Vous êtes sur la page 1sur 15

Rock Mech. Rock Engng.

(1999) 32 (4), 241255


Rock Mechanics
and Rock Engineering
: Springer-Verlag 1999
Printed in Austria
Strike Standard Deviation for Shallow-dipping Rock
Fracture Sets
By
H. H. W. Herda
Department of Mathematics, University of Massachusetts, Boston, MA, U.S.A.
Summary
The well-known problem that the observed strike of a shallow-dipping rock fracture can
vary widely, even if one re-measures the strike in the same location, is dealt with in this
paper. The overall purpose of the paper is to place the statistical estimation of the strike
condence interval for shallow-dipping rock fracture sets on a sound foundation. We have
made series of strike measurements at each of various dips on a wooden laboratory ``frac-
ture'' to allow estimation of strike standard deviation at all shallow dips. This is accurate
for smooth rock and serves as a lower bound estimate for rock with a rougher surface.
These estimates are used, along with two independent iterative statistical procedures, to re-
estimate the dip direction condence interval of an actual set of nineteen fractures. The dip
direction condence interval is more than three times as wide as the corresponding dip
condence interval. This will be made visible in our transformed condence region plot for
the mean pole.
1. Introduction
It is well-known that one cannot accurately measure the strike of a fracture which
dips less than 25

(Baecher, 1972, p. 384). In contrast, the dip of this kind of


fracture can be measured quite accurately. Such a fracture will be dened as
shallow-dipping. The problem worsens rapidly with decreasing dip, and a per-
fectly planar horizontal fracture cannot be assigned any strike at all. Even for
smooth ( joint roughness coecient JRC 02, see Barton and Choubey, 1977,
p. 7) shallow-dipping outcrop fractures in the eld, it is commonly observed
that a second measurement of the strike yields a dierent value. The discrepancy
can exceed 20

or 30

even if the compass is again placed in the same spot.


Physical simulation of the eld measurement process is essential. No purely
mathematical treatment can succeed because of the slight probability of extremely
large variations. It will be shown experimentally that the strike measurement dis-
crepancy is due largely to the shallow dip rather than to measurement error. For a
fracture with a given dip, strike measurement standard deviation (abbreviated as:
strike SD) also depends on the surface smoothness. This will be shown by experi-
ment as well.
Of course, one is especially interested in the mean dip and strike of an entire
fracture set, together with appropriate 99%-condence intervals for dip and strike.
These calculations, performed for Professor Genske's set of 19 shallow-dipping
fractures (shown in Table 3a), will bear out the asymmetry between strike SD (or
the equivalent dip direction SD) and dip SD in this case. Our transformed 99%-
condence region for the mean pole is also shown. This region is elliptical and not
nearly circular.
2. Experimental Apparatus
The MIT Civil and Environmental Engineering (CEE) articial rock fracture (Fig.
1), which is used to teach students how to use a geologic compass, consists of a
3.5 cm. thick and 22.5 cm. 51 cm. inclined surface rectangular wooden board,
hinged to a similar horizontal board. Symmetrically placed vertical aluminum
supports with appropriate holes allow insertion of a horizontal aluminum pin (the
thickest pin in Fig. 2) and enable one to create dips of 4

, 8

, 12

, 24

, 38

, and
47

. By constraining this pin at the supports between the boards, one obtains a dip
of 2

. Mr. Rudolph of the CEE machine shop provided two more aluminum
bedding pins with smaller diameters (see Fig. 2), which similarly allow one to
Fig. 1. Articial rock fracture Fig. 2. Pins for very shallow dip creation
242 H. H. W. Herda
create dips of 0X5

and 1

. The standard position of this fracture on the laboratory


table was marked so that the true strike would not change during a series of
measurements. The smoothness of the board surface is in the range of JRC 02.
All dip and strike measurements were made with a Brunton geological com-
pass (Fig. 3). A brass hinge (see Fig. 3) was assembled and used with the compass
for strike measurements at dips of less than 24

because the bottom of the compass


has a ring-shaped ridge (see Fig. 3), which interferes with strike measurement at
dips less than about 15

