See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/320557135

Investigating the scale-dependency of the

geometrical and mechanical properties of a
moderately jointed rock...

Article in Computers and Geotechnics · October 2017

DOI: 10.1016/j.compgeo.2017.10.002

CITATIONS READS

0 347

4 authors:

Kiarash Farahmand Ioannis Vazaios

Golder Assocaites Queen's University
13 PUBLICATIONS 9 CITATIONS 21 PUBLICATIONS 15 CITATIONS

SEE PROFILE SEE PROFILE

Mark S. Diederichs Nicholas Vlachopoulos

Queen's University Royal Military College of Canada
189 PUBLICATIONS 2,696 CITATIONS 95 PUBLICATIONS 399 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Tunnelling Research - Analysis and Design View project

The Use and Design of Geosynthetics within Geotechnical Engineering Applications View project

All content following this page was uploaded by Kiarash Farahmand on 23 October 2017.

The user has requested enhancement of the downloaded file.

 Computers and Geotechnics xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Computers and Geotechnics

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

Research Paper

Investigating the scale-dependency of the geometrical and mechanical

properties of a moderately jointed rock using a synthetic rock mass (SRM)
approach
⁎
K. Farahmand , I. Vazaios, M.S. Diederichs, N. Vlachopoulos
Department of Geological Sciences and Geological Engineering, Queen’s University, Kingston, ON, Canada

A R T I C L E I N F O A B S T R A C T

Keywords: A synthetic rock mass (SRM) model coupling discrete fracture networks (DFNs) and a discrete element grain-
Synthetic rock mass (SRM) based model (DEM) is used to characterize the mechanical properties of moderately jointed rockmasses under
Discrete fracture networks (DFN) confined and unconfined conditions. The scale dependency of the rockmass properties is investigated using the
Scale-dependency rockmass strength and concept of representative element volume (REV). The numerical results are compared with the estimated values
deformability
from empirical methods. It is determined that the empirical Hoek-Brown criterion correctly estimates the un-
Representative elementary volume (REV)
confined strength of the rockmass with non-persistent joints, while it overestimates the strength under confined
LiDAR
conditions. An S-shaped strength envelope is obtained from the numerical results.

1. Introduction tests to back analyze the rockmass properties, such as plate-loading

Safe and cost-effective design of the geotechnical projects, such as
tunnels, caverns, and mine drifts, is highly dependent on proper
knowledge of the strength and deformability parameters associated
with a given rockmass. Although tremendous efforts have been made to
determine deformability and strength parameters of rockmasses at large
scales, reliable characterization of such properties with an adequate
level of confidence has still remained a challenging issue for rock en-
gineers. The challenge mainly originates from the highly heterogeneous
and discontinuum nature of jointed rocks due to the existence of dis-
tributed discontinuities of various sizes from minute cracks within the
mineral grains to meso and macro-scale features, such as veins, joints
and faults. As a result, the mechanical properties of a rockmass are
controlled by both the behaviour of pre-existing discontinuities and the
stress-fracture behaviour of intact blocks making up the rock bridges
between discontinuities. On the other hand, given the presence of dis-
continuities of various nature and sizes, together with non-linear be-
haviour of intact blocks, the deformability and strength of a specific
rock is highly dependent on the scale of the rockmass being studied.
Various techniques have been applied to measure the mechanical
properties of jointed rock at a representative volume. The most com-
monly used techniques are in situ field tests, closed-form and analytical
solutions, laboratory physical models, empirical methods, back analysis
techniques, and 2D and 3D numerical simulations. Large scale in situ
tests to determine deformability and block shear for strength properties,
are usually time-consuming, expensive to perform and are difficult in
controlling the size of the influenced zone [1]. Analytical solutions (e.g.
[2]), usually fail to capture complicated interactions between blocks
and joints, and their application is only limited to selected simple joint
system geometries (e.g. orthogonal sets of persistent joints) with simple
constitutive behaviour of joints. As a result, these methods might not be
able to estimate with enough accuracy the rockmass modulus and
strength of more complex joint geometries, which are subjected to
various stress paths. It is also debatable as to whether or not generalize
the behaviour of the small-scale physical models of jointed rock to that
of the large-scale rockmass [3,4]. On the other hand, widely used em-
pirical methods [5–8] which infer the rockmass properties from rock
mass classification (e.g. GSI, Q, RMR) often suggest conservative values
of rockmass properties which could result in the overdesign of a project,
especially for the case of massive and moderately jointed rockmasses
(GSI > 65) [9–13]. These methods are mainly developed from ob-
servations of rockmass behaviour near excavation walls with low con-
finement, hence failing to anticipate the confined strength of moder-
ately jointed rock. In addition, the assumption that the prevailing
rockmass characteristics can be described with a unique index that can
anticipate the behaviour of a structurally complex rockmass is deba-
table [11,14,15]. Back analysis based on in situ measurements is only
feasible during or after construction of projects and it does not provide

⁎
Corresponding author at: Department of Geological Sciences and Geological Engineering, Queen’s University, Bruce Wing/Miller Hall 36 Union Street, Queen's University, Kingston,
ON K7L 3N6, Canada.
E-mail address: kiarash.farahmand@queensu.ca (K. Farahmand).

http://dx.doi.org/10.1016/j.compgeo.2017.10.002
Received 31 July 2017; Received in revised form 26 September 2017; Accepted 5 October 2017
0266-352X/ © 2017 Elsevier Ltd. All rights reserved.

Please cite this article as: Farahmand, K., Computers and Geotechnics (2017), http://dx.doi.org/10.1016/j.compgeo.2017.10.002
K. Farahmand et al.
Fig. 1. Scale-dependency of rockmass properties and the concept of Representative
Elementary Volume (REV).
strength and deformability parameters in the feasibility and design
stages of geotechnical professional practice.
Among these rockmass evaluation techniques, numerical modelling
methods, especially discrete element models (DEM), have been proven
to be an alternative approach to derive the rock strength due to its
ability to simulate complex joint system geometries and non-linear
behaviours of both the intact rock and the discontinuities
[13,14,16–21]. Moreover, the scale-dependency of modulus and
strength of jointed rock, due to the presence of discontinuities of var-
ious sizes, can be systematically investigated using a series of numerical
tests. Therefore, the numerical simulation can be employed to estimate
the large-scale rockmass properties using an upscaling process. The
upscaling procedure involves estimating variation of the mechanical
properties with increasing the size of the examined rock volumes up to
a representative elementary volume (REV) in which stationary limits
are reached for the properties under study [22]. Fig. 1 demonstrates the
concept of REV for rockmass properties.
In this paper, the rockmass strength and its scale-dependency are
characterized using a hybrid discrete fracture network (DFN) – grain-
based discrete element method (GDEM). The discrete fracture network
(DFN) is employed to generate the geometry of the pre-existing joints
based on field data collected by LiDAR (Light Detection and Ranging)
scanning. Afterwards, the distinct element code UDEC [23] is used to
simulate the response of the rockmass to the mechanical loading. With
this method, the rock failure is captured either in terms of fracturing of
the rock matrix or displacements (i.e. sliding, opening) of the pre-ex-
isting discontinuities. To simulate initiation and growth of fractures
through intact material, the model domain is discretized into fine
polygonal blocks using a Voronoi tessellation algorithm [24]. The grain
and the grain boundaries respectively represent the rock-forming mi-
nerals and the flaws in the intact rock. As a result, the micro-fracturing
process can be simulated explicitly by development of cracks along
grain boundaries when stresses acting on the interface between the
Voronoi blocks exceed their tensile or shear strengths.
A new cohesive crack constitutive model is implemented into the
UDEC code to control the behaviour of the grain-grain bonds under
2
Computers and Geotechnics xxx (xxxx) xxx–xxx
compression, tension and shear conditions. Implementation of the crack
model in the grain-based simulator aims to enhance the capability of
the UDEC-Voronoi scheme to simulate more realistic micro-cracking
mechanisms, which are similar to the micro-cracking mechanisms ob-
served in physical laboratory experiments. A systematic calibration
procedure is followed to select a suitable set of micro-parameter values
for the model components that simultaneously yields the correct macro-
parameters of deformability and strength as well as the observed frac-
ture response of the intact Lac du Bonnet granite.
In order to create the samples of synthetic rock mass (SRM), DFNs
are constructed first based on joint geometry data (orientation, joint
trace intensity) extracted from LiDAR scanning of the unlined
Brockville Tunnel in Ontario, Canada [25]. The mapped geometrical
data of the joints are used to generate a 60 m × 60 m × 60 m DFN
model. To investigate the scale-dependency of the rockmass properties,
initially, a number of 2D rectangular models of different heights from
0.5 m to 20 m are extracted from this large-cubic DFN model. The
sample specimens are then incorporated into the 2D distinct element
code UDEC to create SRM models.
The calculated elastic deformation and strength properties of the
specimens are then compared with predicted values obtained from the
Hoek-Brown failure envelope [8]. The strength parameters of the Hoek-
Brown criterion are calculated based on values of the Geological
Strength Index (GSI) estimated from blockiness and joint condition
observed in the Brockville Tunnel rock exposure. Additionally, the ef-
fects of increasing confining stress on deformation and strength of
biaxial models are investigated. The applicability of the Hoek-Brown
failure criterion and the GSI system for estimating the unconfined and
confined strength of moderately jointed rock is investigated. Finally,
the simulated failure pattern (depth and extension of failure) forming in
the excavation damage zone (EDZ) around a deep tunnel is compared
when continuum and discontinuum modelling approaches are used.
2. Creation of the discrete fracture network
The use of LiDAR scans, and other remote sending techniques such
as photogrammetry, can assist with the field data collection by enabling
the user to have full access to the exposed rockmass. Field work in-
volves the collection of scanned surfaces using the laser scanner, which
minimizes the time at site, resulting in the creation of a cloud of points
relative to the position of the scanner. After the data has been collected,
3D models of the exposed rockmass can be created by meshing the
point-cloud in order to create digital 3D surfaces. This allows the
geologist or engineer to map the rockmass joints within the 3D surface
model, and therefore have access anywhere the rockmass is exposed
[26–29]. However, while LiDAR can assist with the access to the
rockmass face, it is still limited to the amount of the exposed rockmass
surface.
In such a case, discrete fracture network (DFN) modelling can be of
great value as large-scale 2D and 3D models of joints can be created
[30–35]. The main advantage of DFN modelling is that many models
can be created based on different scenarios for a wide range of input
parameters. This can provide a great insight for the investigation of
rockmasses as the variability of the examined geomaterials and their
effect on the undertaken projects can be taken into account. This can
potentially enhance the modelling taking place at a larger scale and
therefore provide assistance during the design process. When DFN
models are used, orientation, number of joint sets, joint density, and
joint size serve as some of the most important input parameters, since in
reality they control the geometrical characteristics of the natural frac-
ture networks. These quantities can be measured directly in the field by
employing traditional mapping techniques or virtual mapping within
the datasets obtained from laser scanning. DFN modelling provides
many advantages; however, the use of them has limitations which lay
both in the selection of the input parameters and the modelling ap-
proach adopted. Therefore, the user needs to ensure that the generation
K. Farahmand et al. Computers and Geotechnics xxx (xxxx) xxx–xxx

Table 1
Discontinuity orientations of the dominant sets of the examined tunnel section (Modified
after [37]). The dip, dip direction and the constant K of the Fisher distribution are listed
for the major joint planes.

