Computers and Geotechnics 84 (2017) 129–137

Contents lists available at ScienceDirect

Computers and Geotechnics

journal homepage: www.elsevier.com/locate/compgeo

Research Paper

Numerical study on the influence of cross-sectional shape on strength

and deformation behaviors of rocks under uniaxial compression
Yuhang Xu, Ming Cai ⇑
Bharti School of Engineering, Laurentian University, Sudbury, Ont. P3E 2C6, Canada
MIRARCO, Laurentian University, Sudbury, Ont. P3E 2C6, Canada

a r t i c l e i n f o a b s t r a c t

Article history: Cross-sectional shape effect, which has not been well studied, is one of the geometry effects that influ-
Received 11 March 2016 ence rock laboratory test results. In order to investigate the influence of cross-sectional shape on the
Received in revised form 6 October 2016 strength and deformation behaviors of rocks, a comprehensive numerical experiment is carried out to
Accepted 23 November 2016
simulate the deformation responses of circular, square, and rectangular cross-sectionally shaped speci-
mens in uniaxial compression. The validity of the numerical model is first examined by comparing the
uniaxial compressive strengths (UCS) of cylinder and square prism specimens obtained from the numer-
Keywords:
ical modeling with these obtained in laboratory tests. Both the numerical modeling and laboratory test
Cross-sectional shape
Numerical experiment
results show that the cross-sectional shape has a very small influence on the UCS of rocks. However,
Uniaxial compression the numerical results show that the cross-sectional shape affects the post-peak behaviors of rocks con-
Hoop tension effect siderably. It is also concluded that hoop tension contributes little to affecting rock strength. It is revealed
UCS through the numerical study that in the laboratory tests because the square prism specimens with a slen-
Post-peak behavior derness (defined by specimen height divided by specimen width) the same as that of a cylinder specimen
have an equivalent diameter larger than that of the cylinder specimens, a slightly higher strength of the
square prism specimens is thus observed. It is suggested to use the equivalent diameter of a non-circular
cross-section to define the slenderness of a specimen to present laboratory test and numerical simulation
results consistently.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction
Laboratory test results of rocks depend on many factors. Test
results have shown that compared with smaller specimens larger
specimens of the same slenderness (height to width or diameter
ratio) tend to have a lower rock strength [1], and this is known
as size effect. Aside from specimen size, the shape of specimens
can affect laboratory test results. The shape effect includes the
influence of the slenderness and the cross-sectional shape (circular
or square) of a rock specimen on its uniaxial compressive strength
(UCS) and post-peak stress–strain curve. It is documented that due
to the end effect caused by friction at the specimen-platen con-
tacts, the UCS of rocks increases with the decrease of the slender-
ness of the specimens [2,3]. However, the influence of cross-
sectional shape on UCS of rocks is not well studied because circular
rather than square cross-sectionally shaped specimens are often
used in rock mechanical property testing, as suggested by the
⇑ Corresponding author at: MIRARCO, Laurentian University, 935 Ramsey Lake
Road, Sudbury P3E 2C6, Canada.
E-mail addresses: yxu2@laurentian.ca (Y. Xu), mcai@laurentian.ca (M. Cai).
International Society for Rock Mechanics (ISRM) [4]. True triaxial
testing has become popular in recent years and prismatic (square
or rectangular cross-sectionally shaped) specimens are routinely
employed in true triaxial testing [5–10]. Unfortunately, the influ-
ence of specimen’s cross-sectional shape on rock’s peak strength
under true triaxial loading is again not well studied.
The importance of obtaining not only the peak strength but also
the complete stress–strain curve of rocks from laboratory tests has
been recognized because the post-peak behavior of rocks affects
rock stability [11–14]. For instance, a good knowledge of the
post-peak behavior of rocks is needed to estimate the depth of fail-
ure accurately for rock support design [15] and to avoid violent pil-
lar failure [16,17]. Thus, it is also important to study the influence
of cross-sectional shape on the post-peak behavior of rocks.
Few systematic studies focused on the cross-sectional shape
effect. Hoop tension [18–21] can be induced by the geometry of
a cylinder specimen in compression (Fig. 1b) and it may influence
crack propagation and hence the strengths of rocks in laboratory
and in-situ. Hoop tension also exists in square prism specimens
but due to the shape difference, it would be less than that in cylin-
der specimens. For a rectangular prism specimen, hoop tension

