International Journal of Rock Mechanics & Mining Sciences 38 (2001) 167–180

Mining-induced stress change and consequences of stress path on

excavation stability } a case study
P.K. Kaiser*, S. Yazici, S. Maloney
MIRARCO-Mining Innovation, Laurentian University, 935 Ramsey Lake Road, Sudbury, Ont., Canada P3C 2C6
Accepted 24 May 2000

Abstract

For stability assessments as well as for support design, it is important to understand the factors leading to detrimental stress
changes. Stress changes not only influence the demand on the rock support, they also change the support capacity of frictional
support components such as plain cablebolts. Stress and stress changes are commonly predicted by numerical models but it is rarely
possible to verify these predictions. This study presents a practical example illustrating the usefulness of stress change measurements
in providing an accurate picture of the mining-induced stress changes and their value for numerical model calibration. Stress
changes, associated with mining of the 565#6 stope at Winston Lake Mine, were measured by four CSIRO HI stress cells, two in the
hangingwall and two in the back of a sill drift. In order to obtain the full stress history of the hangingwall, stresses at the installation
time were calculated by use of a three-dimensional boundary element program } MAP3D. Two types of rockmass failure were
observed at this mine: hangingwall delamination and collapse leading to ore dilution, and wedge-like failures in the backs of sill
drifts. All stress cells were located such that they were affected by these failure mechanisms. Using measured and predicted stress
paths, this study explains the nature of these two failure modes, as experienced at Winston Lake Mine. Excellent correspondence
between measured and predicted stresses is achieved for both the hangingwall and the back of the sill drifts. Modelled stress
rotations in the hangingwall follow closely the stress rotations determined in the field, except at the end of the monitoring campaign
when the stress cells were affected by inelastic straining of the host rock. These measurements assist in understanding the failure
processes and in defining the limitations of the adopted numerical models. # 2001 Elsevier Science Ltd. All rights reserved.

1. Introduction changes the support capacity of frictional support

components such as plain cablebolts. The bond strength
Mining-induced stress change and related rockmass of such cablebolts decreases as the stress decrease
relaxation can drastically influence the stability of reduces the confinement and thus the strength of the
underground openings. One of the consequence is that steel/grout interface [3,4]. Hence, for stability assess-
a decrease in stress may reduce the critical excavation ments as well as for support design, it is important to
span [1]. Diederichs and Kaiser [2] show that even a understand the factors leading to detrimental stress
few millimeters of hangingwall or abutment relaxation changes. The findings presented here are helpful for
can significantly shift the no-support limit in stiffer both underground mining and civil engineering applica-
rockmasses on the stability chart by as much as one tions, particularly those involving multi-step excavation
order of magnitude and thus lead to the failure of of large caverns.
previously stable spans. If the rockmass is relaxed, Many studies have shown that measuring stress
gravitational forces may dominate over the tensile load change is a practical procedure to observe mining-
bearing capacity of naturally jointed or stress fractured induced stresses. However, it is often difficult to
rockmasses leading to failures. Stress change not only interpret stress change measurements and 3D-models
influences the demand on the rock support, it also are required to properly explain field observations
because each mining step alters the stress field. First,
*Corresponding author. Tel.: +1-705-675-1151x5075; fax: +1-705-
the state of stress at the cell location at the time of cell
675-4838. installation must be determined, taking all excavations
E-mail address: [email protected] (P.K. Kaiser). that affect the local state of stress into account. In large

1365-1609/01/$ - see front matter # 2001 Elsevier Science Ltd. All rights reserved.
PII: S 1 3 6 5 - 1 6 0 9 ( 0 0 ) 0 0 0 3 8 - 1
168 P.K. Kaiser et al. / International Journal of Rock Mechanics & Mining Sciences 38 (2001) 167–180

