ARTICLE IN PRESS

International Journal of Rock Mechanics & Mining Sciences 40 (2003) 893–901

Models for normal fracture deformation under compressive loading

B. Malama, P.H.S.W. Kulatilake*
Department of Mining and Geological Engineering, University of Arizona, 1235 E. North Campus Drive Mines Bldg, #12, Rm 229,
Tucson, AZ 85721 0012, USA
Accepted 1 May 2003

Abstract

A new semi-empirical model that can be used to predict fracture deformation behavior under normal compressive loading is
presented. The development of a simple exponential model is presented first after which a modified and more general exponential
model, with an additional degree of freedom in the model parameters, is obtained. The simple and the modified exponential models
are then compared to available fracture closure models, namely the empirical Barton–Bandis hyperbolic model, and a power-law
model based on Hertzian contact theory, to determine how good they fit the results of fracture closure experiments under
monotonically increasing normal compressive loading. A new parameter called the half-closure stress, s1=2 ; is introduced and is used,
in addition to the maximum fracture closure, Dvm ; in the model fitting procedures for the Barton–Bandis and the simple exponential
model. The half-closure stress is shown to be related to the initial normal stiffness, Kni ; used in the original Barton–Bandis model.
An additional parameter, n; is used in fitting the modified exponential model to the experimental data. Of the models presented
herein, the modified exponential model was found to provide the best fit to the experimental data, for the same values of s1=2 and
Dvm ; over the entire range of compressive stresses. The power-law model based on Hertzian contact theory was found to be
unsuitable for predicting normal fracture deformation behavior.
r 2003 Elsevier Ltd. All rights reserved.

Keywords: Single fractures; Deformability; Normal stress; Experimental; Theoretical; Empirical

1. Introduction include the rock type, weathered state and matedness of

the surfaces, and the spatial and size distributions of
Fracture deformability under the action of normal asperities on the surfaces. The roughness of each
compressive stress is of fundamental importance to the bounding fracture surface is directly related to the size
study of the hydraulic and mechanical behavior of and spatial distributions of the surface asperities,
fractured rock. Fracture deformation directly affects the whereas the aperture size and spatial distributions, and
aperture distribution, contact area distribution, and the contact area distribution are functions of the cross-
spatial connectivity of the apertures, factors that govern correlation of the surface asperity spatial and size
the hydraulic conductivity of single fractures [1–3]. Since distributions [5]. The mechanical strength and deform-
fracture networks govern the hydraulic behavior of ability of the asperities are functions of the rock type
fractured rock masses, it follows directly that deform- and weathered state of the bounding surfaces. It is also a
ability of single fractures due to the action of well-established experimental observation that the shear
compressive stress would affect the hydraulic properties displacement of the bounding surfaces of a fracture,
of a rock mass. It is also generally understood that the which governs fracture mating, drastically affects the
mechanical behavior of rock masses is dominated by the aperture and contact area distributions, even for zero to
deformation of discontinuities [4]. moderate normal loads, for which the asperity spatial
The most fundamental properties of the bounding and size distributions of the individual surfaces are
surfaces of a fracture affecting fracture deformation practically unchanged [6,7].
Fracture closure under normal compressive loading is
*Corresponding author. Tel.: +1-520-621-6064; fax: +1-520-621- conventionally described by stress-deformation relations
8330. in which the normal-stiffness, Kn ; and the maximum
E-mail address: [email protected] (P.H.S.W. Kulatilake). normal joint closure, Dvm ; are used as the characteristic

1365-1609/03/$ - see front matter r 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/S1365-1609(03)00071-6
 ARTICLE IN PRESS
894 B. Malama, P.H.S.W. Kulatilake / International Journal of Rock Mechanics & Mining Sciences 40 (2003) 893–901