. At 24

, strike measurement series were made with and


without the hinge and were statistically similar. All strike measurements were
recorded to the nearest degree. To counteract the eects of local roughness and
warping on strike SD, the inclined wooden board was divided into 6 20 regions
(AF; 120), and pseudorandom numbers were used to generate the series of
measurement positions.
In a second series of strike measurements, an 18 cm. 21 cm. aluminum plate
(Fig. 4) was fastened to the upper end of the inclined articial fracture. We used
the compass on the plate to observe the eect of greater surface smoothness on
strike SD. The aluminum plate was similarly divided into 6 7 regions and mea-
surement positions on it were generated in the same way. To guard against mag-
netic contamination of all series of measurements, variations in the position and
also in the material composition of the laboratory table used were made during the
various series of strike measurements. No magnetic eect was detected.
Fig. 3. Brunton geological compass, shown with ridge and hinge
Strike Standard Deviation for Shallow-dipping Rock Fracture Sets 243
3. Experimental Results
The strike measurement series on the wooden board articial fracture are sum-
marized in Table 1. Note from this summary that, for a xed dip at which several
series of strike measurements were taken, the strike SDs for the various series are
of comparable size, arguing against large measurement error and for inherent in-
accuracy (noise) due to the shallow dip. For each series, the dip, the number of
strike measurements made (= n), and their SD are given. For extremely small dips
one should have many measurements because the strike SD is so large. The 4

dip
lies in the 2

dip to 8

dip transition zone between large strike SD (18

) and
Fig. 4. Aluminum plate on articial rock fracture
Table 1. Strike measurement series on wooden board articial fracture
Dip 0X5

0X5

0X5

Strike SD 62X5

64X6

55X9

40X9

34X3

24X9

22X5

18X7

n 104 104 103 102 103 50 102 103


Dip 4

12

24

38

47

Strike SD 12X2

16X7

11X3

13X5

5X7

3X3

2X3

2X0

2X0

n 102 102 102 52 103 103 103 52 52


244 H. H. W. Herda
small strike SD (6

); therefore the strike was measured especially often there.


Series with bimodal strike distributions (see Fig. 5) are seen only at 0X5

dip. At 1

dip and at higher dip values one encounters only series with skewed unimodal
strike distributions (Fig. 6). There is no other visible trend at any dip angle. The
slight decrease from 104 (or 52) in the number of strike measurements occurs be-
cause at certain dip angles one of the vertical supports occasionally makes it im-
possible to position the compass for measurement.
The series of strike measurements on the aluminum plate are summarized in
Table 2. They show immediately that the plate, being smoother than wood, has
much lower strike SD in each series, compared to Table 1. There are far fewer
series, and fewer measurements in each, because strike SD for the same dip is now
about half (or less) of the value for the wooden board and because this is largely a
comparative exercise, since real rock is usually not so smooth. At 0X5

dip the
strike distribution appears weakly bimodal (Fig. 7). At all higher dip values the
strike distribution is again skewed unimodal (illustrated in Fig. 8 for 1

dip).
The relationship between the dip and the strike SD is shown graphically in Fig.
9. The wooden board articial fracture strike SDs (up to 4

dip inclusive these are


weighted averages from the several measurement series shown in Table 1) are
plotted and augmented into a curve approximating strike SD for all dips between
0X5

and 47

. The aluminum plate strike SDs obtained for each dip are plotted as
small open circles in this gure.
Fig. 5. Histogram of a strike measurement series at 0X5