mines, particularly when substantial parts have been 1989). The mean uniaxial compressive strength of the
mined out, the zone of influence may be quite large. gabbro is 334 MPa perpendicular and 201 MPa parallel
Second, the measured stress change tensor must be to foliation. The corresponding values for the chert are
superimposed on the calculated stress tensor to obtain the 211 and 173 MPa. The ore is stiffer with average elastic
absolute stresses that can then be compared with model properties of E ¼ 66 GPa and n ¼ 0:25, but it is much
predictions. In this manner, numerical models can be weaker with uniaxial compressive strengths ranging
calibrated by comparing principal stress magnitudes and from 89 to 153 MPa depending upon the ore grade.
directions or stress rotations for consecutive mining steps. No stress measurements were made at the mine so the
This study presents a practical example of the virgin in situ stress state was selected based on an
effectiveness of stress change measurements in providing extensive database of stress measurements in the
an accurate picture of the mining-induced stress changes Canadian Shield [7]. The state of stress used for all
and their use for model calibration. In particular, the analyses is: s1 ¼ 35:5, s2 ¼ 26 and s3 ¼ 15:5 MPa. The
results identify the limitations of the model adopted for major principal stress is horizontal in the east–west
this study and draw attention to the need to select the direction, roughly perpendicular to the ore body, the
proper boundary conditions for the model. Further- intermediate stress acts along strike and the minor
more, using measured and predicted absolute stress principal stress is vertical. From Fig. 1a [7] it can be
paths, this study explains the nature of two types of detected that the major principal stress is generally
failures observed at Winston Lake Mine, that is, oriented east–west in the Canadian Shield. Visual
hangingwall delamination and collapse leading to ore observations in the mine included slight evidence of
dilution, and wedge-like failures in the backs of sill stress related failures at the back of drifts oriented in
drifts. north–south direction and minor breakouts on the
north–south sides of the vertical bore holes at the lower
levels of the mine. Hence, these visual observations at
2. Site description the mine suggest that the general state of stress is indeed
valid for Winston Lake Mine [8].
Winston Lake Mine of Inmet Mining Corp. is The mine, originally developed for mechanized cut
situated near Schreiber, Ontario, Canada. The orebody and fill to minimize waste handling, was converted to
consists of a massive, stratiform sphalerite, copper longhole in 1987 to increase productivity [9]. All the
sulphide deposit located within the Big Duck Lake development is located in the hangingwall because of the
volcanic sequence (Fig. 1). It extends downward from a weaker footwall rock. Shaft access levels are spaced at
depth of 300 m to below 600 m, dipping at 40–608 to the approximately 50 m intervals. The intermediate sub-
east and running approximately 450 m along strike. The levels have ramp access. Stopes are 18 m high, on
thickness varies from 2 to 20 m (measured horizontally), average, with a strike length between cross cuts of 60 m.
averaging 7 m (Fig. 1). A gabbroic intrusion forms the Longholes are drilled from the overcut on a
immediate hangingwall except in the thinner ore zones 1.6 m  1.6 m pattern. After opening a slot at one cross
(approximately 60% of the orebody) where a thinly cut, the longholes are blasted, typically three rings at a
banded, bimodal, up to 20 m thick, cherty tuff lies time, retreating to the next cross cut. Following a blast,
intermediate (Fig. 2). The footwall also varies through- ore is mucked out primarily by scooptrams under
out the deposit from an altered mafic flow (30%) to an remote control. The stope is then backfilled with
altered rhyolite. In the massive ore three joint sets are uncemented waste rock from the overcut, although not
present [5]: Set A } striking N/S, dipping 438E, parallel always tight to the free face of the bench and another
to the orebody; Set B } striking E/W, approximately producing cycle is initiated. This mining process is called
vertical; and Set C } striking N/S, dipping 438W, ‘‘modified Avoca method’’.
perpendicular to the orebody. The same sets also exist in Ground control problems manifested themselves in
the chert and in the gabbro, however, jointing is less the early phases of the mine as chert delaminated at the
pronounced and these rocks are much more massive. hangingwall due to gravitational effect on the laminated
Various rock types were classified and described rock (Fig. 2). This led to hangingwall instability; in some
by the rockmass quality index, Q [6]. The modified parts of the mine a total collapse of the chert hanging-
index Q0 neglecting stress and water factors, is on wall caused unacceptable dilution in excess of 100% in
average 8 for the chert, indicating fair rock quality. Q0 narrow stopes. In order to rectify this problem, chert
values of 20 and 66 were determined for the ore and hangingwalls were typically pre-supported by twin
gabbro, respectively, corresponding to good and very 5/8 in, 7-strand plain cablebolts installed in fanning
good ground conditions. Laboratory test results on the rings from sub-parallel cablebolt drifts (Figs. 1 and 3).
gabbro and chert produced similar average elastic The cablebolts were laid out such that a 2 m  2 m
properties: E ¼ 53 GPa, n ¼ 0:15, however, their pattern was achieved at the ore contact over most of the
strengths differ (Boaro, personal communication, hangingwall. However, the central portion of the stope
 P.K. Kaiser et al. / International Journal of Rock Mechanics & Mining Sciences 38 (2001) 167–180 169

Fig. 1. Typical section and plan view of stopes at Winston Lake Mine. (a) Maximum principal stress directions in Canadian Shield (Modified from
Herget, 1987 [7]). (b) Geological section through Winston Lake Deposit. (c) Typical cable bolt pattern for sill drifts. (d) Plan view of hangingwall
cable layout.

hangingwall was less-densely supported. Additional stability. For this purpose, and to determine the stress
support, in the form of standard single cablebolts, was change experienced by the hangingwall and sill drifts,
installed from the sill drifts to stabilize the backs against stress cells and extensometers were installed from the
wedge failures and in the hangingwall to provide cablebolt drifts at Winston Lake Mine (Fig. 3). Specifi-
support continuity between sub-levels (Fig. 1). cally, this instrumentation was intended to investigate
three problems:

3. Stress change instrumentation and modelling * chert delamination and hangingwall failure processes
leading to dilution,
As explained in the introduction, the stress path * continuing chert failure in cablebolted hangingwalls,
and thus the complete stress history of the rockmass is and
required for a rational assessment of excavation * wedge-type failures in the back of sill drifts.
170 P.K. Kaiser et al. / International Journal of Rock Mechanics & Mining Sciences 38 (2001) 167–180