parameters [8]. It has been suggested however, that each time a fracture closure solution is needed [16], and
given the manner in which shear and normal fracture the apparent complexity of the resultant models,
deformation tests are conducted, the displacement and compromise the practicability of using these models
not the stress should be taken to be the dependent that assume the Gaussian or inverted chi-square
variable. In that event the measured fracture properties distribution for peak asperity heights.
are best expressed in terms of the respective shear and In the present work a comparison between the
normal compliances, Cs and Cn ; rather than the hyperbolic model of Bandis et al. [6] and Barton et al.
corresponding stiffnesses [9]. In the simple one-dimen- [7], the power-law model presented by Swan [13], and a
sional normal deformation problem this may seem to be new generalized semi-empirical exponential model de-
a trivial technicality since Cn ¼ 1=Kn : In the general veloped herein, is presented.
case, however, matrix inversion is required given the
tensorial nature of these parameters.
In the study of fracture deformability, it has been 2. Developing a model for fracture closure
generally observed that the relation between fracture
deformation, Dv; and the normal stress, sn ; is non-linear. Given that in the laboratory experiments the depen-
Shehata [10] used a semi-logarithmic empirical model to dent variable is the axial deformation, v; with the normal
describe the relation. Bandis et al. [6] have, however, stress, sn ; being the independent variable, it is more
shown that plots of Dv data against logarithmically appropriate to use intact and jointed rock compliances
scaled sn do not show a linear fit across the entire range [9], which for the one-dimensional problem are the
of normal stresses. Goodman [11,12] suggested a inverses of the respective stiffness values. For intact
hyperbolic relation to describe fracture closure under rock, linear-elasticity is assumed, and therefore, the
normal compressive stress. Bandis et al. [6] and Barton deformation, vi ; as a function of the normal stress, s;
et al. [7] presented a modification of Goodman’s with the subscript n dropped, is given by
hyperbolic model, which was shown to provide a better
fit to experimental data across the whole range of stress vi ¼ Ci s; ð1Þ
and closure values. where Ci is the intact rock compliance, assumed to be a
Alternative models, based on Hertzian contact theory, constant. As has been stated already, for jointed rock,
have also been used to describe the non-linear stress- the relation between the total deformation, v; and the
deformation behavior [9,13,14]. These models suggest compressive stress, s; normal to the fracture plane, is
that the observed non-linear behavior could be attrib- non-linear. The compliance of a jointed rock, Cj ; is
uted to the increasing contact areas and the increasing therefore, not a constant but a continuous function of s:
number of contacts as the normal stress increases. The An examination of experimental [6,7,11] and numerical
asperities are assumed to undergo linear elastic defor- simulation [9,13,15] results obtained by several research-
mation. This is contrary to Goodman’s [11] suggestion ers shows that the jointed rock compliance decreases
that the non-linearity derives from the crushing of with increasing normal compressive stress, from some
asperities. It is the authors’ intuitive view that both initial value to that of the intact rock, Ci : That is, the
factors do contribute to the non-linear behavior and gradient of the v2s curve, Cj ; decreases, monotonically
that the crushing of asperities accounts more for approaching Ci in the limit, as the normal stress
hysteresis and irrecoverable fracture deformation ob- approaches some large upper bound, typically the
served during cyclic fracture closure experiments. uniaxial compressive strength of the intact rock, though
Applying Hertzian contact theory, Swan [13] showed for simplicity it will be assumed to be infinity (Fig. 1).
that if a power-law distribution of the peak asperity Hence, the fractured rock undergoes two deformation
heights was assumed, a power-law relation between Dv phases, namely the non-linear deformation at low
and sn is obtainable. It was also shown that a model for stresses and the linear elastic behavior, observable in
fracture closure of the form presented by Shehata [10] intact rock, at stresses higher than some stress value
and Matsuki et al. [14] follows directly if an exponential related to the mechanical and deformation properties of
distribution was used for the peak asperity height [13]. the fracture.
The power-law model and semi-logarithmic model have The rate of decrease of the jointed rock normal
been shown to offer reasonable approximations only for compliance, Cj ; with increasing normal compressive
a limited range of stresses [6,13]. Numerical procedures stress, s; can be expressed mathematically as follows:
have been implemented in which the Gaussian and
qCj
inverted chi-square distributions were used to describe ¼ f ðsÞ; ð2Þ
peak asperity heights [15]. The results were shown to fit qs
fracture stress-deformation experimental data better for where f ðsÞ is a function which governs the rate of
the inverted chi-square distribution. However, require- decrease of Cj with s; and the negative signifies the
ment of implementing numerical integration schemes decreasing compliance with increasing stress. The
 ARTICLE IN PRESS
B. Malama, P.H.S.W. Kulatilake / International Journal of Rock Mechanics & Mining Sciences 40 (2003) 893–901 895