Set No. of entries Dip (°) Dip direction (°) K (Fisher)

1 103 88 285 86.50

2 101 12 077 43.55
3 58 77 207 48.58

Total 262/402

Table 2
Fracture trace measurements within the LiDAR data for the 10-m section of the Brockville
Tunnel and derived DFN modelling parameters (Modified after [37]).

Set Mapping area A No of Total length ltot Fracture intensity P21

(m2) traces (m) (m/m2)

1 60 65 71.03 1.18
2 54 83 91.61 1.70
3 54 71 62.93 1.17

Fisher constant K for each joint set separately, as listed in Table 1.

Fig. 2. (a) “Virtual” mapping of joint surfaces within the examined 10-m section of the
Brockville Tunnel. The different joint surfaces are highlighted with a different colour. (b)
“Virtual” mapping of the joint traces on the right wall of this specific section. The mapped
traces are highlighted red.
algorithms employed are able to create geologically realistic and ac-
curate DFN models which correspond to the in situ conditions ex-
amined.
In this section, a summary of the selection of the joint input para-
meters, based on LiDAR data obtained from the Brockville Tunnel, and
the DFN modelling approach employed is described. Based on the ac-
quired geometrical data from the performed DFN modelling, SRM
models are created by combining the DFN model with a grain-based
model in UDEC in order to simulate moderately jointed rockmasses.
2.1. Selection of the input parameters
In order to acquire the required discontinuity data to serve as input
parameters for the DFN generation process, LiDAR scanning was em-
ployed. Polyworks [36], which is a general purpose software package
for range-image processing, was used in order to process and assess a 10
m long 3D surface model of the Brockville Tunnel (Fig. 2). The obtained
input parameters include discontinuity orientation and fracture in-
tensity, which were measured directly from the LiDAR data by ex-
amining joint surfaces and joint traces on the tunnel walls and roof. The
measured quantities are summarized in Tables 1 and 2 [37]. It has to be
noted, that the measured areal fracture intensity P21 is defined as the
total trace length per mapped area, as defined by Dershowitz and Herda
[38].
Regarding the mapped joint traces, they serve a dual purpose as input
parameters in the DFN generation process. Their first purpose is to
provide the necessary information to determine the size of the joints
within the model, based on the measured fracture intensity. Random
seeds within a specific domain are created and the employed propa-
gation algorithm creates joint surfaces by using the 2D obtained frac-
ture intensity in order to determine the size of the joint surfaces in 3D
[40]. Their second purpose is to create deterministic joints at the lo-
cations where specific joint traces have been mapped within the ex-
amined area by fitting 3D surfaces on the 2D traces. Therefore, joint
trace measurements are used to create both deterministic and stochastic
joints during the generation process.
The aforementioned 3D surfaces created to represent the joints
within the DFN model consist of a triangulated mesh, which requires
the determination of a few more parameters, apart from the established
joint input parameters, in order to satisfy the applied mathematical
model. In order to achieve this, a calibration process is employed in
order to achieve a match between mapped joints in situ and the de-
terministic simulated joints at the exact locations, as described in [37].
This process is iterative and can be summarized in the following steps:
1. Generation of a deterministic DFN model (Fig. 3a).
2. Extraction of deterministic joint traces from the simulated data
(Fig. 3b).
3. Visual comparison between simulated joint traces and observed in
situ joint traces for a given mapping plane (Fig. 3b).
4. Steps 1–3 are repeated until a match between the deterministic and
mapped joint traces has been achieved.
The calibration process ensures that the parameters required for the
generated triangular mesh forming the joints of the DFN model result in
simulated deterministic joints that match the ones observed and
mapped in situ at specific locations. These joints form the deterministic
part of the DFN model.
2.3. Validation of the DFN model

2.2. Generation of the DFN model
After the required joint input parameters for the generation of the
DFN model were determined, the DFN generator MoFrac [39] was used
for the creation of the DFN models. In order to simulate the joint or-
ientation variability, the Fisher distribution is employed by using the
In order to ensure that numerical artifacts will not interfere with the
model, a large volume and number of realizations need to be employed
in order to eliminate possible boundary effects and the effect of the
stochastic processes taking place. To this end, for the purposes of this
study, a volume of 60 m × 60 m × 60 m was utilized, and 30 different
realizations were created (Fig. 3d). However, the size of the applied

3
K. Farahmand et al. Computers and Geotechnics xxx (xxxx) xxx–xxx

Fig. 3. Generation of the DFN model for the Brockville Tunnel. (a) “Virtual mapping” of the joint traces on the roof of the examined tunnel section on the left, and the extracted joint
traces (right); (b) Deterministic discontinuities generated using MoFrac for Set 1; (c) Joint traces derived from the DFN model overlap the mapped fracture traces (black); (d) Generated
stochastic DFN model for a volume of interest 60 × 60 × 60 m with all joint sets present (Set 1: red, Set 2: blue, and Set 3: yellow) (Modified after [37]). (For interpretation of the
references to colour in this figure legend, the reader is referred to the web version of this article.)

volume of interest and the number of realizations have to ensure that
the required computational time will be within a practical range and
optimization is required [37]. From the total number of the 30 DFN
models generated, one model that was assumed to be the most re-
presentative of the field conditions [37] was selected and processed
further in order to be used in this study.
Another important part of the generation process of a DFN model is
to ensure that the generated stochastic joints correspond to the rock-
mass conditions observed in situ. Therefore, the model must be vali-
dated prior to any further use; otherwise another DFN model has to be
generated until the in situ conditions are met. In order to achieve this,
multiple cross-sections from the generated DFN were extracted (Fig. 4a)
and mapping windows of the same size as the employed virtual map-
ping windows were used. At this point it has to be noted that the
employed mapping windows are created in areas of the DFN that sto-
chastic joints have been generated. As observed in Fig. 4b for Set 1, the
joint trace distribution of the virtually mapped traces in the LiDAR data
matches the joint trace distributions obtained from the stochastic joints
of the DFN model. In a similar fashion, this process was repeated for
Sets 2 and 3 in order to validate the complete DFN model.
3. Generation of the grain-based model
The intact part of the SRM model is represented as an assembly of
polygonal Voronoi grains. As a result, the fracturing of the intact rock
can be simulated by development of cracks along grain boundaries. In
this study, the well-established properties of Lac du Bonnet (LdB)
granite sampled from the 240 m level of the underground Research

Fig. 4. (a) Cross-section from the DFN model with the different joint set traces being highlighted (Set 1: red, Set 2: blue, Set 3: yellow) and (b) joint trace length distributions for Set 1
based on the LiDAR data (red) and the DFN data (black is the average from 10 different mapping windows). (For interpretation of the references to colour in this figure legend, the reader
is referred to the web version of this article.)

4
K. Farahmand et al. Computers and Geotechnics xxx (xxxx) xxx–xxx

Table 3
Micro-material properties for different minerals of Lac du Bonnet.

Mineral phase Abundance (%) Grain size (mm) Density, ρgr (kg/m3 ) Elastic modulus, Egr (GPa) Poisson’s ratio, νgr (–) Mode I fracture toughness, KIC (MPa m m)

Alkali-feldspar 40 3 2560 96.8 0.28 4.18 ± 0.09

P-feldspar 20 3.5 2630 88.1 0.26 4.18 ± 0.09
Quartz 30 1.5 2650 94.5 0.08 8.68 ± 0.18
Biotite 10 0.75 3050 33.8 0.36 3.21 ± 0.05

P-feldspar refers to plagioclase feldspar.

Values of Young’s modulus for different minerals are given from [41].
Values of Young’s modulus, density and Poisson’s ratio for Quartz are given from [42].
Values of KIC for different minerals are given from [43].

histogram in Fig. 5 compares the distribution of various mineral phases

in both real granite and the intact SRM model.

3.1. Cohesive crack model describing the behaviour of the grain

The fracturing behaviour of the contacts is controlled by im-

Fig. 5. Model geometry for UCS testing. Grains with purple, light red, light brown, and
dark red represent K-feldspar, plagioclase feldspar, quartz, and biotite minerals, respec-
tively; and a histogram shows the abundance of different mineral phases in real LdB
granite and the SRM model. (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)
Laboratory in Pinawa, Manitoba, Canada [9], are used to simulate the
intact part of the SRM model forming between DFN joints. The de-
formation properties and density of the common minerals composing
the rock are listed in Table 3. In order to introduce the grain-scale
material heterogeneity of the granite into the model, four material grain
types are distributed in the models based on the abundance of each
mineral type in the composition of the actual rock (as given in Table 3).
An average grain size of 3 cm was selected to generate the models. The
plementing a constitutive model in UDEC, which is referred to as the
cohesive crack model [44]. The cohesive contact binds neighboring
grains together with shear and tensile resistance at the boundary con-
tact [45]. This cohesive model can be regarded as a combination of
those proposed in the literature [45–47], while a bi-linear shear failure
envelope is added to represent the shear strength of the contact during
shearing. According to the cohesive model, contacts may break in
tension (Mode I), shear (Mode II), or a mixed mode I-II depending on
the local stress and relative contact wall displacements. Non-linear
branches of the stress-displacement curves, as illustrated in Fig. 6, de-
fine the behaviour of the contact before breakage, followed by the post-
peak branch representing the softening of the material due to damage
accumulation and the resultant energy dissipation. The non-linear pre-
yield branch is intended to represent the decay of stiffness in the pre-
failure state due to the progression of damage in the contact interface.
According to this cohesive model, the contact boundary between
grains yields when the relative contact displacement reaches critical
values of crack opening, up (Mode I), and shear displacement, s p (Mode
II). Displacements at peak strength are evaluated as
Tc f
up = e. , and s p = e. sh
kn,c k sh0,c (1)
where e= exp(1) = 2.718281 is the base of the natural logarithm, kn0,c
and k sh0,c represent the contact initial normal stiffness and shear stiff-
ness, respectively, and Tc and f sh are the strengths of the contact in
tension (Mode I) and shear (Mode II), respectively.
The peak shear strength of the grain contact (f sh) is determined
based on the following bi-linear criterion (as demonstrated in Fig. 7):
⎧|f sh| = cc + σn. tan(φc ) σn < σn,Transitional point
⎨|f sh| = cc,2nd + σn. tan(φc,2nd ) σn ⩾ σn,Transitional point (2)
⎩
where cc and cc,2nd are the primary and secondary contact cohesions,
while φc and φc,2nd are the primary and secondary contact friction

Fig. 6. Stress-displacement behaviour of the cohesive contact model in (a) Mode I, (b) Mode II, and (c) mixed-Mode I-II.

