International Journal of Rock Mechanics & Mining Sciences 39 (2002) 731–741

Modelling brittle failure of rock

V. Hajiabdolmajida,*, P.K. Kaisera, C.D. Martinb
a
Geomechanics Research Centre, Laurentian University, Sudbury, Ontario, Canada P3E 6B5
b
Department of Civil and Environmental Engineering, University of Alberta, Edmonton, Alberta, Canada
Accepted 30 April 2002

Abstract

Observations of brittle failure at the laboratory scale indicate that the brittle failure process involves the initiation, growth, and
accumulation of micro-cracks. Around underground openings, observations have revealed that brittle failure is mainly a process
of progressive slabbing resulting in a revised stable geometry that in many cases take the form of V-shaped notches. Continuum
models with traditional failure criteria (e.g. Hoek–Brown or Mohr–Coulomb) based on the simultaneous mobilization of cohesive
and frictional strength components have not been successful in predicting the extent and depth of brittle failure. This paper
presents a continuum modelling approach that captures an essential component of brittle rock mass failure, that is, cohesion
weakening and frictional strengthening (CWFS) as function of rock damage or plastic strain. r 2002 Elsevier Science Ltd. All rights
reserved.

1. Introduction Brown constant m for the rock mass, s and a are

constants which depend upon the rock mass character-
Brittle failure is the product of the creation, growth istics, and sci is the uniaxial compressive strength of the
and accumulation of micro- and macro-cracks. Many intact rock.
researchers have reported slabbing and spalling as a The Coulomb criterion, relating normal and shear
dominant failure mode around underground excava- stresses,
tions in massive to moderately jointed rock masses
subjected to high in situ stresses [1–3]. Unlike openings t ¼ c þ s0 tan f ð2Þ
at shallow depth, or at low in situ stress, in which failure
can be determined by the method proposed by Hoek
is controlled by discontinuities, at greater depth, the
and Brown [5]. It is defined by
extent and depth of failure is predominantly a function
of the in situ stress magnitudes relative to the rock mass  0 
sn
stm B
strength [4], i.e., the stress level. t ¼ Asci ; ð3Þ
sci
Traditional approaches of modelling rock mass fail-
ure are often based on a linear Mohr–Coulomb failure where A and B are material constants, s0n is the normal
criterion or on a non-linear criteria such as the Hoek– effective stress, and stm is the ‘tensile’ strength of the
Brown failure criterion. rock mass.
The generalized Hoek–Brown failure criterion for In both criteria, it is implicitly assumed that the
jointed rock masses is defined by cohesive and the normal stress-dependent frictional
 a strength components are mobilized simultaneously,
0 0 s03
s1 ¼ s3 þ sci mb þ s ; ð1Þ i.e., they are assumed to be additive as illustrated by
sci the Terzaghi model in Fig. 1a.
Even when strain-softening models with residual
where s01 and s03 are the maximum and minimum
strength parameters are chosen, the two strength
effective stresses at failure, mb is the value of the Hoek–
components are assumed to be simultaneously mobi-
*Corresponding author. Tel.: +1-705-675-1151x5095; fax: +1-705- lized and then lost in the post-peak range. These
675-4838. approaches with typical strength parameters have not
E-mail address: [email protected] (V. Hajiabdolmajid). been successful in predicting the depth and extent of

1365-1609/02/$ - see front matter r 2002 Elsevier Science Ltd. All rights reserved.
PII: S 1 3 6 5 - 1 6 0 9 ( 0 2 ) 0 0 0 5 1 - 5
732 V. Hajiabdolmajid et al. / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 731–741

simultaneous cohesion
and friction mobilization

τ
Friction

Cohesion τ= c+ σ' tanφ

τ (a) σ'

τ= c(ε)+ σ' (ε) tanφ

(b) σ' (c)

Fig. 1. (a) Terzaghi’s model; (b) Taylor’s model for soil (after Schofield [11]); (c) analog for rock; sequential mobilization of cohesion then frictional
strength.

failed rock in deep underground openings in hard rocks Axial Strain

[6–10] or of borehole breakouts in deep boreholes. 12%
(σ1-σ3)/2 (kg/cm2)

7.5%
1.1. Strain-softening soil
1.2
4%
As Schofield [11] pointed out ‘‘there is no true 1.5%
cohesion on the dry side of critical state’’. In dense soil 1.0
pastes, the peak strength is due to interlock and friction
among particles and not due to the chemistry of bonds. 0.75%
ci 0.8
While friction is immediately mobilized and the fric- Cohesion
tional strength component is proportional to the normal Loss II
cr 0.6
or confining stress, at low normal stress, the interlock
I
resistance can be mobilized and then lost, leading to the 0 0.2 0.4 0.6 0.8 1.0 1.2
typically observed strain-softening behavior of dense σ'3 (kg/cm2)
soils (the Taylor model; Fig. 1b). In other words, Eq. (2)
should be written in a form whereby both strength terms Fig. 2. Bilinear failure envelope for stiff clay (developed as a function
are a function of plastic strain e: of axial strain) illustrating transition from mostly cohesive (I) to an
almost exclusively frictional yield mode (II) (modified after Schmert-
t ¼ cðeÞ þ s0 ðeÞtan f: ð4Þ mann and Osterberg [12]).