Subtracting Eq. (1) from Eq. (6) gives the joint

closure, Dv; as a function of the normal compressive
Total deformation, v Ci stress. Given that lim s-N Dv ¼ Dvm ¼ a=b2 and b ¼
Jointed rock ðln 2Þ=s1=2 ; where s1=2 ¼ sðDvm =2Þ; (Fig. 1(b)) the
equation for fracture closure can then be written as
Ci follows:

∆vm Dv ¼ Dvm ½1  eðs=s1=2 Þln 2 : ð7Þ

Intact rock

The assumption of the exponential decay for the rate

of decrease of the compliance of the fractured rock with
(a) Normal Stress, σ
increasing compressive stress was used for simplicity.
Fig. 2(a) shows how the exponential model given in
Eq. (7) fits to experimental data obtained from a normal
compression test performed on a granodiorite specimen.
∆vm The model fits the data well for low and high stress levels
Fracture closure, ∆v

but shows significant deviation at intermediate stress

levels. To obtain a model that provides an improved fit
to the experimental data for the entire range of stresses,
0.5∆vm a more general exponential model for closure, having

0.6

σ1/2 0.5
(b)
Normal Stress, σ
Fracture closure (mm)

0.4
Fig. 1. (a) A schematic of the observed stress-deformation behavior of
intact and jointed rock under normal compression. (b) Normal
0.3 Experimental data
fracture closure as a function of the applied normal stress.
Exponential model

0.2
function f ðsÞ should satisfy the condition that
lim s-N f ðsÞ ¼ 0 since lim s-N Cj ¼ Ci ¼ constant: An 0.1
exponential function of the following form satisfies this
condition: 0
0 10 20 30 40 50
f ðsÞ ¼ aebs ð3Þ (a) Stress (MPa)
where a and b are constants which give the initial rate of
0.6
decay of the jointed rock compliance and the decay
constant, respectively. As will be shown below, these
0.5
two parameters are related to the maximum fracture
closure, Dvm ; and to the normal stress value at Dvm =2:
Fracture closure (mm)

0.4
Given that Cj ¼ qv=qs; and substituting Eq. (3) in
Eq. (2), one obtains: Experimental data
0.3
q2 v Generalized exponential model
¼ aebs : ð4Þ
qs2 0.2
Eq. (4) can be solved subject to the following
conditions: 0.1

qv
vðs ¼ 0Þ ¼ 0 and lim ¼ Ci ð5Þ 0
s-N qs
0 10 20 30 40 50
yielding the following non-linear relation between the (b) Stress (MPa)
total deformation and the normal stress:
Fig. 2. Fits of (a) the exponential model, and (b) the modified
a
v ¼ 2 ½1  ebs  þ Ci s: ð6Þ (n ¼ 0:7) exponential model to the experimental data for a granodior-
b ite specimen.
 ARTICLE IN PRESS
896 B. Malama, P.H.S.W. Kulatilake / International Journal of Rock Mechanics & Mining Sciences 40 (2003) 893–901