5
K. Farahmand et al.
Fig. 7. Bilinear failure criterion with residual strength and tension cut-off to define the
strength properties of the contacts in shear and tension. Note the purple-colour bi-linear
envelope which defines the peak shear strength of the contact, while the red-colour en-
velope defines the residual strength of the contact after yielding. (For interpretation of the
references to colour in this figure legend, the reader is referred to the web version of this
article.)
angles, respectively, and σn is the normal stress acting across the contact
length. The parameter σn,Transitional point is the normal stress at which the
slope of the shear failure envelope transitions from a steeper to a
shallower slope. Results from laboratory testing on various types of the
rocks [48] suggest that the transition from a steep to shallow slope
occurs at about σn = 30 MPa. For the case of Äspo diorite with ap-
proximately similar mechanical properties (UCSi = 219 MPa, Ti = 15
MPa, Ei = 68 GPa, νi = 0.24) to the LdB granite, the transition from a
steep to shallow slope occurs at about σn = 30 MPa with the friction
angle φc = 70° for σn < 30 MPa and φc,2nd = 18° for σn > 30 MPa . The
estimated friction angles for the two regimes suggest a ratio of φc / φc,2nd
equal to approximately 4 [48]. After shear yielding, the contact
strength drops to its residual value f sh,res .
Upon exceeding the critical peak strength (in Mode I, Mode II, or
mixed-Mode), the contact between grains will break. In the post-yield
state (softening branch), the following stress-displacement (traction-
separation) which links the stress transmitted by the contact to the
contact displacement in tensile and shear states is used [46]:
σ Tc
⎡ τ n ⎤ = χ(Di). ⎡ ⎤
⎣ ⎦ ⎣ f sh ⎥
⎢ ⎦
where σn and τ represent the normal and shear stresses acting across the
contact length, respectively.
In Eq. (4), χ(Di) is a softening function that defines the decay of
contact strength in the post-yield regions. Evans and Marathe [49]
formulated the following softening equation based on experimental
results:
A + B−1 A + C × B ⎞⎤
χ (Di ) = ⎡1− ·exp ⎛Di ⎜ ·[A (1−Di ) + B (1−Di )C ]
⎟
⎢ A + B ⎝ A + B (1−A−B ) ⎠ ⎥
⎣ ⎦
(4)
where A, B, and C are experimentally determined constants equal to
0.63, 1.8, and 6.0, respectively; and Di ( i= I,II,I−II) is a damage vari-
able with a value between 0 and 1. This equation defines the shape
(slope and curvature) of the softening branch of the stress-displacement
curves in both Mode I and Mode II of fracturing.
The damage variable for a contact subjected to Mode I [46], Mode
II, and mixed-Mode I–II [47], displacements are defined, respectively,
as
2 2
u−up s−s p ⎛ u−up ⎞ + ⎛ s−s p ⎞
DI = ,DII = , and DI−II = ⎜ ⎟ ⎜
ures − up sres − sp ⎝ ures−up ⎠ ⎝ sres−s p ⎠
where u and s are the values of contact relative displacement in normal
Computers and Geotechnics xxx (xxxx) xxx–xxx
and shear directions, respectively. By substituting the damage variables
in Eq. (4), the shape of the softening branch of the stress-displacement
curves in both Mode I and Mode II can be achieved.
The values of ures and sres in Eqs. (6) and (7) can be calculated ac-
cording to following relations:
3GIc 3GIIc
ures = up + , and sres = s p +
Tc f sh (6)
where GIc and GIIc are defined as the energy that is required to extend
the crack surface of a unit area in tension and shear modes, respectively
[47]. The GIc in the plane strain condition can be estimated according
to the following equation proposed by [50]:
K2Ic
GIc = . (1−ν2gr )
Egr (7)
where KIc is fracture toughness of crack in Mode I, and Egr and ν gr are
the Young’s modulus and Poisson’s ratio of the material surrounding the
crack. In this study, the value of GIIc is considered to be 2 times the
value of GIc . Additionally, the behaviour of the contacts under com-
pressive normal loading is represented by the hyperbolic function
suggested by [51].
3.2. Calibration of the grain-based model
In the micro mechanical-based models such as DEM-Voronoi, it is
the properties of the micro-constituents (grains and their contacts) of
the system that controls the macroscopic mechanical behaviour of the
model. As a result, calibration of micro-parameters is a primary re-
quirement for GBM modelling.
In this paper, a systematic calibration procedure, as outlined in the
flowchart presented in Fig. 8, is followed in order to determine a set of
values for microscopic model parameters enabling the synthesized rock
to replicate the macroscopic characteristic of the material of interest.
The calibration procedure presented herein is an iterative process that
involves running a series of Brazilian indirect tensile, unconfined
compressive strength (UCS), and biaxial compressive strength simula-
tions to determine a set of input parameters that yields a mechanical
response of the tested rock that matches physical laboratory test be-
haviour. The step-by-step calibration process is discussed in detail in
[52]. The calibration process begins with assigning the exact mineral
(3) deformability properties to the model grains based on reported data in
the literature (e.g. [41]). Next, adjustments to the contact stiffness
values (kn0,c and k sh0,c) continue until the desired Young’s modulus and
Poisson’s ratio for the material are achieved. The strength properties
assigned to contacts must be adjusted until the stress-strain curves for
the simulated Brazilian and UCS tests resemble the laboratory-derived
curves. By doing so, the model simultaneously yields the correct macro-
mechanical values for Ei , νi , Ti , UCSi , CI (crack initiation threshold
[53]), CD (crack damage threshold [53]). At this stage, the simulated
fracture patterns and failure mode must be inspected to be consistent
with those observed in laboratory samples. After that, a series of biaxial
test simulations under varying confining stresses are be modelled to
adjust the contact friction angle, φc . The iterative refinement of the
values of the Cc , φc and Tc micro-parameters must be carried out until
the simulated biaxial strength envelope and the resultant macro-me-
chanical Ci and φi values match those obtained from biaxial laboratory
tests. Finally, the emergent macro-mechanical strength and deform-
ability of the model using UCS and Brazilian simulations must be
checked again to confirm that they remain in agreement with the ex-
perimental values.
Since the models are composed of four mineral phases, ten contact
types must be calibrated. Calibrated contact properties for the ten dif-
⎟
ferent interfaces of the model are given in Table 4. Figs. 9 and 10 show
(5)
the resultant stress–strain curves of the Brazilian and unconfined
compression tests, respectively. The strength properties determined by
6
K. Farahmand et al. Computers and Geotechnics xxx (xxxx) xxx–xxx

Fig. 8. Flow chart illustrating the calibration procedure for

the combined DFN-grain based models.

the model such as tensile strength, compressive peak strength, crack
damage, and crack initiation thresholds demonstrate a very good
agreement with those of the experimental data (Table 5).
4. Numerical results
In order to examine the scale-dependency of mechanical properties
of a moderately jointed rock, multiple SRM models of various sizes were
created based on the calibrated DFN and grain-based intact models. The
structural features extracted from the DFN simulation are introduced
into the UDEC-Voronoi model as illustrated in Fig. 11. The shear be-
haviour of the joints in the synthesized rockmass is controlled by a
Coulomb slip model with residual strength. The residual strength si-
mulates displacement-weakening of the joints by loss of cohesion and
friction at the onset of shear yielding. The properties of assigned to the
joints are based on the laboratory tests results reported by [55]. Ac-
cording to the laboratory testing results, the joints are simulated as
frictional features with no cohesion or tensile strength. The friction
angle, dilation angle, initial normal stiffness, and shear stiffness of
joints are defined as 25°, 5°, 434 GPa/m, and 434 GPa/m, respectively,
and the critical shear displacement for dilation is assumed to be 3 mm
[55]. A hyperbolic function proposed by [51] is used to simulate the
normal stress-normal displacement (closure) response of the joints
under compressive loading condition. The maximum joint closure is
assumed be to be equal to 5 m [16]. In the following section, the results
of the numerical simulations including uniaxial and biaxial compres-
sion testing are discussed. The calculated mechanical properties are
compared to the properties obtained from a rockmass characterization
system approach based on GSI.
4.1. The scale-dependency of mechanical properties
A series of compression tests under unconfined conditions were si-
mulated on 60 UCS specimens with heights ranging from 1 m to 10 m,

Table 4
Calibrated micro-mechanical properties for contacts.

Contact type Tc,Tc,res (MPa) Cc (MPa) φc,φc,res (degree) kn0,c (GPa/m) k sh,c/kn0,c (–) GIc (J/m2 )

Feld/Feld 35.0, 0.0 95.0 65, 15 12,000 0.67 286.0

Feld/Feld 35.0, 0.0 95.0 65, 15 12,000 0.67 286.0
Feld/Quartz 28.2, 0.0 82.5 58, 15 12,000 0.67 229.0
Feld/Biotite 22.4, 0.0 55.0 51, 15 12,000 0.67 172.0
Feld/Feld 35.0, 0.0 95.0 65, 15 12,000 0.67 286.0
Feld/Quartz 28.2, 0.0 82.5 58, 15 12,000 0.67 229.0
Feld/Biotite 22.4, 0.0 55.0 51, 15 12,000 0.67 172.0
Quartz/Quartz 35.0, 0.0 95.0 65, 15 12,000 0.67 901.0
Quartz/Biotite 22.4, 0.0 55.0 51, 15 12,000 0.67 541.0
Biotite/Biotite 25.3, 0.0 45.0 45, 15 12,000 0.67 521.0

a
The fracture energy release of the contact is assumed to be proportional to its tensile strength.
Therefore GFeld/Quartz
IC = GFeld/Feld
IC
Feld/Quartz Feld/Feld
. (TC /TC )

