Você está na página 1de 53

Engineering Geology Elsevier Publishing Company, Amsterdam

MEASUREMENT
COMPRESSION

OF TENSILE STRENGTH

Printed in The Netherlands

BY D I A M E T R A L

OF DISCS AND ANNULI

MALCOLM MELLOR AND IVOR HAWKES

U.S. Army Cold Regions Research and Engineering Laboratory, Hanover, N.H. (U.S.A.)
(Received January 6, 1971)
(Resubmitted June 22, 1971)

ABSTRACT
MELLOR, M. and HAWKES,I., 1971. Measurement of tensile strength by diametral compression
of discs and annuli. Eng. GeoL, 5: 173-225.
In the first part of the paper, the validity of diametral compression tests for indirect
measurement of tensile strength is investigated theoretically and experimentally. Linear elastic
theory for diametral compression of discs and armuli by opposed strip loads is reviewed, and the
significance of failure criteria in fracture initiation and test interpretation is considered. Results
of careful tests are given for three types of rock, two plastics, glass, and ice, and the experimental
results are compared with theoretical expectations. Close consideration of the conflicts between
theory and experiment leads to the conclusion that, while there are very serious objections to the
ring test, the Brazil test is capable of giving a good measure of uniaxial tensile strength for Griffithtype materials. In the second part of the paper, practical problems involved in diametral compression testing are considered in some detail. Special attention is given to contact stresses under the
applied loads, and a design is given for a loading jig that reduces contact stresses. Specimen dimensions, size effects, loading rate, force readout, and specimen preparation are discussed, and
some recommended practical procedures for Brazil tests are outlined.
INTRODUCTION
A s a n alternative to the uniaxial tensile test, which is difficult to p e r f o r m
to acceptable s t a n d a r d s for brittle materials, d i a m e t r a l c o m p r e s s i o n o f discs a n d
annuli has s t r o n g p r a c t i c a l appeal. In principle, the stress field which induces
tensile failure when a disc o r a n n u l u s is c o m p r e s s e d d i a m e t r a l l y can be tully
d e t e r m i n e d , p r o v i d e d t h a t the m a t e r i a l m a i n t a i n s perfect linear elastic b e h a v i o r
u p to the p o i n t o f failure. I n reality, however, m a n y brittle materials such as rocks
have n o n - l i n e a r stress/strain characteristics, a n d their i n h e r e n t defect structures
have d i m e n s i o n s which are n o t very m u c h smaller t h a n certain d i m e n s i o n s o f
test specimens. T h u s there m a y b e serious discrepancies between theoretical a n d
a c t u a l behavior.
I n the fields o f concrete testing, r o c k mechanics a n d ceramics technology,
d i a m e t r a l c o m p r e s s i o n tests a n d their s h o r t c o m i n g s have received considerable
attention, b o t h theoretical a n d experimental, a n d it has b e c o m e evident t h a t they

Eng. GeoL, 5 (1971) 173-225

174

M. MELLOR AND I. HAWKES

can never adequately replace the direct uniaxial tensile test, which is actually the
basis for definition of tensile strength. Nevertheless, it is equally clear that good
direct tests are still too difficult and expensive for routine application to large
numbers of specimens, and diametral compression tests appear to offer the most
desirable alternative. The Brazil test, made on solid discs, is particularly appealing
for its simplicity in specimen preparation, but it produces failure in a biaxial,
rather than a uniaxial, stress field. The ring test causes failure to initiate in a
uniaxial stress field, but the crack must propagate through steep stress gradients,
and only a small volume of material is critically stressed. The ring test tends to
have good reproducibility with simple technique, but specimens are more difficult
to prepare than Brazil specimens.
In the present study some of the conflicts between theory and experiment
are explored, the relative merits of ring and Brazil tests are weighed, and an
attempt is made to establish optimum standards for test procedure.
GENERAL PRINCIPLES

Idealized stress distribution in discs and annufi


The starting point for a study of failure in discs and annuli under diametral
compression is determination of the stress field, assuming that the material is
homogeneous, isotropic, and linearly elastic.
Diametral compression of a solid disc. A stress solution for a disc or cylinder
compressed normally by line loads along diametrically opposite generators was
obtained in 1883 by H. R. Hertz, and the problem was pursued some years later
by J. H. Michell. A refinement to account for distribution of the loads over strips
of finite width was made by Hondros in 1959.
A complete stress solution for the case load is distributed over finite arcs,
valid for conditions of both plane stress (discs) and plane strain (cylinders), is
given by HONDROS (1959). Magnitudes and directions of principal stresses for
the case of line loading have been mapped graphically by COLBACK(1966). Since
the Brazil test is only valid when primary fracture occurs along the loading diameter, the stress distribution along that diameter is of greatest interest. The
stress component normal to theloading diameter (tr0)and the stress component along
the loading diameter (ar) are principal stresses; Hondros's solution gives them as:

(7o=

(T r ~

P {
[1-(r/R)2]sin2ct
[
7rRt~ 1 - 2(r-/-R~-~sos-2~ +-(r/R) 4 - tan-I 11 - (r/R) 2 tan a
P
_