http://dx.doi.org/10.1016/j.compgeo.2016.11.017
0266-352X/Ó 2016 Elsevier Ltd. All rights reserved.
130 Y. Xu, M. Cai / Computers and Geotechnics 84 (2017) 129–137
expansion under
compression
hoop tension
crack
(a) (b)
Fig. 1. Illustrations of hoop tension and its influence on crack propagation: (a) crack
propagation on a non-circular shaped cross section (e.g., near an excavation
boundary); (b) crack propagation on a circular shaped cross section (e.g., in a
cylinder specimen).
would be even less. If hoop tension does influence rock strength, it
can be hypothesized that the strength of a cylinder specimen
should be higher than the strength of a square or a rectangular
prism specimen.
Uniaxial compression tests using cylinder and square prism
specimens, sampled from a large block of Beishan granite, were
conducted by Zhao et al. [22] and the test results indicate that
there is no significant cross-sectional shape effect on the UCS of
the rock. Due to the difficulty in preparing and conducting UCS
tests using rectangular prism specimens, the conclusion on the
hoop tension effect was inclusive. Furthermore, the post-peak
deformation behavior of the rocks was not investigated in the lab-
oratory tests.
Inspired by the laboratory test conducted by Zhao et al. [22], we
plan to use numerical experimental approach to study the cross-
sectional shape effect in the present study. First, the strength
parameters of Beishan granite [22,23] are obtained for the numer-
ical modeling through a detailed model calibration. Rock strengths
of cylinder and square prism specimens are then obtained from the
numerical modeling and compared with the laboratory results
[22]. Subsequently, more numerical experiments are conducted
to investigate the influence of cross-sectional shape on the UCS
by adding results of rectangular prism specimens with different
slenderness. Finally, the influence of cross-sectional shape on the
post-peak behavior of rocks is studied.
2. Influence of cross-sectional shape on UCS of rocks
2.1. Numerical model and model parameters
A numerical experiment using the ABAQUS/Explicit FEM tool is
carried out to study the cross-sectional shape effect in uniaxial
compression tests. ABAQUS/Explicit is a powerful tool in solving
highly nonlinear structure system problems under transient loads
by employing the explicit numerical scheme. It is also robust to
solve problems involving complex boundary conditions with effi-
cient contact convergence and oscillation control.
The material properties of Beishan granite are calibrated first to
simulate the uniaxial compression tests. The pre-peak behavior of
the rock specimens in compression is simplified as linear elastic,
and the elastic properties (Poisson’s ratio and Young’s modulus),
obtained from the uniaxial compression test results, are summa-
rized in Table 1. The peak strength and the post-peak behavior of
the rock specimens are governed by strength parameters. Mohr-
Coulomb failure criterion with a tension cut-off is employed. Based
on the fitting equation for the triaxial compression test results of
Beishan granite under low confinement (0–5 MPa) [23], the fric-
tional strength parameter of the rock is obtained (Table 1). Then,
based on the UCS of Beishan granite [22], cohesive strength param-
eters of the rock (Table 2) are calibrated and used in this numerical
experiment. Tensile strength was not provided in the test results of
Beishan granite; thus the calibration for tension cut-off was based
on the data compiled in [24,25], and the strength ratio of UCS to
tensile strength was taken as 20.
Steel loading platens used to apply a constant loading velocity
onto the ends of the specimen are modeled to honor laboratory test
conditions; steel property (E = 200 GPa, m = 0.3) is assigned to the
platens. 3D simulation models of the specimens with standard
slenderness (H/W = 2 or H/D = 2, where H is the height of the spec-
imens, W is the width of the square prism specimen, and D is the
diameter of the cylinder specimen) subjected to uniaxial compres-
sion by two steel platens are shown in Fig. 2. The geometry of the
cylinder specimen is 50 mm in diameter and 100 mm in height
(Fig. 2a), and that of the square prism specimen is 50 mm in width
and 100 mm in height (Fig. 2b), which are same as the dimensions
of the laboratory test specimens [22].
In the laboratory tests, the ends of the specimens were lubri-
cated with a thin layer of Vaseline to reduce the end effect. A coef-
ficient of friction (l) of 0.1, recommended from some researchers
[26,27], is used for the specimen-platen contacts in the numerical
modeling. In addition to the simulation of the cylinder and square
prism specimens used in the laboratory tests, a rectangular prism
specimen is considered in our numerical modeling to study the
shape effect. The cross-section of the rectangular prism specimen
is 70 mm in length and 35 mm in width, with a cross-sectional area
of 2450 mm2, which is very close to the cross-sectional area of the
square prism specimen (2500 mm2). The height of the rectangular
prism specimen is 100 mm, which is the same as that of the cylin-
der and the square prism specimens.
2.2. Modeling results
Fig. 3 presents the modeling results of the UCS of the cylinder,
square prism, and rectangular prism specimens with different slen-
derness varying from 1.0 to 2.5, along with the UCS obtained from
the laboratory tests for a slenderness of 2.0 [22]. Because there is
no agreed definition of the slenderness of a rectangular prism spec-
imen, the slenderness of the rectangular prism specimens pre-
sented in the figure is defined by the specimen height divided by
the specimen’s equivalent width that results in the same cross-
section area as the square prism specimen. Due to the end effect
that can activate confined zones near the specimen ends (Fig. 4),
the UCS of the specimens increases as the slenderness decreases.
It is seen that there is no significant difference of UCS among differ-
ent cross-sectionally shaped specimens with the same slenderness.
The UCS of the square prism specimens is, in fact, slightly higher
than that of the cylinder specimens of the same slenderness. The
difference of the UCS between the cylinder and the square prism
specimens increases with the decrease of the slenderness due to
the increased end effect.
The numerical simulation results are in good agreement with
the laboratory results of Zhao et al. [22]. In their laboratory tests,
eight specimens of each shape with a slenderness of 2.0 were
tested and the mean UCS and the standard deviations of the cylin-
der and the square prism specimens were 132.1 MPa, 4.58 MPa,
and 135.8 MPa, 7.20 MPa, respectively. The laboratory results
showed that using carefully prepared specimens for testing, the
average UCS of the cylinder and the square prism specimens were
close to each other. In fact, the average UCS of the square prism
specimens is 2.7% higher than that of the cylinder specimens,
and our numerical modeling result is in good agreement with the
test result. Therefore, it is proven both experimentally and numer-
ically that the cross-sectional shape has a limited influence on the
UCS of rocks.
 Y. Xu, M. Cai / Computers and Geotechnics 84 (2017) 129–137 131