A plan view of the 565#6 stope, showing the locations

of monitoring equipment and the excavation history,
as well as four cross-sections through the stope are
presented in Fig. 3. Stress changes, associated with
mining of the 565#6 stope, were measured by four
CSIRO HI stress cells, two in the hangingwall (Cells 1
and 2) and two in the back of the sill drift (Cells 3
and 4). Stress changes were obtained by converting
the strain changes using the stress reduction program
MIMCELL, a variation of the CSIRO software
(Dzik, personal communication, 1988). This program
systematically rejects gauges giving the poorest fit to
the strains corresponding to the previously calculated
stress field as it iterates to a ‘best-fit’ solution. Strain
data was corrected for creep prior to calculating
the stress change. Constant rock deformation properties
of Poisson’s ratio n ¼ 0:12 and elastic modulus
E ¼ 50 GPa were used. As a consequence, stress change
predictions are expected to become excessive when
the rock response becomes non-linear at low stresses.
This is illustrated in Fig. 4 which shows that for small
strain changes De1m or De2m (increase or decrease), the
actual or predicted stress changes are Ds1ap , Ds2ap .
However for large negative strain changes De3 (exten-
sion or relaxation) the assumed high tangent modulus
leads to predicted stress changes ðDs3p Þ that exceed the
Fig. 2. A photograph showing the laminated chert. actual stress changes ðDs3a Þ in the field. Thus, absolute

Fig. 3. 565#6 stope, instrumentation locations and excavation history.

 P.K. Kaiser et al. / International Journal of Rock Mechanics & Mining Sciences 38 (2001) 167–180 171

Fig. 4. Effect of non-linear rock response during unloading on stress change calculations.

tensile stresses may be predicted when, in fact, the stress form, at Cells 1 and 3 when excavation of the 565#6
is near zero. stope (approximately at 550 m depth) began were added
At the time of stress cell installation, the initial stress to the measured field stress change tensors to determine
state had already been somewhat altered by nearby the absolute stress tensors at various stages of stope
mining. Hence, in order to obtain the full stress history excavation. Model stresses were calculated at the stress
of the hangingwall, the primal stresses were calculated cell locations for 11 mining steps, representing the
by use of a three-dimensional boundary element excavation sequence from September 5 to December 6,
program } MAP3D [10] assuming a continuous, 1989 for comparison with the monitoring results. The
isotropic, homogenous and linearly elastic rockmass. proximity of the active mining face relative to the cell
MAP3D is a general, fictitious stress, three-dimensional locations, Cell 1 in the hangingwall and Cell 3 in the
boundary element modelling tool widely used in under- back of the sill drift is shown in Fig. 3. Figs. 6 and 7
ground mining. Such elastic modelling tools are often present the magnitudes of the measured and predicted
preferred over other codes because of their simplicity stresses associated with mining of the 565#6 stope. The
when modelling complex geometries of mine-wide mining front reaches Cell 1 on November 13 (Step 8)
models including stopes, accompanying drifts and and Cell 3 on November 25 (Step 10).
crossings. For reasons explained earlier (Fig. 4), stress
change predictions using elastic model are found to be
reasonably reliable as long as the rockmass is in a 4. Discussion
compressive stress state, e.g., ahead of the advancing
stope face. Behind the face, when delamination and 4.1. Hangingwall stability
relaxation occur, elastic models tend to over estimate
stress relaxation but the observed trends (stress change A gradual stress build-up was recorded in the
reversals) can be obtained from such models (see later, hangingwall (Fig. 6, Cell 1) as the free face approached
e.g., Fig. 7). In a companion study, the same rockmass along the strike of the orebody. A stress increase of
was modelled using a 2D discontinuous, jointed rock- about 15 MPa in s1 occurred by the end of October
mass model [12]. While this model properly simulated when the face was 7 m from the cell location (Step 6).
the delamination process, far from the stope face, it does After the face passed the cell location, s1 was observed
not provide stress changes ahead of and near the face. and predicted to decrease by 30 MPa. The model shows
The three-dimensional MAP3D model of the 565#6 that the maximum increase in s1 should be reached in
stope area containing the existing openings, cross cuts, Step 7 (Fig. 6a) slightly after the observed peak. This
top sill and the cablebolt drifts is presented in Fig. 5. suggests that some mining-induced stress redistribution
The stress tensors calculated by MAP3D, in Cartesian due to rockmass non-linearity occured ahead of the
172 P.K. Kaiser et al. / International Journal of Rock Mechanics & Mining Sciences 38 (2001) 167–180

Fig. 5. Winston Lake Mine MAP3D model focusing on the 565#6 stope; insert shows entire mine model geometry.

