rial characteristics:
softening material models. The key shortcoming being the
mechanical response is strongly mesh size dependent [32]. In
(1) Elasticity; a physical theory this is unacceptable, and therefore, correc-
(2) Mohr–Coulomb plasticity; tive methods have been applied to ensure mesh-independent
(3) Rankine tensile failure. behaviour, the most common approach involving inclusion
of a mesh size length scale at the material level adjusting the
Shales are typically laminated and hence possess a pre- softening slope, a process known as regularisation [31].
ferred structure with elastic anisotropy. The Mohr–Coulomb A typical continuum extensional uniaxial stress–strain
plasticity and Rankine tensile failure constitutive models are response for a quasi-brittle material is shown in Fig. 3 (ε0
captured using a combined single surface yield envelope as and εf are the uniaxial yield and failure strains, respectively).
shown in Fig. 2, where σ 1 , σ 2 and σ 3 are the three princi- The pre-yield response is governed by elasticity parameters,
pal stresses and f t1 , f t2 and f t3 are the corresponding tensile Young’s modulus E and Poisson’s ratio ν. Only two para-
strengths [31]. In this paper it is assumed for simplicity that meters are necessary to characterise the post-yield response,
the shale is isotropic. tensile strength f t and fracture energy G f . To ensure objective

123
Comp. Part. Mech. (2016) 3:229–248 233

energy dissipation in arbitrary meshes, the softening slope H

is dependent on an element characteristic length Cl [31]. This
approach is popular and simple to implement in finite/discrete
element codes and yields global system responses that are
mesh independent. Further applications of this technique can
be found in the literature [33,34].

2.4 Fracture fluid flow

In hydraulic fracturing of tight shales typically two fracking

fluids are used: either

(1) Slickwater or
(2) Cross-linked gels.

Due to the low viscosity, slickwater is often preferred to

save on overall energy usage to drive the fractures. However, Fig. 4 A typical dynamic fluid loss experiment
slickwaters are particularly poor at transporting proppant
grains through the opening narrow channels. Populating the ulation of a tight gas shale reservoir [35]. The exact cause
fracture with a uniform distribution of proppant is thought of this loss is not known, but some hypotheses include the
essential for the overall production rates of the reservoir. In migration of fracturing fluid into fissures adjacent to the main
contrast, viscous cross-linked gels are very good at trans- propagating fracture and capillary action due to the inher-
porting proppant grains but often do not produce the long ent small pore throat radii of the shale grains. Whatever the
fractures required to significantly increase the overall perme- root cause, one effect of this fluid loss is a drop in frac-
ability of the fractured region. In practice, hydraulic fracture ture fluid pressure as the fracture propagates, so mimicking
pump schedules use a combination of slickwaters and cross- this behaviour is extremely important in order to capture the
linked gels in sequence to ensure that both a long fracture is correct fracture volume and length generated during stimula-
created and subsequently proppant grains are transported to tion. This behaviour is represented using a 1D Carter leak-off
the tip of the fracture, so in essence using the best qualities model (transversal flow) which has an in-built decay of leak-
of both fracturing fluid types. off rate due to the assumed formation of a filter cake on the
Macroscopically the slickwater exhibits Newtonian fluid exposed fracture surface over time [5]. The model assumes
characteristics and the crossed-linked gels non-Newtonian an initial volume loss Vsp per unit area over a spurt time tsp
behaviour. In an effort to reduce the viscosity at higher strain followed by a constant leak-off coefficient C as:
rates, a shear-thinning cross-linked gel is commonly adopted.
Constitutively, this behaviour can be captured using the non- Vsp
t − texp < tsp ; ql = ,
Newtonian power law model: tsp
C
t − texp ≥ qtsp ; ql = , (6)
τ = K γ n, (5) t − texp

where τ is the fluid shear stress, γ is the fluid shear strain where t is the current model time, texp is the time at which a
rate, K is the consistency index and n is the power law expo- fracture surface is exposed for leak-off and ql is the 1D nor-
nent. For the simulation of fluid flow within the fractures, the mal leak-off velocity. The spurt volume is determined from
following assumptions are applied [29]: fluid loss experiments (see Fig. 4, [36]). In addition, Fig. 4
shows the controlling leak-off mechanisms as they evolve
(1) The fluid is incompressible; through the experiment. Initially, the flow characteristics of
(2) The flow is locally equivalent to the flow between two the reservoir (e.g., the reservoir fluid viscosity or its intrinsic
smooth, parallel plates; permeability) are dominant in controlling the degree of fluid
(3) The flow is laminar with a low Reynolds number. leak-off which invades the host rock. However, once a filter
cake has formed on the exposed fracture surface, this con-
2.5 Fracture fluid leak-off trols the fluid loss and the rate reduces, eventually tending to
a steady-state fluid loss.
The petroleum industry commonly finds that approximately The 1D Carter leak-off model in Eq. (6) is independent
50–80 % of injected fluid is lost during typical hydraulic stim- of the pressure difference between the fracture fluid pressure