Table 1
Laboratory test data for model parameter calibration.

Elastic properties Mohr - Coulomb strength parameters

Poisson’s ratio: 0.22 (0, 145.99) (1, 154.63)
(r2 = r3, r1)
Young’s modulus 38 (2, 163.83) (3.5, 182.44)
(MPa):
(GPa): (5, 206.91)
UCS (MPa): 132 Fitting equation: r1 = 12.129 r3 + 142.86
Data source: [22] Data source: [23]

Table 2 modeling, it is hence not possible to compare the post-peak

Strength and deformation parameters of
Beishan granite used in the numerical
experiment.
Parameters Value
Poisson’s ratio 0.22
Young’s modulus (GPa) 38
Cohesion (MPa) 19
Residual cohesion (MPa) 1
Tension cut-off (MPa) 7 of the same rock type, depending on the servo-control loading
Residual tension (MPa) 0.01
Friction angle (°) 58
3. Influence of cross-sectional shape on post-peak behavior of
rocks
3.1. Strain-softening behavior of rocks
In this section, we study the influence the cross-sectional shape
of a specimen on its post-peak behavior. Comparing with the pre-
peak deformation behaviors and peak strengths for rock specimens
with the same geometry in the laboratory results, the post-peak
stress–strain curves are variable. This is because that the mechan-
ical behavior of rocks is influenced by rock heterogeneity, and this
is especially the case in the post-peak deformation stage, where
localized failure normally takes place [29,30]. Obviously, it is diffi-
cult to investigate the problem using laboratory tests and the
numerical experimental approach is used. Moreover, as justified
later that instead of using the circumferential strain-controlled
loading as that used in the laboratory testing, the axial strain
controlled-loading is used in the simulation. Because different
loading methods were used in the laboratory test and numerical
stress–strain curves.
The post-peak behaviors of rocks in uniaxial compression are
usually classified into Class I and Class II failure types [31]. Class
I failure type shows a strain-softening behavior. Class II failure
type, on the other hand, shows that the post-peak strength
decreases with the decrease of axial strain. Both Class I and Class
II failure types can be observed in uniaxial compression test results
methods (axial strain control or lateral or circumferential strain
control) used. Therefore, the post-peak behavior of rocks is loading
condition dependent [2,32,33]. The risk of violent rock failure often
forces tests to be conducted using lateral or circumferential strain
controlled-loading. Thus, Class II type stress–strain curves are
commonly seen in most laboratory test results.
Class I failure type is considered for the Beishan granite speci-
mens in this numerical experiment. This is because that we want
to focus on investigating the influence of cross-sectional shape
on rock deformation behaviors, and it is better to keep the loading
condition simple. Class II failure type can only be obtained using
the lateral or circumferential-strain-controlled loading, but it is
simple to use the axial-strain-controlled loading in a numerical
experiment. More importantly, using the axial strain-controlled
loading might be more appropriate to reflect the actual loading
condition in situ. For instance, the loading condition in a pillar is
axial deformation controlled, and the in situ complete load–defor-
mation curves of pillars obtained by field testing so far are all Class
I [34,35].
Strain-softening behavior is assumed in the numerical experi-
ment to capture the post-peak behaviors of different cross-
sectionally shaped specimens in uniaxial compression tests. This
study focuses on studying the post-peak stress–strain curves of
rocks, and the simulation of the explicit fracturing process is

ram
steel
platen

rock
specimen

steel
platen
ram
(a) (b) (c)
Fig. 2. 3D models of (a) cylinder, (b) square prism, and (c) rectangular prism specimens for uniaxial compression test simulation.
132 Y. Xu, M. Cai / Computers and Geotechnics 84 (2017) 129–137