~Rt~

[1 -

(r/R) 2] sin 2~t

1 - 2(r/R) 2 cos 2ct + (r/R) 4

]1
1t

tan 1L --+'rJ"
(r/R)2 tan ~t

(la)

(lb)

Eng. Geol., 5 (1971) 173-225

MEASUREMENT OF TENSILE STRENGTH BY DIAMETRAL COMPRESSION

175

where P is the applied load, R is the disc radius, t is the disc thickness, 2~ is the
angular distance over which P is assumed to be distributed radially (usually ~< 15 o),
and r is distance from the center of the disc. Tensile stress is taken positive. Fig.1
gives values of a0 and a, for ~ = tan -1 1/12 = 4 46'.
0.2

0
-0.2
-0.4

-0.6

(P/'n'Rta)
-0.8

-1.0

-I.2
-1.4

-16

0.2

0.4

0.6

0.8

1.0

Fig.1. Distribution of stress along the loading diameter of a solid disc. Tension positive.
(After HONDROS,1959.)
According to the Griffith criterion, the exact center of the disc is the only
point at which the conditions for tensile failure at a value equal to the uniaxial
strength are met. The principal stresses there are:
P (sin2~
a0 = + -n--/~-

P (sin2~t
ar =

7rRt

--

P
~ + 7rR-----/-

)
+ 1

3P
,~

(2a)

(2b)

7rRt

With a 15 o arc of contact, the error introduced by use of the approx!mate


expression for a0 is 2 ~o.
D i a m e t r a l c o m p r e s s i o n o f an annulus. A stress solution for an annulus compressed

by equal and opposite line loads acting along diametrically opposite generators
was obtained in 1910 by Timoshenko, who combined the solution for a solid
disc with a solution for a hollow cylinder loaded internally by a system of forces
necessary to annul the shear and normal forces which would exist at the corresponding section of a solid disc (TIMOSHENKOand GOODIER,1951). The problem
Eng. Geol., 5 (1971) 173-225

176

M. MELLOR AND I. HAWKES

was considered further by Timoshenko's students, and in particular by C. W.


Nelson. Another early treatment of the ring problem was by FILON (1924). The
first analysis directed explicitly to indirect tensile testing of brittle materials was
made by RIPPERGERand DAVIDS(1947).
A new solution, using complex variable methods in place of the previous
stress function and superposition approach, was obtained by JAEGERand HOSKINS
(1966) for the case in which the external load is distributed over finite areas.
It appears that in the published work of JAEGERand HOSKINS(1966) there
are some typographical errors in the equations, since numerical computations by
the writers based upon them show some violation of boundary conditions. However, computations based on a solution derived by NEVEL(1969), and given in the
Appendix, confirm the published results of Jaeger and Hoskins.
In a properly conducted ring test, primary fracture invariably occurs along
the loading diameter, and the crack initiates at, or near, the surface of the hole.
Thus interest centers on the distribution of tensile stress along the loading diameter, and particularly on the critical points where the loading diameter intersects
the hole boundary. The distribution of tangential stress along the loading diameter
is shown in Fig.2, which gives results of the Jaeger and Hoskins analysis for a
range of values of p, where p is the ratio of internal to external radius. The critical

~/R:P

i.o

0.5

(P/2RI~)
0.1

0.2

/
0.3

/
p=0.4

0.5
-2

0.6
-3 -t
Fig.2. Distribution of or0 along the loading diameter for annuli. Compression positive.
(After JAEOER and HosKrNs, 1966.)

Eng. Geol., 5 (1971) 173-225

MEASUREMENT OF TENSILE STRENGTH BY DIAMETRAL COMPRF.SSION

177

tensile stress at the intersection of the loading diameter with the hole can be
written as:
P
ae = + K ~rR---t

(3)

where P is the applied load, K is a stress concentration factor, R is the outside


radius of the annulus, and t is thickness of the annulus. K is a function o f p, where
p is the ratio of internal to external radius. At the surface of the hole, o r and z,~
are, of course, zero. Eq. 2a, which gives ae at the center of a solid disc, can be
regarded as a special case of eq.3 with K ~ 1.
Fig. 11 and 12 (pp. 188-9) give values of K obtained from various analytical
solutions of the problem. They have been confirmed by photoelastic studies
(see discussion of RIPPERGER and DAVIDS, 1947). The values obtained by Jaeger
and Hoskins for load distributed over 15 o arcs are not significantly different,
for most practical purposes, from the values obtained for line loads.