stope face. Considering the fact that the ore is much questionable after Step 10. Nevertheless, the predicted
weaker than the host rock, this is not unexpected. The and observed stress changes and rotations, as will be
intermediate stress s2 is predicted to remain essentially shown later, correspond well up to Step 11.
unchanged whereas it was observed to gradually Figs. 8a and b present polar plots of the measured
increase by about 10 MPa before falling back to the and predicted principal stress vectors for Cell 1 located in
original state as the face passed the cell. the hangingwall. The vector orientations are plotted on
In December, the field data shows an extensive stress the lower hemisphere together with the mining step
decrease in all stress components. The minor principal numbers. The stress paths are reproduced in Figs. 8c and
stress s3 is predicted to drop as soon as the mining d to assist data interpretation. A stress vector rotating in
face reaches the cell location. In reality, however, a plane follows a great circle. This can be detected for
the measurement shows that s3 has already dropped large parts of the predicted and observed stress paths.
to almost zero stress before Step 7. A dramatic drop of It should be noted that lines sequentially connect points
s3 to 35 MPa occurred during Step 8. As explained of the same principal stress (minor, major or inter-
earlier this is an artefact of the constant rock modulus mediate). Hence, if two principal stresses switch position,
used to calculate the stress changes. If reduced unload- e.g. because s1 becomes s2 , this is reflected by an
ing modulus of 25 GPa (instead of 50 GPa) is used to apparent rotation by 908. A careful study of Fig. 8a
represent the effect of opening of fractures for measure- reveals a s1 versus s3 flip and a stress rotation to align
ments that fall below the original state of stress, the with the hangingwall (Fig. 9). s1 rotated 908 to the north,
measured s3 corresponds well with the predicted as the face approached without a change in dip,
(Fig. 6b). Two principal stresses (s2 and s3 ) become eventually becoming parallel to the strike of the orebody
tensile after Step 11 with s3 drops to 520 MPa and (Steps 7 and 8, 4–13 November, Fig. 8a). In the model s1
s2 to 57 MPa (Fig. 6b). The unreasonably large, rotates to north–south direction during Step 9 (Fig. 8b).
calculated tensions can thus be explained by non- After the face passed Cell 1, s1 rotated further, to an
linearity of the rock during unloading. east–west direction (14–17 November) but dipped (57–
The hangingwall delamination process cannot be 638) nearly parallel to the orebody. The model shows the
reproduced by the elastic continuum model without same trend (Fig. 8b } Step 10). Between 14 and 20
tension cut-offs. Hence, the predicted stress reductions December, after 565#6 stope was excavated, s1 rotated
are not valid after delamination starts. It is shown that back to a northwest direction with a shallower dip of 15–
around 25–80 mm of cumulative displacement were 208. This behaviour is not observed in the model (shown
recorded by the extensometers in December [8]. This with a shaded line in Fig. 8c) and can be attributed to
suggests that both model and field observations are non-elastic straining of the stress cells as explained earlier.
 P.K. Kaiser et al. / International Journal of Rock Mechanics & Mining Sciences 38 (2001) 167–180 173

Fig. 6. Variation of principal stress magnitudes with time for Cell 1 in hangingwall when the initial stress state is obtained from modelling.
(a) Assuming constant E ¼ 50 GPa. (b) For E ¼ 50 GPa to October 20 and E ¼ 25 GPa for lower stress reducations after October 20.

Fig. 7. Variation of principal stress magnitudes with time for Cell 3 at the back of the sill drift.

The intermediate principal stress s2 first rotated from principal stress s3 , initially vertical, dipped to 508 as the
north to south 1808 but, ultimately, settled nearly excavation approached. As the stope face passed, it
horizontal and parallel to the strike of the ore body. suddenly dipped to nearly 308 aligned almost perpendi-
The model predictions show the same trend. The minor cular to the orebody, as shown schematically in Fig. 9.
174 P.K. Kaiser et al. / International Journal of Rock Mechanics & Mining Sciences 38 (2001) 167–180

Fig. 8(a–d). Principal stress orientations during 565#6 stope excavation; Cell 1 in hangingwall: (a) observations, (b) MAP3D model prediction.
Numbers indicate mining steps (refer to Fig. 3).

Fig. 9. Principal stress orientation during the excavation of 565#6 stope. (a) Initial condition; (b) after excavation.

This is not unexpected as the stress field on the the measurements for a reduced unloading modulus
excavation wall always has no shear and no normal (E ¼ 25 GPa; Fig. 10c), it can be seen that the observed
stress, hence the excavation wall is always a principal and predicted stress paths correspond well (note: tensile
stress plane and two stress components must be parallel stresses in excess of 5 MPa tension are plotted at this
to the wall. Again, it must be remembered that the final assumed threshold limit because of the laminated
observations are somewhat questionable because of the structure of chert, with rather limited tensile strength).
assumed linearity discussed earlier. Nevertheless, an In the model, the largest increase in s1 occurs
detailed comparison of Figs. 8a, c and b, d shows a following Step 7 when the face is approximately 10 m
excellent agreement between predicted and observed away from Cell 1 (Fig. 6). In the field, the maximum
stress rotations. increase is observed when the face is 7–12 m away from
Figs. 10a and b present the predicted and measured Cell 1 (Steps 5 and 6 between 23 and 27 October). Figs. 6
stress paths with a constant rock modulus (E ¼ 50 GPa) and 10 show a gradual increase of s1 from the primary
for both s1 and s2 versus s3 at Cell 1. After correction of stress of 35 MPa to a maximum of 53 MPa during Steps
 P.K. Kaiser et al. / International Journal of Rock Mechanics & Mining Sciences 38 (2001) 167–180 175

Fig. 10. Maximum and intermediate principal stress versus minimum principal stress at Cell 1 in hangingwall of 565#6 stope. (a) Assuming constant
E ¼ 50 GPa; (b) model predictions; (c) for E ¼ 50 GPa to October 20 and E ¼ 25 GPa for lower stress.