160 3.2. Modeling results

cylinder (laboratory)
square prism (laboratory) Fig. 5 presents the complete stress–strain curves of different
cylinder (simulation)
cross-sectionally shaped specimens in uniaxial compression. It is
150 square prism (simulation)
rectangular prism (simulation)
seen that the peak strengths of different cross-sectionally shaped
specimens are similar to each other. However, the post-peak
UCS (MPa)

behaviors of the specimens are influenced by their cross-

140 sectional shapes. The post-failure slope of the square prism speci-
men is the smallest, the rectangular prism specimen is the largest,
and the cylinder specimen is between these two.
130

120
1 2 3 4.1. End effect for peak load
Slenderness
Fig. 3. UCS of circular, square, and rectangular cross-sectionally shaped specimens
with different slenderness obtained from uniaxial compression test simulation.
Laboratory test results for a slenderness of 2.0 are also shown [23].
c
c ends. The elements in these ‘‘highly” confined zones have higher
Fig. 4. Illustration of end effect in specimens with different slenderness in rock
uniaxial compression tests (redraw based on [28]).
beyond the scope this research. Because the fracturing process dur-
ing the post-peak deformation stage leads to a cohesion loss [36],
the strain-softening behavior of the rock can be defined by degrad-
ing the rock’s cohesion strength as a function of plastic strain
(defined in Table 3).
A fundamental problem of incorporating strain-softening mate-
rial models in a standard continuum model such as ABAQUS is the
inherent mesh size sensitivity, which is an unsettled problem in
numerical modeling in continuum mechanics [37,38]. The mesh
sizes for all the specimens with different cross-sectional shapes
4. Discussions
The modeling results demonstrate that there is no significant
influence of cross-sectional shape on the peak strength of rocks,
which is supported by the laboratory test results [22]. An advan-
tage of numerical modeling is that it allows a detailed investigation
of the mechanism that governs the observed phenomenon. Fig. 6
presents the distributions of confined elements (r3 < 0, compres-
sive stress is negative in ABAQUS) in the specimens (the 1st
row), the contours of the minimum principal stress (r3) on the
side, top, and two vertical surfaces revealed from a quarter cut of
the specimens (the 2nd row), and the distribution of confined ele-
ments with relatively high confinements (r3 < 1.5 MPa) in the
specimens (the 3rd row) at peak load. The distributions of the con-
fined elements and the r3 contours in different cross-sectionally
shaped specimens are similar at their peak loads because their
slenderness and contact friction are the same. Consequently, spec-
imens with different cross-sectional shapes have similar peak
strengths. The confined elements whose confinements (absolute
values) are greater than 1.5 MPa at peak load are lumped near
the specimen’s ends, forming cone-shaped confined zones, and
the tops of which are approximately 15 mm from the specimen’s
peak strengths due to increased confinements. Therefore, hour-
glassing failure mode is normally observed in uniaxial compression
tests [39,40]; it can also be observed in pillar failure in the field
[41] because of strong end constraint to the pillars.
Both the numerical simulation and the laboratory test results
show that the UCS of a square prism specimen with a slenderness
of 2.0 is higher than that of a cylinder specimen of the same slen-
derness. Table 4 presents the percentages of the confined elements
(r3 < 0) to the total number of elements in each specimen, along
with the average minimum principal stresses (r3) in the whole
specimen, in the portion of 15 mm to the specimen’s end and in
50 mm
140
50 mm

are relatively fine and the mesh shapes (hexahedron) are the same
in each model. In this fashion, numerical errors associated with 120
mesh size are minimized and a relative comparison of the simula- 50 mm
Stress (MPa)

100
tion results can be made because all the models are within the
same order of numerical errors. 80

60

Table 3 40 70 mm
35 mm

Strain-softening parameters of rock

used in numerical modeling. 20

Cohesion (MPa) Plastic strain 0

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
19 0
Strain (%)
16 0.008
9.5 0.035
Fig. 5. Stress–strain curves of different cross-sectionally shaped specimens in
1 0.090
uniaxial compression.
 Y. Xu, M. Cai / Computers and Geotechnics 84 (2017) 129–137 133

Confined
elements with 3
< 0 in specimens
(compressive
stress is negative
in ABAQUS)

Contour of 3
(unit: Pa)

Confined
elements with 3
-1.5 MPa in
specimens
(compressive
stress is negative
in ABAQUS)

(a) (b) (c)

Fig. 6. Distributions of confined elements at two r3 thresholds and r3 contours in different cross-sectionally shaped specimens at peak load: (a) cylinder, (b) square prism,
and (c) rectangular prism specimens.