the form given below was adopted: Hence, functional relations could be developed between
ðs=s1=2 Þn ln 2 these parameters and the joint compressive strength
Dv ¼ Dvm ½1  e ; ð8Þ
(JCS), the mean asperity height, the variance of the
where 0ono1:0: Fig. 2(b) shows how this modified and asperity height and the cross-correlation of the asperity
more general exponential model (Eq. (8)) fits to the size and spatial distributions of the bounding surface.
experimental data. It is clearly seen that the generalized Bandis et al. [6] have developed empirical relations
model provides a better fit than the simple exponential between the parameters Kni and Dvm ; and the average
model given by Eq. (7). It can be shown that the aperture thickness, JCS and joint roughness coefficient
generalized exponential model could be obtained by (JRC). Other workers have shown that surface asperity
assuming that the decay function, f (s), has the following height variability or the surface roughness could be
form: described using self-affine fractals [17–20]. In such cases
n two fractal parameters should be used to quantify
f ðsÞ ¼ aðsÞebs : ð9Þ
stationary roughness [18–20] instead of the JRC or the
The governing curve fitting parameters, Dvm and s1=2 ; mean and variance of the asperity height distribution. In
used for the generalized exponential model are the same the present work, however, no attempt is made to
as those used for the original exponential model. develop functional relations between the joint closure
The hyperbolic model for fracture closure presented parameters, s1=2 and Dvm ; and the mechanical properties
by Bandis et al. [6] could be obtained by assuming that and size distribution of the asperities on the fracture
the decay function, f (s), has the following form: surfaces. The two parameters are estimated directly
c from experimental data.
f ðsÞ ¼ ; ð10Þ The power-law model for fracture closure given by
ðd þ sÞ3
Swan [13] has the following form:
where c and d are constants that are related to Dvm and
s1=2 : The solution to the resulting governing differential Dv ¼ asb ; ð12Þ
equation is obtained subject to the conditions given in
Eq. (5). It can be shown that lim s-N Dv ¼ Dvm ¼ c=2d where a and b are constants determined empirically,
and d ¼ s1=2 : Hence, the assumption of the decay with bo1: It should be noted at the outset that this
function given by Eq. (10) leads to the following model model does not satisfy the condition lim s-N Dv ¼ Dvm ;
for fracture closure: a condition that has been repeatedly observed by several
Dvm s researchers. In fact for this power-law relation,
Dv ¼ : ð11Þ lim s-N Dv ¼ N; which may explain the observation
s þ s1=2
made by Swan [13] and Sun et al. [9] that the power-law
The original hyperbolic model of Bandis et al. [6] and provides a good fit to experimental data only at low
Barton et al. [7] is expressed in terms of the initial stress levels. The power-law model also suggests infinite
normal stiffness, Kni and the maximum joint closure, initial fracture compliance, which may be appropriate
Dvm : It can be shown, by comparing Eq. (11) to the only for some highly compliant rock fractures.
original form of the Barton–Bandis model, that Kni ¼ In the sections that follow, we describe the fracture
s1=2 =Dvm : Given the relative ease with which s1=2 closure experiments performed during the course of the
could be estimated from laboratory results compared current research and present a comparison of how well
to Kni ; s1=2 will be used, in addition to Dvm ; as the the generalized exponential, hyperbolic [6] and the
primary property governing normal fracture closure power-law [13] models fit to fracture closure data. The
in the analysis of experimental data presented in this fits to the experimental data presented herein are for a
work. given set of parameter values indicated in Table 1 for the
It should be emphasized that the parameters s1=2 and generalized exponential and hyperbolic models in
Dvm depend on the mechanical properties and the Eqs. (8) and (11), respectively, and a set of values of a
distribution of the asperities on the fracture surfaces. and b in Eq. (12) for the power-law model.

Table 1
Parameter values for rock samples tested in the laboratory experiments

Rock sample Side dimensions (mm) Dvm (mm) Ci (mm/MPa) s1=2 (MPa) n sC =s1=2 sm =s1=2 a b

Grd-1 93  106  93 0.505 0.020 1.45 0.70 15.0 20.7 0.154 0.40
Grd-2 87  99  95 0.481 0.015 1.65 0.70 15.0 18.2 0.148 0.38
Grd-3 103  104  89 0.683 0.025 1.60 0.80 10.7 12.5 0.222 0.44
Dr-1 94  105  70 0.380 0.017 2.25 0.75 13.6 9.78 0.103 0.42
 ARTICLE IN PRESS
B. Malama, P.H.S.W. Kulatilake / International Journal of Rock Mechanics & Mining Sciences 40 (2003) 893–901 897