1–7. In the model, s3 is almost constant at 20 MPa stress (see Fig. 6), then the predicted and observed stress
during this loading phase whereas in the field, s3 first paths correspond even better (Fig. 10c).
decreased before it increased in value. Without correction, these stress paths presented in
After 20 October, when the face passed the cell Fig. 10 demonstrate that the deviatoric stresses (s1 2s3
location (Figs. 3 and 5, Step 8 in the model), s3 or s1 2s2 ) do not exceed 45 MPa representing 22% of
decreased rapidly beyond zero (became tensile) while s1 the minimum uniaxial compressive laboratory strength
dropped by an amount equal to the loss of confinement of the hangingwall rock. Hence, the stresses never
(Ds1 was roughly equal to Ds3 ). s2 , on the other hand, is exceed the compressive damage initiation limit of
predicted and was observed to remain almost constant 0:33sc defined by Martin [11] and, consequently, failure
during this loss of confinement. If an unloading of the chert hangingwall could not have occurred by
modulus of 25 GPa is assumed for near zero states of overstressing of the chert near the mining face. As a
176 P.K. Kaiser et al. / International Journal of Rock Mechanics & Mining Sciences 38 (2001) 167–180

matter of fact, it is not even likely that the hangingwall collapse. To prevent this type of failure, it is
was damaged by these compressive stresses. Failure paramount to adopt mining sequences and use stope
must have occurred when the minor principal stress sizes or shapes that create stable abutments. As
exceeded the tensile strength of the chert. According to Diederichs and Kaiser [2] show, it takes only a few
the numerical model this would have occurred during millimeters of displacement in stiff rocks to create
Steps 9 or 10 when the stope face was between 5 and sufficiently large stress drops to cause failure.
10 m past the cell location. Stress measurements suggest Extreme care must be taken to ensure optimal
that failure was actually initiated up to 5 m ahead of the displacement control in such situations. Undercutting
stope face during Steps 7 or 8 (Fig. 6). This is consistent of the hangingwall, either by sill drifts or by dilution
with delamination observations from the extensometers in previously mined stopes, must be prevented. Ore
located in the hangingwall [8]. accesses or drawpoints and cablebolt patterns should
Several practical implications follow from a compar- be staggered (level to level) to ensure that potentially
ison of measured stress changes and those predicted by unstable areas in an overlying stope are well
elastic continuum models: constrained by sound stope walls below. Top-down
mining would, at least theoretically, create more
* Because the stresses never exceed the compressive stable abutment conditions in this case.
failure initiation threshold, the elastic continuum
model represents well the stress conditions observed 4.2. Stability of the backs of sill drifts
at the mine prior to the development of tensile
stresses. Only when failure occurs, i.e., at the As was mentioned earlier, two structurally distinct
transition from continuum to discontinuum beha- types of rockmass failure were experienced at the
viour, after Step 10 or 11, does the numerical model mine: The hangingwall failure due to abutment relaxa-
fail to represent reality. tion, described in the previous section, and wedge-like
* Under such conditions and for hangingwalls consist- failures in the back of sill drifts as schematically
ing of jointed rock with little or no tensile strength illustrated in Fig. 11. The influence of stress change on
capacity, failure can be predicted by determining this latter mechanism will now be demonstrated using
when and where confinement is lost as s3 drops to the measurements made during the mining of 565#6
zero or below. stope.
Unfortunately, a criterion of s3 ¼ 0 alone is not The stress paths at Cell 3 in the back of the 565#6 top
sufficient to predict delamination in moderately jointed sill were investigated and verified with the help of
rocks with some tensile strength (st > 0) or in situations boundary element modelling. Plots of s1 and s2 versus
where arching is possible (e.g. when Voussoir arch s3 are presented for both predicted and measured
action is possible; [2]. In such cases, s1 or s2 could build- stresses in Fig. 12. Measured tensile stresses of
up until a stable situation is achieved. Measurements at 55 MPa are again plotted at 5 MPa (tension). The
Cell 1, however, show that this is not the case in the temporary transition through tension gives insight into
hangingwall of the Winston Lake Mine. After Step 10, the failure phenomenon. First from A to B (Fig. 12), the
all three principal stresses suddenly drop below zero, ground behaves non-elastic, but once the discontinuity
indicating loss of arching action. During early attempts controled wedge failure process (Fig. 12) has stabilized,
to model the hangingwall failure using UDEC, this a compressive stress field is reestablished at D.
interpretation was confirmed [12]. In the UDEC model, Fig. 12 indicates that all three predicted principal
the blocky hangingwall could not be brought to failure, stresses steadily increase to reach a maximum at Step 11
even with negligible interblock friction unless unreason- whereas the measured stresses peak at Step 10 (Point A).
ably thin slab thicknesses were assumed. Only when the The minor principal stress increases sufficiently to
boundary condition was changed in the model to allow produce a stress path pointing consistently to the
boundary movements parallel to the hangingwall, did right, toward more stable conditions. The intermediate
the major principal stress rapidly drop, as observed, and stress, both predicted and observed, falls well below
hangingwall collapse occur in the model [12]. This leads the damage initiation threshold for the ore, however, the
to another important practical conclusion: major principal stress exceeds the damage initiation
threshold as soon as the sill drift is excavated.
* Even discontinuum models cannot properly predict For Q0 ¼ 20, corresponding to a GSI of approxi-
the rockmass response if the dominant boundary mately 71 for the ore, Hoek et al. [13] and Hoek and
condition is not properly represented; in this case, Brown [14] recommend s ¼ 0:04 and mb ¼ 5:3 (assum-
freedom of movement parallel to the hangingwall due ing mi ¼ 15). This criterion is shown in Fig. 12. If these
to mining in the lower stopes (with hangingwall parameters were valid, failure should not occur because
dilution). It is the de-stressing parallel to the the predicted stress path never exceeds this strength
hangingwall, reducing s1 or s2 , that leads to the envelope. Indeed, the predicted safety margin is
 P.K. Kaiser et al. / International Journal of Rock Mechanics & Mining Sciences 38 (2001) 167–180 177