dip with a bimodal strike distribution


(n = 104, wooden board)
Strike Standard Deviation for Shallow-dipping Rock Fracture Sets 245
4. From Experiments to Revision of Theory
This section is intended to be read as a link between the preceding experimental
sections (2 and 3) and the following sections treating a particular shallow-dipping
fracture set (Section 5), the development of statistics (Section 6), which in turn are
used in order to correctly calculate the 99%-dip direction condence interval for
this fracture set (Section 7). The purpose is to give a proper background for the
calculation of strike condence intervals for shallow-dipping fracture sets, rather
than to make a eld study.
The 99%-condence region for the location of the mean pole of a fracture set
on the unit sphere of geology is often pictured essentially as a circle (see the bc, d1,
and d2 sets in Fig. 11). This is certainly inaccurate in the case of the bc set in Fig.
11 because all its fractures are shallow-dipping and therefore the 99%-condence
interval for the location of the true mean strike (a latitude interval centered at
the mean of the observed strikes) is much wider than the 99%-condence interval
for the location of the true mean dip (a longitude interval centered at the mean
Fig. 6. Histogram of a strike measurement series at 1

dip with a skewed unimodal strike distribution


(n = 103, wooden board)
Table 2. Strike measurement series on aluminum plate
Dip 0X5

12

24

38

47

Strike SD 30X8

14X3

9X2

3X2

4X6

2X6

1X5

1X3

1X2

1X1

n 52 50 50 50 26 26 25 26 26 26
246 H. H. W. Herda
observed dip). Both centers coincide. The resulting elliptical 99%-strike/dip con-
dence region can be rotated upward 90

to provide the 99%-condence region for


the location of the mean pole of the bc fracture set (call it the bc pole condence
region), which is seen in Fig. 12.
The preceding experiments provide strike SD data at various dips. These are
needed for calculation of the strike (or dip direction) condence intervals as out-
lined above. This will be done in Section 7. The statistical derivations in Section
6 are non-parametric (they do not depend on particular distributions of strike
measurements) and lead to two strike (or dip direction) SD's which are compared
(see Table 5) and ultimately yield a conservative estimate for the bc pole con-
dence region. Only the 13 steepest-dipping (and therefore most reliable) of the 19
strike measurements are used, based on the statistics, in order to limit the strike
SD condence interval width. This also aects the geometry of the bc pole con-
dence region.
5. Genske's Ahr River Valley bc Fracture Set
Professor Genske investigated probabilistic concepts concerning the relative safety
of unsupported slate escarpment formations near the Rhine river. One of these
scarps, the Altenahrer Felsboeschung, is located in Germany's Ahr river valley
(50X5

N, 7

E; Genske, 1988, pp. 176179). Its quasi-horizontal bc fracture set
Fig. 7. Histogram of a strike measurement series at 0X5

dip with a weakly bimodal strike distribution


(n = 52, aluminum plate)
Strike Standard Deviation for Shallow-dipping Rock Fracture Sets 247
Fig. 8. Histogram of a strike measurement series at 1

dip with a skewed unimodal strike distribution


(n = 50, aluminum plate)
Fig. 9. Relationship between the dip and the corresponding strike SD
248 H. H. W. Herda
(n = 19), all of whose fractures are tectonically part of the same fracture system
(Genske, written communication), is shown in Fig. 10 and has reported dips and
corresponding dip directions (dip direction always diers from strike by 90

) as
shown in Table 3a. For a set of dip directions, the dip direction SD is the same as
the strike SD of the set of corresponding strikes. Hence our laboratory results are
Fig. 10. The Altenahr escarpment: all quasi-horizontal fractures belong to the bc fracture system
Table 3a. Genske's Ahr River Valley bc fracture set (n = 19): reported dips and dip directions
Dip 1

16

10

22

Dip direction 290

290

305

332

333

73

87

88

110

Dip 13

10

10

14

15

13

14

17

15

13

Dip direction 114

141

144

148

154

161

161

161

168

260

Table 3b. Genske's bc fracture set (n = 19): equivalent dips and dip directions
Dip 1

16

10

22

Dip direction 110

110

125

152

153

73

87

88

110

Dip 13

10

10

14

15

13

14

17

15

13

Dip direction 114

141

144

148

154

161

161

161

168

80

Strike Standard Deviation for Shallow-dipping Rock Fracture Sets 249


statistically applicable to Genske's dip direction measurements. For the Schmidt
net plot of the poles, see Fig. 11. Some equivalent negative dips were introduced
(changing the dip direction by 180

