Practical Method for Prediction of Geomechanical Failure-time

Azania MUFUNDIRWA
Candidate for the Degree of Doctor of Philosophy
Supervisor: Prof Yoshiaki FUJII
Division of Solid Waste, Resources and Geoenvironmental Engineering

Introduction
The issue of predictability of landslides and rock slope
failures, which are major geo-hazards, is of great
concern. It is a challenge to date in the rock mechanics
field to precisely predict failure-time of geo-hazards,
and geo-hazards still pose major threat to life and major
loss in terms of economics. For a preview, in the land
transport sector, landslides are most prevalent and
deaths account for numbers less than 1, 000 deaths per
year and can cause damage amounting to few billion
dollars (Boll, 2002). For example, in Japan, much of the
land is slopy ground, and because of a large population
on dwindling available land, it is becoming inevitable to
built habitats for humanity near or on slopy ground
susceptible to failure. Therefore, there is a need to
reduce the amount of damage to property, fatalities and
financial losses due to geo-hazards.
In the geotechnical field, structures are monitored to
ascertain their stability, but the question, When is
geomechanical failure going to occur? is still an issue.
Various monitoring equipment and devices such as
Global Positioning System (GPS), Slope Stability Radar
(SSR), extensometers, survey stations, and others are
used yet somehow failure still occurs unanticipated.
Needless to mention, monitoring the behaviour of
landslides and rock slopes is an important aspect to
mitigate failure or accidents, and is vital in successfully
forecasting the time of slope failure.
Conventionally, displacement is of primary
importance to the rock engineer because it is a good
indicator to movement of rock mass. Some rock mass
exhibits time-dependent failure, from primary to tertiary
creep. In-situ monitoring is used in geotechnical fields
to anticipate failure in time-dependent deformations.
However, it is crucial to appreciate that, in reality, some
slope failures are largely dominated by processes or
mechanisms that are not effected by creep (e.g.,
structural failures). Hence there is a need, to develop
simple, quick and reliable geomechanical failure-time
prediction methods that can find extensive application
under different failure mechanisms, geometrical and
hydrogeological complexity, etc. Needless to mention, it
is of paramount importance to understand the causative
agents (pre-failure processes) affecting rock mass
deformation; this helps us verify, why it failed?
However, chiefly, this study seeks to address the
question, When will geomechanical failure occur?
In brief, I am proposing a new geomechanical
failure-time prediction method that utilizes the
divergence phenomenon of measured displacement or
strain prior failure, so that I can predict
geomechanical failure-time irrespective of failure
mechanism, lithology etc. The proposed method is

applied to predict failure-time of case histories (e.g.,

Asamushi and Vaiont landslides) and laboratory studies
(creep tests). Of essence, processes causing rock mass
deformations prior failure are also considered. This
included a typical case study from Japan involving
continuous monitoring of rock slope deformations. To
reveal the precursors of rock slope deformations, a new
and simple method was proposed to reduce the apparent
displacements (masking the true rock mass
displacements) so that I could enhance accurate
displacements. Furthermore numerical analysis on rock
mass model was incorporated to investigate the causes
of rock slope deformations prior failure.

Failure-time prediction methods

To develop prediction methods with a broad range of
application, it would be reasonable to assume that
accelerating displacements or strains of slopes prior to
failure are analogous to the terminal phase in tertiary
creep. Under this assumption, I could possibly
encompass prediction of failure-time for diverse
geomechanical failures caused by structural failures,
creep and other precursory failure mechanism.
To begin with, in my endeavour to address the
above-mentioned issues, I adopted Eq. (1) proposed by
Fukui and Okubo, (1997) representing strain divergence
in the terminal phase of creep failure in rocks.

= B log(Tf t ) + C

(1)

where : strain, t: time, Tf: failure time, Tf t: life

expectancy, B and C: constants. In our derivations, I
substituted displacement u instead of strain and
differentiated both sides of Eq. (1) with respect to time
t:

du
B ,
=
dt
Tf t

(2)

where (du/dt) is the displacement rate. Re-arranging Eq.