7
K. Farahmand et al.
Fig. 9. Axial stress-strain curve obtained from the calibrated Brazilian specimen.
Fig. 10. Intact model response during uniaxial compression test; (a) Axial stress-Axial
strain curve and associated incremental accumulation of tensile cracks and shear cracks
(cumulative distribution of cracks); (b) Volumetric strain-Axial strain curve.
8
Computers and Geotechnics xxx (xxxx) xxx–xxx
and a constant height to width ratio of 2.5 to investigate the influence
of specimen size on the mechanical properties of SRM samples. The
mechanical REV size was determined based on the change in degree of
variability of unconfined strength (UCSrm ) and deformability (Erm, νrm)
of the SRM models. To load the sample in uniaxial boundary condition,
the SRM sample is subjected to a constant displacement rate to induce
stresses until failure is achieved. In order to achieve this, the rockmass
model is located between two stiff platens. The upper and lower platens
move toward each other with constant velocities to apply a compressive
load to two ends of the rock.
The axial stress-axial strain behaviour is monitored during testing
by using user-defined FISH functions embedded in UDEC [23]. Com-
pressive stresses are measured in upper platens and three domains
within the sample. Axial strain is calculated based on the displacement
of the two points at the very top and bottom of the specimens for the
compression test. In addition, 14 sets of measuring points along ex-
ternal lateral sides of the sample are selected to monitor the lateral
strain, and the average of the lateral strain in the monitoring points
represents the overall lateral strain of the rock under compression
(Fig. 11c).
The modelling results indicate significant variation of Young’s
modulus as a function of the model size. As illustrated in Fig. 12, the
variation of the Young’s modulus decreases as the specimen size in-
creases. In general, the scattering of stiffness of the rockmass decreases
with increasing the sample size, as the larger samples contain a larger
population of joints. For specimen heights of less than 3 m, the range of
the values is notably wider. However, the range of the emergent
Young’s modulus narrows significantly for sample heights larger than 7
m, indicating that the minimum REV of 7 m can be justified for the
deformability of this specific rockmass. In this study, a REV size for a
property is established, if the coefficient of variance (CoV) for a specific
property falls below 20%. The value of the CoV for the 7 m high spe-
cimen is approximately 15%, which is lower than the acceptable CoV
value of 20% for establishing the REV size. The CoV of specimens
higher than 7 m is also lower than 20%; hence the minimum REV size
based on the estimated Erm is 7 m. The rockmass Young’s modulus (Erm)
at the REV size varies between 50% to 72% of the intact Young’s
modulus of the granite with the mean value of Erm = 40 GPa.
In order to have an estimate of strength and deformability based on
an empirical approach, the Geological Strength Index (GSI) was used to
evaluate the rockmass from the observed structural features in the
Brockville Tunnel and the joint surface conditions based on data in the
literature [55]. In order to quantify the degree of blockiness of the
rockmass, three parameters including the RQD, joint spacing, and block
volume are considered [56,57], based on information collected in the
field using LiDAR scanning and the generated DFN geometries. Varia-
bility of these parameters were measured by examining the 3D master-
volume DFN and taking a large number of measurements in various
locations within the DFN. The use of the RQD, joint spacing, and block
volume to infer the GSI range provides a more objective quantitative
index to represent the rockmass condition. According to the data re-
ported in [55], it is assumed that the joint condition is fair with the
smooth, moderately weathered and altered joint surface. Details on the
procedure followed to measure these geometrical features can be found
in [37].
The measurements on the DFN model showed that the RQD is
within the range of 95–100%, while joint spacing varies approximately
from 0.50 m to 0.90 m. The blocks formed in the synthesized rockmass
have an average range of volumes from 0.23 m3 to 2.30 m3. Based on
the RQD-based GSI chart reported in [57] and given that the value for
the RQD ranges from 95% to 100%, a range of 67–77 is predicted for
the GSI according to Eq. (8):
GSI = 1.5JCond89 + RQD/2 (8)
In Eq. (8), RQD/2 is used to quantify the degree of blockiness and
the 1.5JCond89 gives a value for the joint surface condition. The value of
K. Farahmand et al. Computers and Geotechnics xxx (xxxx) xxx–xxx

Table 5
Comparison between experimental and simulation result for the LdB granite (values of
experimental data are from [54]).

Experimental value Model result

Young’s modulus (Ei ) (GPa) 69 ± 5.8 70.4

Poisson’s ratio (νi ) 0.22 ± 0.04 0.22
Crack initiation stress (CI) (MPa) 90 (41% of UCSi ) 92 (42% of UCSi )
Crack damage stress (CD) (MPa) 172 (86% of UCSi ) 160 (73% of UCSi )
Uniaxial comp. strength UCSi (MPa) 221 (201 ± 22) 220
Tensile strength (T) i (MPa) 9.3 ± 1.3 10
Cohesion (Ci ) (MPa) 30 27.5
Friction angle (φi ) (° ) 59 58
mi (–) 28–30 29

JCond89 can be determined by referring to [58]. In this study, a range

for JCond89 between 13 and 20 is considered for the surface condition of
the pre-existent joints.
However, when the rockmass blokiness degree is postulated on the
basis of mean joint spacing and block volume according to the GSI chart
proposed by [56], different values of GSI are obtained compared to the
Fig. 12. Young’s modulus obtained from the SRM simulation and the empirical Hoek-
GSI range estimated from the RQD-based chart. By using the Cai et al.
Diederichs equation for different specimen sizes.
[56] GSI chart and given that the mean joint spacing ranges between
0.50–0.90 m and the block volumes range is between 0.23 and 2.30 m3,
the estimated GSI ranges from approximately 50–65 and 50–67, re-
spectively.
Fig. 12 compares the Young’s modulus obtained for the SRM si-
mulations and the one calculated based on Eq. (9) [1]. A disturbance
factor of D = 0 is considered in order to estimate the Erm.

D
⎛ 1− 2 ⎞
Erm = Ei ⎜0.02 + ⎟
⎝ 1+e ( (60 + 15D) − GSI
11 ) ⎠ (9)

In general, the RQD-based GSI result in a larger stiffness for the

rockmass (considering GSI = 67–77), however, the joint-spacing-based
GSI and block volume-based GSI (considering GSI = 50–65) yield a
closer approximation to the Young’s modulus resulted from SRM si-
mulations. Note that Eq. (9) tends to overestimate the Young’s modulus
of the moderately jointed rockmasses (GSI > 65), whereas it under-
estimates the modulus of highly jointed rocks (GSI < 30). Fig. 13. Estimated Poisson’s ratio (νrm ) for the SRM samples of different sizes. The 2 m
The value of Poisson’s ratio (νrm ) does not significantly vary with the tall SRM sample with a pre-existing joint extending throughout the specimen is illustrated
size of the samples except for the 2 m sample, as shown in Fig. 13. The with an extreme strain ratio influenced by two steep discontinuities corrupting the lateral
strain reading. Each green dot represents the result of one simulation per specimen size.
calculated Poisson ratio for the 2 m sample is close to 0.8 which is not
(For interpretation of the references to colour in this figure legend, the reader is referred
mechanically sound as an elastic parameter for a pseudo-continuum. to the web version of this article.)
This measured lateral/axial strain ratio does not represent a true
Poisson's ratio but is the result of the presence of two joints which are

Fig. 11. (a) Samples of various sizes for the development of

the SRM models (white dotted rectangles) extracted from
the DFN model, and (b) a 10 m × 4 m SRM specimen
comprising the DFN joint and the intact material re-
presented by the DEM Voronoi grains. (c) Configuration of
the measuring points and zones for monitoring the stress
and strain within the sample. Black dots indicate the mon-
itoring points for the strain measurements. The stresses are
monitored within the Domains 1 (black box), 2 (green da-
shed box), and 3 (red dashed box). (For interpretation of the
references to colour in this figure legend, the reader is re-
ferred to the web version of this article.)