Comparison of solutions for the solid disc and the annulus. Comparison of eq.2a
with eq.3 shows that direct analysis for the solid disc leads to a stress concentration
factor ot unity Co = 0, K = 1), whereas the analysis for the annulus yields a limiting
value K --. 6 as p -~ 0. There is, then, a question as to whether results of tests on
solid discs should be interpreted with K = 1 or K = 6.
The reason for the discrepancy is fairly obvious. At the center of a solid
disc under diametral compression the ratio of principal stresses is - 3 (see eq.2).
I f a hole is present in such a stress field the stress concentration in the direction
of the compressive stress, b y stress superposition from the two stresses, will be
X 6. In the Brazil test any such stress concentrations are produced by naturally
occurring pores or cracks and are, of course, ignored in determining the stress
field at failure. However, in the case of the ring test the material under test is
assumed to be free of all flaws and completely homogeneous, so there is no dimensional limitation on the stress concentration produced by introduced flaws
(holes).
When the central hole is large compared to the sca!e of the naturally occurring flaws in the material, there will still b e significant differences in the stress
fields at the points where failure is expected to initiate. These are as follows:
(1) In the annulus, the stress a 0 ( = a t ) is uniaxial at the point of failure
(i.e., a r = a 3 = 0), and 3a 1 + a s > 0 for a finite distance from the point of failure
initiation. By contrast, in the disc the stress field is biaxial at the critical point;
3tr~ + tr 3 = 0 at the critical point r = 0, but 3a~ + as < 0 for finite values of
r. Thus the simple Griftith conditions for the tensile strength to be equal to the
tensile principal stress are met strictly at a point in the solid disc, but over a finite
distance in the annulus.
Eng. Geol., 5 (1971) 173-225

178

M. MELLORAND I. HAWKES

(2) At the critical point of the annulus (r = a, 0 = 0) there is a steep gradient


of (70 along the loading diameter, i.e. ~(7o/d, is strongly negative at r = a, 0 = 0.
In the disc, d(To/ar = d(7,/ar = 0 at r = 0, 0 = 0; (Te remains constant for a considerable distance along the loading diameter, but (Tr (compressive) increases in
magnitude as r increases.
(3) In the annulus, only a very small volume of material is subjected to
tensile stresses which approach the peak stress of the critical point (see Fig.2 and
33). In the disc, however, a considerable volume of material is subjected to the
peak tensile stress and, as FAIRHURS'[ (1964) has shown, the influence of compressive stress (Trdoes not greatly reduce the effective severity of stress along the loading
diameter for r < R/2.
According to the statistical theory of strength (WEIBULL, 1939, 1952; see
also HAWKES and MELLOR, 1970), all three of these factors may be expected to
contribute to some systematic differences between tensile strength measured using
discs and annuli.

Failure criteria and fracture initiation


The foregoing relationships give the distribution of stress along the lines
where cracks are known to initiate and propagate. However, in order to decide
where failure will begin in a real material it is necessary to introduce a failure
criterion, i.e., a characteristic relationship between the principal stresses at failure.
Failure in solid discs. It is generally accepted that the Griffith failure criterion is
the most satisfactory explanation of the fracture of brittle materials.
According to this criterion, failure occurs when:
(71 = aT

if

3(7t + o'3 >I0

(4)

and when:
(0" 1 -- (73) 2 "4- 8 (7r((71 '[- (73) = 0

if

3 (71 "4- (73 ~ 0

(5)

where (71 is the major (tensile) principal stress, (73 is the minor principal stress,
and (TTis the uniaxial tensile strength of the material (tension positive). It is assumed that the intermediate principal stress (72 has no influence on failure.
In the computation of results from Brazil tests, it is usually assumed that
failure occurs according to condition 4, with failure initiating at the center of the
disc. However, the fact that the ratio of uniaxial compressive strength to uniaxial
tensile strength may vary considerably from the theoretical value of 8 indicated by
Griffith theory appears to indicate that real materials may not conform to the
simple Gritfith criterion.1 Furthermore, real rock materials have grain, pore and
1 The simplistic interpretation of compression tension strength ratios is examined elsewhere
(HAwKESand MELLOR,1970).

Eng. Geol., 5 (1971) 173-225

MEASUREMENTOF TENSILESTRENGTHBY DiAMETRALCOMPRESSION

179

crack structures, so that they can only be considered homogeneous in volumes


that are large compared with the dimensions of these structures. There is also a
characteristic distribution of these defect structures such that the probability of
encountering larger, and therefore more critical, defects increases with increasing
sample volume. Thus, while the most severe stress conditions may exist at a
certain point in the specimen, the most critical defects in a small specimen may lie
at some distance from this point. This being so, failure may initiate not at the
theoretically critical point, but at another point where a critical combination of
stress and defect structure occurs. The effects of these factors on Brazil test results
have been investigated theoretically by FAIRHURST (1964).
Fairhurst generalized the Griffith criterion to account for variation of the
compression/tension strength ratio n from the theoretical value n --- 8, so that the
conditions for failure according to the generalized Griffith criterion become1:
failure when:
O-1 = O-T if

m(m

--

(6)