(2), I get Eqs. (3) and (4). This was done so that I could
increase my chances of developing a suitable alternative
prediction method.

du
du
= Tf
B,
dt
dt

t Tf .
dt
=
du
B

(3)

(4)

Using non-linear approximation (NLA), Tf can be found

from the (du/dt)-t curve from Eq. (2), (Mufundirwa and
Fujii, 2008, 2010). And, using simple arithmetic, Tf is
evaluated as the slope of t(du/dt)-du/dt curve for Eq. (3);
this is my newly developed method termed the SLO.
And in the other method Tf is evaluated as the xintercept of (dt/du-t) curve for Eq. (4). Fukada et al.
(1999) showed that Eqs. (2) and (4) tend to give delayed
or unsafe (Fig. 1) predicted failure-time Tfp and Eq. (3)
gives earlier/safe predicted failure-time. From our
derivation, the latter is called the INV method,
conventionally inverse-velocity method (Fukuzono,
1985). Apparently, t can be measured from an arbitrary
initial time.
Reasonably, using measured displacement or strain;
I can compute deformation rates (displacement or strain
rate) using the data filtering method Eq. (5), so as to
allow for smoothing of short-term deformation
deviations that could be insignificant or cause false
alarms. The data filtering method consists of using the
nth observation (sampling value) to calculate the rate:

In this section, I will present case histories used in my

study, namely rock mass failure, Asamushi and Vaiont
landslides.

Rock mass failure in open-pit mine

In open pit mining, rock slopes are created when mining
activities progress. The stability of rock slopes is critical
to mining production, safety of mining personnel and
equipment. In this case study, rock mass failure (500
m3) occurred on a rock slope at a limestone mine in
Japan (Fig. 2). Fortunately, no injuries or damage to
equipment were reported.
Rupture plane

Direction of
slip

du ui ui n
(i=n + 1, n + 2,.,m), (5)
=
dt i ti ti n
where (du/dt)i are the computed displacement rate
points, tm and um are the time and displacement at the
instant of prediction, respectively (Rose and Hungr,
2007). Sampling value, n, was selected so as to yield
positive rates of displacement or strain only.

T f - t0

Unsafe

Asamushi Landslide

T fp > T f

(+)

The 100,000 m3 landslide occurred at Asamushi at

22:12 in July 1966, on the Tohoku line, Japan,
interrupting railroad traffic for 26 days and burying 80
m length of track (Saito, 1969). The geologic structure
of the zone of landslide consisted of near surface joints
in liparitic tuff.

Actual life expectancy

Tfp - t m

(T f - t m ; T f = T fp )

t m < T fp< T f
c
t0 O
d

(-)

Fig. 2. Rock mass failure in an open-pit limestone mine,

Japan (2007). Aerial view of mine site and (top right)
outline of rock mass failure.
denotes the region of
rock mass failure (courtesy of the mine).

Safe
B

tm

Tf

No prediction
T fp < t m

Vaiont Reservoir Landslide Disaster

The catastrophic failure of approximately 270 million
m3 occurred at 23:39 on 9 October 1963 in northeastern
Italy, killing about 2000 people (Genevois and Ghirotti,
1963).

Observational approach and predicting

Fig. 1. A simple conceptual model representing
predicted life expectancy (Tfp tm) as a function of tm
(time at instant of predicting) as failure-time Tf
approaches. Note: t0 is initial arbitrary time representing
commencement of prediction and Tf t0 is initial actual
life expectancy.

Case histories in geomechanical failure

For illustration purpose, I will use the case study of rock

mass failure. Since the rock mass was dominated by
structural failure, using observation and sound
engineering judgement, I managed to estimate the onset
of accelerating displacements similar to the terminal
phase in tertiary creep. As a limitation, I acknowledge
that it is difficult to locate precisely the boundary
between steady and accelerating displacements.