123
234 Comp. Part. Mech. (2016) 3:229–248

and the rock formation pore fluid pressure. More sophisti- It should be noted that the seepage field component is only
cated leak-off models which are functions of the pressure included to satisfy completeness of the equilibrium equation
difference can be found in the literature [24] but will not be which is a function of the total and not-effective stress ten-
discussed further here. sor. The finite element discretised structure field equation is
A single phase analysis is assumed between fluids in both given by
the fracture and porous media regions. In practice this states
that the fracturing fluid and the tight reservoir gas cannot BTu σ  ∂Ωu − BTu αmNs ∂Ωs ps = fu , (8)
mix. A special treatment is invoked which allows the fluid to Ωu Ωs
be extracted from the fracture and stored in a separate storage
where Ωu is the structure domain, Ωs is the seepage domain
block (i.e., fluid is not transferred into the rock formation).
and fu is the mechanical load vector. The effective stress ten-
The extraction of fluid mass is important as this is a key driver
sor σ  is computed using the combined Mohr–Coulomb and
in the final hydraulic fracture length with lengths significantly
Rankine cap material model. The finite element discretised
reduced for those fractures experiencing high degrees of leak-
network field equation is given by
off. During the flowback stage the stored fluid is available for
extraction, so mass conservation of the fracturing fluid in the
k fr ∂ pn
global system is maintained. BTn Bn pn ∂Ωn + NnT S fr Nn ∂Ωn = fn , (9)
Ωn μn Ωn ∂t

where Ωn is the network domain and fn is the fracture fluid

3 Numerical algorithms flow load vector. The structure and network fields coupled
governing equations can be written in matrix form as
3.1 Overall methodology
M 0 ü 0 0 u̇
Two solutions are used to update the main governing equa- +
0 0 p̈n QTn Sn ṗn
tions:
K 0 u fu
+ = , (10)
(1) For the evolution of the mechanical stresses an explicit 0 Hn pn fn
solution method is used. This method is most advanta-
geous when dealing with material strain softening where with matrices and vectors defined as
implicit schemes can struggle to attain convergence dur-
ing the material stress update; M= (Nu )T ρB Nu ∂Ωu ,
(2) For evolution of fracture fluid pressure an implicit solu- Ωu

tion method is adopted. As the fracture flow region is K= (∇Nu )T D∇Nu ∂Ωu ,
limited to a small part of the problem domain the com- Ωu
putational overhead is low. In addition, the convergence
Qn = (∇Nu )T mNn ∂Ωn ,
rates are robust for a wide range of fluid properties and Ωn
fracture aperture widths.
Sn = (Nn )T S fr Nn ∂Ωn ,
Ωn
(11)
e2
3.2 Discretised geomechanical equations Hn = (Nn )T ∇Nn ∂Ωn ,
Ωn 12μn
The two governing equations for the mechanical (structure fu = (Nu )T ρB b∂Ωu + (Nu )T tΓu ,
field) and fracture fluid flow (network field) are semi- Ωu Ωu
 
discretised using the FEM. It is assumed that the shape e2
fn = − (Nn )T ∇ T ρ fn b ∂Ωn
functions can be independent for structure Nu , seepage Ns Ωn 12μn
and network Nn fields, respectively.
+ (Nn )T qΓn ,
Ωn
Bu = Lu Nu , Bs = Ls Ns and Bn = Ln Nn , (7)
where D is the material stiffness matrix, Γu and Γn are the
where Lu , Ls and Ln are the gradient operators for the struc- boundary regions of the structure and network fields, respec-
ture, seepage and network fields, respectively, and finally tively, t is the external traction load which in the present
Bu , Bs and Bn are the shape function spatial gradient matri- context is the fluid pressure along the exposed fracture sur-
ces for the structure, seepage and network fields, respectively. face and q is the fracturing fluid flux. This sink term is used

123
Comp. Part. Mech. (2016) 3:229–248 235

Fig. 5 Coupling between structure, seepage and network fields

Fig. 6 Graphical representation of fluid pressure update in both struc-

in conjunction with the 1D leak-off when fracturing fluid loss ture and network fields between successive coupling times tc1 and tc2
is a significant part of the analysis.
3.4 Coupling strategy: communication scheme
3.3 Coupling strategy: an overview
Figure 6 shows a graphical representation of the network
pressure update in both structure and network fields.
A staggered coupling scheme is adopted in which the
A note on the notation, the superscripts i and e refer to
mechanical governing equation is solved explicitly [11] and
the implicit and explicit solvers, respectively. The fluid pres-
the fracture fluid flow governing equation is solved implic-
sures at a coupling time are usually not equal in structure and
itly [25,26,30,37–40]. This implies that in practice there are
network fields, i.e., ( pne )c1 = ( pni )c1 . The implicit network
many more explicit time steps per implicit time step, and
field analysis first marches forward in time between coupling
hence, on the mechanical field the fracture fluid pressure
times tc1 and tc2 yielding fluid pressures ( pni )c1 and ( pni )c2 .
needs to be updated using intermediate values between cou-
Figure 7 presents the communication between implicit
pling times. The coupling between structure, seepage and
and explicit solvers which takes place across successive
network fields is shown in Fig. 5.
coupling times tc1 and tc2 . More details regarding the indi-
The motion of the fracture aperture is computed in the
vidual implicit and explicit solution updates can be found in
structure field and this information is transferred to the net-
Sects. 3.5 and 3.6, respectively
work field to update the permeability characteristics of the
propagating fracture. Likewise, the fracturing fluid pressure
3.5 Time integration: network field implicit solution
is computed in the network field and this is transferred to the
structure field where it acts as an external traction load on
From Eq. (10) the network field at the current time tk+1 can
the fracture surface. In this paper the structure and seepage
be written as (where the subscript k + 1 implies that the cor-
fields are considered uncoupled.
responding matrix or vector term is computed at the current
A key feature in the proposed formulation is the treatment
time)
of the network fluid pressure in both structure and network
fields. Both fields compute locally their own network fluid  
QTn (u̇)k+1 + (Sn )k+1 ( ṗn )k+1 + (Hn )k+1 ( pn )k+1
pressure and coupling is maintained via a predictor–corrector k+1
scheme as outlined in Sect. 3.6. = (fn )k+1 . (12)
Since this class of problem investigates tight gas shale
reservoir with isotropic Young’s moduli in the range of 30– A generalised Newmark procedure GN11 is performed to
60 GPa [41], the time steps could be quite small unless update the nodal network pressure pn between successive
remedial measures are undertaken. Mass scaling techniques time steps tk and tk+1 :
[42] are performed on relevant parts of the finite element
mesh whenever an explicit time step is liable to drop below a ( pn )k+1 = ( pn )k + ( pn )k timp + β0  ṗn timp , (13)
threshold value. These are typically near the fracture tip with ( ṗn )k+1 = ( ṗn )k + β1  ṗn , (14)
a maximum scaling factor of 100. Global system kinetic and
strain energies are performed to ensure a quasi-static solution where timp = tk+1 − tk is the implicit solution time incre-
is attained. ment. In addition β0 and β1 are the integration parameters

123
236 Comp. Part. Mech. (2016) 3:229–248

Fig. 7 Communication
between network and structure
fields during successive
coupling times

set as β0 = 0 and β1 = 1.0. Equation (12) can be rewritten increment. This implies that the storativity and permeability
as matrix terms are also constant.