2)O-1 + O-3 ~ 0

and failure when:

if

I)20-T [ 1 +

2(m

(O-I - - O-3) 2
O-i O-3

O-i + o'3
m(m

--

2)al + O-3 ~< 0

(7)

where m = x/n + 1
A parameter called the "stress severity" S was defined as the ratio of the
theoretical load at failure of the specimen at the most highly stressed point to the
load required to cause failure at any point being considered. The failure conditions
6 and 7 were restated in such a way that S could be calculated as a function of
position along the loading diameter. Since O-1 = O-0and O-3 = 0"r, Hondros's stress
solution (eq.la and lb of this paper) gave the required stresses.
It was pointed out that failure may occur in one of two regions of the loading
diameter, either region (i), in which condition 6 holds, or region (ii), in which
condition 7 holds. Depending upon the values of n and ~, failure occurs as follows:
(a) Failure in region (i) when:
O-im.~ = O-T in region (i)

(8)

with

O-~max(o-1
O-S)< 2(m -- I)z [ I +o-l+
- 20-~m,x
- (0-~- +O-3)2
O-3I(~ --~)2 - -

1]

(9)

in region(ii)
1 In the original paper the conditions are given erroneously as m (2m-l) cri + ~8 ~ 0. The error
stems from an algebraic mistake in eq. A10 of the derivation.

Eng. GeoL, 5 (1971) 173-225

180

M. MELLOR AND 1. HAWKES

(substituting almax for o.r, where o.lm,~ is the maximum value of el in region (i).)

With failure in region (i), the stress severity S is:


in region (i)
S =

o.1
o.lmax

(10)

in region (ii)
-

(O.1

0"3)2

o.lmax(o.1 + o.3)

2(m-1) 2 1 + - -

(11)
- 1

o"1 -k- 0"3

(b) Failure in region (ii) when:


o.lmax = flo.T (0 < fl < 1)

in

region (i)

(12)

with
- ( o . 1 - a3) 2
2(m
1)2[
2o.xmax { ( ~ _ . . 1 ) 2
}]
o.lmax(o.t + '3) ~
1 + fl(o.1 + o'3)
- 1

(13)

in region (ii)
Failure occurs at the point in region (ii) where fl in eq. 13 is a minimum, and the
stress severity is:
in region (i)
S = Pmi.

O.1

(14)

O'lmax

in region (ii) 1
S =

flmin

(15)

//is obtained for each point by solution of eq,13.


Fig.3 gives Fairhurst's results for S as a function of position on the loading
diameter for three values of n and two values of a, the half-angle of the platen
contact. These indicate that in order to assure fracture initiation near the center
of a homogeneous specimen it is necessary to spread the applied load over an
appreciable arc of contact (20 or more). They suggest that with a narrow contact
strip (g '~ 5 o) there will be a pronounced tendency for off-center fracture initiation
in rocks which have a low value of n. They also show that if n is relatively small
and the contact strip is narrow, there will be a systematic underestimation of
tensile strength if fracture does occur near the center of the specimen.
t The fight-hand side of this equation is printed inverted in the original paper.
Eng. GeoL, 5 (1971) 173-225

MEASUREMENTOF TENSILESTRENGTHBY DIAMETRALCOMPRESSION


/

'

181
/

0.5

3
0

'

'

Fig.3. Distribution of "stress severity" along the loading diameter of a solid disc for
materials with different compression/tension strength ratios. Results are given for two values
of ct. (After FAIRHURST,1964.)
ADDINALL and HACKETT (1964, 1965) reached similar, though less general,
conclusions by semi-graphical methods. They found that fracture in a classical
Griffith material (n = 8) was likely to initiate well off center for values of ~ less
than about 9 o, but for wider contact arcs the point of origin of fracture moved
quite abruptly towards the center of the specimen. They obtained a limited amount
o f experimental evidence in support of their conclusions.

Failure in rings. Unless the value of p is very small, it is most unlikely that there
will be any uncertainty about the origin o f fracture in an annulus, since Im(m-2)a~l
is very much larger than la31 in the vicinity o f the hole wall. However, if p is very
small (p ,,~ 0.01) and the absolute size of the hole is comparable to the size of
grains, pores or cracks in the rock, then the ring tends to behave like a solid disc.
The small hole may serve to initiate fracture at the center of this "disc", and the
effective stress severity for this point can be obtained from the foregoing analysis
for solid discs.
Comparison of experimental results with theoretical predictions
Results of some diametral compression tests. Diametral compression tests were
made on discs and annuli of three rocks (Berea sandstone, India limestone, Barre
granite), two plastics (Lucite, CR39), 1 soda-lime glass, and fine-grained ice.
Careful measurements of uniaxial tensile strength were made for the rocks and
the ice.
x Lucite = methyl methacrylate. CR39 = allyl diglycol carbonate.