The pioneering work of Schmertmann and Osterberg

[12] showed that in some soils, these two strength
components (cohesional and frictional) are not necessa- confinement, where the cohesional strength component
rily mobilized simultaneously. They showed that the is gradually lost when the rock is strained beyond its
maximum of the cohesional component of strength cðeÞ peak strength. This is illustrated by the gap model of
was mobilized early in the test, while the frictional Fig. 1c, representing an analog for brittle rock. This
component s0 ðeÞtan f required 10–20 times more strain- analog illustrates that the cohesion at the bottom of the
ing to reach full mobilization as shown in Fig. 2. sliding wedge must be overcome before the frictional
strength can be mobilized when the gap between the two
1.2. Strain-softening rock wedges is closed. Only after this gap-closure deforma-
tion has taken place, will the normal stress, symbolized
Similarly, there is no ‘‘true’’ and permanent cohesion by the spring, be activated and a strain-dependent
in rocks, at least not in brittle rocks at relatively low effective stress build up to create a frictional resistance.
 V. Hajiabdolmajid et al. / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 731–741 733

Consequently, the shear strength equation must be with a particle flow code (PFC) model by Diederichs
written in the form of Eq. (4), with a strain-dependent [14], showing internal stress variations at a constant
cohesion and a strain-dependent effective stress terms. external confinement of 20 MPa. Hence, the frictional
In the low confinement range, the stress path will retract strength is not everywhere fully mobilized to the level
after reaching the cohesive strength surface as illustrated determined by the average stress represented by the
in Fig. 1b. center of the cone [15,29].
Martin and Chandler [13] demonstrated that the In this paper results from continuum models using
frictional strength component of granite is only mobi- conventional failure criteria and the strain-dependent
lized after a significant amount of the rock’s cohesional cohesion weakening-frictional strengthening (CWFS)
strength is lost. Originally, it was thought that the concept are compared.
friction rather than effective contact stresses, producing
the frictional strength component, needed to be mobi-
lized. In otherwords, variable friction at a constant 2. Brittle failure around a circular test tunnel
stress was implied rather than variable stress with
constant friction. Between 1990 and 1995 Atomic Energy of Canada
This difference in interpretation is important as it is Limited carried out a Mine-by Experiment at the URL,
not the friction that depends on strain but the effective in Pinawa, Man., Canada. This well documented
normal stress s0 (e) causing a gradual development of experiment involved the excavation of a 3.5-m-diameter
the frictional strength component. Damage is induced in circular test tunnel in massive granite [16]. The
brittle rock when it is stressed beyond a damage primary objective of the experiment was to investigate
initiation threshold. As a result, the effective normal brittle failure processes. To achieve this objective
stress s0 inside the rock is highly variable and at some the tunnel was excavated by 0.5–1 m rounds using
locations, less than the stress applied at the boundary. a line-drilling technique and displacements, strains,
This is illustrated by the stress cone in Fig. 3, produced stress changes and micro-seismic emissions were
monitored with state-of-the art instruments. Martin
et al. [3] reported observations of the brittle failure
process resulting in classic V-shaped notches, in the
region of maximum compressive stress (Fig. 4). Read
[17] showed in an extensive characterization report
of the damage zone around the tunnel that the extent of
the compressive stress-induced damage was confined
to the notch regions. Outside the notch little damage
could be visually observed. Martin [8] noted that the
extent of the damage in the roof was greater than
that observed in the floor because of the confining stress
provided by the muck in the floor and also because
the stress path in the floor was significantly different
from that in the roof. Based on low P-wave velocities
and acoustic emission measurements, Read [17] noted
that the tensile zones in the side walls of the tunnel
were damaged, even though no cracking or fracturing
could be visually observed.
One of the objectives of the Mine-by Experiment was
to assess the predictive capability of numerical models in
capturing the extent and shape of the failed zone. For
this purpose, the in situ stresses near the tunnel were
determined accurately: s1 ¼ 6073 MPa, s2 ¼ 457
4 MPa, s3 ¼ 1172 MPa [18]. Extensive laboratory
testing was carried out and the Hoek–Brown failure
parameters were defined: (sc ¼ 224 MPa, m ¼ 28:11;
s ¼ 1). This failure criterion is shown in Fig. 5 together
with the stress required to initiate damage (acoustic
emissions) in the laboratory (sci ¼ 71 MPa+1.5s3) and
Fig. 3. Internal stress variations at an external confinement of 20 MPa
(PFC) model by Diederichs [14]. Variable stress flow patterns between in situ (s1 2s3 ¼ 70 MPa). Given the well-defined stress
particles lead to local stress variations producing internal localized, state, the simple circular geometry of the excavation, the
low or tensile confining stress zones. essentially intact, massive granite, predicting the extent
734 V. Hajiabdolmajid et al. / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 731–741