ψ itn
k+1 = (fn )k+1 − (Sn )k+1 ( ṗn )k+1
itn itn itn
3.6 Time integration: structure field explicit solution
 itn
− (Hn )itn
k+1 ( pn )k+1 − Qn
itn T
(u̇)itn
k+1 Time integration of the structure field adopts the standard
k+1
= 0, (15) central difference scheme to update the nodal accelerations,
velocities and displacements [10,31].
where the superscript itn refers to the iteration number. Since in practice compared with the implicit solution
Clearly, convergence is achieved when ψ itn scheme the explicit solution requires many more time inte-
k+1 → 0. The
unknown parameter in Eq. (15) is  ṗ and this is computed gration steps, it is necessary to update the network field fluid
using the Newton–Raphson method: pressure at intermediate steps between coupling times (tc1
and tc2 , respectively, see Fig. 6). The following presentation
builds on the parameters shown in Fig. 6 and they are defined
∂ψ itn
k+1
− d ( ṗn )itn = ψ itn
k+1 . (16) in more detail here. As the implicit network field advances
∂ ṗn ahead of the explicit structure field then from the perspective
of the structure field the network fluid pressures ( pni )c1 and
Linearisation of Eq. (15) leads to the final transient equa- ( pni )c2 at successive coupling times tc1 and tc2 are known
tion that is then used to update the nodal network pressures: values. The structure field network fluid pressures ( pne )c1 are
  itn also known at the start of the coupling interval tc1 . The goal
β1 (Sn )itn
k+1 + β0 t (Hn )k+1 d  ṗnn
itn is to compute the network fluid pressure between coupling
times on the explicit structure field using only these values.
= (fn )itn
k+1 − (Sn )k+1 ( ṗn )k+1
itn itn
It is assumed that this value can be decomposed into three
 i
− (Hn )itn components as:
k+1 ( pn )k+1 − Qn (u̇)itn
k+1 .
itn T
(17)
k+1
   
Once  ṗn is equated, the fluid pressure terms in Eqs. (13) pne ist+1
= pni + ps + pd , (18)
int
and (14) are updated to the current problem time tk+1 . In the
present formulation, the implicit solution initially advances where ( pne )ist+1 is the network fluid pressure at the interme-
ahead of the explicit solution of the structure field, and dur- diate step ist + 1 between coupling times tc1 and tc2 . The
ing this stage the aperture and aperture strain rate terms number of intermediate steps is dependent on the ratio of the
are assumed constant over a single implicit solution time explicit time step size to the coupling time interval. ( pni )int is

123
Comp. Part. Mech. (2016) 3:229–248 237

the linearly interpolated network pressure from the implicit region, thus avoiding the need for a computationally exhaus-
e
values at time tist+1 : tive global remesh.
At the material level each element response follows con-
    te − tc1  i     tinuum damage theory (see Sect. 2.3) and the key constitutive
pni = pni + ist+1 pn − pni . parameters like stress, strain and damage are spatially located
int c1 tc2 − tc1 c2 c1
(19) at the element Gauss point. The damage indicates the degree
of softening which an element has undergone during loading,
The incremental change in network fluid pressure due to with a value of 0 indicative of no damage and 1 indicative of
the storativity and aperture change is given by fully damaged. In term of the physical response this repre-
sents the coalescence of micro-flaws, leading to the creation
  of two new fracture surfaces.
ps = − 1/S fr eε , (20) A typical fracture propagation model is shown in Fig. 8a–
c with the fracture length almost doubling in size during fluid
where eε is the change in aperture strain computed between injection. The propagating fracture is not pre-seeded due to
successive explicit time steps ist and ist + 1 as model constraints and it is freely able to follow the stress
state as dictated by the model (refer to Fig. 8b, c).
eist+1 − eist A schematic of the damage process zone and its numerical
eε = , (21)
0.5(eist+1 + eist ) treatment is shown in Fig. 9a–c. This shows the opening of
a crack mouth with a tensile damage zone ahead of the frac-
e
where eist+1 and eist are the apertures extracted at times tist+1 ture tip. The elements at the tip are fully damaged with the
e , respectively. The third component is a measure of
and tist remaining elements below this threshold and still capable of
the difference between the implicit and explicit computed supporting load (refer to Fig. 9a). The predicted fracture sur-
network pressures at the start of a coupling interval. This face is not forced to follow the element edges and can follow
difference is linearly ramped off over the coupling interval: the stress state as dictated by the simulation (see Fig. 9b).
Due to the mode-1 assumption of failure, the failure direc-
tc2 − tist+1
e     
pd = pni − pne c1 . (22) tion is defined as orthogonal to the maximum tensile principal
tc2 − tc1 c1 stress. When the failure path exceeds a user-specified length,
all points along the path are used to form a geometric entity,
The coupling interval must be appropriately set; if it is not and this finally leads to the fracture surface (Fig. 9b, c). The
tight enough there will be an unacceptable drift between the fracture prediction algorithm is shown in Table 1.
two values and, conversely, too tight a coupling interval can
lead to increased run-time without improved accuracy of the
solution. 3.9 Local remeshing around fracture tip