) in order to produce a more compact dip di-


rection set. The dip direction SD (and the pole location) is the same for a dip and
its equivalent negative. The equivalent set of dips and dip directions is shown in
Table 3b.
We are now going to assume that measuring on our laboratory board is com-
parable to measuring on Genske's bc fracture set. Indeed, some of Genske's rock
requires the use of at circular measuring plates of various diameters as bases for
the measuring compass (Genske, 1988, p. 85). His dip direction SDs are much
larger, leading to a larger dip direction condence interval. We also assume
(Genske, written communication) that the true dip directions at his 19 measure-
ment locations are identical.
Fig. 11. Schmidt net plot of poles of the bc fracture set (among others) in the Altenahr escarpment
250 H. H. W. Herda
6. Lindeberg's Central Limit Theorem; Best Linear Unbiased Estimators
In this section, we use the term ``strike'' (rather than ``dip direction'') throughout.
Our goal is to nd the approximate SD of the mean strike of Genske's bc
fracture set. Since the strike distributions of the individual fractures are not known
in detail, we need a theorem which does not require them. Lindeberg's theorem
(Feller, 1966, pp. 256257) assumes that X
1
Y X
2
Y F F F Y are mutually independent
one-dimensional random variables with distributions F
1
Y F
2
Y F F F Y with expectations
E(X
k
) = 0, and variances Var(X
k
) = s
2
k
. The strike mean is
m
a
= (X
1
X
2
X
n
)anX (1)
Set s
2
n
= s
2
1
s
2
2
s
2
n
. If a certain integral condition is met (as it is in
our case, with n = 19) then the distribution of the normalized sum S
+
n
=
(X
1
X
2
X
n
)as
n
tends to the normal distribution N, with E(N) = 0 and
Var(N) = 1. We need the variance of the mean instead:
Var(1an)(X
1
X
2
X
n
) = (1an
2
) Var(X
1
X
2
X
n
)
= (s
2
n
an
2
) Var(1as
n
)(X
1
X
2
X
n
)
qs
2
n
an
2
= (s
2
1
s
2
2
s
2
n
)an
2
X (2)
The standard deviation of the strike mean is thus approximately the linear un-
biased Lindeberg SD
s
a
= (1an)(s
2
1
s
2
2
s
2
n
)
0X5
X (3)
We now derive the best linear unbiased estimator, or BLUE, of the strike mean
and SD by specializing the following formula (see Box and Tiao, 1992, p. 113,
(2.7.1) or Veneziano, 1978, p. 287, (65)) in the theory of multivariate linear
regression:
y
B
= Xy
B
e
B
X (4)
Here, y
B
is the n 1 vector of observations, X is an n k matrix of known con-
stants, y
B
is a k 1 vector of regression coecients, and e
B
is an n 1 vector of
random variables (errors) having zero means. In our case, y
B
is the n 1 vector of
strike observations s
j

n
j=1
, X is an n 1 vector of 1's, y
B
is the true strike (an un-
known 1 1 scalar, assumed to be the same for all strike observations) and e
B
is an
n 1 vector of random variable errors having zero means and variances s
2
i
. In
each coordinate, the formula thus becomes:
s
i
= y e
i
X (5)
The linear unbiased estimator for y is

y =

n
j=1
a
j
@ A
1

n
j=1
a
j
s
j
Y which implies
V(

y) = Var(

y) =

n
j=1
a
j
@ A
2

n
j=1
a
2
j
s
2
j
X (6)
Strike Standard Deviation for Shallow-dipping Rock Fracture Sets 251
To minimize this variance, we let a
i