During data collection, precise recording was done at

11,755 min (Tf) prior to failure, and Fig. 3a shows
displacement during the last stages of failure (June
2007). Using the observational approach, I estimated the
onset of accelerating displacement at t = 11,440 min (u
= 1.6 mm) as illustrated in Fig. 3a.
Notably,
displacement gradually increased until t = 11,682 min (u
= 4.6 mm), and thereafter, displacement steeply
increased towards failure, Tf = 11,755 min (u = 18.6
mm) as shown in Fig. 3a. The sudden increase in
displacement towards failure was possibly due to the
rupture plane within the rock mass (Fig. 2).

20

failure
Tf = 11,755 min

at this point the corresponding number of data set is nfp

= 1, by simply counting (Fig. 3b). However, as the time
t increases towards actual failure-time Tf = 11,755 min,
the number of data sets increases. For example, at t =
11,562 and 11,751 min, the corresponding number of
data sets is nfp = 24 and 213, respectively. In reiteration,
using observation and sound engineering judgement,
data set for initial prediction was selected by estimating
the point where displacement rate progressively
increased, and this corresponds to nfp = 24 (Fig. 3).
Thereafter, during prediction, data sets were then
gradually increased (nfp = 24, ...213) until the period just
prior failure Tf = 11,755 min, and Fig. 4 shows typical
results from SLO and INV at nfp = 213.
SLO
Tf = 11755 min (actual failure-time)
10000
nfp = 213
tm = 11751 min

10

t(du/dt) (mm)

Displacement (mm)

failure approaching

11500

11600
11700
t (min)

11,440

Tfp = 11753.8 min

Tfp

5000
1

11800

safe prediction tm <Tfp < Tf

Onset of accelerating displacement

0
sampling value, n = 100

1.5

Tf = 11,755 min
failure

0.2

0.4
0.6
du/dt (mm/min)

0.8

INV
Tf = 11755 min (actual failure-time)
nfp = 213

nfp = 213

10
dt/du(min/mm)

du/dt (mm/min)

R = 0.99

failure approaching
0.5
nfp = 24

11500

11600
11,562
nfp = 24

11700

tm = 11751 min
unsafe prediction Tfp > Tf
2

slight convexity

11800

t (min)

11,539 increasing data ranges nfp

nfp = 1

R = 0.93

Tfp = 11774.42 min

11,751
nfp = 213

Fig. 3. (a) Displacement as a function of time t, 5.25

hours before rock mass failure (b) displacement rate as a
function of time t.
To explain my approach in forecasting geomechanical
failure-time, I will utilize the deformation data from
rock mass failure as an example (Fig. 3). Taking a
closer look at Fig. 3a and b, I can observe that after
sampling data (n = 100 in Eq. (5)), the onset of
accelerating displacement rates is t = 11,539 min. And

11500

11600

11700 11800
t (min)

11900

12000

Fig. 4. Typical plots used for prediction of rock mass

failure using (a) SLO (b) INV.

Table 1

Table 3

Summary of results of in situ failure-time predictions

using SLO and INV

Summary of failure-time predictions for Inada granite

from Brazilian creep tests

Prediction method
SLO

Asamushi
Landslide
Vaiont
Landslide

Time before failure (min)

Safe prediction (%)
Time before failure (hrs)
Safe prediction (%)

23/31; 74.2%

SLO

1/31; 3.2%
80.49

8/14; 57.1%

Time before failure (days)

Safe prediction (%)

Prediction method

INV
193

0/14; 0%
130

9/12; 75%

9/12; 75%

Time before failure (s)

INV
3114

0.85

Creep stress ratio (CSR)

Rock mass
failure

Safe prediction (%)

4/8; 50%

Time before failure (s)

3/8; 37.5%
1440

0.90
Safe prediction (%)

8/10; 80%

10/10; 100%

Case histories in geomechanical failure were predicted,

namely the rock mass failure, Asamushi and Vaiont
landslides to investigate the prediction performance of
the proposed SLO and the INV method. Monitored
displacements or strains were used to compute Tfp using
SLO and INV methods, and predicted life expectancies
from both methods were compared focusing mainly on
the safe and unsafe errors. Findings reveal that the rock
mass failure, Asamushi and Vaiont landslides were
predicted at 193 min, 80.49 hrs and 130 days before
failure, respectively, with SLO having most safe
predictions (Table 1).