3.7 Time integration: coupling parameters A very important modelling aspect is the capability to deal
with meshing the new geometry in a computationally attrac-
The present formulation exhibits good convergence proper- tive manner. A local meshing methodology is employed
ties with the implicit solution scheme typically converging which makes redundant the need for a traditionally expensive
within a few iterations. For hydraulic fracture design cases global remesh, which is also likely to introduce dispersion of
pumping times are usually measured in the order of thou- key material variables such as stress, strain and damage indi-
sands of seconds and through investigation it has been found cators. The local meshing zone is defined via a patch region
that a coupling interval of 1.0 s is typically appropriate for which is adjacent to the fracture tip, and it is always very
this class of problem. small compared to the problem size, so in relative terms the
computational cost is low (refer to Fig. 11a).
3.8 Geometry-based fracture prediction It is very important that the remesh is only performed
locally at a fracture tip. For example, in a typical indus-
The geometry insertion technique is a means of introducing trial scale mesh of many 100,000’s elements typically only a
new geometry lines (in 2D hydraulic stimulation models) 100 elements are remeshed. The mesh in all other regions of
into the finite element discretisation, which do not necessar- the problem domain remains unchanged (refer to Fig. 11b).
ily follow the edges of the finite element mesh. The newly Clearly, fracturing is a very dynamic process in that the model
introduced geometry lines simulate fracture growth during is constantly changing, so this procedure of following the
hydraulic stimulation. In addition, a local remesh algorithm fracture tip via a patch region and subsequently only remesh-
is implemented, which only operates around the fracture tip ing locally is continually being updated during hydraulic

123
238 Comp. Part. Mech. (2016) 3:229–248

Fig. 8 a–c Evolving discrete fracture

Fig. 9 a–c Fracture prediction and insertion

Table 1 Fracture prediction There is a parent–child relationship between the network

and fracture surface nodes which allows, for example, the
For each fracture tip fluid pressure from the network field to be transferred as an
(a) Extract material state for elements within patched region (see external load to a corresponding node on the fracture surface
Fig. 10a, b)
of the structure field. During a local remesh around a frac-
(b) Interpolate damage variable to element nodes (see Fig. 10c)
ture tip this mesh topology must be maintained, and this is
(c) For those nodes which have surpassed the damage threshold
achieved by initially stitching the mesh back to a non-discrete
of 1.0 as set by the material model, construct a best-fit linear
line from fracture tip through damaged nodes (see Fig. 10d) body, performing the local remesh on the resulting bonded
(d) If line length matches that specified by the user, then the line domain and then displacing the elements back to their orig-
is marked ready for fracture insertion inal spatial location after the remesh. This is achieved by
monitoring the displacements of the fracture surfaces such
that they can be returned to their position after the remesh.
stimulation. This is extremely important for this class of prob- Since mode-1 failure rather than shear failure mechanisms is
lem as it has been observed that for some tight gas shales considered in the present application, typically there is very
stimulated fracture lengths reach values of many hundreds little relative slip between opposing fracture surfaces during
of metres [43]. hydraulic stimulation.

123
Comp. Part. Mech. (2016) 3:229–248 239

Fig. 10 a–d Predicting fracture length and direction

Mapping of the displacements is very important to ensure

that the correct aperture is maintained after a remesh since the
implicit solution network element permeability is a function
of the current aperture. The newly inserted network elements
are initialised to the initial reservoir fluid pressure and the
increased fluid pressure from further fluid injection typically
results in additional fracture opening.

4 Geomechanics and microseismicity

4.1 Overall methodology

The concept behind integrating geomechanics and microseis-

mic prediction for the FEM [45] is based on the approach
described in [46]. The integrated geomechanics and micro-
seismic prediction was applied to a North Sea field, where
subsidence prediction was used to calibrate the coupled
flow-geomechanical model of the field [47]. The paper was
interested in monitoring high shear stress regions in the reser-
voir layer with the potential increased risk of microseismicity.

4.2 Implementation
Fig. 11 a and b Local remeshing around fracture tip

The approach followed in this paper is outlined in the paper by

Key nodal and element variables, such as displacements Angus et al. [45]. Seismic events are monitored in space and
and stresses, are mapped between old and new meshes using time, where a microseismic event is predicted to occur within
standard mapping techniques (see [44] for more details). an element when the effective stress satisfies the Mohr–

123
240 Comp. Part. Mech. (2016) 3:229–248

Coulomb yield envelope. Based on the differential stress

tensor Angus calculates a pseudo-scalar seismic moment
(i.e., stress drop) which can be used to infer microseismicity.
Future work will include a method outlined by Lisjak et al.
[20] for transferring the material strain energy into a seismic
energy signal for mode-1 type failure.

5 Numerical examples

5.1 Validation model

To ensure the validity of the proposed formulation, the soft-

ware is compared against a well-known analytical solution
in the hydraulic fracture mechanics literature. The Khris-
tianovitch and Zheltov, Geertsma and De-Klerk (KGD)
analytical solution assumes

(1) The relationship between crack or fracture surface

pressure and fracture aperture is given by Sneddon’s for-
mulation;
(2) Steady-state fluid flow inside the fracture.