3. Joint deformation experiments The normal compressive loads were applied using a
Structural Behavior Engineering Laboratory (SBEL)
The normal joint closure experiments were performed CT-500 test frame, equipped with a 447 constant-
on jointed diorite and granodiorite rock specimens displacement servo controller. The maximum applied
collected from surface rock exposures in Arizona. The stresses to the rock specimens were in the range of 25–
specimens came from an open-pit mine in which regular 50 MPa. These loads were sufficient to cause maximum
blasting activities are conducted. The normal deform- joint closure given that all the deformation that could be
ability tests were performed on rectangular blocks of attributed to the joint was observed to occur at lower
rock, each with a single planar joint oriented normal to loads, typically so20 MPa: A simple ‘‘rule-of-the-
the loading direction to within 1 . All the joints can be thumb’’ will be presented in this paper that could be
considered as interlocked and un-weathered. The sur- used to determine the stress value above which joint
face spatial variations of all the joint specimens are deformation makes only a negligible contribution to the
shown in Fig. 3. The specimens can be arranged in the total rock specimen deformation.
following increasing order of large-scale roughness or The normal displacement was measured by means of
major undulations: Grd-1 & Grd-2; Grd-3; Dr-1. Small- an LVDT attached to the bottom plate on which the
scale roughness of the specimens increase in the rock specimen is placed and having an accuracy of
following order: Dr-1; Grd-1 & Grd-2; Grd-3. The 71 mm, which was connected to a digital data acquisi-
dimensions of the joint specimens are summarized in tion system with an adjustable data-recording rate. The
Table 1. The respective values of the parameter n that normal joint closure, Dv; was obtained by subtracting
provided the best fit to the data for each fractured rock the deformation of an equivalent intact specimen, vi ;
specimen are included in Table 1. The observed from the total deformation of the associated jointed
variability in the values of n for the rock specimens specimen. The deformation of the equivalent intact rock
suggests that n may be a function of rock fracture was computed using Eq. (1) with the assumption that
surface properties as well as the loading history of the the intact rock compliance, Ci ; is given by the high-stress
fracture. Also included in Table 1 are the estimated compliance of the jointed specimen, i.e. lim s-N Cj ¼
values of the maximum fracture closure, Dvm ; the intact Ci : Thus, total jointed rock deformation–stress data
rock compliance, Ci ; and the half-closure stress, s1=2 : were used to estimate the value of Ci assuming that the
The parameters a and b for the power-law model fits are condition lim s-N Cj ¼ Ci was approximately satisfied
also included. in the stress ranges used in the tests.

Grd-1 Grd-2

Dr-1
Grd-3

Fig. 3. Spatial variation of the bottom surfaces of the respective fracture specimens indicated.
 ARTICLE IN PRESS
898 B. Malama, P.H.S.W. Kulatilake / International Journal of Rock Mechanics & Mining Sciences 40 (2003) 893–901

4. Analysis of normal joint deformation data compressive stress is increased beyond some character-
istic value. In the closure models presented, this
The vj 2s and Dv2s curves for all the jointed rock approach to linearity is controlled by respective non-
specimens tested showed typical nonlinear behavior linear functions, which become clear when the total
(Fig. 4). The compliance of the jointed rock, Cj ; showed deformation equations for the generalized exponential
monotonic decay to an asymptotic value of Ci as and hyperbolic models are cast into the following
depicted in Fig. 5 for one of the granodiorite specimens. dimensionless forms, respectively:
The shown jointed rock compliance, Cj ; at a given stress  
vj Ci n
value, was computed using the approximate relation ¼ 1þ s  eðs=s1=2 Þ ln 2 ; ð13Þ
Dvm Dvm
Cj ¼ dvj =ds: It is also clear from the vj 2s curves that
 
there is an approach to linearity as the normal vj Ci s1=2
¼ 1þ s  : ð14Þ
Dvm Dvm s þ s1=2
1.6 The term in parenthesis in Eqs. (13) and (14)
constitutes the linear part that is approached as the
1.4 joint undergoes deformation, while the second term
constitutes the nonlinear contribution to the total
1.2 deformation. This latter term governs the rate of
Total deformation, v j (mm)

approach to linearity. This nonlinear term decays with

1 increasing stress, with the decay rate being exponential
in Eq. (13) and hyperbolic in Eq. (14). The value of the
0.8
stress at which the non-linear term becomes negligible
0.6
will be taken to be that value above which the
Grd-1 contribution of this term to the total deformation, vj ;
0.4 Grd-2 is less than 1% of the maximum possible joint
Grd-3 deformation, Dvm : For the generalized exponential
0.2 model, this characteristic non-dimensional stress is
given by
0  
0 10 20 30 40 50 60 sC ln 100 1=n
¼ : ð15Þ
Stress (MPa) s1=2 ln 2
Fig. 4. Nonlinear deformation–stress behavior for jointed rock Hence, the characteristic non-dimensional normal
specimens of granodiorite under normal compression loading. compressive stress above which the contribution of the
nonlinear term is negligible has the approximate value of
15.0 for the generalized exponential model with n ¼ 0:7;
0.45 and the approximate value of 100.0 for the hyperbolic
model. Given that n is not a constant for different rock
0.4 specimens, the value of this non-dimensional stress has
corresponding variability as indicated in Table 1. It
0.35 should be noted that the values of the characteristic
dimensionless stress given in the table are those
Cj = δv/ δσ (mm/MPa)