n
i=1
vary, and we compute dV(

y)ada
i
= 0
for i = 1Y 2Y F F F Y n, for each i obtaining
2

n
j=1
a
j
@ A
3

n
j=1
a
2
j
s
2
j

n
j=1
a
j
@ A
2
2a
i
s
2
i
= 0X
Then
2

n
j=1
a
j
@ A
1

n
j=1
a
2
j
s
2
j
2a
i
s
2
i
= 0X (7)
The rst term is the same for each i, hence it is constant. Therefore a
i
is directly
proportional to s
2
i
, always with the same proportionality constant
k =

n
j=1
a
j
@ A
1

n
j=1
a
2
j
s
2
j
X (8)
Substituting ks
j

2
for a
j
everywhere in

y and V(

y), we obtain

y =

n
j=1
(1as
2
j
)
@ A
1

n
j=1
(s
j
as
2
j
) and V(

y) =

n
j=1
(1as
2
j
)
@ A
1
X (9)
Re-name

y as mean strike estimator m
b
and V(

y)
0X5
as the corresponding strike
BLUE SD
s
b
=

n
j=1
(1as
2
j
)
@ A
0X5
X (10)
7. Dip-direction and Dip Condence Intervals for Genske's Fracture Set
We want to compare the results of applying the two linear unbiased estimators of
the mean strike (or dip direction) m
a
, m
b
and their associated linear unbiased
strike (or dip direction) SD's s
a
, s
b
as derived above. Because of the likelihood of
dip direction measurement noise at extremely small dips we proceed with these
calculations iteratively, for the k largest dips among Genske's set of 19 dip/dip-
direction measurements. In Table 4, these dip measurement groups are sorted
from largest to smallest, together with the multiplicity at each dip value and the
count of the k largest dip measurements. The lower bound estimates of dip direc-
tion SDs for Genske's dips are obtained from Fig. 9 and are also inserted for each
dip value.
Let s
i
denote the ith dip direction measurement made in the eld, and let s
i
denote its estimated SD, found in the laboratory experiment. We then dene:
m
a
(k) = (1ak)

k largest dips
s
j
Y implying s
a
(k) = (1ak)

k
j=1
s
2
j
2 3
1a2
Y
252 H. H. W. Herda
and
m
b
(k) =

k
j=1
(s
j
as
2
j
)
0

k
j=1
(1as
2
j
)Y implying s
b
(k) =

k
j=1
(1as
2
j
)
@ A
1a2
X
s
a
(k) and s
b
(k) are listed for comparison in Table 5, for k = 1 to 19. This
arrangement will allow us to reject the most uncertain dip direction SDs found
at the smallest dips. There is very good agreement between s
a
(k) and s
b
(k) until
k = 13 inclusive. We believe that the subsequent sharp increase in s
a
(k) is due
to inherent dip direction measurement uncertainty at extremely small dip, since
the smallest dips of the set were arranged to occur for 13 ` k 19. We will pick
s
a
(13) = 0X90

as the conservative choice for estimated dip direction SD, and we


will similarly use Table 3b to nd m
a
(13) = 130X9

as the mean dip direction of the


13 largest dips. Since the distribution of the dip direction mean can be assumed to
be nearly normal, we nd that the 99%-condence interval for the true dip direc-
tion is approximately 130X9

q2X3

.
Table 5. Genske's bc fracture set (n = 19): separately
derived dip direction SDs s
a
, s
b
for k largest dips
k Largest dips Lindeberg SD
s
a
(k)
BLUE SD
s
b
(k)
1 2X40

2X40

2 1X81

1X79

3 1X52

1X51

5 1X23

1X22

7 1X07

1X06

10 0X92

0X91

13 0X90

0X85

15 1X24

0X85

16 1X55

0X84

18 2X13

0X84

19 2X73

0X84

Table 4. Genske's bc fracture set (n = 19): dip measurement groups sorted largest to smallest
k Largest dips 1 2 3 5 7
Multiplicity 1 1 1 2 2
or dip 22

17

16

15

14

Dip direction SD 2X4

2X7

2X8

2X9

3X0

k Largest dips 10 13 15 16 18 19
Multiplicity 3 3 2 1 2 1
or dip 13

10

Dip direction SD 3X1

4X2

10X2

16X5

20X6

35X1

Strike Standard Deviation for Shallow-dipping Rock Fracture Sets 253


Under the conservative assumption that the SD of each of the 13 largest dip
measurements is 1