The implementation proposed in this paper assumes tran-

sient flow, so it is only possible to partially capture the
assumptions in the analytical solution. A further validation
comparison is presented via a semi-analytical solution in
which the mechanical response of the fracture width is once
again dictated by Sneddon’s equation, but the flow part is
solved via a 1D numerical solution using a finite difference
scheme, allowing the transient effects to be captured. From
Sneddon’s equation, the width as a function of net pressure
Fig. 12 a and b Comparison of software against KGD solution (ana-
and fracture length is given by [1]
lytical and semi-analytical solutions)

(1 − ν 2 )
W ∝4 L P, (23)  0.25
E 6QμE 3
psol = L 0.25 , (25)
8H 4 (1 − ν 2 )3
where W is the maximum fracture width or aperture, E is the  0.25
isotropic Young’s modulus, ν is the Poisson’s ratio, L is the 96Qμ(1 − ν 2 )
Wsol = L 0.25 , (26)
fracture length and P is the net pressure along the fracture. E
Assuming parallel plate flow theory for 1D laminar flow, the
relationship between fluid velocity and flow rate is given by where psol and Wsol are known as the pressure solution and
width solution, respectively.
dp 12Qμ Figure 12a, b shows the comparison between the software
∝− , (24)
dx HW3 and both KGD analytical and semi-analytical solutions. It
can be seen that the trends for both ‘pressure versus fracture
where p is the fluid pressure, x is the 1D spatial dimension, half-length’ and ‘fracture width versus fracture half-length’
Q is the fluid injected flow rate, μ is the fluid viscosity and are captured with the semi-analytical KGD solution provid-
H is the height of the fracture. Substituting Eq. (23) into ing a better match due to its capability to mimic transient
Eq. (24) and solving the resulting ordinary differential equa- flow inside the fracture. Sneddon’s solution assumes elas-
tion via a simple separation of variables leads to the following tic behaviour, so in the numerical solution fluid was only
solutions: injected prior to inelastic deformation at the fracture tip.

123
Comp. Part. Mech. (2016) 3:229–248 241

Fig. 13 Case 1: model

geometry of intact reservoir
(plan view), Case 2: model
geometry of naturally fractured
reservoir with two DF sets (plan
view)

(a) Case 1 (b) Case 2

5.2 Demonstration models Table 2 Pre-production stresses and pore pressure (assumed uniform)
Pre-production parameters Values (MPa)
The demonstration cases investigate hydraulic fracturing
in both intact (see Fig. 13a) and naturally fractured (see Minimum horizontal effective stress 10
Fig. 13b) shale reservoirs. The reservoir is modelled in Maximum horizontal effective stress 15
plan view as a 2D plane strain domain. Hence, the model Overburden effective stress 20
is simulating horizontal fracturing with an assumed con- Reservoir pore pressure 30
stant extruded height or reservoir layer thickness. The outer
boundaries of the model are fixed. The natural fractures Table 3 DFN set 1
are modelled with Mohr–Coulomb stick-slip contact regions
DFN parameters Values
[48,49]. The pre-production stresses and pore pressures are
specified in Table 2. The initial stresses are not aligned with Orientation 90◦
the global axis but are rotated clockwise by 40◦ relative to Fracture spacing 80 m
north (i.e., positive y-axis). This is indicative of the case Fracture length 40 m
where the horizontal wellbore is not drilled exactly parallel Persistence 80 m
with the minimum principal stress.
The natural fractures in Fig. 13b are specified via a statis- Table 4 DFN set 2
tical variation of four key DFN parameters; these are
DFN parameters Values
• Orientation (relative to north, i.e., y-axis);
Orientation 330◦
• Fracture spacing;
Fracture spacing 30 m
• Fracture length;
Fracture length 40 m
• Persistence.
Persistence 50 m
For the naturally fractured case, two DFN sets are specified
with DFN parameters as stated in Tables 3 and 4, respectively.
Table 5 Shale elastic properties
In reality the natural fractures are not perfectly aligned
according to a uniform DFN parameter value, and so a small Material parameters Values
standard deviation is applied to each parameter to give a small Young’s modulus 32,000 MPa
variation about the input mean (see Fig. 13b). Poisson’s ratio 0.2

5.3 Mechanical and fracture fluid properties

Table 6 Shale fracture mechanics properties

Isotropic elasticity is assumed for the shale rock with the Material parameters Values
material parameters given in Table 5. The fracture mechanics
Tensile strength 1.0 MPa
material parameters and fracking fluid properties are stated
Fracture energy 50 N m
in Tables 6 and 7, respectively.

5.4 Case 1: hydraulic stimulation of intact rock schedule as outlined in Table 8. It is assumed that the fluid
loss from the fracture into the reservoir is very low.
An intact shale of uniform layer thickness 10 m is pumped Figure 14 shows the orientation of the initial mini-
with a slickwater fracturing fluid along with a pumping mum and maximum effective horizontal principal stresses.

123
242 Comp. Part. Mech. (2016) 3:229–248

Table 7 Fracturing fluid properties its most rapid during the initial pumping stages and gradually
Fluid properties Values reduces over the pumping schedule.

Viscosity 1.67E−3 Pa s
5.5 Case 2: hydraulic stimulation of a naturally
Bulk fluid modulus 2000 MPa
fractured reservoir

A key design question to the hydraulic fracture engineer is

Table 8 Pumping schedule the influence of natural fractures on propagating fractures.
Flow rate (m3 /s) Volume (m3 ) Duration (s) From coring the fracture density is known in terms of the
number of fractures per core length, but this provides limited
0.125 75 600 information on the orientation of the natural fractures, so in
effect the required information is only partially known to the
engineer and any remaining DFN parameters form part of a
sensitivity analysis.
Figure 15a–c shows the evolution of the propagating fracture. In this case study the exact same material properties, pre-
In particular, it can be observed that the fracture propagates production stresses and pumping schedules are applied as
towards the maximum compressive stress and maintains this in Case 1; the only difference now being the inclusion of
direction throughout the fluid injection phase. In practice the two DFN sets. Therefore, this contrasts hydraulic fracture
initial stress field in the reservoir is most likely heteroge- prediction in an intact reservoir rock with one which contains
neous, and this would influence the propagating fracture, multiple DFN sets.
leading to a potentially complex fracture pathway [50] which The evolution of the bottom hole pressure (BHP) is often
can be modelled. Figure 16 shows the evolution of fracture analysed during the post-appraisal phase of a hydraulic frac-
volume and length. The rate of fracture length increase is at ture design to assess the accuracy of the initial model input