Microseismic
Event Failed Zone
σ3
σ1

r f = 1.
Acoustic
Emission

m
75
3a

1.
=
Tensile

a
Zone

Scale
0

Fig. 4. Shape of the failed zone observed around the circular test tunnel. Also shown are micro-seismic events locations in the notch area (+), and
acoustic emission locations (
) in the tensile failure zone (after Read [17]).

800 3. Modelling brittle failure

Hoek-Brown
σc = 224 MPa
Lab Peak m = 28.11 Hoek and Brown [20] provide guidelines for rock
700 s =1
mass modelling based on the geological strength index
(GSI) and recently Hoek [21] included massive rock in
600 this index. The following Lac du Bonnet granite
parameters were derived for the Mine-by Experiment
σ1 (MPa)

500 in Lac du Bonnet granite by Martin [8]:

Rock type Lac du Bonnet granite

400
Intact compressive strength sci ¼ 224 MPa
Intact tensile strength sti ¼ 10 MPa
300 Hoek–Brown Constant mi ¼ 28:11
Microseismic GSI GSI ¼ 90
Events Friction angle f ¼ 481
200 Lab σci = 71 MPa + 1.5 σ3
Cohesive strength c ¼ 25 MPa
Hoek–Brown Constant mb ¼ 19:67
100
s ¼ 0:329
Microseismic
In situ σ1 − σ3 = 70 MPa
Compressive strength of rock scm ¼ 128 MPa
Events
mass
-10 0 10 20 30 40 50 60 Tensile strength of rock mass stm ¼
3:7 MPa
σ3 (MPa) Rock mass modulus E ¼ 60 GPa
Fig. 5. Hoek–Brown failure parameters for Lac du Bonnet granite, Poisson’s ratio n ¼ 0:2
and the stress required initiating damage (after Martin [8]).
The common approach to simulate brittle rock failure
is to adopt an elastic-brittle-plastic or strain-softening
of failure should be a trivial task. However, as many model [4] (Fig. 6). The rock mass strength parameters
attempts [19] have shown and will be documented below provided above are used to explore the post-peak
this is not the case. response for modelling the brittle failure observed in
 V. Hajiabdolmajid et al. / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 731–741 735

75 MPa
Plastic
Failed Zone Profile
100
125 50

σ1=150 MPa 25
σ

Strain
0
Softening
Brittle

ε
Fig. 6. Various post-peak responses used in continuum models.
75 MPa

Fig. 7. Major principal stress distribution around the circular test

the Mine-by tunnel. This approach is then compared to tunnel.
a strain-dependent strength mobilization model utilizing
the numerical code FLAC2d [22].
Failed Zone Profile
3.1. Elastic models

One simple way of estimating the depth and extent of

the failed zone is by an elastic stress analysis determining
the induced stresses and comparing them with the rock
mass strength based on GSI. Fig. 7 shows the elastic
major principal stress distribution around the test
tunnel, which reaches a maximum value of approxi-
mately 150 MPa at the roof. Considering the rockmass
uniaxial compressive strength scm ¼ 128 MPa; this stress
is sufficient to create a very thin failed zone in the region Failed elements in shear
of the notch, which significantly underestimates the
depth, and extent of the actual breakout or failed zone
(Figs. 4 and 7).
One method that is often used to overcome the
limitation of elastic analyses is to simulate the progres-
sive nature of slabbing and spalling by successive
removals of failed elements [23]. This approach was
used by Read [16] and Martin [8] with different criteria
for element removal but in both cases the depth of Fig. 8. Extent of damage with an elastic-perfectly-plastic constitutive
breakout zone was overestimated by a factor of 2–3. model;  indicate elements presently in the yielding state in shear, +
indicate elements that previously yielded.