Table 4 the average confinement in the specimen’s end zones is much

Percentages of confined elements to the total number of elements and average higher than that in the middle portions of the specimens, showing
minimum principal stresses (r3) in various portions of the specimens at peak load.
that the end effect has a large influence on the stress condition
Cylinder Square Rectangular near the specimen’s end which can influence the peak load [42].
specimen prism prism
specimen specimen
Total number of elements 8844 9537 9108 4.2. End effect for post-peak behavior
Number of confined elements 6836 8531 7178
Percentage of confined elements at 77% 89% 79% As seen in Fig. 5, the post-peak behavior of a specimen depends
peak load
on its cross-sectional shape. Fig. 7 presents the distributions of
Average r3 (MPa) in the whole 1.22 1.40 1.15
specimen at peak load confined elements at two r3 thresholds and the r3 contours in
Average r3 (MPa) in the portion of 3.95 4.00 3.70 the three cross-sectionally shaped specimens at axial strain
15 mm to the specimen end at e = 0.5% in the post-peak deformation stage. It is seen that due to
peak load
the end effect, the square prism specimen has the highest local
Average r3 (MPa) in the 70 mm 0.04 0.27 0.04
middle portion of the specimen
confinement in the post-peak deformation stage and the rectangu-
at peak load lar prism specimen has the lowest local confinement. Table 5 pre-
sents the percentages of confined elements and the average
minimum principal stresses in the three cross-sectionally shaped
specimens at e = 0.5%. The results show that locally confined ele-
the 70 mm middle portion of the specimen at peak load. The r3 ments in a specimen are reduced (comparing with the numbers
value in each confined element due to end constraint controls shown in Table 4) due to the progressive failure of the rocks, and
the overall strength enhancement of the specimen. The square the degrees of decrease of the confined elements in the post-
prism specimen has the highest percentage of confined elements peak deformation stage is different for the three specimens. The
in its volume and the highest confinement (absolute value, or the square prism specimen has the highest percentage of confined ele-
lowest r3 value) than the other two specimens. This is why it ments and the smallest tensile stress r3; hence, its strength reduc-
has a higher UCS than the other two specimens. For all specimens, tion in the post-peak deformation stage is the smallest. In contrast,
134 Y. Xu, M. Cai / Computers and Geotechnics 84 (2017) 129–137

Confined
elements ( 3 <
0) in specimens
(compressive
stress is
negative in
ABAQUS)

Contour of 3
(unit: Pa)

Confined
elements
( 3 -1.5 MPa)
in specimens
(compressive
stress is
negative in
ABAQUS)

(a) (b) (c)

Fig. 7. Distributions of confined elements at two r3 thresholds and r3 contours in different cross-sectionally shaped specimens at e = 0.5% in the post-peak deformation
stage: (a) cylinder, (b) square prism, and (c) rectangular prism specimens.