Fig. 14 Case 1: pre-production minimum and maximum principal directions (red minimum, blue maximum)

Fig. 15 a–c Case 1: propagation of hydraulic fracture

123
Comp. Part. Mech. (2016) 3:229–248 243

the initial stress state. The overriding propagating fracture

direction is unchanged even though there is noticeable con-
nectivity with both DFN sets 1 and 2. The overall fracture
length is only slightly smaller than the value computed for
the intact reservoir case (compare Figs. 15c, 18d).
Figure 19a, b shows the evolution of the most tensile
principal stress during the initial pumping stage (i.e., t = 22–
32 s) in a region adjacent to the stimulated section. The tip
of the fracture is in tension, and this allows the propagat-
ing fracture to develop and finally connect with adjacent
DFN sets. The ELFEN TGR software ensures that there is
a fine mesh around the tip region to adequately capture the
sharp stress gradients. This is particularly important when the
hydraulic fracture approaches DFN sets where large element
Fig. 16 Case 1: evolution of fracture volume and fracture length sizes could smear out the stresses over an unduly large area
and result in the propagating fracture erroneously connecting
with a DFN set.
Figure 20a, b shows the reservoir relative maximum hor-
izontal stress σHR , which is computed as

σH
σHR = , (27)
σH0

where σH and σH0 are the current and pre-production maxi-

mum horizontal stresses, respectively. This parameter offers
a number of insights into the mechanical behaviour of the
reservoir. Regions where the value is effectively equal to
1 indicates reservoir sections with unchanged stress, so it
defines the limits of the stress influence of the propagating
fracture. In addition, regions with perturbed values outside of
1 are very important as these can indicate whether a natural
Fig. 17 Cases 1/2: evolution of bottom hole pressure fracture is open or closed relative to its initial configuration.
From Fig. 20a, b the natural fractures corresponding to DFN
set 2 are in a state of increased compression (i.e., σHR > 1.0).
and also to develop a better understanding of the whole frac- This suggests that these natural fractures are now in a closed
turing process. The BHP evolution for both Case 1 (intact state and less conducive for porous flow, which is a key factor
reservoir) and Case 2 (naturally fractured reservoir) is shown in obtaining sufficient hydrocarbon recovery from a gas-filled
in Fig. 17. BHP starts from reservoir pressure and increases tight shale reservoir.
due to fluid injection and reaches a peak value, known as One of the key indicators of fracture propagation is not
breakdown pressure [1], at which time the reservoir rock only the initial stress state but also the evolving stress field
tensile strength is overcome and a fracture forms. Due to the in terms of both magnitude and direction as the hydraulic
sudden increase in fracture volume, there is a decrease in fracture propagates in a reservoir potential populated with
the BHP and this eventually tends to a near constant value a high density of natural fractures. This is where numerical
known as the propagating pressure. Once the fracture reaches techniques become a powerful tool, for analysing systems
a significant length, the BHP required to maintain fracture with many complex non-linear components like continuum
propagation is significantly reduced; in this study the BHP damage mechanics and Mohr–Coulomb frictional slip along
drop from breakdown to propagating pressure is approxi- natural fractures. Figure 21a, b shows the evolution of the
mately 10 MPa. principal stress vectors near the stimulation region during
Figure 18a–d shows the evolution of the hydraulic fracture the first few tens of seconds of the pumping stage. It can
throughout hydraulic stimulation. A very important observa- be observed from Fig. 21a that adjacent to the propagat-
tion is the propensity of the hydraulic fracture to follow the ing fracture the maximum stresses rotate such that they
far-field maximum horizontal principal stress direction even are orthogonal to this propagation direction. Interestingly,
when connecting with natural fractures which are oblique to the extent of the rotation of the principal stresses from the

123
244 Comp. Part. Mech. (2016) 3:229–248

Fig. 18 a–d Case 2:

propagation of hydraulic
fracture

Fig. 19 a and b Case 2: evolution of minimum principal stress (Pa) near injection region (positive is tensile: red contour)

123
Comp. Part. Mech. (2016) 3:229–248 245

Fig. 20 a and b Case 2: evolution of maximum horizontal principal stress near injection region (relative value = current value/initial value)

initial far-field state is relatively small, so in effect only tion is attained. This is often referred to as the ‘creaking
a small region surrounding the hydraulic fracture under- of the rock’ and the change in stress field could affect the
goes a change in stress and much of the reservoir stress fracture propagation direction. Figure 22a, b shows a com-
remains unchanged. This is the case in point for the DFN parison between the slip predicted by the geomechanical
sets, material properties and pumping schedule as defined in model (Fig. 22a) and the corresponding inferred seismic-
this paper. Further investigations are required to determine ity (Fig. 22b) 650 s after pumping started. Beach balls are
whether this observation is more universal for this class of a graphical representation of the source mechanism [45,47]
problem. and gave information on both the potential failure mecha-
nism in terms of seismic moments (e.g., shear or tension)
and the magnitude of the seismic event. It can be observed
5.6 Case 2: link between geomechanical model and that the stress changes remain local to the propagating frac-
microseismicity ture and do not strongly influence potential slippage along
DFN sets at a distance from the fracture tip. This behaviour is
During fluid injection the reservoir undergoes local changes confirmed from both the geomechanical model and inferred
in the stress field due to the propagating hydraulic fracture. seismicity. Indeed the inferred seismicity predicts a large ten-
As the fracture spreads into regions with a high density sile event at the fracture tip and low shear slippage along
of DFN’s, it is instructive to assess the potential for slip DFN sets adjacent to the propagating fracture and this corre-
along the surface of the DFN’s. Slip results in a change sponds with localised stress changes observed as the fracture
of the local stress field as a new equilibrium configura- propagates.