9
K. Farahmand et al.
Fig. 14. Unconfined compressive strengths (UCSrm ) obtained from the SRM simulations
and Hoek-Brown criterion for different specimen sizes. Input GSI parameter in the Hoek-
Brown criterion was estimated based on the average block volume, mean joint spacing
and RQD measured in the Brockville site, as well as joint condition data from [55].
extended throughout the model at a critical dip, so that the shearing of
these joints results in a high amount of lateral deformation and as a
result high value of νrm . Therefore, a Poisson’s ratio range approxi-
mately between 0.25 and 0.35 is recommended to be used as an
equivalent input parameter into continuum-based numerical models.
The results of calculated values of unconfined compressive strength
(UCSrm ) from multiple numbers of SRM models, at different sizes are
illustrated in Fig. 14. Similar to the stiffness (Fig. 12), the variance of
calculated UCSrm decreases as the specimen size increases, and the
UCSrm values maintain approximately constant ranges after a certain
sample size. At the sample height of 1 m, the UCSrm of the model ranges
from 5 MPa to 221 MPa depending on the population, orientation, and
persistence of the pre-existing joints in the model. This scattered data
implies that the model at this size cannot be described as statistically
homogeneous because a stable range of the properties cannot be
achieved. However, for the sample sizes higher than 7 m, the scattering
of results narrows to an average UCSrm of approximately 51 MPa. The
value of the CoV for the 7 m high specimen is approximately 19%,
which is lower than the acceptable CoV value of 20% for establishing
the REV size. The CoV of specimens higher than 7 m is also consistently
lower than 20%; hence establishing the REV at 7 m. The mean strength
of the models at REV size of 7 m × 2.8 m is approximately 23% of the
granite intact strength which is approximately 51 MPa.
Fig. 14 also compares the estimated unconfined rockmass strengths
obtained from SRM simulation and those determined from the gen-
eralized Hoek-Brown criterion with strength parameters determined
using RQD, joint-spacing, and block-volume-based GSI. The unconfined
strength of the rockmass is estimated using Eq. (10) [8]:
UCSrm = UCSi ·sa (10)
where UCSrm and UCSi are the unconfined strength of the jointed
rockmass and the unconfined strength of the intact rock, respectively.
The s and a are parameters which depend on the characteristics or
quality of the jointed rockmass, and can be calculated using Eqs. (11)
and (12), respectively [8]:
GSI−100 ⎞
s= exp ⎛
⎝ 9−3D ⎠ (11)
1 − GSI − 20
a= 0.5 +
6 (
e 15 −e 3 ) (12)
As can be seen, under unconfined conditions, the simulation pre-
dicts that the rockmass fails at about 43 MPa (at 7 m high sample size).
The joint spacing- and block volume-based GSI approaches
(50 ≤ GSI ≤ 67) consistently underestimated the unconfined strength
10
Computers and Geotechnics xxx (xxxx) xxx–xxx
Fig. 15. Areal fracture intensity (P21) as measured from the DFN model for various rec-
tangular specimen sizes (blue dots). The mean values for each specimen size (black line)
and the corresponding CoV values (purple line) are also illustrated. (For interpretation of
the references to colour in this figure legend, the reader is referred to the web version of
this article.)
of the rockmass by predicting UCSrm ranges from 24 MPa to 35 MPa.
This could be attributed to the underestimation of the GSI value when
the joint spacing and block volume are used to define the degree of
blockiness for the rockmass. A range of 35–63 MPa is determined for
the unconfined strength when the RQD index is used (67 ≤ GSI ≤ 77).
This implies that, the predicted strength value using the RQD to esti-
mate the GSI results in the mean strength value of 49 MPa which is
closer to the UCSrm values (∼43 MPa) obtained from the SRM simu-
lation. The Hoek-Brown criterion with the GSI range suggested by the
RQD values was able to successfully capture the variability in the un-
confined strength of the rockmass.
Additionally, the areal fracture intensity P21 (defined as the total
joint trace length divided by the mapping area) is measured from the
DFN model. As shown in Fig. 15, for specimen sizes larger than 7 m, the
CoV drops below 20% and a geometrical REV of 7 m is established. The
P21 at the REV size is within the range of 2–3 m/m2 with a mean value
of 2.5 m/m2.
The mechanical and geometrical properties used to determine the
REV sizes are listed in Table 6. The minimum REV sizes (REV = 7 m) is
used for the further analysis on anisotropy of strength, and failure en-
velope for the studied rockmass. In addition, the estimated rockmass
macro-mechanical properties (UCS, Young’s modulus, and Poisson’s
ratio) at the REV size are considered as equivalent continuum proper-
ties, and therefore, they can be used as input parameters into con-
tinuum-numerical codes [16,20].
Fig. 16 compares the final fracture patterns for the UCSrm models of
different sizes. In the 1 m high sample, accumulation of damage in the
direction sub-parallel to the applied loading axis creates macroscopic
fracture patterns in form of axial splitting (Fig. 16b). In this case, most
of the propagated micro-cracks are isolated cracks. These are not in-
itiated from the tips of the pre-existing joints and the coalescence of
these cracks leads to failure of the specimens. Therefore, the failure
mechanism of this specimen size is very similar to the failure of the
Table 6
Summary of the Representative Elementary Volume (REV) determination.
Property Mean value REV size (m)
Geometrical
Mechanical Young’s modulus, Erm (GPa)
2
Areal fracture intensity, P21 (m/m ) 2.5
40
7.0
7.0
Poisson’s ratio, νrm 0.35 2.5
Unconfined comp. strength, UCSrm 56 7.0
(MPa)
Minimum REV size 7.0
K. Farahmand et al.
intact granite reported in [53]. However, for the larger samples, stress
concentration at the tips of the joints causes the propagation and the
formation of wing cracks (Fig. 16a). The wing cracks propagate in the
direction of the maximum external load and as the number of induced
cracks increases by increasing the applied stress, cracks will begin to
coalesce and interact. These results are in agreement with experimental
observations discussed in [10,59,60] in which two types of cracks,
namely primary and secondary cracks, were observed during uniaxial
loading of brittle rocks, with primary (or wing) cracks appearing first.
These are tensile cracks that start at the tips of the flaws and propagate
in a curvilinear path as the load increases. Wing cracks grow in a stable
manner since an increase in load is necessary to extend the cracks and
align with the direction of the highest compressive load. Secondary
cracks, which initiate as shear cracks, appear later and are responsible
for sample failure. These shear cracks in most cases initiate in a
11
Computers and Geotechnics xxx (xxxx) xxx–xxx
Fig. 16. Comparison between fracture patterns of SRM
specimens and those observed in experimental tests. (a)
Fracture patterns in UCS specimens of different sizes at their
peak strength, and (b) Isolated micro-cracks observed in a
10 mm thin section of LdB granite compared with the iso-
lated micro-crack geometries formed parallel to the direc-
tion of maximum principal stress in the numerically testes
intact sample. (c) Comparison between the induced frac-
tures that occurred during uniaxial loading of marble sam-
ples with two pre-fabricated flaws [61] and a 7 m tall SRM
model. Note that the two coalescence cracks in marble are
marked by a pair of horizontal arrows. The extent of the
spalling zones that occurred between two pre-existing flaws
are shown in both the SRM model and marble sample.
direction coplanar to the flaw. The accumulated numbers of induced
cracks demonstrate that the majority of created cracks are tensile,
which suggests the overall mode of the sample failure can be considered
to be tensile failure.
4.2. Anisotropy of rockmass strength and deformability
It has been well recognized that mechanical behaviour of intact and
jointed rocks can be highly anisotropic [61]. In the case of a jointed
rockmass, it is the presence of joints distributed in preferential or-
ientations that cause the material to deform differently depending on
loading direction. In this section, the anisotropy of Young’s modulus
and strength under unconfined condition is investigated. A number of
rotated UCS specimens with a height of 7 m and a height to width ratio
of 2.5 were extracted from the master DFN model at three different
K. Farahmand et al.
orientations (Plane 1-(0°), Plane 2-(45°) and Plane 3-(90°) relative to
the alignment of the Brockville Tunnel), as illustrated in Fig. 3. The
unconfined strength and rockmass modulus values are shown in polar
diagrams in Fig. 17. The eccentricity of the polar plots is indicative of
the degree of anisotropy of the material behaviour. In the case of
Fig. 17a, the plot indicates that the unconfined strength is greatest and
least in the 0° and 45° directions, respectively, with its maximum and
minimum values of 55–4 MPa. Similarly, in Fig. 17b the rockmass
modulus has its lowest and highest values of 42 GPa and 15 GPa at 135°
and 0° angles, respectively. The very low UCSrm values obtained from
the 45-degree rotated sample (Fig. 17) is due to the presence of joints,
which are extended relatively throughout the model at a critical dip
with respect to the direction of the applied load. The shear slip along
these planes of weakness causes the samples to prematurely fail.
4.3. Determining the failure envelope for the moderately jointed rockmass
The inability of the Mohr-Coulomb and Hoek-Brown strength
12
Computers and Geotechnics xxx (xxxx) xxx–xxx
Fig. 17. Anisotropy of the rockmass properties with respect
to direction of loading for three different sampling planes;
(a) polar plot of UCSrm values for the rotated specimens; (b)
the resultant estimates of the rockmass modulus as a func-
tion of the direction of loading in degrees.
criteria to predict the maximum depth of breakout and the onset of
brittle failure in hard rocks is reported in the literature [9,10,62,63].
Both models tend to replicate an incorrect breakout shape and a smaller
depth of spalling due to the inadequate incorporation of a brittle da-
mage mechanism. Field observations (e.g. the ONKALO rock char-
acterization facility [64], the Niagara Falls Tunnel Project [65], and the
AECL-URL Test Tunnel [66]) indicate that progressive slabbing of near
excavation material is responsible for the formation of V-shaped not-
ches in the areas around the excavation with the highest magnitude of
tangential stress. It has been reported that spalling failure initiates at
excavation boundaries where the low confinement allows the cracks to
freely propagate and form parallel slabs. The onset of spalling corre-
sponds to 0.3–0.4 of the UCSi (crack initiation stress CI threshold) of
the intact rock measured in physical laboratory tests. The yielding of
the rockmass and the corresponding development of the notch stops
when the notch geometry provides sufficient confinement to stabilize
the process zone formed at the notch tip. In fact, the cracks are free to
propagate adjacent to the excavation where confinement is relatively
K. Farahmand et al.
Fig. 18. Schematics of the Generalized Hoek-Brown criterion (yellow curve) [8] and the
DISL model (the black curve represents the damage initiation envelope and the red curve
represents the spalling limit envelope) [11] for determining the in-situ strength of
rockmasses. The blue curve represents the strength envelope of the intact rock. (For in-
terpretation of the references to colour in this figure legend, the reader is referred to the
web version of this article.)
low. However, further growth of the cracks will be suppressed beyond a
certain depth due to reduction in deviatoric stresses at the tip of the
cracks. As a result, the rapid increase in the in situ rock strength with
increasing confinement can be attributed to the diminished ability of
cracks to grow under high normal stresses acting on their surfaces.
From a continuum mechanics point of view, this phenomenon can be
interpreted as a strength envelope with a shallow-slope first branch
indicating a low friction angle under low confinement condition. A
second branch with higher friction angle (steeper slope) is replicating
the influence of confinement which leads to an abrupt increase in the
rockmass strength.
Based on such field and laboratory characterization of brittle failure
mechanisms, Diederichs [11] developed a brittle Damage Initiation
Spalling Limit (DISL) constitutive model to be able to predict, with
mechanistic reasoning, the shape and depth of the yielded zones ob-
served in situ. This model defines a bilinear failure envelope to the
rockmass strength (Fig. 18). The first branch of the envelope accounts
for the unstable crack growth under low confinement conditions ex-
perienced by the near excavation material. The second branch of the
model, known as the spalling limit, captures the rapid increase in the
rock strength with increasing confinement. As a result, the sensitivity of
the material strength to confinement can be realistically modelled by
applying this approach.
In this section, the results of a series of simulated biaxial compres-
sion tests on a 7 m high sample are assessed for different ranges of
confining stresses in order to derive a strength envelope for the mod-
erately jointed rockmass. The stress-strain curves recorded for the
biaxial compression simulations are illustrated in Fig. 19 and the ob-
tained strength envelopes are plotted in Fig. 20.
In general, for the sparsely and moderately jointed rockmasses,
where the joints are discontinuous and not of sufficient length to create
individual blocks, the failure of the material can be mainly attributed to
tensile fracture of intact rock bridges initiating from the tips of pre-
13
Computers and Geotechnics xxx (xxxx) xxx–xxx
existing structures. The ability of these propagating fractures to grow
further is highly dependent on the level of confinement acting normal
to the surface of the fracture. Therefore, by increasing the external
confining stress (σ3) on the rockmass, a higher ratio of applied principal
stresses (σ1/σ3) is required for a fracture to be able to initiate and grow
through intact rock bridges. This dependency between the length of a
propagating fracture and the applied confinement is the main reason for
the increase in the strength of the jointed sample with increasing the
confining stress, as exhibited in Fig. 19.
The strength envelope as suggested by the Hoek-Brown criterion is
compared to the envelope obtained from the SRM modelling. The ob-
jective is to derive a failure envelope that is able to estimate the
strength of moderately jointed rock more accurately for hard rock-
masses (UCSi > 100 MPa) with GSI ranging from 65 to 75 and rough,
dilatant joints. As shown, the unconfined strength calculated from the
Hoek-Brown criterion is in good agreement with the strength predicted
from the SRM models. However, under confined conditions the pre-
dicted strength from the SRM models does not match the strength
predicted from the criterion. As previously mentioned, field observa-
tions suggest that the rock damage initiates at a lower stress state than
the stress suggested by the Hoek-Brown criterion. On the contrary, the
SRM estimated envelope is closer to the in situ rockmass strength data
observed in the Pinawa Underground Research Lab (URL) [54].
Using the simulated peak strengths, the suggested S-shaped failure
envelope consists of three sections of curves that fit to the numerical
data. Components of the fitted envelope are expressed using the Hoek-
Brown and the Mohr-Coulomb parameters as listed in Table 7. The
fitted envelope clearly differs from the strength envelope suggested
based on the conventional shear-based Hoek-Brown parameters. The
first section of the criterion corresponds to peak strength of jointed rock
under confining stresses up to 12 MPa. In this range of confining stress,
the block rotation and eventual yielding of the rockmass is facilitated
by intact rock bridge fracturing. The second section, referred to as
spalling limit, demonstrates the transitional curve for rock strength
under intermediate confining stress (12 MPa < σ3 < 20 MPa). At this
range of intermediate confinement, the stress acting normal to the
surface of a propagating crack prevents the cracks from extending
further. This resistance to crack growth causes the induced cracks to
require higher shear stresses to lengthen compared to the case in which
lower σ3 applied to the boundaries of the SRM specimens. As a result,
the intact rock bridges between the pre-existing joints will fail at higher
stress levels. This strength increase expresses itself with a higher value
of friction angle (φrm) and mb achieved for the second section of cri-
terion compared to the first and third sections. For the simulated
rockmass, the critical confinement which causes this rapid strength
enhancement corresponds to σ3 = 12 MPa (≈ UCSi /20).
The third section of the fitted envelope demonstrates a transition to
shear failure mechanism under sufficiently high confinement
(σ3 ≈ UCSi /10). In the case of the simulated rockmass, this transition
occurs when σ3 approximately equals to 20 MPa. The fitted curve to the
peak strength data for σ3 > 20 MPa shows a significant increase in the
apparent cohesion (crm). This apparent increase in the cohesion of the
material with non-persistent joints can also be experimentally observed
when at sufficient confinement the shear failure through intact rock
bridges leads to a significant increase in the apparent cohesion. At high
confinement, the strength of the rockmass is very close to that of the
intact rock.
The results of a series of simulated biaxial compression tests on a 7
m high sample when the sample is rotated at 45° are shown in Fig. 20
(pink diamonds). As shown, the strength of the rotated SRM samples are
significantly lower than the strength of the non-rotated samples. Unlike
the non-rotated models, the failure envelope obtained from the rotated
models does not follow an S-shaped envelope. This suggests that the
occurrence of an S-shaped criterion depends on the direction of loading.
K. Farahmand et al.
Fig. 19. Stress-strain curves for the 7 m high SRM specimens loaded under confining stresses (σ3 ) ranging from (a) 0 to 4.5 MPa, (b) 5 to 12 MPa, and (c) 14 to 35 MPa.
4.4. Excavation damage zone (EDZ) determination using continuum and
discontinuum modelling
In this section, the anticipated behaviour of the ground in deep
underground tunnels is compared when continuum and discontinuum
modelling analyses are used. The continuum-based finite element
method (FEM) code RS2 [67] and discontinuum-based UDEC code are
used to simulate the breakout around a tunnel excavated in Lac du
Bonnet granite. The in situ stress state and geometry of the simulated
tunnel were selected based on the well-characterized case history of the
Mine-by experiment tunnel in the Pinawa URL, Canada [54]. The Mine-
by experiment tunnel was excavated in massive Lac du Bonnet granite
(GSI ∼ 75) with a 3.5 m diameter while the magnitudes of the in situ
principal stresses are 60 MPa, 49 MPa, and 11 MPa, respectively. The
maximum principal stress is oriented at 11° with respect to the hor-
izontal direction. The mechanical properties of the intact Lac du Bonnet
granite are listed in Table 5.
Finite element analyses were performed with strength properties of
14
Computers and Geotechnics xxx (xxxx) xxx–xxx
the continuum rockmass model defined by the shear-based Generalized
Hoek-Brown criterion and the brittle DISL model [11]. The strength
parameters used in each analysis are listed in Table 8.
Potyondy and Cundall [68] showed that the calibrated micro-
properties of discontinuum PFC2D models [69] should be adjusted by
application of a strength reduction factor of 0.6 to the calibrated
properties of the laboratory-scale model in order to successfully simu-
late the brittle spalling notch development around tunnels to a similar
extent observed in situ. Similarly, Farahmand [70] showed that the
micro-mechanical contact strength properties should be uniformly re-
duced to approximately 0.4–0.5 times the calibrated micro-properties
to match the long-term instead of the short-term strength of the rock,
and also to include the impact of the application of different stress paths
that a rock experiences in the laboratory testing and in situ. As a result,
a uniform strength reduction is performed to adjust the micro-para-
meters of the tunnel model (Fig. 21) by multiplying both the tensile and
shear strengths of all contacts by a factor of 0.5.
An failure zone approximately 0.43 deep developed in the roof of
K. Farahmand et al. Computers and Geotechnics xxx (xxxx) xxx–xxx