0.3
predicted by the generalized exponential model as given
0.25 in Eq. (15). Hence, for values of the non-dimensional
normal stress greater than the characteristic values
0.2 specified in Table 1, for the respective values of n in
the case of the generalized exponential model, or
0.15 sC =s1=2 > 100 for the Barton–Bandis model, the vj 2s
curves could be used to determine the intact rock
0.1
compliance, Ci : However, for this fact to be useful, the
0.05
value of s1=2 has to be known a priori. This would
Ci
require the knowledge of the functional dependence of
0 s1=2 on the mechanical strength, and on the size and
0 10 20 30 40 50 spatial distribution of asperities on the joint surfaces,
Normal stress (MPa) subjects that are not explored in this paper. Hence, an
Fig. 5. Monotonic decay of jointed rock compliance, Cj ; with alternative procedure of estimating s1=2 from experi-
increasing normal stress, to the asymptotic intact rock value of Ci : mental data is required. The suggested procedure
 ARTICLE IN PRESS
B. Malama, P.H.S.W. Kulatilake / International Journal of Rock Mechanics & Mining Sciences 40 (2003) 893–901 899

involves estimating Dvm and the intact rock compliance, 0.8

Ci ; first from the high-stress linear portion of the vj 2s

data plot, where Dvm is the vj -intercept and Ci is the
slope (Fig. 1(a)). It should be noted that both Barton–
Bandis model and the exponential models presented in 0.6

this paper reduce to the same linear equation for large

Fracture closure (mm)

values of s: Having estimated Dvm and Ci ; and using the
assumption of linear elasticity for intact rock, joint
closure data could be obtained from which the value of 0.4

s1=2 is then determined using the normal joint deforma-

tion–stress (Dv2s) curves as depicted in Fig. 1(b).
As already stated, Table 1 includes the values of the
parameters Dvm and s1=2 ; estimated using the procedure 0.2 Experimental data
Exponential model
outlined above for the jointed rock specimens for which Hyperbolic model
the experimental results are presented herein. The table Generalized exponential model
also includes the mean intact rock compliances, Power-law model
0
estimated from the respective vj 2s curves, and the
ratios sm =s1=2 ; where sm is the minimum value of the 0 20 40 60
Normal stress (MPa)
measured stresses over the linear interval of the vj 2s
data plot over which the parameters Dvm and Ci are Fig. 7. Experimental results for a granodiorite fractured rock speci-
estimated for each jointed rock specimen. The men, Grd-2, tested under normal compressive loading, and model fits
using the parameter values given in Table 1.
rock specimens Grd-1, Grd-2 and Grd-3 were subjected
to sm =s1=2 values that are greater than their res-
pective sC =s1=2 -values predicted by the generalized
exponential model, whereas Dr-1 was subjected to 1

sm =s1=2 osC =s1=2 : For Dr-1 the contribution of the

nonlinear term to the total rock deformation at sm was
about 0:022Dvm ; which is still sufficiently a low 0.8
contribution. It should be noted also that in the case
of the hyperbolic model of Barton and Bandis,
Fracture closure (mm)

sm =s1=2 osC =s1=2 ; for all the rock specimens subjected 0.6
to normal fracture closure tests, since sC =s1=2 E100:

0.4
0.8

Experimental data
0.2 Exponential model
Hyperbolic model
0.6 Generalized exponential model
Power-law model
Fracture closure (mm)

0 10 20 30
Normal stress (MPa)
0.4
Fig. 8. Experimental results for a granodiorite fractured rock speci-
men, Grd-3, tested under normal compressive loading, and model fits
using the parameter values given in Table 1.