Eng. Geol., 5 (1971) 173-225

182

M. MELLOR AND I. HAWKE$

Rock specimens were cut with standard diamond tools, taking care to
avoid any contamination, and moisture contents were standardized (HAWKES
and MELLOR,1970). Glass specimens were cut from plate glass slabs with sinteredtip glass cutting drills, using cover plates to avoid surface spalling, but the machined
surfaces were not etched or treated in any way. Plastic specimens were cut and
drilled on a standard lathe. Disc and ring samples of ice were cut from cylinders,
which themselves were made by filling Lucite molds with sieved and vibrated ice
grains, saturating with degassed distilled water, and freezing. The ice cylinders
were sliced on a bandsaw, and the resulting discs were drilled on a lathe with
steel twist drills, all machining being carried out at -10C. Flat surfaces were
cleaned and smoothed by skimming with a metal plate warmed a few degrees
above the melting point.
All materials except ice were loaded in a high-capacity (150-ton) press,
which had relatively "stiff" characteristics for the small loads required. For the
first series of tests, load was applied to the specimens through a special jig, in which
the lower platen was fixed and the upper platen was freely suspended by rubber
bands. Force was applied to the upper platen through a steel ball. After unsatisfactory trials with machined contact arcs, fiat steel platens were used, with 0.025
inch thick cardboard between platen and specimen (CoLBACK, 1966). After each
test the cardboard cushions were rubbed with soft pencil to check contact width
and uniformity of loading, as indicated by taper of the imprint. Most of the rock
samples were tested to failure at a nominal press speed of 1.45 inches/min, as
high loading rate was expected to evoke a material response more nearly elastic
than would be obtained at low loading rates.
Ice samples were tested in a smaller machine ( ~ 1/2-ton capacity) housed
in a cold room regulated to give a temperature of - 7 C (_+ 1 C) at the platens.
They were loaded through flat Lucite platens faced with fine emery cloth. Some
of the specimens were used to study the effect of strain rate on failure, while the
remainder were loaded to failure at the highest speed attainable on the machine
(2 inches/min).
The glass and plastic discs and annuli were intended mainly for photoelastic observations and studies of crack propagation in both highly elastic and
viscoelastic materials.
Results of the tests are summarized in Fig.4-7, where (P/rcRt) is plotted
against p, and the direct uniaxial tensile strength is shown when available. It
may be seen from these plots that there is no apparent discontinuity between the
results for solid discs and results for annuli with small values of p.
In Fig.8 tensile strength calculated from the foregoing test results is plotted
against p for the three rocks, and the measured uniaxial strength is given for two
of the rocks. Ring tensile strength is calculated from eq.3, using values of K given
by the analysis of JAEGER and HOSKINS (1966). As a direct measure of uniaxial

Eng. Geol., 5 (1971) 173-225

183

MEASUREMENT OF TENSILE STRENGTH BY DIAMETRAL COMPRESSION


2400

2000

1600

~\
\
\

Specimen Data

P
TrRt

(Ibf/in z)

Diameter: 2125 inches


Thickness: I O inch
Temp: 70F
Loading Speed: 1.45 in/rain.
Air Dry

\
\\Borre Granite

1200

~
800

\\X

it t x\

,}

,
\

\\

400

Indiana

'~

Limestone

Beret. \{,.

Sandstone

~ -......

"~-..

0.2

0.4

0.6

0.8

1.0

Fig.4. Results of ring and disc tests on air-dry Berea sandstone, Indiana limestone and
Barre granite at room temperature. Specimens pressed between fiat steel platens with cushions
of 0.025 inch thick cardboard.

,20

0.2

0.4

UnioxJal Tensile Strenqth: 2 9 8 Ibf/in z


Diameter: 2.0 inches
Thickness: 1.0 inch
T e m p : - 7eC

7rRt
P 80 I
(Ibf/in2)40 t? ~\t~Fine-grainedPolycrystallineIce
o

Specimen[)ata i

Machine Speed: 5.0 cm/min (I.g7 in/nit

I
P

0.6

0.8

1.0

Fig.5. Results of ring and disc tests on fine-grained polycrystalline ice (equant randomly
oriented grains) at --7C. Specimens pressed between fiat Lucite platens faced with fine emery
cloth.

Eng. Geol., 5 (1971) 173-225

184

M. MELLOR AND I. HAWKES


5000

Specimen Dato
Suspected
Platen

4000

Diometer: 231 inches


Thickness: 0.984 inches
Ternp: 70=F
Nominally Dry

\Foilure
\\

R--'-~
7r
3000

( I bf/in z)

\\Gloss
\
\
\
\

2000

I000

0.2

0.4

0.6

0.8

1.0

P
Fig.6. Results of ring and disc tests on soda-lime glass at room temperature. Specimens
2.31 inches outer diameter, 0.984 inch thick. Nominal machine speed 1.45 inches/min. Specimens pressed between flat steel platens with cushions of 0.025 inch thick cardboard.

------Apparent

Uniaxial Tensile Strength from Lucite (slob w/hole) Tests - - - -

12000
. . . . .

8000

TRt

UnioxiolTensile Strength (Rohm and Haas Co., Philo, Pa.)

.....

t: Deformation
Gross
ond
Platen Failure
"disc

(I bf/in z )
4000

\
\
~.*'disc~\

I
0

I
0.2

I
0.4

I
0.6

I
0.8

--I .....
1.0

P
Fig.7. Results of ring and disc tests on Lucite at room temperature. Specimems 2 and 4
inches outer diameter, 1 inch thick. Machine speed 1.45 inches/rain.