Rock failure was predicted at 3114 sec, 1440 sec and 99

sec at CSR = 0.85, 090 and 0.95, respectively. Notably,
SLO exhibited a higher level of safe predictions (> 50%
safe predictions) in all the three CSRs. However,
pragmatically, the dominance of the brittle nature posed
a limitation to the prediction performance of the SLO
and INV.

Experimental studies

Effects on Prediction Performance

Laboratory uniaxial compression creep tests were

carried out on Shikotsu welded tuff (SWT). The tests
were performed on SWT cylindrical specimens (60 mm
length, 30 mm diameter), and the recorded strains (a;
axial strain and c; circumferential strain) were used in
prediction using the SLO and INV method.
Furthermore, laboratory Brazilian creep tests were
carried out on dry Inada granite to explore the ability to
predict extensile failure-time for rock specimens using
the measured strains. Summary of prediction results are
as below.

To test the efficacy of the prediction methods, I

evaluated the effects of time and spatial scales of
failures and failure mechanisms on reliability of the
methods. I found that, both methods, particularly SLO,
were applicable (safe predictions) under different time
scale and volumes of failure. In essence, I have also
noted that the methods seem useful under different
failure modes. Some empirical relations from prediction
results were deduced.

Prediction method
SLO
c

Time before failure (s)

Safe prediction (%)

7/8; 87.5%

5/8; 62.5%

10

8
Vaiont landslide

6
Asamushi landslide

Rock mass failure

358.99
15/25; 60%

Time before failure (s)

Safe prediction (%)

INV

Safe prediction (%)

99

Note: 4/8 = 4 safe predictions out of 8 total predictions

log t (s)

Table 2
Summary of failure-time predictions using c and a for
SWT.

0.95

Time before failure (s)

21/25; 84%

259.99
0/25; 0%

Lab. uniaxial creep test

Brazilian creep test

t = 6310 V

0.35

0/25; 0%

Note: 15/25 = 15 safe predictions out of 25 total

predictions
Predictions using circumferential strain c and axial
strain a were done, and c gave earlier, reliable and
largely safe predictions at 358.99 sec before failure
compared to 259.99 sec from a comprised of 0% safe
predictions. Both the SLO and INV had safe predictions
c, at 60% and 84% respectively.

-4 -2 0

4 6 8 10 12 14 16 18 20
3
log Volume (m )

Fig. 5. Proportionality between t (= Tf t0; Tf is

failure-time and t0 is estimated time for onset of
accelerating deformations in the terminal phase of
failure; initial time representing commencement of
prediction) and volume of failure reveals scale
independency of SLO (newly proposed) and INV
prediction methods. Safe predictions were obtained
from laboratory to real-life failures.

Wyllie & Munn. (1979) & MacRae (1982)

Vaiont landslide

Rock mass failure

0
Cruden & Masoumzadeh (1987)

To minimize displacement components that are

proportional to temperature, the following equations
were used:

Asamushi landslide

-6

-4 -2 0

4 6 8 10 12 14 16 18 20
log Volume (m 3)

Fig. 6. Relations between critical deformation rate

(du/dt)crit and volume of failure from laboratory to reallife failures. In this case, critical deformation means
deformation rates at failure or just before failure.

Pre-failure process affecting rock mass

deformation
Potential rock slope failures are a major threat if they
are located near a roadway or a railway. A case study
from a cool temperate region in Iwate prefecture, Japan
is presented. The first part focuses on field
measurements of natural rock slope deformation across
fractures predominantly in a chert rock mass, which
were monitored using six surface fracture displacement
sensors. The second part is devoted to numerical
analysis of rock slope deformations simulate field
conditions. Weather conditions namely, air temperature,
humidity, snowfall, and rainfall were recorded from 9
November 2006 to present. Data till 14 December 2008
were analyzed.