Fig. 20. Comparison of the S-shaped strength envelope fit to SRM simulation results of 7
m high specimens loaded under different confining stresses and the predicted strength
using the Hoek-Brown criterion. Note that the strength data illustrated with pink dia-
monds are the SRM simulation estimates of peak strength when the rotated model at 45
degrees is subjected to biaxial compression.

Fig. 21. Damage predicted by the SRM modelling around a circular tunnel in UDEC using
Table 7 the calibrated joints and intact rock properties. Note that the pink features in the model
The calculated Hoek-Brown (H-B) and Mohr-Coulomb (M-C) components of the S-shaped represent the pre-existing joints while the white features illustrate the progression of
failure envelope. damage through the intact material due to the joint slippage or extension and fracturing
of the intact blocks.
Failure envelope Properties

Adjusted H-B parameters Adjusted M-C parameters failure. This can be known as one of the reasons why the brittle failure
process in massive rocks occurs at stress levels much lower than what
mb s a crm (MPa) φrm (° )
the Hoek-Brown and Mohr-Coulomb criteria would predict. The role of
1st section 2.1 0.015 0.35 17.2 37.4
the discontinuities in facilitating spalling failure in underground
2nd section (Spalling 7.0 0.0022 0.75 9.2 53.6 openings is reported in the literature [71,72].
limit) In general, the continuum-based FEM models do not consider the
3rd section 3.4 0.1889 0.50 24.2 34.9 influence of rockmass heterogeneity that exists due to presence of pre-
existing joints. As a result, the discontinuum SRM modelling can pro-
vide a better understanding of the potential depth and extension of
Table 8
failure zone around the excavation when compared to continuum
The failure envelope parameters used in FEM continuum modelling.
modelling results. It can be seen that the inclusion of joints in the SRM
Peak envelope Residual envelope modelling provides enhanced insight into the potential response of the
rockmass to excavation. Having the knowledge of potential rockmass
mb s a mb s a
behaviours that may occur during excavation is very useful for many
Hoek-Brown 12.3 0.062 0.5 12.3 0.062 0.5 aspects of engineering design. This can be specifically helpful when
model dealing with hard moderately jointed rockmasses which are exposed to
Damage initiation envelope Spalling Limit envelope high in situ stresses (i.e. deep mining or tunnelling in massive and
moderately jointed rockmasses).
mb s a mb s a The inability of the Generalized Hoek-Brown (H-B) strength cri-
terion to realistically simulate the geometry of developed spalling
Brittle DISL model 1 0.033 0.25 7 0 0.75
notches around excavations in moderately jointed and massive rock-
masses with high in situ stress ratio (Ko ≈ 4.5) is shown in Fig. 22. In
the simulated SRM tunnel in UDEC, which corresponds to the size of the these FEM models, the conventional H-B criterion produces an incorrect
failure zone observed in the URL test tunnel, as illustrated in Fig. 21. breakout shape and a smaller depth of spall breakout in the roof of the
The model also shows potential for the formation of a much larger excavation. As mentioned previously, failed attempts for reproducing
breakout in the roof as well as at the side walls through yielding of the realistic breakout development using the Hoek-Brown were also re-
rock bridges between the joints. It can be seen that the model suc- ported in literature [9,10,62,63]. However, the FEM models shown in
cessfully simulates the progressive formation of the parallel slabs driven Fig. 22 that use the brittle DISL strength model realistically predict the
by extension cracking. Similar to the spalling mechanism observed in shape and depth of the yielded zones as well as the onset of damage
reality, the formation and detachment of each slab near the excavation initiation. When discrete joints are explicitly simulated in the dis-
wall allows the confinement (σ3 ) acting on the material immediately continuum UDEC model, the presence of the joints influences both the
behind the slabs to reduce. As a result, the induced deviatoric stresses stress distribution around the excavation (Fig. 22c), and the extent and
encourage the extensile cracking of the material and the creation of a shape of the excavation damage zone (EDZ) (Fig. 22d).
newly developed slab. Moreover, the failure process is facilitated by the
presence of the preferentially oriented joints around the excavation. 5. Discussion
These structural features act as stress risers so the stress concentrated at
the tips of the joints is able to initiate and promote spalling in areas A step-by-step procedure for the creation and calibration of 2D
around the tunnel where the known stress should not yet initiate Synthetic Rock Mass (SRM) models, as illustrated in Fig. 8, provides a
systematic framework for SRM model development. Implementing

15
K. Farahmand et al. Computers and Geotechnics xxx (xxxx) xxx–xxx

(a) (b)
(a),(b)

(c) (d)

(c)

σ1

0m 3.5m 7.0m

Fig. 22. Comparison of the depth and extent of the Excavation Damage Zone (EDZ) around a 3.5 m diameter circular excavation predicted by (a) continuum finite-element modelling
using RS2 code with the Generalized Hoek-Brown failure criterion defining the strength of materials showing the damage extent (black dashed line); (b) continuum finite-element model
RS2 code with the brittle DISL model showing the damage extent (black and white dashed lines); (c) discontinuum model UDEC code simulating the presence of pre-existing dis-
continuities; (d) discontinuum model with joints and fractures visible (highlighted pink) with the extent of the potential failure zone demonstrated (red dashed lines). The contours of
maximum principal stress (σ1) for Figures (a), (b), and (c) are shown. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this
article.)