3.2. Elastic-perfectly plastic model

depth of failure is deeper, this approach still does not
An elastic-perfectly plastic constitutive law includes predict the failed zone.
the effects of plastic straining and related stress
redistribution on the depth of failure, but is hardly 3.3. Elastic-brittle model
appropriate for a brittle rock because the obvious
material weakening is ignored. It therefore provides an As suggested by Hoek et al. [4], an elastic-brittle
upper bound (minimum depth of failure with stress material model should be best suited to represent brittle
redistribution). Fig. 8 shows the predicted failed zone failure. In this model the slabbing process is simulated
for rock without dilation (c ¼ 01). While the extent and by decreasing the Hoek–Brown parameter m and s to
736 V. Hajiabdolmajid et al. / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 731–741

very small values in the post-peak range to represent (a) developed around a circular tunnel in massive, brittle
a rapid loss in cohesion of the rock mass to about 20% granite. None of these traditional approaches could be
of peak, and (b) a reduction in friction to the rock basic used to predict the failed zone a priori with any degree
friction angle. of confidence.
Fig. 9 shows results from FLAC2d with mr ¼ 1 and
sr ¼ 0:01: As with the elastic-perfectly plastic model this
approach underestimate the depth of failure. However,
it overestimates the lateral extent of failure. Read and 4. Strain-dependent CWFS model
Martin [24] combined this approach with an element
removal scheme, which over predicted the depth and Hajiabdolmajid [10] adopted a constitutive model in
extent of failure by a factor of 2. which the plastic strain-dependencies of various strength
In summary, the conventional approaches commonly components in brittle failing rocks are considered. This
adopted for rock failure modelling all failed to predict is illustrated by the schematic diagram in Fig. 10. Fig. 10
both the shape and extent of the failed zone that also schematically illustrates the cohesion loss and
frictional strength mobilization in compression tests.
Localization of micro-cracks and formation of a macro-
shear plane (shear band) leads eventually to the full
Failed Zone Profile mobilization of the frictional strength after the initial
cohesion (ci ) is lost and reaches its residual value (cr ).
The process of slabbing around underground openings
cannot directly be compared with the shear banding
process in laboratory compression tests, even though it
also involves a process of cohesion loss and frictional
strength mobilization. The cohesional component of
strength is the predominant strength component at the
early stage of brittle failure and cohesion loss is the
predominant failure process leading to the observed
brittle behavior. The cohesive strength is gradually
Failed elements in shear
destroyed by tensile cracking and crack coalescence. The
normal stress-dependent frictional strength can only be
fully mobilized after the cohesional component of
strength is significantly reduced, much damage has
accumulated, and when the rock fragments can move
relative to each other in shear. The delay in frictional
Fig. 9. Extent of damage with an elastic-brittle plastic constitutive strength mobilization of the frictional strength is the
model with mr ¼ 1 and sr ¼ 0:01: characteristics of the brittle failure in geomaterials in

Compression Test

Onset of σ3=σt to 0
Microcracking σ1
III σ3
Axial stress

IV
II I II III IV
Around Opening
Cohesive
Strength cr
I
ci
Frictional Strength

εcp εfp
Axial strain
Elastic Cohesional Cohesional
+ Frictional
(a) (b)
Fig. 10. Mobilization of the strength components in the CWFS model: (a) in the laboratory compression tests, (b) around underground openings: ci
and cr are the initial and residual cohesion, respectively, and epc and epf represent the plastic strain components when the frictional and cohesive
strength components have reached these ultimate values [10].
 V. Hajiabdolmajid et al. / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 731–741 737