Eng. Geol., 5 (1971) 173-225

MEASUREMENT OF TENSILE STRENGTH BY DIAMETRAL COMPRESSION


"

'

'

'

185

12,000

I0,000

8000

\ 6~ ;r,~e

6000

K=593
o

K=593

4000

Tadiano

-r

2000

~ ^ Berea ~,,,~estone
. . . . . . . . . . .
tK~_"O___ ~
~

~ran_~ite

1
-~.Tensii.
Direct

j s,r.n,th
0

0.2

04

06

0.8

I
IO

P
Fig.8. Calculated tensile strength as a function of p for three r o c k types.

tensile strength the ring tensile values are patently absurd, but the disc results
appear to give a reasonable value for tensile strength.
The results for ice (Fig.5 and 9) are in contrast to those for rocks. Whereas
calculated ring tensile strength is higher than the uniaxial strength for rocks
(Fig.8), it is less than the uniaxial strength for ice, and disc tests on ice greatly
underestimate the uniaxial tensile strength when a K factor of unity is used
(Fig.9).
The results for Lucite (Fig.7) show characteristics intermediate between
those for rocks and those for ice: calculated values of ring tensile strength for
small
culated

values

or p are higher

that

the uniaxial

tensile

strength,

while

values

cal-

from disc tests are only about half the uniaxial strength.

Relation of disc test results to uniaxiai tensile strength. In the case of rocks, it
seems that a properly conducted Brazil (solid disc) test yields a good approximation
Eng. GeoL, 5

(1971) 1 7 3 - 2 2 5

] 86

M. MELLOR AND I. HAWKI~


I

'

'

300

Uniaxial

Tensile

Strength

o=5.93
\
\

K.s.93 \ \ o

NID 2 0 o

\Nice
\

'~o~

0
~- I00

~}K=I.O

I
0

0.1

Temp: - 7 *C
J

I
0.3

0.2

0.4

Fig.9. Calculated tensile strength as a function of p for polyerystalline ice at --7C.

to the uniaxial tensile strength when the usual analysis is applied with a value of
K = 1.0.
In Fig.10 Brazil tensile strength is plotted against uniaxial tensile strength
for two rocks which were tested both air-dry and saturated with water. The
uniaxial data are taken from tests made by techniques described elsewhere (HAw2400

'

'

Borre
Granite /

1600

oO,(

/// /

ODry

owet

==.

/
/

1:1//

m 1200
=

'

2000

".s
.

oDrY~
//,~O,y
~

Indiana
Limestone

800

400

/
,

400

/ Rwot

[
t
I
I
J
1200
1600
2000
Unioxial Tensile S t r e n g t h , I b f / i n 2

800

2400

Fig.lO. Correlation of Brazil tensile strength with uniaxial tensile strength. Each plotted
point represents mean values for at least five replications for each type of test.

Eng. Geol., 5 (1971) 173-225

MEASUREMENTOF TENSILESTRENGTHBY DIAMETRALCOMPRKSSION

187

KES and MELLOR, 1970). Available data for Berea sandstone could not be used
because the anisotropic rock was stressed in different directions for the uniaxial
and the Brazil tests.
The limited data of Fig.10 suggest a direct 1:1 correlation between Brazil
tensile strength and uniaxial tensile strength. For Indiana limestone the ratio of
uniaxial compressive strength to uniaxial tensile strength is low (7.4 dry, 8.1 wet),
but there is no evidence that this causes the Brazil test to underestimate the tensile
strength, with the technique used. FAIRHURST (1964) concluded that a Brazil
test in which the platen contact width is 2~t ~ 10 (ct = t a n - 1 1/12) would underestimate tensile strength by 30 ~o for a rock with a ratio of compressive to tensile
strength of 8.
For ice the situation is entirely different from that prevailing for rocks.
While the Brazil test data for ice are of lower quality than the rock data, they
suffice to shown an important difference in behavior between rock and ice. For
ice, the usual formula for the solid disc, eq.3, gives values of tensile strength that
are far lower than the uniaxial tensile strength, whereas the ring formulas for
p -~ 0 give values which compare reasonably well with the direct tensile strength.
This is not a new observation; it has been commented on by ASSUR (1960) and
BUTKOVlCH(1959), and the reality of the effect is further confirmed by unpublished
data (Hanson, ca. 1958) analyzed by the writers. It has long been a practice in ice
testing to use a value of K ~ 6 whenever the Brazil test is used. FRANKENSTEIN
(1959) recommends using K = 5.2 for Brazil tests On sea ice, this value being
chosen to give compatibility with ring tests in which p = 0.166 and K = 7.09.
Theoretical and experimental values o f the factor K. For comparison of theoretical
and experimental values of the stress concentration factor K i n eq.3, it is convenient
to normalize K, first with respect to the "zero-hole" values for tests and analyses,
and secondly with reference to the absolute (direct uniaxial) tensile strength
o f the material where this is available. In Fig.12 the theoretical ratios K/Kp ~ o
and K/Kp = o are shown as functions o f p (Kp _. o ~ 6, Kp = o ~ 1). On the same
graph, the experimental values of K/Kp _. o = K/Kp = o = Po/Pp for three rocks
and for ice are shown (Po is failure load for a disc, Pp is failure load for a ring).
In Fig.l 1 the theoretical ratios of K = trvr/(P/~Rt), where ~rDTis direct uniaxial
tensile strength, are compared with corresponding experimental values for two
rocks and for ice.
Fig. 12 shows that for no material tested was there any discontinuity between
the results for solid discs and results for discs with a small central hole. The loadcarrying capacity of a disc did not abruptly drop by a factor of 6 when a small
hole was bored at its center, as would be predicted by simple theory for ideal
materials. The experimental results converge with the theoretical curve for K/Kp _., o
as p ~ 0, but they show a tendency to cross over towards the theoretical curve
for K/Kp = o as p --, 1. This implies that while a very small hole has no significant
Eng. Geol., 5 (1971) 173-225