Table 5 tion of tensile stress (r3 > 0) at different loading stages in the three
Percentages of confined elements to the total number of elements and average cross-sectionally shaped specimens. Tensile stress becomes more
minimum principal stress (r3) in various portions of the specimens at e = 0.5% in the
prominent with the increase of axial deformation in the post-
post-peak deformation stage.
peak deformation stage. At peak load, tensile stress appears only
Cylinder Square Rectangular near the edge in the central areas of the specimens. In the post-
specimen prism prism
peak deformation stage, the minimum principal stresses in the
specimen specimen
cone-shaped confined zones seen before at peak load gradually
Total number of elements 8844 9537 9108
become tensile as deformation increases. As mentioned above,
Number of confined elements 3978 5582 3670
Percentage of confined elements at 45% 59% 40% the lateral expansion rate of the rock is higher than that of the steel
e = 0.5% platens, and the elements under the platens are restricted by the
Average r3 (MPa) in the whole 1.54 0.12 2.68 steel platens and are subjected to compressive stresses if there is
specimen at e = 0.5% no rock failure. On the other hand, tensile stresses are generated
Average r3 (MPa) in the portion of 1.36 0.18 3.05
15 mm to the specimen end at
in these elements in the post-peak deformation stage because their
e = 0.5% lateral contraction rate is higher than that of the steel platens as
Average r3 (MPa) in the 70 mm 1.62 0.09 2.62 the load decreases. However, the evolutions of the minimum prin-
middle portion of the specimen cipal stress in the post-peak loading stage are different for different
at e = 0.5%
cross-sectionally shaped specimens. The rectangular prism speci-
men shows the largest tensile stress zones in the specimen, while
the cylinder and the square prism specimens show relatively smal-
the rectangular prism specimen has the lowest percentage of con- ler tensile stress zones.
fined elements and the largest tensile stress r3, and its post-peak The slenderness effect becomes more significant in uniaxial
curve is the most brittle. compression tests when the slenderness of a specimen decreases
As can be noticed from Tables 4 and 5, the average r3 in each (Fig. 4). In essence, it is the change of the geometry of the specimen
specimen changes from compression at peak load to tension in that leads to an increase or a decrease of the influence of the end
the post-peak loading stage at e = 0.5%. Fig. 8 presents the evolu- constraint on the rock deformation behaviors. In other words, the
 Y. Xu, M. Cai / Computers and Geotechnics 84 (2017) 129–137
tensile 3 (unit:
Pa) at peak load
3 at = 0.4% in
the post-peak
deformation stage
3 at = 0.5% in
the post-peak
deformation stage
3 at = 0.6% in
the post-peak
deformation stage
(a)
Fig. 8. Evolutions of tensile minimum principal stresses (r3 > 0 contours) in different cross-sectionally shaped specimens at different deformation stages: (a) cylinder, (b)
square prism, and (c) rectangular prism specimens.
geometry effect of a rock specimen is essentially manifested by the
distribution of confined zones caused by the end effect. The cross-
sectional shape effect can be seen clearly in the post-peak defor-
mation stage, when progressive failure of elements gradually
decreases the effective bearing area of a specimen and conse-
quently changes the effective specimen geometry.
Failures occur first locally in elements whose confinement is
small, and the locations of these elements depend on the cross-
sectional shape of the specimen. The degree of the effective speci-
men geometry change is also dependent on the specimen’s cross-
sectional shape. Therefore, the confined zones in a specimen are
more dependent on the specimen’s cross-sectional shape in the
post-peak deformation stage and so is the strain-softening curve.
For instance, the elements of the rectangular prism are non-
uniformly confined (Fig. 7c) and there are many of them subjected
to tensile minimum principal stresses (Fig. 8c); hence, the overall
post-peak strength of the rectangular prism specimen is lower
than these of the other two specimens. The cylinder and square
prism specimens are uniformly restricted (Fig. 7a and b) and there
are less elements whose minimum principal stresses are tensile
within the specimens (Fig. 8a and b) in the post-peak deformation
stage. Therefore, the post-peak strengths of these two specimens
are relatively high. Furthermore, the post-peak strengths of the
cylinder and the square prism specimens are different because
135
(b) (c)
the r3 contours in the two specimens are not the same. Thus, it
is concluded that the cross-sectional shape affects the post-peak
behavior of rocks.
4.3. Hoop tension effect
Because a continuum numerical tool is used in this study, it is
not possible to capture explicitly the crack initiation and propaga-
tion processes that lead to discontinuous failure of rocks [43].
Although the influence of hoop tension on a dilating crack cannot
be simulated explicitly, it does not inhibit us from comparing the
confined elements and the r3 values in the cylinder specimen with
those in the prism specimens. Based on the hypothesis of the hoop
tension theory, it can be reckoned that the number of confined ele-
ments and the average compressive r3 value in a cylinder speci-
men should be greater than that in a square or a rectangular
prism specimen. Hence, the strength of the cylinder specimen
should be the highest because the circular shape favors hoop ten-
sion development.