relatively low confinement environments, when the

Cohesion
brittle fracturing (extensile cracking) is the dominant Maximum

Frictional Strength
mode of failure. Initial Cohesion Frictional Strength
A typical stress–axial strain curve from simulated Lac
du Bonnet granite using PFC2d by Diederichs [14] is
shown in Fig. 11. The stress–strain curve shows the Residual Cohesion
characteristics inflection in slope at about 0.7–0.8 of the p
peak strength and rapid strain weakening immediately εf =0.5 %
after peak. Also shown in Fig. 11 are the incremental
p Plastic Strain Plastic Strain
‘‘snap-shots’’ of tensile and shear crack growth, which εc =0.2 %
are very similar to the laboratory findings reported by
Fig. 12. Illustration of the cohesion-loss and frictional strength
Lockner et al. [25]. The accumulative numbers of tensile
mobilization as a function of plastic strain.
and shear cracks are also shown in Fig. 11. Even though
the sample is confined with 20 MPa, the total amount of
tensile cracking dominates shear cracking by a ratio of
approximately 50:1 and that there is very little new crack in situ failure cases. For the Mine-by tunnel model, the
growth after the macro-scale failure zone has formed laboratory damage-controlled tests on Lac du Bonnet
(between 0.5% and 0.6% axial strain). Vasak and Kaiser granite reported by Martin [27] and Martin and
[26] used this concept to simulate the failure processes Chandler [13] were used to establish the plastic strain
near horse-shoe-shaped tunnels subjected to dynamic limit for cohesion loss. From back analyses of the failure
loading (rock burst). zone by slabbing around the Mine-by tunnel, it was
In the CWFS model the plastic strain limits at which found that the plastic strain (or damage) necessary for
the cohesional component of strength reaches a residual the destruction of the cohesional strength (epc ) is in
value, and the frictional strength component mobilizes general lower than the plastic strain required for the full
are two material properties that in reality depend mobilization of the frictional strength (epf ). For Lac du
primarily on heterogeneity and grain characteristics. Bonnet granite the strain-dependent cohesive and
However, they should be calibrated on laboratory and frictional strength mobilization were linearized as
illustrated in Fig. 12 and introduced into the continuum
modelling code FLAC2d. Hajiabdolmajid [10] argued
that while the plastic strain limit for cohesion loss can be
Incremental Shear Cracks considered a true material property, the circumstances
(strain limit) under which the frictional strength reaches
its full mobilization depends to some extent on the
loading system characteristics (geometry and loading
Incremental Tensile Cracks
rate). He attributed the very low strength observed
around the Mine-by tunnel to a delayed mobilization of
the frictional strength (i.e., in the epf > epc ), compared to
300 10000
the high strength obtained in the laboratory compres-
sion tests in which the frictional strength reaches its full
Crack rate (Events per 0.005% strain)

Axial Tensile
250 stress mobilized capacity with less damage or cohesion loss
cracks
1000 (i.e., in the laboratory epf oepc ). This strain-dependent
Total number of cracks

0.81 Peak
200 Final
Damage
strength mobilization can be used to determine the
brittleness of the failing rockmass [28].
The plastic strain limits listed in Table 1 together with
Axial stress (MPa) &

150 100
Initial
Damage
Shear the initial and ultimate strength parameters were used to
cracks
100
simulate the brittle failure of Lac du Bonnet granite,
10
near the Mine-by tunnel. The cohesion was reduced
Crack rate (AE) from its initial peak value of 50 MPa to its residual value
50
15 MPa after 0.2% plastic strain. The initial cohesion
was taken as 78% of scm =2 or the anticipated long-
0 1
0 0.2 0.4 0.6 0.8 1.0
term cohesion without friction mobilization. Dilation
Axial strain (%) was held constant at 301, and the tension cut off was
10 MPa. FISH programming language in FLAC2d was
Fig. 11. Axial stress versus axial strain in PFC test on material with
equivalent properties of Lac du Bonnet granite (after Diederichs [14]). used to introduce the effect of non-simultaneous
Also shown are the number of tensile and shear cracks as well as the mobilization of frictional and cohesional components
crack rate per unit strain. of strength in to a continuum model.
738 V. Hajiabdolmajid et al. / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 731–741

Table 1 Fig. 13 shows the simulated failed zone, using the

Parameters of cohesion weakening-frictional strengthening CWFS model with the parameters listed in Table 1.
Cohesion (MPa) Friction angle Dilation angle Despite several simplifications, such as ignoring stiffness
softening, the predicted depth and extent of the failed
Initial 50 01 301
zone is in excellent agreement with the measured failed
Residual 15 481 301
Plastic strain epc ¼ 0:2% epf ¼ 0:5% 0% zone shown in Figs. 4 and 13. As previously mentioned,
the non-symmetry in observed notch shapes in the roof
and floor is attributed to excavation effects (muck in the
floor and different stress path in three-dimensional
sense) that are not properly reflected in the two-
dimensional symmetric model adopted here.
Profile of Failed Zone
Contours of plastic strain (i.e., cohesion loss or
damage) inside the notch are shown in Fig. 14. As
expected the induced damage decreases when moving
from the tunnel boundary towards the notch tip. This is
in general agreement with the characterization results
reported by Read [17] which demonstrated that outside
the notch the rock mass was essentially undamaged.
Most importantly, this approach properly predicts the
arrest of the failure process that is difficult if not
Failed elements in shear
Failed elements in tension impossible to simulate with traditional models. The
arrest of the observed slabbing process after a new, more
stable, geometry is reached, can be explained by an
increase in confinement (progressive frictional strength-
ening), coupled with a decrease in the induced damage
(plastic strain) and thus a decrease in cohesion loss
arresting the failure process and the growth of the notch.
In Fig. 14 the mobilized frictional strength s% n tanðfÞ is
calculated using principal stresses from the FLAC2d
model and the strain-dependent friction angle. Beyond
the damaged zone (beyond the notch in the intact rock)
Fig. 13. Prediction of the failed zone around AECL’s Mine-by test where there is no plastic straining there is no mobilized
tunnel using FLAC2d and CWFS model, with parameters listed in frictional strength and the cohesive strength is not
Table 1; o indicates elements failed in tension. affected.
c %

a)
th (MP
c= =0

Streng
εp

nal
Frictio
i
10
20
30
40
50
60
70
80

c> cr
0.1% Notch Profile

c=cr
Plastic Strain
n (φ)