Fig. 11. Stress-related wedge-like failure at the back of sill drifts; predicted stresses are shown. (a) After the excavation of sill drift; (b) after the
excavation of 565#6 stope.

parameters: s ¼ 0:11 (for sc ¼ 28 MPa) and mb ¼ 1:5.

Hence, for this stress path to ever reach failure, the
criterion would have to be represented by s > 0:11 and
mb 51:5, values that deviate considerably from those
recommended by Hoek and Brown [14].
The measured stress path deviates drastically from
that predicted. After Step 6, the minor principal stress s3
first decreases to about 5 MPa and then increases rapidly
to in excess of 20 MPa with an equal increase in s1 (i.e.,
s1 2s3 remains constant). This observed stress path falls
inside the damage initiation criterion for the ore and
suggests that, during damage initiation, rapid lateral
straining (dilation) causes s3 to build up more rapidly
than predicted by a non-dilatant, elastic material model
with a Poisson’s ratio of n ¼ 0:25. After Step 7, s1
increases quickly to 80 MPa (Point A) while s3 remains
more or less constant. Both the modelled and measured
stresses reach the same maximum s1 , though at different
mining steps (Step 11 in the model and Step 10 in the field).
Once the stress path (predicted or measured) reaches
the lower limit of the zone labelled ‘‘unstable failure
propagation’’ in Fig. 12, a wedge-type failure occurred.
This is clearly evidenced by the sudden loss of
confinement (drop of s3 to below 0 MPa) with
Fig. 12. Maximum and intermediate principal stress versus minimum simultaneous drops in s1 and s2 of more than 50 and
principal stress at Cell 3 in back of 565#6 stope. 30 MPa, respectively (Fig. 7). It is interesting to note,
from Fig. 12, that during this failure, both stress changes
Ds1 and Ds2 (Point A–B) are roughly equal to the
observed to steadily increase as the stress path diverges corresponding change in s3 , or that the measured stress
from the failure envelope. The predicted major principal paths follow a ðs1 2s3 Þ=constant path. In soil me-
stress (Fig. 12, full diamonds) follows an almost linear chanics terminology, this would be called an undrained
path described by s1 ¼ 28 þ 2:7s3 , to a maximum of failure process represented by a lateral shift of the Mohr
80 MPa during Step 11 when the ring immediately circles (Fig. 13a).
beneath Cell 3 is taken (Point A in Fig. 12 at Since there was no water involved, the effective
s1 ¼ 80 MPa, s3 ¼ 20 MPa). This path can approxi- stresses must be equal to the total stresses and
mately be described by the following Hoek–Brown the observed shift in stress circle cannot be explained
178 P.K. Kaiser et al. / International Journal of Rock Mechanics & Mining Sciences 38 (2001) 167–180