(7)

where u is measured displacement (one-day movingaverage), u is the corrected displacement, A is thermal

correction coefficient, Ta is air temperature (one-day
moving-average), Ua is displacement amplitude, umax is
maximum corrected displacement, and umin is minimum
corrected displacement.

Corrected displacements
Essentially, after correction at ch1 in Fig. 8 (u using A =
+0.007 mm/oC), the steep increase in fracture
displacement (at t 3640 weeks) was observed, which
could be related to fracture growth. As an important
feature, it was noted that the computed coefficients
(+0.005 to +0.007 mm/oC) are close to the determined
laboratory coefficient (+0.00721 mm/oC), except for
ch2. This means that fractures in chert rock mass have
little or insignificant deformations with temperature
variation.
ch1 (A = +0.007)
0.3

30

0.2
20
0.1
0

10

-0.1
0

-0.2
-0.3

20

40

60

80 100

-10

ch2

30

0.2
20
0.1
0

10

-0.1
0
-0.2
-0.3

20

(A = +0.005)
30

Fracture opening

0.2

20
0.1
0

10

-0.1

Long bolting

ch3

-0.2
-0.3

ch4

20

40

60

80 100

Ta- Air temperature ( oC )

ch4
0.3

-10

20
0.1
0

10

-0.1
0
-0.2
0

20

40

60

80 100

-10

ch3

(A = +0.007)
30

0.2
20
0.1
0

10

-0.1
0
-0.2
-0.3

20

40

60

80 100

-10

ch6
0.3

(A = +0.005)
30

Fracture closing

0.2

20
0.1
0

10

-0.1
0
-0.2
-0.3

20

40

60

80 100

-10

t (week)

t (week)

u'

Fig. 7. Fracture displacement sensors (ch1ch6)

installed on surface of natural rock slope in Iwate
prefecture (Japan).

-10

u' - Corrected displacement (mm)

(A = +0.006)
30

0.2

-0.3

80 100

t (week)

Ta- Air temperature ( oC )

ch6

u' - Corrected displacement (mm)

ch5

ch5

60

0.3

t (week)

0.3

40

t (week)
u' - Corrected displacement (mm)

ch2
Large aperture

u' - Corrected displacement (mm)

t (week)

ch1

(A = -0.003)

0.3

Ta- Air temperature ( oC )

Lab. uniaxial creep test

Brazilian creep test

U a = u ' max u 'min

Ta- Air temperature ( oC )

0.47

(6)

Ta- Air temperature ( oC )

-3

(du/dt)crit = 1.26 x 10 V

-4

u ' = u + AT a

Ta- Air temperature ( oC )

Zavodni & Broadbent (1980)

u' - Corrected displacement (mm)

-2

u' - Corrected displacement (mm)

log Critical deformation rate - (du/dt)crit (mm/min)

Effects of temperature

Ta

Fig. 8. Temporal variations in corrected displacement u

(moving averaged) and air temperature Ta (moving
averaged) for ch1ch6.

Ch3 and 5, exhibit very small temperature-induced

permanent fracture displacements (due to thermal
change) over a period of more than two years. Ch4 was
gradually opening whilst ch6 was generally closing.
Insignificant influences of weather conditions on
fracture/rock mass movement were observed.

442

Rock surface

Fracture
y
x

5m

Numerical simulation of rock slope

deformation

5m

To get a better insight into the deformation of a fracture

under temperature variation, numerical analyses for a
rock mass with a one-metre deep fracture under
temperature variations were carried out using (5 m 5
m) two-dimensional plane strain analysis coupled with
the fundamental heat transfer equation using the CrankNicolson method (Crank, 1975). Two boundary
conditions were used; confined and unconfined
conditions to simulate different in-situ conditions under
elastic and elasto-plastic finite element analysis as
shown in Fig. 9. To simulate temperature in the field,
two patterns of transient heat analysis (variable
temperature boundary) were exclusively applied on the
top rock surface under confined and unconfined
conditions for duration of 52 weeks. The first
temperature pattern had a positive temperature variation
(7.832.2 oC), with an initial rock mass temperature
(Tref) of 20 oC applied. The second temperature pattern
had positive and sub-zero temperatures (2.222.2 oC),
with an initial temperature of 10 oC applied as the
temperature of rock mass. With reference to Fig. 10, we
defined:

= T

(8)

where is the thermal strain due to temperature

variation, is the thermal expansion coefficient of rock;
T is the change in temperature. It was assumed that
rock expands by f from T1 to T2 due to freezing of pore
water, where T1 is the temperature at which freezing
starts; T2 is the temperature at which freezing finishes
(Fig. 10).
In brief, there was tensile failure of elements at
fracture tip as temperature lowered in both cases (Fig.
11). Freezing effects on deformation of rock slope
(predominantly in chert rock mass) are little or
insignificant,
and
minor
permanent
fracture
deformations occur under temperature variation across 0
o
C. We tentatively suggest that the permanent fracture
deformations were dominantly caused by thermal
fatigue.

Concluding remarks
Attempts to predict failure-time Tf of rock mass failure,
Asamushi landslide, Vaiont reservoir landslide and
Shikotsu welded tuff (SWT), Inada granite were done.
SLO developed in this study is reliable predictive tool
that proved consistent and was validated in most cases.
We suggest that permanent fracture deformations were
dominantly caused by thermal fatigue.

Symmetrical axis
Case 1: side AB is confined.
Case 2: side AB is unconfined.

Fig. 9. Finite element mesh configuration used in

numerical simulation for confined and unconfined
boundary conditions: Note: 442 is node number that
represents the beginning of the fracture at the rock
surface. The fracture has some aperture.
Strain,

Tref (Reference/Initial temperature)

1
T1
T2

()

Temperature, T

(+)

Expansion due to
freezing of pore
water

Fig. 10. Model for expansion due to freezing of pore

water.
a

Tensile failure at 7.8 oC

Six elements failed in tension in

the vicinity of fracture tip

Tensile failure at 2.2 oC

Four elements failed in tension

in the vicinity of fracture tip

Fig. 11. Calculated results of Elasto-plastic confined FE

model: (a) Element failure at minimum temperature 7.8
o
C and (b) Element failure at minimum temperature
2.2 oC. Note: Deformation is magnified 1000.

References
Boll, A., 2002. Landslides and Rock fall. Swiss Federal Institute of Forest, Snow and
Landscape Research.
Crank, J., 1975. The mathematics of diffusion. Oxford university press, Oxford, pp. 144146.
Fukui, K., Okubo, S., 1997. Life expectancy and tertiary creep for rock. In: Proc. of fall
meeting of mining and materials processing institute of Japan, pp. 9194 [in Japanese].
Fukuzono, T., 1985. A new method for predicting the failure time of a slope. In: Proceedings
of the fourth international conference and field workshop on landslides. Tokyo: Japan
Landslide Society, pp. 145150.
Genevois, R., Ghirotti, M., 2005. The 1963 Vaiont Landslide. Giornale di Geologia Applicata
1, pp. 4152.
Mufundirwa, A., Fujii, Y., 2008. New methods for prediction of geomechanical failure-time.
Proc. of the Korean Rock Mechanics Symposium, Gwanju, pp. 183190.
Mufundirwa, A., Fujii, Y., 2010. Prediction of rock mass failure-time of geo-hazards. Proc. of
the ISRM European Rock Mechanics Symposium, Lausanne, pp. 567570.
Rose, N.D., Hungr, O., 2007. Forecasting potential rock slope failure in open pit mines using
the inverse-velocity method. Int. J. Rock Mech. Min. Sci., 44, pp. 308320.
Saito, M., 1969. Forecasting time of slope failure by tertiary creep. Proc. of 7th International
Conference on Soil Mechanics and Foundation Engineering, Mexico City, 2, pp. 677683.