0.3%
(Damage)
0.4%
σ n ta

c=cr

c=cr εp=0.5%

0.6%
Tunnel
Wall

Fig. 14. Contours of plastic strain (full cohesion loss at epc =0.2% to cr ). The progressive frictional strengthening inside the notch leading to failure
arrest at the notch tip is illustrated by the profile of (s% n tan f) superimposed on the contour plot.
 V. Hajiabdolmajid et al. / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 731–741 739

1.0 Pn Direct shear test

0
Ft

3=
σ
0.8 Experimental results
φr 2400
Cohesion weakening
(σ1-σ3)/2c*

Peak strength 2200 frictional model

0.6

Average shear stress (kPa)

2000
1800
0.4 1600 φ=370
Residual strength
1400

0.2 1200
Frictional strength
1000
800
0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 600
(σ1+σ3)/2c* 400
Cohesional strength
200
Fig. 15. Bilinear yield envelope developed resulting from the cohesion
0
weakening-frictional strengthening model. c ¼ 2c: 0 500 1000 1500 2000 2500 3000
Average normal stress (kPa)

The strength envelopes that are commonly used to 0 1000 2000 3000 4000 5000 6000

assess stability of underground openings are either the Normal stress (kPa)
linear Mohr–Coulomb envelope or the non-linear Fig. 16. Results from direct shear laboratory tests using solid blocks
Hoek–Brown envelope with a downward curvature. containing voids, laboratory data from Lajtai [30]; model result form
The CWFS model results in an upward-bent bilinear Hajiabdolmajid [10].
envelope of the form shown in Fig. 15. Fig. 15 was
obtained by tracking the history of several grid points frictional strength mobilization characteristics was
located in the V-shaped notch region of the test tunnel. adopted [10].
The peak strength of this region is only a function of The consequence of non-simultaneous mobilization of
the rock mass cohesion while the post-peak strength strength components again is a ‘‘bilinear’’ upward-bent
degrades toward the residual frictional strength. Fig. 15 failure envelope. At low confinement or normal stresses,
shows that some points (open squares) in the notch the semi-elliptical failure envelope (Fig. 16) corresponds
region have reached the residual level with others still to the zone in which the cohesional component of
possess higher cohesion strength component. strength is mobilized and consumed by tensile failure.
However, at elevated normal stress, the cohesional
component of strength is significantly reduced or
eliminated and the strength is mainly due to the
5. Modelling brittle failure in jointed rock frictional strength component, illustrated by the shaded
region.
Lajtai [30] carried out direct shear tests on solid In Fig. 16, the second horizontal axis represents the
plaster blocks containing two voids to represent cracks true normal stress in the rock bridge as there is no
or fracture. The rock bridges between the voids make up contact stress at the ‘‘open joints’’. It can be seen that all
50% of the plane through the voids (Fig. 16). The of the cohesive strength is lost when the actual normal
plaster material had a tensile strength of approximately stress reaches the uniaxial compressive strength of the
1.1 MPa, a uniaxial compressive strength of 4.1 MPa rock (4100 kPa). At this point, rock bridges are
and a basic friction angle 371. The test results of Lajtai completely crushed under the applied normal stress
[30], shown in Fig. 16, illustrate the transition from (before application of shear) and only the residual,
predominantly cohesional to frictional strength con- frictional strength can be mobilized. This provides
trolled failure. Lajtai’s results showed that at low further credence to the argument of cohesion loss in
normal stresses, rock subjected to direct shear loading, the left portion of the bilinear failure envelope.
fails by tension-induced damage or cohesion loss and at The notion that the yield envelope for cohesive
high normal load in shear with full frictional strength materials is bilinear is not new. Schofield and Worth
mobilization. This data clearly support the notion of a [31] demonstrated that this type of yield envelope was
bilinear failure envelope for brittle and jointed rock appropriate for stiff over-consolidated clays and used
masses. The direct shear test as shown in Fig. 16 was this notion to lay the foundations for critical state soil
simulated, using the continuum-modelling code FLAC2d mechanics. Taylor [32] also suggested this type of yield
with the CWFS model. Instead of a bilinear model for envelope for interlocked sands. Fig. 17 illustrates that
the frictional strengths and cohesion loss, a non-linear the bilinear yield envelope is applicable for a range of
model for full cohesion loss and relatively rapid geomaterials including brittle rock.
740 V. Hajiabdolmajid et al. / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 731–741

back-analyses of excavations that have experienced

brittle failure.
Cohesion loss zone φr(rock)
in intact rock
Shear stress (τ)