M. MELLOR AND I. HAWKES

188
l

52

2 d

/'

//

O'oT
P/~rR t . K
/
16 i

/Fine- groined Ice


at -70C

Exp~

,~

/
8-

"~oskins

Timoshenko. Filon,
Nel son, Ripperge
0rid 0ovlds

imestone

~/

Granite J

Theoretical

0.2

0.4

0.6

08

P
Fig.ll. Stress concentration factor K as a function of p. Values of K normalized with
respect to the zero-hole values of K, i.e., with respect to the value of K given by the solid disc
analysis 6o = 0) or the value of K given by the limit of the ring analysis 6o ~ 0).
effect as a stress raiser in material with natural pores or flaws, a very large hole
(relative to outside diameter) may produce an effect comparable to that predicted
by theory.
Fig.11 provides final confirmation that a very small hole Co ,-~ 0.01) has no
significant stress concentrating effect in the rocks tested, and that holes remain
less effective than is predicted by theory for higher values of p. The results for
ice are anomalous in this respect.

Initiation and propagation of cracks. The results of the ring tests on rock give no
reason to doubt that cracks initiate at or near the points predicted by simple
theory. They prove quite conclusively that, in a properly conducted test, cracking
does not start at the platen contacts, since in many tests, both ring and Brazil,
the diametral crack terminated about D/10 from the outer boundary. The diametral
cracks in some solid discs tended to " g a p e " in the mid-section, suggesting that
failure probably started in the mid-section. The crack surfaces of failed ring specimens in Lucite, CR39 and glass provided clear evidence in the form of "ray
patterns" (GRAUBERr, 1965) that failure originated near the hole walls. Some
local cracking was seen beneath the platen contact in ice specimens with small
values of p, but the individual cracks (which were of roughly the same size as the

Eng. Geol., 5 (1971) 173-225

MEASUREMENT OF TENSILE STRENGTH BY DIAMETRAL COMPRESSION


l

189

'

K
32

Kp,o

Timoshenl~o, Filon,
Nelson, Ril:~ergert
and Davids /t

jaeger
and
Hoskins

24

E x pr e r i m e n t a l _

,Q. K__. K___


K

Kp_o

Kp.o

%
e) Data points for ice

16
Limestone

Theoretical
K
Kp-O
I

0.2

0.4

0.6

0.8

P
Fig.12. Stress concentration factor K as a function of p. The K values are those given by
the ratio of uniaxial tensile strength a n T to (P/aRt).

ice grains) showed no tendency to propagate. However, in two of the ice specimens
with p = 0.03 the primary crack actually bypassed the central hole.
The primary cracks propagated along the diametral loading plane as predicted by theory (Fig.14, 15). Ring tests on Lucite gave fracture surfaces which
appeared perfectly plane when tested with a straight edge. With suitable equipment and technique, primary cracks in both disc and ring specimens ceased to
propagate when they reached the platen contact zones. In ring tests it was possible
to initiate cracks in rock specimens without immediately driving those cracks
through to the contact zones; in Lucite specimens it was not possible to create and
then halt a crack, although a pre-formed crack could be driven progressively
through the specimen.
A variety of secondary crack patterns were formed in disc and ring specimens
when loading was inadvertently continued after primary failure (Fig.13). It was
demonstrated that cracks other than the main diametral crack were, in fact,
secondary, and it was shown that with suitable technique failed specimens remained substantially free from secondary cracking (Fig.14, 15). Various combinations of primary and secondary cracks ("crescentic", "hourglass", "triple
cleft") have sometimes been regarded as "characteristic" fracture patterns, but
Eng. Geol., 5 (1971) 173-225

190

M. MELLORAND I. HAWKES

it should be recognized that secondary cracks are actually characteristic of imperfect testing technique.

Discussion of results
The Brazil test. As far as can be ascertained from the present tests, diametral
compression of a solid disc gives a good measure of the uniaxial tensile strength
of rocks, even though there are significant differences between the stress fields
in Brazil specimens and uniaxial tensile specimens. The results show a convincing
1:I correlation between Brazil and uniaxial tensile strengths for the rocks tested,

Fig.13. Typical crack patterns in ring specimens tested by simple methods using fiat
platens. The rock shown is Indiana limestone; identical crack patterns were formed in Berea sandstone and Barre granite. In all cases the "vertical" diametral cracks are primary cracks, and all
other breaks are secondary.