0.2 Experimental data

Exponential model
Hyperbolic model Hence, according to the fracture closure behavior
Generalized exponential model
Power-law model predicted by the Barton–Bandis model, the procedures
0 outlined above cannot be used to determine the
0 10 20 30 40 50 appropriate model fitting parameters given in Table 1,
Normal stress (MPa) for the range of normal stresses used in the experiments
Fig. 6. Experimental results for a granodiorite fractured rock speci- presented in this work. It is, however, clearly evident
men, Grd-1, tested under normal compressive loading, and model fits from the results obtained in these experiments that the
using the parameter values given in Table 1. fractured rock specimens closely approach linear vj 2s
 ARTICLE IN PRESS
900 B. Malama, P.H.S.W. Kulatilake / International Journal of Rock Mechanics & Mining Sciences 40 (2003) 893–901

0.4 1.6

0.3 1.2
Fracture closure (mm)

Fracture closu re (mm)

0.2 0.8

0.1 Experimental data 0.4 Experimental data

Exponential model Exponential model
Hyperbolic model Hyperbolic model
Generalized exponential model Generalized exponential model
Power-law model
0 0

0 5 10 15 20 25 0 10 20 30 40 50
Normal stress (MPa) Normal stress (MPa)

Fig. 9. Experimental results for a diorite fractured rock specimen, Dr- Fig. 10. The total deformation–stress (vj 2s) behavior for Grd-1,
1, tested under normal compressive loading, and model fits using the together with the model fits for the simple exponential, modified
parameter values given in Table 1. exponential and the hyperbolic models.

behavior for the range of normal stresses used, and that transitional behavior from the fracture-dominated non-
the requirement of sm =s1=2 > 100 is too stringent. linear deformation to the intact rock-dominated linearly
The parameters a and b used for fitting the power-law elastic deformation. This is made clear in Fig. 10 which
model to the experimental data are also included in shows the total deformation–stress (vj 2s) behavior for
Table 1. These parameters are estimated by way of Grd-1, together with the model fits for the simple
linear regression analysis of log ðDvÞ2log ðsÞ data. As exponential, generalized exponential and the hyperbolic
shown below, the power-law model did not provide models. The hyperbolic model of Barton–Bandis does
good fits to the experimental data to warrant further not predict an approach to the asymptotic fracture
consideration of the model. Using the parameter values closure value, Dvm ; which is as rapid as what is observed
given in Table 1, the exponential, generalized exponen- in the test results, for the set of parameters values of Dvm
tial, hyperbolic and power-law models were compared and s1=2 obtained from the parameter estimation
on how best they fit the experimental data. The results of procedure discussed in the previous section. Thus, for
the model fitting procedure are shown in Figs. 6–9. the range of stresses used in the research reported here,
the hyperbolic model could not fit the experimental data
for values of closure close to Dvm : It underestimates the
5. Discussion and conclusions amount of fracture closure values in the high stress
region (Figs. 6–9), since it predicts transitional behavior
An examination of experimental results presented between the nonlinear and linear deformation regimes
herein shows that, for the fractured rock specimens that spans a very wide and prolonged range of stresses
tested in this research work, the modified or generalized (Figs. 6–10). It is important also to note that, as was
exponential model for fracture deformation given by stated in previous section, the hyperbolic model predicts
Eq. (8) provides the best fit when compared to the that a characteristic dimensionless compressive stress of
simple exponential model of Eq. (7), the Barton–Bandis sC =s1=2 E100 is required to cause contribution of the
hyperbolic model, and the power-law model. For joint deformation to the total rock deformation to
monotonically increasing loading experiments, it has decrease to a negligible amount. Clearly, this is contrary
been observed that the simple exponential model of to what is suggested by the experimental results
Eq. (7) predicts an approach to the maximum fracture obtained in this work.
closure value, Dvm ; at a rate faster than is generally The parameter n in the generalized exponential model
observed in the experimental results. This causes the shows variability from one rock specimen to another.
model to deviate from observed closure behavior at This suggests that this parameter could be a function of
intermediate stress values as can be seen in Figs. 2 and rock material type, the roughness and weathered state of
6–9. The model therefore, fails to accurately predict the the fractures, and the loading history of a given rock
 ARTICLE IN PRESS
B. Malama, P.H.S.W. Kulatilake / International Journal of Rock Mechanics & Mining Sciences 40 (2003) 893–901 901