References
φr(soil)
[1] Kaiser PK, McCreath DR, Tannant DD. Canadian support
handbook. Geomechanics Research Centre, Laurentian Univer-
sity, Sudbury, Canada, 1996. 314p.
Overconsolidated Clay/ [2] Ortlepp WD. Rock fracture and rockbursts—an illustrative study.
Interlocked Sand Monograph Series M9. The South African Institute of Mining
and Metallurgy, Johannesburg, 1997. 98p.
Normal stress (σn) [3] Martin CD, Read RS, Martino JB. Observations of brittle failure
around a circular test tunnel. Int J Rock Mech Mining Sci
Fig. 17. Schematic yield envelope for geomaterials. 1997;34(7):1065–73.
[4] Hoek E, Kaiser PK, Bawden WF. Support of underground
excavations in hard rock. A.A. Balkema, Rotterdam, 1995.
215p.
6. Conclusions [5] Hoek E, Brown ET. Underground excavations in rock. London:
The Institution of Mining and Metallurgy, 1980. 527p.
[6] Wagner H. Design and support of underground excavations in
Brittle failure results from the growth and accumula- highly stressed rock. In: Herget G, Vongpaisal S, editors.
tion of tensile cracks. Around underground openings Proceedings of the Sixth ISRM Congress on Rock Mechanics,
this progressive failure process manifests itself in the Montreal, A.A. Balkema, Rotterdam, vol. 3, 1987. p. 1443–57.
form of spalling or slabbing. This transition from [7] Pelli F, Kaiser PK, Morgenstern NR. An interpretation of ground
continuum to discontinuum behavior is extremely movements recorded during construction of Donkin–Morien
tunnel. Can Geotech J 1991;28(2):239–54.
difficult to capture in numerical models despite advances [8] Martin CD. Seventeenth Canadian geotechnical colloquium: the
in discontinuum modelling [33]. Traditional continuum effect of cohesion loss and stress path on brittle rock strength.
modelling approaches to this class of problems assume Can Geotech J 1997;34(5):698–725.
that the mobilization of the cohesional and frictional [9] Hajiabdolmajid V, Martin CD, Kaiser PK. Modelling brittle
strength components is instantaneous. This approach failure of rock. Proceedings of the Fourth North American Rock
Mechanics Symposium, Seattle, Washington. A.A. Balkema,
overlooks a fundamental aspect of brittle failure, that Rotterdam, 2000. p. 991–8.
the formation of tensile cracks precedes failure in shear. [10] Hajiabdolmajid VR. Mobilization of strength in brittle failure of
The constitutive model introduced in this paper with rock, Ph.D. thesis, Department of Mining Engineering, Queen’s
cohesion weakening and frictional strengthening University, Kingston, Canada, 2001. 268p.
(CWFS) implicitly captures this phenomena by making [11] Schofield AN. Don’t use the C word—Mohr Coulomb error
correction. Ground Engineering, 1998;29–32.
the mobilization of the strength components function of [12] Schmertmann JH, Osterberg JH. An experimental study of the
damage or plastic strain. Physically, the delayed development of cohesion and friction with axial strain in
mobilization of the frictional strength component is saturated cohesive soils. Research Conference on Shear Strength
attributed to the delayed development of internal of Cohesive Soils. Boulder, CO, New York, American Society of
Civil Engineers, 1960. p. 643–94.
normal stress (shear contact), for example, due to
[13] Martin CD, Chandler N. The progressive fracture of Lac du
material heterogeneity. Bonnet granite. Int J Rock Mech Mining Sci 1994;31(6):643–59.
The CWFS model was implemented into FLAC2d [14] Diederichs MS. Instability of hard rockmasses: the role of tensile
using its internal FISH functions. Adopting a con- damage and relaxation. Ph.D. thesis, Department of Civil
tinuum modelling approach to model a discontinuum Engineering, University of Waterloo, Waterloo, Canada, 1999.
process will certainly not capture all the subtleties of 566p.
[15] Kaiser PK, Hajiabdolmajid V. Recent advances in rock engineer-
brittle failure. However, in underground excavation ing for underground excavations, Keynote lecture at the Nineth
practice, what is of paramount importance to the International Colloquium on Structural and Geotechnical
designer is the maximum extent of brittle fracturing as Engineering, Cairo, Egypt, 2001. p. 166–77.
it is directly related to the support requirements. The [16] Read RS. Interpreting excavation-induced displacements around
proposed modelling approach is capable of filling this a tunnel in highly stressed granite. Ph.D. thesis, Department of
Civil and Geological Engineering, University of Manitoba, Man.,
need reliably. Canada, 1994. 328p.
The concept captured both the notch formation and [17] Read RS. Characterizing excavation damage in highly stressed
its stabilization around AECL’s Mine-by tunnel in Lac granite at AECL’s Underground Research Laboratory. In:
du Bonnet granite and the failure of artificial, jointed Martino JB, Martin CD, editors. Proceedings of the International
Conference on Deep Geological Disposal of Radioactive Waste.
rock. Different rocks are expected to have different
Winnipeg, Toronto: Canadian Nuclear Society, 1994. p. 35–46.
strain limits for cohesion weakening and frictional [18] Martin CD, Read RS. AECL’s Mine-by experiment: a test tunnel
strengthening. These strain limits for the CWFS model in brittle rock. In: Aubertin M, Hassani F, Mitri H, editors.
can be established by laboratory tests and from Proceedings of the 2nd North American Rock Mechanics
 V. Hajiabdolmajid et al. / International Journal of Rock Mechanics & Mining Sciences 39 (2002) 731–741 741