123
246 Comp. Part. Mech. (2016) 3:229–248

Fig. 21 a and b Case 2: evolution of principal stress vectors (Pa) near injection region (red minimum, blue maximum)

6 Conclusions new technology was shown to predict hydraulic fractures of

some 500 m in length which is consistent with field obser-
This paper has presented a novel combined finite element vations by the petroleum industry. The interaction between
and discrete element approach to investigate hydraulic frac- DFN sets and far-field stresses was also shown in the naturally
ture stimulation in TGRs complemented by microseismic fractured reservoir case. In the latter case inferred microseis-
analysis. micity confirmed that the anticipated degree of slip along
The method combines coupled techniques to assess the DFN sets was minimal and only those DFN sets which were
interplay between injected fluid and the mechanical response near the propagating fracture were affected.
of the reservoir. The propagating fracture is a complex Future work for the present tool includes the capability to
response combining non-linearities at many levels, including simulate transport of the proppant grains inside the fracture
material behaviour and the insertion of new fracture surfaces. regions (i.e., the slurry stage). This leads to a propped con-
An innovative approach has been implemented to deal ducive zone inside the fracture which is required for sufficient
with fracture insertion based on new and dynamically hydrocarbon drainage of a TGR [51].
updated geometrical entities rather than the traditional The 2D hydraulic stimulation model presented in this
approach of splitting along a finite element edge. By basing paper is most suitable for thin reservoir layers and homoge-
fracture insertion on geometrical entities a degree of control neous stress states. The reservoirs are often very thin, perhaps
is maintained over the quality of the evolving finite element in the region of 30–40 m thick, so the 2D approximation is
mesh. Indeed, with the present method it was observed dur- suitable in this instance. However, for reservoirs with highly
ing an analysis that the time step size change was very small heterogeneous stress states, a 3D model is essential. The tech-
which is testament to the geometric approach. niques described in this paper will be extended to include a
Two demonstration cases were presented; hydraulic stim- fully functioning 3D hydraulic stimulation and production
ulation of both intact and naturally fracture reservoirs. The modelling software.

123
Comp. Part. Mech. (2016) 3:229–248 247

Fig. 22 a and b Case 2: link between slip prediction from geomechanical model and inferred microseismicity (a snapshot of the reservoir 650 s
after start of fluid injection)

References 9. Yew CH, Weng X (2015) Mechanics of hydraulic fracturing. Gulf

Professional Publishing, Waltham
1. Economides MJ, Martin T (2007) Modern fracturing enhancing 10. Belytschko T, Liu WK, Moran B (2000) Nonlinear finite elements
natural gas production. Energy Tribune, Houston for continua and structures. Wiley, Chichester
2. King GE (2010) Thirty years of gas shale fracturing: what have we 11. Wu SR, Giu L (2012) Introduction to the explicit finite element
learned? In: Proceedings of the SPE annual technical conference method for nonlinear transient dynamics. Wiley, Hoboken
and exhibition, Florence, Italy 19–22 September, SPE 133456 12. Alqahtani NB, Miskimins JL (2010) 3D finite element modeling of
3. Cipolla CL, Lolon EP, Erdle JC, Rubin B (2010) Reservoir mod- laboratory hydraulic fracture experiments. In: Proceedings of the
elling in shale-gas reservoirs. In: Proceedings of the SPE eastern annual conference, Barcelona, Spain, 14–17 June, SPE 130556
region meeting, Charleston, West Virginia, USA, 23–25 Septem- 13. Nassir M, Settari A, Wan H (2012) Prediction and optimization
ber, SPE 15530 of fracturing in tight gas and shale using a coupled geomechani-
4. Khoei AR (2015) Extended finite element method: theory and cal model of combined tensile and shear fracturing. In: Hydraulic
applications. Wiley, Chichester fracture technology conference, The Woodlands, Texas, USA, 6–8
5. Economides MJ, Nolte KG (2000) Reservoir stimulation. Wiley, February, SPE 152200
New York 14. Cipolla CL, Warpinski NR, Mayerhofer MJ, Lolon EP, Vincent
6. Matthews HL, Schein G, Malone M (2007) Simulation of gas MC (2010) The relationship between fracture complexity, reser-
shales; they’re all the same—right? In: Proceedings of the hydraulic voir properties and fracture-treatment design. In: Annual technical
fracturing technology conference, College Station, Texas, USA, conference and exhibition, Denver, Colorado, USA, 21–24 Sep-
29–31 January, SPE 106070 tember, SPE 115769
7. Mayerhofer MJ, Lolon EP, Warpinski NR, Cipolla CL, Walser D, 15. Weng X, Kresse O, Cohen C, Wu R, Gu H (2011) Modeling of
Rightmire CM (2008) What is stimulated reservoir volume? In: hydraulic-fracture-network propagation in a naturally fractured
Proceedings of the shale gas production conference, Fort Worth, formation. In: Hydraulic fracturing technology conference, The
Texas, USA, 16–18 November, SPE 119890 Woodlands, Texas, USA, 24–26 January, SPE 140253
8. Jones J, Britt JZ (2005) Design and appraisal of hydraulic fractures. 16. Kresse O, Cohen C, Weng X, Wu R, Gu H (2011) Numerical mod-
Society of Petroleum Engineers, Richardson eling of hydraulic fracturing in naturally fractured formations. In:

123
248 Comp. Part. Mech. (2016) 3:229–248

45th US rock mechanic/geomechanics symposium, San Francisco, 37. Zienkiewicz OC, Taylor RL, Zhu JZ (2005) The finite ele-
CA, 26–29 June ment method: its basis and fundamentals. Butterworth-Heinemann,
17. McClure MC (2012) Modeling and characterization of hydraulic Oxford
stimulation and induced seismicity in geothermal and shale gas 38. Huang M, Yue ZQ, Tham LG, Zienkiewicz OC (2004) On the
reservoirs. PhD Thesis, Stanford University stable finite element procedures for dynamic problems of saturated
18. Munjiza A (2004) Combined finite-discrete element method. porous media. Int J Numer Methods Eng 61:1421–1450
Wiley, Chichester 39. Minkoff SE, Stone CM, Bryant S, Peszynska M, Wheeler MF
19. Gordeliy E, Peirce A (2013) Coupling schemes for modelling (2003) Coupled fluid flow and geomechanical deformation model-
hydraulic fracture propagation using the XFEM. Comput Method ing. J Pet Sci Eng 38:37–56
Appl Mech Eng 253:305–322 40. Huang M, Wu S, Zienkiewicz OC (2001) Incompressible or nearly
20. Lisjak A, Liu Q, Zhao Q, Mahabadi OK, Grasselli G (2013) incompressible soil dynamic behaviour—a new staggered algo-
Numerical simulation of acoustic emission in brittle rocks by rithm to circumvent restrictions of mixed formulation. Soil Dyn
two-dimensional finite-discrete element analysis. Geophys J Int Earthq Eng 21:169–179
195:423–443 41. Barree RD, Gilbert JV (2009) Stress and rock property profil-
21. Kolditz O, Delfs J-O, Burger C, Beinhorn M, Park C-H (2008) ing for unconventional reservoir stimulation. In: Proceedings of
Numerical analysis of coupled hydrosystems based on an objected the hydraulic fracturing technology conference, The Woodlands,
compartment approach. J Hydroinform 10:227–244 Texas, USA, 19–21 January, SPE 118703
22. Kolditz O, Gorke U-J, Shao H, Wang W (2012) Thermo– 42. Olovsson L, Simonsson K, Unosson M (2005) Selective mass scal-
hydro-mechanical–chemical processes in fractured porous media. ing for explicit finite element analyses. Int J Numer Methods Eng
Springer 63:1436–1445
23. Wang W, Kolditz O (2004) Object-oriented finite element analysis 43. Davies RJ, Mathias S, Moss J, Hustoft S, Newport L (2012)
of thermos-hydro-mechanical (THM) problems in porous media. Hydraulic fractures: how far can they go? Mar Pet Geol 37:1–6
Int J Numer Eng 69:162–201 44. Peric D, Hochard CH, Dutko M, Owen DRJ (1996) Transfer oper-
24. ELFEN TGR manual (2014) Rockfield Software Ltd ators for evolving meshes in small strain elasto-plasticity. Comput
25. Kim J (2010) Sequential methods for coupled geomechanics and Methods Appl Mech Eng 137:331–344
multiphase. PhD Thesis, Stanford University 45. Angus DA, Kendall JM, Fisher QJ, Segura JM, Skachkov S, Crook
26. Bai M, Elsworth D (2000) Coupled processes in subsurface defor- AJL, Dutko M (2010) Modelling microseismicity if a producing
mation, flow, and transport. ASCE Press, Reston reservoir from coupled fluid-flow and geomechanical simulation.
27. Diersch HJ (2013) FEFLOW: finite element modeling of flow, mass Geophys Prospect 58:901–914
and heat transport in porous and fractured media. Springer, Berlin 46. Hazzard J, Young R (2002) Moment tensors and micromechanical
28. Lobao MC (2007) Finite element modelling of hydraulic fracture models. Tectonophysics 356:181–197
flow in porous media. PhD Thesis, Swansea University 47. Angus DA, Dutko M, Kristiansen TG, Fisher QJ, Kendall J-M,
29. Valko P, Economides MJ (1995) Hydraulic fracture mechanics. Baird AF, Verdon JP, Barkved OI, Yu J, Zhao S (2015) Integrated
Wiley, London hydro-mechanical and seismic modelling of the Valhall reservoir:
30. Lewis RW, Schrefler JZ (1998) The finite element method in the a case study of predicting subsidence, AVOA and microseismicity.
static and dynamic deformation and consolidation of porous media. Geomech Energy Environ 2:32–44
Wiley, New York 48. Wriggers P (2006) Computational contact mechanics. Springer,
31. Klerck PA (2000) The finite element modelling of discrete fracture Heidelberg
in quasi-brittle materials. PhD Thesis, Swansea University 49. Zhong ZH (1993) Finite element procedures for contact-impact
32. Bazant PZ, Planas J (1997) Fracture size effect in concrete and problems. Oxford University Press, New York
other quasi-brittle materials. CRC Press 50. Soliman MY, Augustine J (2010) Fracturing design aimed at
33. Crook AJL, Willson SM, Yu JG, Owen DRJ (2006) Predictive enhancing fracture complexity. In: Proceedings of the annual con-
modelling of structure evolution in sandbox experiments. J Struct ference, Barcelona, Spain, 14–17 June, SPE 130043
Geol 28:729–744 51. Medeiros F, Ozkan E, Kazemi H (2008) Productivity and drainage
34. Crook T, Willson S, Yu JG, Owen R (2003) Computational area of fractured horizontal wells in tight gas reservoirs. In:
modelling of the localized deformation associated with borehole Proceedings of the Rocky Mountain oil and gas technology sym-
breakout in quasi-brittle materials. J Pet Sci Eng 38:177–186 posium, Denver, Colorado, USA, 16–18 April, SPE 108110
35. Bai M (2011) Improved understanding of fracturing tight-shale gas
formations. In: Proceedings of the production and operation sym-
posium, Oklahoma City, Oklahoma, 27–29 March, SPE 140968
36. Williams BB (1970) Fluid loss from hydraulically induced frac-
tures. J Pet Technol 22:882–888

123