these calibration steps is critical to ensure the resulting model con-
sistently produces reasonable rockmass behaviour. As shown, the DFN
and GBM models can be generated and calibrated separately. Following
the creation of the two separate models, the DFN geometry is imported
into the calibrated GBM model in order to create SRM models. The SRM
models can then be utilized in order to assess the macro-mechanical
properties and response of the rockmass under study.
An upscaling procedure was performed to extract the macro-me-
chanical properties of the rockmass. The obtained properties at this size
are assumed to be representative of the rockmass behaviour at a larger
scale (i.e. the excavation scale) by eliminating associated size effects. To
establish the REV size, the variability of UCSrm , Erm , νrm with increasing
the sample size was investigated. The overall REV model specimen
height of 7 m with a height to width radio of 2.5 is established based on
the variability of the aforementioned deformation and strength para-
meters. The extracted mechanical properties at this scale are considered
to be equivalent continuum properties and can be used as input para-
meters for the excavation scale modelling of jointed rockmasses. In this
study, the SRM sample with a height of 7 m (REV scale) was used to
examine the influence of confining stress and directional loading on the
rockmass strength and deformability.
To quantify a range of GSI values for the rockmass, three methods
based on the RQD, mean joint spacing, and mean block volume were
applied to the virtual 3D LiDAR model of the Brockville Tunnel. In the
RQD-based method, the rockmass condition was inferred based on the
one dimensional index of RQD and the joint surface condition while the
methods based on joint spacing and block volume quantify the degree
of interlocking of the jointed medium on the basis of the measured
spacing between discontinuities and the mean volume of blocks, re-
spectively. The degree of blockiness for the GSI estimation
(50 ≤ GSI ≤ 67) obtained from the joint spacing and block volume
values underestimates the degree of blockiness observed in the
Brockville Tunnel where the rockmass condition clearly indicates that
the rockmass is massive to moderately jointed (GSI > 65). However,
quantifying the degree of blockiness based on the RQD seems to give a
better estimate of the UCSrm . A GSI range from 67 to 77, estimated
according to the measured RQD, corresponds more closely to the field
observations made at the site.
In general, the Hoek-Diederichs rockmass modulus equation (Eq.
(9), [1]) tends to overestimate the rockmass modulus when compared
to the stiffness values obtained from synthesized rockmass (Fig. 12).
However, substituting the lower bound of GSI (GSI = 67) obtained
from the RQD into Eq. (9) yields a value for the rockmass modulus
which is much closer to the SRM modelling prediction. In the case of
predicting the strength, the closest approximation to the values of un-
confined compressive strength of SRM samples is made when a GSI
range of 67–77 is used in the Hoek-Brown criterion. This suggests that
the RQD can provide better estimates for the GSI of slightly to mod-
erately jointed rockmasses. Hence, it is recommended to use the RQD-
based quantified GSI chart reported in [57] to determine the GSI range
for practical purposes. Regarding the compressive strength, the Hoek-
Brown failure criterion seems to offer a realistic value of the rockmass
strength under an unconfined condition. This conclusion is made by
comparing the UCSrm values obtained from the empirical approach and
SRM models, and assuming that the calibrated model is able to produce
correct results for the strength of the material.
The overall mechanical properties of the rockmass are shown to be
highly anisotropic due to the spatial distribution of pre-existing joints,
the joint set geometrical characteristics (i.e. joint size, orientation and
density), and their interaction. As a result, the unconfined compressive
strength of the rockmass varies from 4 MPa to 55 MPa depending on the
direction of loading. A similar response is observed for the rockmass