Symposium, Vol. 1. Montreal. Rotterdam: A.A. Balkema, 1996. Applications in the Mineral Industry. Montreal: McGill Uni-
p. 13–24. versity, 1995. p. 238–47.
[19] Martin CD, Kaiser PK, McCreath DR. Hoek–Brown parameters [27] Martin CD. The strength of massive Lac du Bonnet granite
for predicting the depth of brittle failure around tunnels. Can around underground openings. Ph.D. thesis, Department of Civil
Geotech J 1999;36(1):136–51. and Geological Engineering, University of Manitoba, Winnipeg,
[20] Hoek E, Brown ET. Practical estimates of rock mass strength. Int Canada, 1993. 278p.
J Rock Mech Mining Sci 1997;34(8):1165–86. [28] Hajiabdolmajid V, Kaiser PK, Martin CD. Mobilization of
[21] Hoek E. The Second Glossop Lecture: putting numbers to strength in brittle failure of rock-in laboratory vs. in situ. In
geology—an engineer’s viewpoint. Quarterly J Eng Geol 1999; Proceedings of the Fifth North American Rock Mechanics
32(1):1–19. Symposium, Toronto, Canada, 2002, 8p.
[22] Itasca. FLAC—fast Lagrangian analysis of continua. Modelling [29] Kaiser PK, Diederichs MS, Martin CD, Sharp J, Steiner W.
software. Version 3.3. Itasca Ltd., 1995. Underground works in hard rock tunneling and mining. Keynote
[23] Zheng Z, Kemeny J, Cook NGW. Analysis of borehole break- lecture at GeoEng2000, Melbourne, Australia, Technomic Pub-
outs. J Geophys Res 1989;94:7171–82. lishing Co. 2000, vol. 1. p. 841–926.
[24] Read RS, Martin CD. Technical summary of AECL’s Mine-by [30] Lajtai EZ. Mechanics of second order faults and tension gashes.
experiment phase 1: excavation response. AECL Report AECL- Geol Soc Am Bull 1969;80:2253–72.
11311, Atomic Energy of Canada Limited, 1996. [31] Schofield AN, Worth CP. Critical state soil mechanics. Maiden-
[25] Lockner DA, Byerlee JD, Kuksenko V, Ponomarev A, Sidorin A. head: McGraw-Hill, 1966.
Observations of quasi-static fault growth from acoustic emissions. [32] Taylor DW. Fundamentals of soil mechanics. New York: Wiley,
In: Evans B, Wong TF, editors. Fault mechanics and transport 1948, 700p.
properties of rocks. London: Academic Press, 1992. p. 3–31. [33] Napier JAL, Peirce SS. Simulation of extensive fracture formation
[26] Vasak P, Kaiser PK. Tunnel stability assessment during rock- and interaction in brittle materials. In: Rossmanith H-P, editor.
bursts. In: Mitri HS, editor. Proceedings of CAMI’95: In Proceedings of the Second International Conference on Me-
Proceedings of the 3rd Canadian Conference on Computer chanics of Jointed and Faulted Rock, 1995. p. 63–74.