The simulation results show that for the same slenderness
(defined by H/D and H/W for the cylinder and square prism spec-
imens, respectively) of different shaped specimens, it is the square
prism specimen rather than the cylinder specimen that has the
highest percentage of confined elements and the highest confine-
136 Y. Xu, M. Cai / Computers and Geotechnics 84 (2017) 129–137
ment (or the lowest tension), both at peak and in the post-peak
loading stages (Tables 4 and 5). In addition, both the laboratory
and the numerical results show that the strength of the square
prism specimen is higher than that of the cylinder specimen
(Fig. 3). The strength of the rectangular prism specimen is also very
close to that of the cylinder and the square prism specimens. Thus,
it is concluded that hoop tension contributes little to affecting rock
strength.
A major difference between the strengths obtained from labora-
tory tests (using specimens) and field data interpretation is that
the interpreted field rock strength depends on the interpretation
model used. When a numerical model reflects the field condition
better, the interpreted rock strength of massive rocks is closer to
that obtained from laboratory tests and this was demonstrated
by Cai and Kaiser [44] using the Mine-by tunnel case history.
Although the height (100 mm) and the slenderness (2.0) of the
cylinder (H/D) and the square prism (H/W) specimens are the same
in the laboratory tests, the cross-sectional area of the square prism
specimen (50  50 = 2500 mm2) is larger than that of the cylinder
specimen (p  252 = 1963 mm2). In such a case, the equivalent
diameter of the square specimen is 56.4 mm, resulting in an equiv-
alent slenderness of H/D = 1.77 that is squatter than the cylinder
specimen with H/D = 2.0. Therefore, the end effect has a greater
influence on the strength of the square prism specimen. Additional
numerical simulation results shown in Fig. 9 reveal that if a square
prism specimen with a width of 44 mm is used in uniaxial com-
pression tests, it gives almost the same cross-sectional area
(1936 mm2) as that of a cylinder specimen with a diameter of
50 mm. In such a case, the UCS of the square prism specimens is
very close to that of the cylinder specimens.
5. Conclusions
The influence of the cross-sectional shape of a rock specimen on
its strength in uniaxial compression tests was studied using the
numerical experimental approach. It is found that the cross-
sectional shape has a very small influence on the peak strength
of rocks, but it affects the post-peak behavior of rocks significantly.
The study revealed that it is the relation between the distribution
of confined and tensile zones during the post-peak deformation
process that actually contributes to the cross-sectional shape
effect. It is shown in our modeling results, which are supported
by laboratory test results, that the strength of a square prism spec-
imen is actually higher than that of a cylinder specimen for the
160
cylinder (D = 50 mm)
150 square prism (W = 44 mm)
UCS (MPa)
140
130
120
1 2 3 [17] Wiles T. Loading system stiffness-a parameter to evaluate rockburst potential.
Slenderness
Fig. 9. UCS of circular (diameter D = 50 mm) and square (width W = 44 mm) cross-
sectionally shaped specimens with different slenderness obtained from uniaxial
compression test simulation.
same slenderness calculated by the specimen height divided by
the width (square specimen) or the diameter (cylinder specimen).
The strength of a cylinder specimen is not significantly higher
than that of a rectangular prism specimen, implying that hoop ten-
sion contributes little to affecting rock strength. The measurable
strength difference between the cylinder specimens and the square
prism specimens observed in the laboratory test and the numerical
simulation results is attributed to the relatively large cross-section
areas of the square prism specimens used. If the slenderness of a
square prism specimen is calculated by dividing the specimen
height by an equivalent diameter showing the same cross-section
area as a cylinder specimen, then the square prism specimen will
have the same strength as the cylinder specimen.
Insight gained from this study can assist in interpreting labora-
tory test results more objectively. One suggestion is to use the
equivalent diameter of a non-circular cross-section to define the
slenderness of a specimen. In this fashion, consistent presentation
of the test results can be achieved. For the same cross-section area
and slenderness, a rectangular pillar has the most brittle post-peak
behavior. Hence, it is beneficial not creating rectangular cross-
sectionally shaped pillars in underground mines to avoid violent
pillar failure.
Acknowledgement
The work presented in this paper was financially supported by
NSERC (Natural Science and Engineering Research Council of
Canada).
References
[1] Hoek E, Brown ET. Underground excavations in rock; 1980.
[2] Hudson JA, Brown ET, Fairhurst C. Shape of the complete stress-strain curve for
rock. Stability of Rock Slopes: ASCE; 1972. p. 773–95.
[3] Das MN. Influence of width/height ratio on post-failure behaviour of coal. Int J
Min Geol Eng 1986;4:79–87.
[4] Fairhurst C, Hudson J. Draft ISRM suggested method for the complete stress-
strain curve for intact rock in uniaxial compression. Int J Rock Mech Min Sci
1999;36:279–89.
[5] Bobet A. Influence of the loading apparatus on the stresses within biaxial
specimens. ASTM Geotech Test J 2001;24:256–72.
[6] Mogi K. Fracture and flow of rocks under high triaxial compression. J Geophys
Res 1971;76:1255–69.