16
K. Farahmand et al.
modulus with its highest and lowest values of 42 GPa and 15 GPa, re-
spectively. These anisotropic properties should be considered when si-
mulating the excavation scale response of the rockmass using con-
tinuum models and the derivation of direction-dependent input
equivalent parameters. It is possible to investigate strength variability
for a given rockmass by simulating tension and compression tests on the
DFN specimens with various joint densities and orientations. Then, the
probabilistic design using continuum numerical codes can be performed
based on strength variability obtained from the numerical modelling.
It has been shown that the strength of non-persistent hard rock-
masses under confined condition follows an S-shaped failure envelope.
The estimated strength envelope is in agreement with the in situ
strength of the ground measured in various underground projects. It
was also found that the GSI approach fails to correctly predict the
strength of the highly interlocked jointed rockmass under confined
loading condition where the material behaviour tends not to be con-
trolled by interblock shear strength of discontinuities. It was found that
the intercept of the third branch of the failure envelope corresponds to
the long-term strength (CI threshold) of the intact Lac du Bonnet
granite. Under confined conditions the obtained strength from the SRM
models is lower than the predicted strength from the Hoek-Brown cri-
terion. The empirical Hoek-Brown criterion was originally developed
and verified through back calculating the behaviour of the persistently
jointed rocks which are located in low confinement regions near ex-
cavation surfaces. Therefore, the applicability of the GSI system is only
verified to predict the strength of rockmasses with closely spaced joints
and relatively smooth surfaces under low confining stresses. This may
explain the inability of the criterion in accurately predicting the
strength of the non-persistent jointed rockmass under confined condi-
tions.
Finally, by including the presence of structural features such as
joints in models and simulating the effect of stress concentration on the
premature or excess failure of a rockmass, we can identify the areas
around an excavation that are not driven only by brittle failure of intact
rock, but also by interaction between new and existing fractures.
6. Summary and concluding remarks
In this paper, a quantitative procedure to estimate the excavation
scale mechanical properties of a moderately jointed rock has been
presented. It can be confirmed that the S-shaped brittle failure envelope
offers a more realistic strength envelope for the confined, moderately
jointed rock considered in this study. The conclusions of this study are
as follows:
• The response of the Synthetic Rock Masses during simulations of
UCS, biaxial, and Brazilian indirect tensile tests provided significant
insight into the stages of brittle failure, which involved joint slip,
growth and coalescence of new cracks. It has been observed that the
failure of the massive and moderately jointed rockmasses can be
attributed to the stress-driven fracturing of the rock intact bridges
combined with block separation, rotation and/or shear slip de-
pending on the level of confinement.
• The impact of scale of the rockmass deformability and strength was
successfully characterized by generating SRM samples of different
sizes and subjecting them to standard stress paths. Based on the
geometrical and mechanical properties of the rockmass a minimum
REV size was determined at 7 m.
•
For many practical problems, continuum modelling is still largely a
predictive tool for providing guiding ideas for feasibility assessment
and design of rock engineering structure. Reliable input parameter
(including anisotropy properties of rockmass) in this type of the
models is the key for successful application of this technique.
Therefore, the equivalent continuum properties of rockmass (in-
cluding the anisotropic properties) at a representative size can be
characterized using the discontinuum modelling. These up-scaled
17
Computers and Geotechnics xxx (xxxx) xxx–xxx
properties then can be served as input parameters in the continuum
codes.
• Under unconfined conditions, the strength estimated by the SRM
modelling is in good agreement with the strength predicted by the
empirical Hoek-Brown criterion. However, under confined condi-
tions, the empirical criterion tends to overestimate the strength of a
rockmass containing few non-persistent joints while the SRM mod-
elling results match more closely to the in-situ strength. This study
provides more insight on the limits of applicability of commonly
used empirical methods for the estimation of strength of highly in-
terlocked rockmass.
• The strength of massive and moderately jointed rockmasses can be
expressed using the S-shaped strength envelope. The strength re-
sponse of the rockmass under confined and unconfined conditions
can be explained by a failure envelope consisting of three separate
sections based on the level of confinement.
• The direct use of the SRM technique is recommended for the ana-
lysis and design of geotechnical projects. However, the computa-
tional efficiency of the model in terms of practical solution time for
excavation scale problems remains as a limiting factor.
Acknowledgements
The authors wish to thank the Nuclear Waste Management
Organization (NWMO) of Canada, the National Science and
Engineering Research Council (NSERC), the Canadian and the Ministry
of National Defence (DND), and the RMC Green Team for funding this
research work. The discussions with Dr. Jennifer J. Day, Mr. Timothy
Packulak, Dr. Matthew Perras, Dr. Alireza Baghbanan, and Mr. Felipe
Duran Del Vale helped with the preparation of this paper.
References
[1] Hoek E, Diederichs M. Empirical estimation of rock mass modulus. Int J Rock Mech
Min Sci 2006;43(2):203–15.
[2] Amadei B, Goodman R. A 3-D constitutive relation for fractured rock masses. Stud
Appl Mech, B 1981;5:249–68.
[3] Einstein H, Hirschfeld R. Model studies on mechanics of jointed rock. J Soil Mech
Found Div 1973;99.
[4] Ribacchi R. Mechanical tests on pervasively jointed rock material: insight into rock
mass behaviour. Int J Rock Mech Rock Eng 2000:243–66.
[5] Bieniawski Z. Determining rock mass deformability: experience from case histories.
Int J Rock Mech Min Sci Geomech Abstr 1978;15(5):237–47.
[6] Palmstrøm A. Characterizing rock masses by the RMi for use in practical rock en-
gineering, part 2: some practical applications of the rock mass index (RMi). Tunn
Undergr Space Technol 1996;11(3):287–303.
[7] Barton N. Some new Q-value correlations to assist in site characterization and
tunnel design. Int J Rock Mech Min Sci 2002;39(2):185–216.
[8] Hoek E, Carranza-Torres C, Corkum B. Hoek-Brown failure criterion-2002 edition.
5th North American rock mechanics symposium and 17th tunneling association of
Canada conference: NARMS-TAC; 2002. p. 267–71.
[9] Martin C, Kaiser P, McCreath D. Hoek-Brown parameters for predicting the depth of
brittle failure around tunnels. Can Geotech J 1999;36(1):136–51.
[10] Diederichs M. Instability of hard rockmasses: the role of tensile damage and re-
laxation. Ph.D. thesis. University of Waterloo; 2000.
[11] Diederichs M. The 2003 Canadian Geotechnical Colloquium: mechanistic inter-
pretation and practical application of damage and spalling prediction criteria for
deep tunnelling. Can Geotech J 2007;44(9):1082–116.
[12] Carter T, Diederichs M, Carvalho J. Application of modified Hoek-Brown transition
relationships for assessing strength and post yield behaviour at both ends of the
rock competence scale. 2008;108(6):325–38.
[13] Bahrani N, Kaiser P. Strength degradation of non-persistently jointed rockmass. Int
J Rock Mech Min Sci 2013;62:28–33.
[14] Esmaieli K, Hadjigeorgiou J, Grenon M. Estimating geometrical and mechanical
REV based on synthetic rock mass models at Brunswick Mine. Int J Rock Mech Min
Sci 2010;47(6):915–26.
[15] Kaiser PK. Ground support for constructability of deep underground excavations.
Avignon: ITA; 2016.
[16] Baghbanan A, Jing L. Stress effects on permeability in a fractured rock mass with
correlated fracture length and aperture. Int J Rock Mech Min Sci
2008;45(8):1320–34.
[17] Elmo D, Stead D. An integrated numerical modelling–discrete fracture network
approach applied to the characterisation of rock mass strength of naturally frac-
tured pillars. Rock Mech Rock Eng 2010;43(1):3–19.
[18] Ivars DM, Pierce ME, Darcel C, Reyes-Montes J, Potyondy DO, Young RP, et al. The
K. Farahmand et al.
synthetic rock mass approach for jointed rock mass modelling. Int J Rock Mech Min
Sci 2011;48(2):219–44.
[19] Khani A, Baghbanan A, Norouzi S, Hashemolhosseini H. Effects of fracture geometry
and stress on the strength of a fractured rock mass. Int J Rock Mech Min Sci
2013;60:345–52.
[20] Farahmand K, Baghbanan A, Shahriar K, Diederichs M. Effect of fracture dilation
angle on stress-dependent permeability tensor of fractured rock. In: Proceedings of
49th US rock mechanics/geomechanics symposium. American Rock Mechanics
Association; 2015.
[21] Duran F. A numerical investigation of stress path and rock mass damage in open
pits. MASC thesis. Queen's University; 2017.
[22] Jing L, Stephansson O. Fundamentals of discrete element methods for rock en-
gineering: theory and applications. Elsevier Science; 2007.
[23] Itasca. Itasca Consulting Group Inc., The Universal Distinct Element Code (UDEC),
Ver. 6.0., Minneapolis, USA; 2014.
[24] Lorig LJ, Cundall PA. Modeling of reinforced concrete using the distinct element
method. In: Shah SP, Swartz SE, editors. Fracture of concrete and rock. New York,
NY: Springer; 1989.
[25] Diederichs MS, Carter MA, Lato M, Bennett JB, Hutchinson DJ. Lidar surveying for
liner condition, rock stability and reconditioning assessment of Canada’s oldest
railway tunnel in Brockville, Ontario. Proceedings of GeoMontreal, Montreal; 2013.
[26] Lato MJ, Voge M. Automated mapping of rock discontinuities in 3D Lidar and
photogrammetry models. Int J Rock Mech Min Sci 2012;54:150–8.
[27] Voge M, Lato MJ, Diederichs MS. Automated rockmass discontinuity mapping from
3-dimensional surface data. Eng Geol 2013;164:155–62.
[28] Fekete S, Diederichs MS, Lato MJ. Geotechnical and operational applications for 3-
dimensional laser scanning in drill and blast tunnels. Tunneling Underground Space
Technol 2010;25(5):614–28.
[29] Sturzenegger M, Stead D. Close-range terrestrial digital photogrammetry and ter-
restrial laser scanning for discontinuity characterization on rock cuts. Eng Geol
2009;106:163–82.
[30] Baecher GB. Statistical analysis of rockmass fracturing. Math Geol
1983;15(2):329–48.
[31] Dershowitz WS, Einstein HH. Characterizing rockjoint geometry with joint system
models. Rock Mech Rock Eng 1988;21(1):21–51.
[32] Dowd PA, Xu C, Mardia KV, Fowell RJ. A comparison of methods for the stochastic
simulation of rock fractures. Math Geol 2007;39:697–714.
[33] Xu C, Dowd P. A new computer code for discrete fracture network modelling.
Comput Geosci 2010;36:292–301.
[34] Davy P, Le Goc R, Darcel C. A model of fracture nucleation, growth and arrest, and
consequences for fracture density and scaling. J Geophys Res: Solid Earth
2013;118:1393–407.
[35] Xu C, Dowd PA. Stochastic fracture propagation modelling for enhanced geo-
thermal systems. Math Geosci 2014;46:665–90.
[36] Innovemtrics. Polyworks V 11.0.4. Quebec City: Innovemtrics; 2016.
[37] Vazaios I, Vlachopoulos N, Diederichs MS. Integration of lidar-based structural
input and discrete fracture network generation for underground applications. Geol
Geotech Eng 2017;35(5):2227–51.
[38] Dershowitz WS, Herda HH. Interpretation of fracture spacing and intensity. Santa
Fe, NM, Proceedings of the 32nd US rock mechanics symposium; 1992. p. 757–66.
[39] MoFrac software version alpha 2015, MIRARCO Mining Innovation. < http://
www.mofrac.com > .
[40] Srivastava RM. Field verification of a geostatistical method for simulating fracture
network models. Golden, Colorado, Proceedings of the 41st US rock mechanics
symposium; 2006.
[41] Bass J. Elasticity of minerals, glasses, and melts. Mineral physics and crystal-
lography: a handbook of physical constants; 1995. p. 45–63.
[42] Mavko G, Mukerji T, Dvorkin J. The rock physics handbook: tools for seismic
analysis of porous media. Cambridge University Press; 2009.
[43] Mahabadi O, Randall N, Zong Z, Grasselli G. A novel approach for micro-scale
characterization and modeling of geomaterials incorporating actual material het-
erogeneity. Geophys Res Lett 2012;39(1).
[44] Farahmand K, Diederichs M. Implementation of a cohesive crack model in grain-
based DEM technique for simulating fracture in quasi-brittle geomaterial. In: The
proceedings of the GeoQuebec 2015 CGS conference, the Canadian Geotechnical
Society, Quebec, Canada; 2015.
[45] Kazerani T, Yang ZY, Zhao J. A discrete element model for predicting shear strength
and degradation of rock joint by using compressive and tensile test data. Rock Mech
18
View publication stats
Computers and Geotechnics xxx (xxxx) xxx–xxx
Rock Eng 2012;45(5):695–709.
[46] Munjiza A. The combined finite discrete element method. Chichester, UK: John
Wiley and Sons Ltd; 2004.
[47] Tatone S. Investigating the evolution of rock discontinuity asperity degradation and
void space morphology under direct shear. Doctoral dissertation. University of
Toronto; 2014.
[48] Backers T. Fracture toughness determination and micromechanics of rock under
mode I and mode II loading. Doctoral dissertation. University of Potsdam; 2005.
[49] Evans RH, Marathe MS. Micro-cracking and stress-strain curves for concrete in
tension. Mater Struct 1968;1(1):61–4.
[50] Griffith AA. The phenomena of rupture and flow in solids. Philos Trans Roy Soc
London. Ser A, Containing Papers Math Phys Charact 1921;221:163–98.
[51] Bandis S, Lumsden AC, Barton NR. Fundamentals of rock joint deformation. Int J
Rock Mech Min Sci Geomech Abstr 1983;20(6):249–68.
[52] Farahmand K, Diederichs M. A calibrated synthetic rock mass (SRM) model for
simulating crack growth in granitic rock considering grain scale heterogeneity of
polycrystalline rock. San Francisco, USA, Proceedings of the 49th US rock me-
chanics symposium; 2015.
[53] Diederichs MS, Martin CD. Measurement of spalling parameters from laboratory
testing. In: Proceedings of Eurock 2010, Lausanne, Switzerland; 2010. p. 323–6.
[54] Martin C. The strength of massive Lac du Bonnet granite around underground
openings. Ph.D thesis. Winnipeg, Canada: University of Manitoba; 1993.
[55] Nirex. Evaluation of heterogeneity and scaling of fractures in the Borrowdale
Volcanic Group in the Sellafield area. Harwell, UK: Nirex Report SA/97/028; 1997.
[56] Cai M, Kaiser PK, Uno H, Tasaka Y, Minami M. Estimation of rock mass deformation
modulus and strength of jointed hard rock masses using the GSI system. Int J Rock
Mech Min Sci 2004;41(1):3–19.
[57] Hoek E, Carter TG, Diederichs MS. Quantification of the geological strength index
chart. In 47th US rock mechanics/geomechanics symposium. American Rock
Mechanics Association; 2013.
[58] Bieniawski ZT. Engineering rock mass classifications: a complete manual for en-
gineers and geologists in mining, civil, and petroleum engineering. John
Wiley & Sons; 1989.
[59] Bobet A, Einstein H. Fracture coalescence in rock-type materials under uniaxial and
biaxial compression. Int J Rock Mech Min Sci 1998;35(7):863–88.
[60] Wong L, Einstein H. Crack coalescence in molded gypsum and Carrara marble: part
1. Macroscopic observations and interpretation. Rock Mech Rock Eng
2009;42(3):475–511.
[61] Amadei B. Importance of anisotropy when estimating and measuring in situ stresses
in rock. Int J Rock Mech Min Sci Geomech Abstr 1996;33(3):293–325.
[62] Pelli F, Kaiser P, Morgenstern N. An interpretation of ground movements recorded
during construction of the Donkin-Morien Tunnel. Can Geotech J 1991;28:239–54.
[63] Castro L, McCreath D, Kaiser P. Rockmass strength determination from breakouts in
tunnels and boreholes. 8th ISRM Congress. Tokyo, Japan; 1995. p. 531–6.
[64] Siren T, Hakala M, Valli J, Kantia P, Hudson JA, Johansson E. In situ strength and
failure mechanisms of migmatitic gneiss and pegmatitic granite at the nuclear waste
disposal site in Olkiluoto, Western Finland. Int J Rock Mech Min Sci
2015;79:135–48.
[65] Perras MA. Tunnelling in horizontally laminated ground: the influence of lamina-
tion thickness on anisotropic behaviour and practical observations from the Niagara
tunnel project, MASc Thesis. Kingston, Canada: Queen’s University; 2009.
[66] Read RS. 20 years of excavation response studies at AECL's Underground Research
Laboratory. Int J Rock Mech Min Sci 2004;41(8):1251–75.
[67] RocScience Inc. RS2 – Phase2 Version 9.010 64 bit., Toronto, ON, Canada; 2015.
< http://www.RocScience.com > .
[68] Potyondy D, Cundall P. A bonded-particle model for rock. Int J Rock Mech Min Sci
2004;41(8):1329–64.
[69] Itasca. Particle Flow Code in 2 dimensions (PFC2D), Ver. 4.0. Minneapolis, USA;
2008.
[70] Farahmand K. Characterization of rockmass properties and excavation damage zone
(EDZ) using a Synthetic Rock Mass (SRM) approach. Ph.D. thesis. Kingston, Ontario,
Canada: Queen’s University; 2017.
[71] Everitt R, Lajtai E. The influence of rock fabric on excavation damage in the Lac du
Bonnett granite. Int J Rock Mech Min Sci 2004;41(8):1277–303.
[72] Kaiser P. Tunnel stability in highly stressed, brittle ground – Rock mechanics con-
siderations for Alpine tunnelling. In: Geologie und Geotechnik der Basistunnels am
Gotthard und am Lötschberg; 2005. p. 183–202.