, and computing the usual mean of these dips, we nd similarly


that the 99%-condence interval for the true dip is 12X0

q0X7

. Hence, the dip


direction condence interval is more than three times as wide as the dip condence
interval.
The poles (n = 19) of Genske's bc data set are plotted using a rectangular
projection in Fig. 12 (pole direction = Table 3b dip direction; pole inclination =
Table 3b dip 90

). The six poles within the dashed band correspond to the six
paired measurements, which were excluded from the condence interval calcu-
lations. The 99%-bc-mean pole condence region is also shown in Fig. 12. It is
elliptical and is just the translated form of the corresponding 99%-dip/dip direc-
tion condence region.
8. Conclusion
It has been shown that for rather at and smooth (JRC 02) rock fractures with
xed shallow dip (0X5

to 47

) the SD of the strike measurements can be estimated


experimentally. For rougher rock, these estimates provide lower bounds for strike
SD. On repeating this experiment at the same dip with an aluminum plate, the SD
Fig. 12. Rectangular projection plot of the poles of Genske's Altenahr bc data set with 99%-condence
region for the mean pole
99%-bc-mean pole condence region
70
pole direction, degrees
p
o
l
e
i
n
c
l
i
n
a
t
i
o
n
,
d
e
g
r
e
e
s
254 H. H. W. Herda
of the strike measurements was found to be about half as large. An aluminum
plate is smoother than most rock. An Ahr river valley bc fracture set (with mean
dip nearly horizontal) was used together with a result from a central limit
theorem and a second unbiased linear estimator in nding 99%-condence inter-
vals for the true dip direction and dip of this set. The rst of these intervals was
estimated to be more than three times as wide as the second. This was illustrated
using an associated pole plot featuring the 99%-mean pole condence region,
which is elliptical.
Acknowledgements
The author wishes to thank Professor Daniele Veneziano of the Department of Civil and
Environmental Engineering at MIT for advice on statistics. He is also indebted to Professor
Dieter D. Genske of the Institute of Environmental Engineering at Ecole Polytechnique
Federale de Lausanne for providing him with a copy of his Ph.D. thesis, the dip/dip direc-
tion measurements of the Ahr river valley bc fracture set, and the Projekt Ahrtal Schmidt
net plot. He wants to thank Professor Yuzo Ohnishi of the School of Civil Engineering at
Kyoto University for being his host during the sabbatical year during which he rst thought
about this problem.
References
Baecher, G. B. (1972): Site exploration: a probabilistic approach. Ph.D. Dissertation,
Department of Civil Engineering, Massachusetts Institute of Technology, 515 pp.
Barton, N., Choubey, V. (1977): The shear strength of rock joints in theory and practice.
Rock Mech. 10, 154.
Box, G. E. P., Tiao, G. C. (1992): Bayesian inference in statistical analysis. Wiley, New
York, 587 pp.
Feller, W. (1966): An introduction to probability theory and its applications, vol. 2. Wiley,
New York, 626 pp.
Genske, D. D. (1988): Ansatz fuer ein probabilistisches Sicherheitskonzept ungesicherter
Felsboeschungen im Rheinischen Schiefergebirge. Bericht No. 8. Grundbau, Boden-
mechanik und Unterirdisches Bauen. Ph.D. thesis, Bergische Universitat, Gesamt-
hochschule Wuppertal, Fachbereich Bautechnik, 210 pp.
Veneziano, D. (1978): Course notes on random processes for engineering application.
Massachusetts Institute of Technology, 304 pp.
Author's address: Prof. Hans Herda, M.I.T., 77 Massachusetts Avenue, Room 1330,
Cambridge, MA 02139 U.S.A.
Strike Standard Deviation for Shallow-dipping Rock Fracture Sets 255

Vous aimerez peut-être aussi