The failure at the back of the sill drifts can be

explained using the cohesion/friction mobilization
concept:
* Damage to the sill drift back was initiated when
ðs1 2s3 Þ exceeded about 40 MPa and rockmass
degradation continued with little friction mobiliza-
tion until a failure threshold at ðs1 2s3 Þ ¼ sc (ore;
rockmass)=62 MPa was reached (Point A).
* Due to rapid confinement loss, the unconfined stress
state (Point B, Fig. 12) is reached, without loss of
rockmass cohesion but also with little or no friction
mobilization (Fig. 13a).
* From Points B to C, as damage accumulates and
cohesive strength is lost, the unconfined compressive
strength of the ore drops to sc (damaged ore)=33 MPa
(Figs. 12 and 13b).
* Because the failing rockmass was well supported at
this location, such that the friction could be
mobilized while the confinement built up to 5 MPa
(Figs. 12 and 13b), the rockmass could be further
loaded to Point D inside the shaded area (Fig. 12). At
this stage, the friction angle fD is approximately 508
(Fig. 13b), If it is assumed that no further cohesion
loss occurred during the confinement build-up (C–D).
The mobilized friction plus dilation angle fD could
exceed 508 if further cohesion loss occurred between
stages C and D.
The result of the above-presented logic is summarized
in Fig. 14. The rockmass in the sill drift back fails when
the deviatoric stress ðs1  s3 Þ reaches the unconfined
rockmass strength of the ore at 62 MPa. Little or no
friction is mobilized at that point because the confining
pressure drops rapidly to near zero values. As the
Fig. 13(a–c). The interpretation of stress path during the measured
unloading at Cell 3. rockmass is further deformed by hangingwall-footwall
convergence, the rockmass’ intrinsic strength or cohe-
sion is lost while friction is mobilized and both the
as undrained behaviour. The rapid transition to confining stress s3 and the major principal stress s1
an unconfined failure state must have occurred increase until a new equilibrium is reached. If the
without friction mobilization (fA=0, Fig. 13). This rockmass were unsupported, confinement could not
is consistent with the cohesion/friction mobiliza- build-up and failure would propagate until a more
tion concept proposed by Martin et al. [15] and Martin stable shape was achieved (breakout formation).
[11] which suggests that a significant portion of the If elastic models are used to design excavations in
rockmass cohesion must be lost before friction can be brittle failing rock, a failure criterion that predicts Point
fully mobilized. These researchers demonstrated that A as failure (Fig. 12) must be adopted. In other words, a
during brittle failure the peak cohesion and the full pure cohesive or frictionless criterion must be used to
friction are not mobilized simultaneously; most of the predict the outset of failure. This case study further
cohesion is lost before the peak friction is mobilized supports the brittle rockmass parameters, m ¼ 0 or
(Fig. 13c). f ¼ 0, recommended by Martin et al. [16]. However, for
The measured stress path in Fig. 12 reaches an adequately supported excavations, allowing full friction
unconfined state at s1 ¼ 62 MPa suggesting that the mobilization in the failed rockmass, conventional
unconfined rockmass strength of the ore is 40–65% of Hoek–Brown parameters (with m 6¼ 0) with a damaged
the ore’s laboratory compressive strength. In other rockmass strength properly describe the ultimate state of
words, s should be between 0.15 and 0.45 which again is the rock (Figs. 13 and 15).
inconsistent with the recommended value for a good- From the previous discussions, it can be concluded
quality rockmass. that the rockmass was gradually damaged and
 P.K. Kaiser et al. / International Journal of Rock Mechanics & Mining Sciences 38 (2001) 167–180 179

eventually failed by breaking up in a more or less changes is reflected in the consistent stress tensor
unconfined state. Only after significant deformation had rotations described by the polar plot presented in
accumulated was the frictional shear strength of the Fig. 8. The CSIRO cells proved to be an effective means
rockmass mobilized. This interpretation is consistent for stress change monitoring. Model stress rotations in
with qualitative observations of the debris pile at other the hangingwall follow closely the stress rotations
locations in sill drifts at Winston Lake Mine. When determined in the field except at the end of the
collapse occurred the rockmass broke-up into block size monitoring campaign when the stress cells were affected
pieces that could unravel between the cablebolts [8]. by inelastic straining of the host rock. As anticipated,
stresses rotated by as much as 908 in the hangingwall.
These large stress rotations can disturb pre-existing
5. Conclusions discontinuities under shear, cause fracture propagation,
open-up joints or induce failure of rock bridges,
Two types of failure, relaxation-induced hangingwall contributing to the degradation of the rockmass.
delamination (1) and stress-driven wedge-type back Particularly, in laminated rock, stress rotation alone
failures (2), were carefully monitored with stress cells can lead to a loss in the tensile load bearing capacity of
at Winston Lake Mine. Fortunately, these cells were the hangingwall rock. It is our interpretation that this
located in the failing rockmass and produced reliable degradation weakened the lamination planes and thus
results for comparison of observed and predicted stress contributed to a hangingwall failure process whereby
changes. The high quality of the measured stress stress relaxation at stress levels well below the damage
initiation threshold and without crossing of the yield
surface of the rockmass lead to failure (Path 1 in
Fig. 15).
The data shows that this stress rotation and related
strength degradation occurs ahead of or close to the
advancing stope face. For this type of failure, elastic
3D models are adequate to determine areas of low or
tensile stress, but care must be taken to properly re-
present model boundaries, particularly those that can
move due to nearby mining or undercutting. If these re-
laxation causing mechanisms, as observed at Winston
Lake Mine, are neglected, models incorrectly generate
apparently stable conditions. This mode of failure is
explored in more detail by Diederichs and Kaiser [2].
For the stress driven wedge-type failure (Path 2 in
Fig. 15), observed and predicted stress paths differed
Fig. 14. Schematic representation of measured and predicted stress drastically. It was observed that significant stress
paths at Cell 3. redistribution (loss of confinement as s3 ! 0) occured