[7] Takahashi M, Koide H. Effect of the intermediate principal stress on strength
and deformation behavior of sedimentary rocks at the depth shallower than
2000 m. In: ISRM international symposium. International Society for Rock
Mechanics; 1989.
[8] Haimson B, Chang C. A new true triaxial cell for testing mechanical properties
of rock, and its use to determine rock strength and deformability of Westerly
granite. Int J Rock Mech Min Sci 2000;37:285–96.
[9] He M, Miao J, Feng J. Rock burst process of limestone and its acoustic emission
characteristics under true-triaxial unloading conditions. Int J Rock Mech Min
Sci 2010;47:286–98.
[10] Sheng D, Westerberg B, Mattsson H, Axelsson K. Effects of end restraint and
strain rate in triaxial tests. Comput Geotech 1997;21:163–82.
[11] Lockner D, Byerlee J, Kuksenko V, Ponomarev A, Sidorin A. Quasi-static fault
growth and shear fracture energy in granite. Nature 1991;350:39–42.
[12] Zipf Jr RK. Analysis of stable and unstable pillar failure using a local mine
stiffness method. In: Proceedings of the workshop on coal pillar mechanics
and design. Washington (DC): U.S. Bureau of Mines; 1992. p. 128–43.
[13] Hoek E, Brown E. Practical estimates of rock mass strength. Int J Rock Mech
Min Sci 1997;34:1165–86.
[14] Joseph T, Barron K. The post-failure characteristics of rock. CIM Bull
2003;96:66–74.
[15] Cai M, Kaiser P, Tasaka Y, Minami M. Determination of residual strength
parameters of jointed rock masses using the GSI system. Int J Rock Mech Min
Sci 2007;44:247–65.
[16] Salamon MDG. Stability, instability and design of pillar workings. Int J Rock
Mech Min Sci Geomech Abstr 1970;7:613–31.
In: First international seminar on deep and high stress mining, Australian
Centre for Geomechanics. Perth; 2002.
[18] Henkel D, Gilbert G. The effect measured of the rubber membrane on the
triaxial compression strength of clay samples. Geotechnique 1952;3:20–9.
[19] Meyerhof G. The bearing capacity of concrete and rock⁄. Mag Concr Res
1953;4:107–16.
 Y. Xu, M. Cai / Computers and Geotechnics 84 (2017) 129–137
[20] Geertsma J. Some rock-mechanical aspects of oil and gas well completions. Soc
Petrol Eng J 1985;25:848–56.
[21] Vernik L, Zoback MD. Estimation of maximum horizontal principal stress
magnitude from stress-induced well bore breakouts in the Cajon Pass
Scientific Research borehole. J Geophys Res: Solid Earth (1978–2012)
1992;97:5109–19.
[22] Zhao X, Cai M, Wang J, Li P. Strength comparison between cylindrical and
prism specimens of Beishan granite under uniaxial compression. Int J Rock
Mech Min Sci 2015:10–7.
[23] Zhao X, Cai M, Wang J, Ma L. Damage stress and acoustic emission
characteristics of the Beishan granite. Int J Rock Mech Min Sci
2013;64:258–69.
[24] Brook N. The measurement and estimation of basic rock strength. Compr Rock
Eng 1993;3:41–66.
[25] Cai M. Practical estimates of tensile strength and Hoek-Brown strength
parameter mi of brittle rocks. Rock Mech Rock Eng 2010;43:167–84.
[26] Grote K-H, Antonsson EK. Springer handbook of mechanical
engineering. Springer Science & Business Media; 2009.
[27] Rashed G, Peng SS. Change of the mode of failure by interface friction and
width-to-height ratio of coal specimens. J Rock Mech Geotech Eng 2015;7
(3):256–65.
[28] Van Vliet M, Van Mier J. Experimental investigation of concrete fracture under
uniaxial compression. Mech Cohesive-Frict Mater 1996;1:115–27.
[29] Rudnicki JW, Rice J. Conditions for the localization of deformation in pressure-
sensitive dilatant materials. J Mech Phys Solids 1975;23:371–94.
[30] Bobet A, Einstein H. Fracture coalescence in rock-type materials under uniaxial
and biaxial compression. Int J Rock Mech Min Sci 1998;35:863–88.
[31] Wawersik WR, Fairhurst Ch. A study of brittle rock fracture in laboratory
compression experiments. Int J Rock Mech Min Sci Geomech Abstr
1970;7:561–75.
137
[32] Labuz J, Biolzi L. Class I vs Class II stability: a demonstration of size effect. Int J
Rock Mech Min Sci Geomech Abstr 1991:199–205.
[33] Bieniawski ZT, Denkhaus H, Vogler U. Failure of fractured rock. Int J Rock Mech
Min Sci Geomech Abstr 1969:323IN29331–30IN32341.
[34] Van Heerden W. In situ complete stress-strain characteristics of large coal
specimens. J South Afr Inst Min Metall 1975;75:207–17.
[35] Cook NGW, Hodgson K, Hojem JPM. A 100 MN jacking system for testing coal
pillars underground. J South Afr Inst Min Metall 1971;71:215–24.
[36] Martin CD. Seventeenth Canadian geotechnical colloquium: the effect of
cohesion loss and stress path on brittle rock strength. Can Geotech J
1997;34:698–725.
[37] Needleman A. Material rate dependence and mesh sensitivity in localization
problems. Comput Meth Appl Mech Eng 1988;67:69–85.
[38] Cook RD. Concepts and applications of finite element analysis. John Wiley &
Sons; 2007.
[39] You M. True-triaxial strength criteria for rock. Int J Rock Mech Min Sci
2009;46:115–27.
[40] Xu Y, Cai M. Numerical simulation of end constraint effect on post-peak
behaviors of rocks in uniaxial compression. In: 49th US rock
mechanics/geomechanics symposium. San Francisco, USA.
[41] Martin C, Maybee W. The strength of hard-rock pillars. Int J Rock Mech Min Sci
2000;37:1239–46.
[42] Brady B. Effects of inserts on the elastic behavior of cylindrical materials
loaded between rough end-plates. Int J Rock Mech Min Sci Geomech Abstr
1971;8:357–69.
[43] Senent S, Jimenez R, Reyes A. Numerical simulation of the influence of small-
scale defects on the true-triaxial strength of rock samples. Comput Geotech
2013;53:142–56.
[44] Cai M, Kaiser P. In-situ rock spalling strength near excavation boundaries. Rock
Mech Rock Eng 2014;47:659–75.