specimen. The dependence of this parameter, n; on [8] Goodman RE, Taylor RL, Brekke TA. A model for the
loading history could be made particularly evident by mechanics of jointed rock. J Soil Mech Fdns Div, Proc Am Soc
performing cyclic loading experiments. Civ Eng 1968;94(SM3):637–59.
[9] Sun Z, Gerrard C, Stephansson O. Rock joint compliance tests
for compression and shear loads. Int J Rock Mech Min Sci
Geomech Abstr 1985;22(4):197–213.
Acknowledgements [10] Shehata WM. PhD thesis (1971), quoted in Sharp JC and Maini
YNT, in fundamental considerations on the hydraulic character-
Acknowledgement is made to the Donors of The istics of joints in rock. Proceedings of the Symposium on
Petroleum Research Fund, administered by the Amer- Percolation Through Fissured Rock, paper no. T1-F, Stuttgart,
1972.
ican Chemical Society and the US National Science
[11] Goodman RE. The mechanical properties of joints. Proceedings
Foundation (Grant no. CMS-0085059) for supporting of the Third Congress on ISRM, Denver, vol. 1A, 1974.
this research. The authors would also like to express Washington, DC: National Academy of Sciences. p. 127–40.
their gratitude to Robert Morgan and Mijia Yang for [12] Goodman RE. In: Methods of geological engineering in
their assistance and input during the performance of the discontinuous rock. New York: West, 1976. p. 172
[13] Swan G. Determination of stiffness and other joint properties
fracture closure experiments.
from roughness measurements. Rock Mech Rock Eng 1983;16:
19–38.
[14] Matsuki K, Wang EQ, Sakaguchi K, Okumura K. Time-
References dependent closure of a fracture with rough surfaces under
constant normal stress. Int J Rock Mech Min Sci 2001;
[1] Iwai K. Fundamental studies of fluid flow through a single 38(5):607–19.
fracture. PhD thesis, University of California, 1976. 208pp. [15] Brown SR, Scholz CH. Closure of random elastic surfaces in
[2] Tsang YW, Witherspoon PA. Hydromechanical behavior of a contact. J Geophys Res 1985;90(B7):5531–45.
deformable rock fracture subject to normal stress. J Geophys Res [16] Yoshioka N. Elastic behavior of contacting surfaces under
1981;86(B10):9287–98. normal loads: a computer simulation using three-dimensional
[3] Tsang YW, Witherspoon PA. The dependence of fracture surface topographies. J Geophys Res 1994;99(8):15549–60.
mechanical and fluid flow properties on fracture roughness and [17] Brown SR, Scholz CH. Broad band width study of the
sample size. J Geophys Res 1983;88(3):2359–66. topography of natural rock surfaces. J Geophys Res 1985;90:
[4] Hungr O, Coates DF. Deformability of rock joints and its relation 12575–82.
to rock foundation settlements. Can Geotech J 1978;15:239–49. [18] Shirono T, Kulatilake PHSW. Accuracy of the spectral
[5] Lanaro F. A random field model for surface roughness and method in estimating fractal/spectral parameters for self-
aperture of rock fractures. Int J Rock Mech Min Sci 2000; affine roughness profiles. Int J Rock Mech Min Sci 1997;34(5):
37:1195–210. 789–804.
[6] Bandis SC, Lumsden AC, Barton NR. Fundamentals of rock [19] Kulatilake PHSW, Um J, Pan G. Requirements for accurate
joint deformation. Int J Rock Mech Min Sci Geomech Abstr quantification of self-affine roughness using the variogram
1983;20(6):249–68. method. Int J Solids Struct 1998;35(31–32):4167–89.
[7] Barton NR, Bandis SC, Bakhtar K. Strength, deformation and [20] Kulatilake PHSW, Um J. Requirements for accurate quantifica-
conductivity coupling of rock joints. Int J Rock Mech Min Sci tion of self-affine roughness using the roughness-length method.
Geomech Abstr 1985;22(3):121–40. Int J Rock Mech Min Sci 1999;36(1):1–18.