Fig. 15. Stress paths for (1) relaxation and (2) stress-driven failure.
180 P.K. Kaiser et al. / International Journal of Rock Mechanics & Mining Sciences 38 (2001) 167–180

long before the conventional failure envelope (assuming References

simultaneous mobilization of cohesive and frictional
resistance components) is reached. This loss of confine- [1] Kaiser PK, Falmagne V, Suorineni FT, Tannant DD. Incorpora-
ment leads to an actual stress path that curves towards tion of rockmass relaxation and degradation into empirical stope
design. 97th Annual Meeting of CIM, Vancouver, 1997.
the unconfined rockmass strength of the undamaged [2] Diederichs MS, Kaiser PK. Tensile strength and abutment
rockmass, which is higher than suggested by conven- relaxation as failure control mechanisms in underground excava-
tional failure criteria (e.g. [14]. As the rockmass is tions. Int J Rock Mech Min Sci Geomech Abstr 1999;36(1):69–96.
strained and damage accumulates, the cohesive strength [3] Kaiser PK, Diederichs M, Yazici S. Cable bolt performance
component is gradually lost and the frictional strength during mining induced stress change } three case examples. In:
Kaiser, McCreath, editors. Proceeding of the Rock Support in
component is mobilized leading to damaged strength Mining and Underground Construction, 1992. p. 377–84.
that can be described by conventional failure criteria. [4] Kaiser PK, Yazici S, Nosé J. Effect of stress change on the bond
However, this strength can only be reached if the strength of fully grouted cables. Int J Rock Mech Min Sci
rockmass is sufficiently confined by an effective support Geomech Abstr 1992;29(3):293–306.
[5] Milne D, Jocob D. Winston Lake Mine support study data -
system. In brittle, hard rock, this is difficult to achieve
Phase I } geotechnical mapping and borehole camera logging.
because the support resistance is very small compared to Noranda Technology Centre, Internal Report, 1990.
the strength of the rock and the loads induced by the [6] Barton N, Lien R, Lundu J. Engineering classification of rock
buckling rockmass during failure. Hence, the strength masses for the design of tunnel support. Rock Mech
envelope of the damaged rockmass can seldom be 1974;6(4):183–236.
reached, except in the low confining range. The [7] Herget G. Stress assumptions for underground excavations in
Canadian Shield. Int J Rock Mech Min Sci Geomech Abstr
conclusions from this case study are consistent with 1987;24(1):95–7.
the brittle Hoek–Brown criteria presented by Martin [8] Maloney S, Kaiser PK. Field investigation of hanging wall
et al. [15]. support by cable bolt pre-reinforcement at Winston Lake Mine.
Unpublished Report to Winston Lake Mine, Geomechanics
Research Centre, Laurentian University, 1993. p. 140.
Acknowledgements [9] Werniuk J. The birth of a zinc mine. Can Min J 1998;119(1):13–20.
[10] Wiles T. MAP3D Version 36 User’s Manual, Mine Modelling
This research was financially supported by a provin- Pty. Limited Box 637, Mt. Eliza, Victoria, 3930, Australia, 1996.
[11] Martin CD. Seventeenth Canadian Geotechnical Colloquium: the
cial grant to the Chair for Rock Mechanics and Ground
Effect of Cohesion Loss and Stress Path on Brittle Rock Strength.
Control at Laurentian University, by eleven Canadian Can Geotech J 1997;34:698–725.
mining companies through the Mining Research Direc- [12] Kaiser PK, Yazici S, Maloney S. Mining induced displacement
torate of Ontario (today CAMIRO), and the Natural and support performance } a case study. Unpublished Geome-
Sciences and Engineering Research Council of Canada chanics Research Centre Report, 1998.
[13] Hoek E, Kaiser PK, Bawden WF. Support of underground
(NSERC). The field experiments were facilitated by
excavations in hard rock. Rotterdam, Brookfield: A.A Balkema,
John Nosé, Edward Biskup and Ruth Fearon, and 1995.
preliminary modelling was carried out by Luigi Cotesta. [14] Hoek E, Brown ET. Practical estimates of rock mass strength. Int
Their contribution was of great importance for this J Rock Mech Min Sci Geomech Abstr 1997;34(8):1165–86.
research and is thankfully acknowledged. Lastly, the [15] Martin D, Chandler NA. The progressive fracture of Lac du
Bonnet granite. Int J Rock Mech Min Sci Geomech Abstr
authors wish to sincerely thank Joe Boaro, Bill Brosko 1994;31(6):643–59.
and Peter Ingram along with everyone else at Winston [16] Martin CD, Kaiser, PK, McCreath DR. Hoek–Brown parameters
Lake Mine for their kind co-operation throughout the for predicting the depth of brittle failure around tunnels, Can
course of this work. Geotech J 1998, submitted.