Eng. Geol., 5 (1971) 173-225

<

Fig.14. Crack patterns in annuli tested with the curved-jaw jig using refined technique.
l~rea sandstone speciruens are shown in the left column. The bottom specimen is photographed
as taken from the jig, with the diametral crack almost invisible. The upper two specimens of the
left column have been deliberately abraded and rubbed with carbon black to show the diametral
crack and the absence of secondary cracks. Indiana limestone specimens are shown in the right
column. The bottom specimen is photographed as taken from the jig, with the diametral crack
almost invisible and no secondary cracks detectable. The upper two specimens of the right
column have suffered secondary cracking by overloading; note that the diametral crack does not
extend to the periphery.

Eng. Geol., 5 (1971) 173-225

Fig.15 Brazil specimens tested using the curved-jaw jig and refined loading technique.
Specimens in the right column are as taken from the jig; note that the diametral crack does not
extend to the periphery. Specimens in the left column have been broken apart for clearer definition
of the diametral crack. Rocks are, from top to bottom: Berea sandstone, Barre granite, Indiana
limestone.
Eng. Geol., 5 (1971) 173-225

MEASUREMENTOF TENSILESTRENGTHBY DIAMETRALCOMPRESSION

193

with no indication of systematic underestimation in Brazil tests on a rock with low


compression/tension strength ratio. Problems of off-center fracture initiation
were probably avoided with the technique used, but in any case there is no experimental evidence that off-center fracture invalidates test results. HUDSON (1969)
deliberately induced off-center fracture, and found no significant change in the
load-bearing capacity of the disc.
At the highest machine speed available during the present tests (2.3 inches/
min), it was impossible to make valid Brazil tests on Lucite at room temperature.
Gross deformation ( > 10~o strain of the loading diameter) occurred prior to
failure, and final collapse of the specimen took place when a pair of symmetrical
crescentic cracks formed between the outer edges of the platen contact, which
was of approximate width R at failure. There was no diametral crack, but the
loading diameter had a residual strain of about 3 ~ after failure. CR39 also
failed by crescentic cracking.
The mode of failure of glass discs could not be determined. Fastax photography at 6000 frames per second was carried out in the hope that it might "catch"
crack initiation, but the results were inconclusive. In one frame which showed a
specimen failing there was a straight crack along the loading diameter, with a
figure-eight or hourglass crack pattern disposed about it. After Brazil tests on
glass the center portion of the specimen was always completely shattered, and two
outer segments with crescentic crack surfaces were left intact. It is suspected that
failure initiated at the platen contacts, since the surface flaws of the machined
perimeter were obviously more severe than the internal flaws of the glass.
The results of Brazil tests on polycrystalline ice are puzzling, since the accepted Brazil test analysis gives strength values that are far too small by comparison
with uniaxial data. BUTKOVICI-I(1959) and ASSUR (1960) simply assume that the
bubbles present in this type of ice have the same effect as a small hole, ignoring
the fact that the same bubbles are present in uniaxial test specimens. For consistency, it has to be recognized that if a bubble gives a critical stress concentration
factor of 6 in the biaxial stress field at the center of a disc, it will give a critical stress
concentration factor of 3 in uniaxial tension, so that there ought to be a factor
of 2 discrepancy between the two tests, not a factor of 6. However, if the bubble
argument is valid, it ought to apply also to porous rocks which have spherical
or cylindrical pores, and so far no evidence has been found to suggest that this
is the case.
In the hope of shedding further light on the ice anomaly, FAIRHURST'S
(1964) analysis was applied, taking for ice loaded at 2 inches/rain a value o f n = 4.
This value was obtained from uniaxial tests on ice of the same type at the same
temperature (HAWKESand MELLOR,1971). Fig. 16 gives the resulting distribution
of the stress severity S as a function of position along the loading diameter for
two values of contact width. Since the loading width for Brazil tests on ice was
approximately 1)/10 (a .~, R/10), the applicable curve in Fig.16 is that for ~ =

Eng. GeoL, 5 (1971) 173-225

194

M. blELLOR A N D I. HAWKES

1.0

0.8

mr
R

0.6

0.4

0.2

0.4

0.8

1.2

Fig.16. Distribution of "stress severity" S along the loading diameter of a Brazil specimen
for two contact angles and a compression/tension strength ratio of 4.
tan -1 (1/12). It can be seen that, if fracture initiated near the center of a disc
of this ice, then the Brazil test would systematically underestimate the tensile
strength by a factor of approximately 2. This may explain part of the discrepancy
between Brazil und uniaxial results, although there is no guarantee that the discs did
actually break near the center.
A more probable explanation for the discrepancy between the uniaxial
tensile strength of ice and that calculated from the Brazil test is that the compressive stresses in the disc play a part in the failure process. In other words the Griffith
failure criterion, as expressed by eq.4 and 5, and in modified form by eq.6 and 7,
does not apply for ice under tensile failure conditions. There is no direct evidence
for this, as no data on the strength of ice in mixed stress biaxial and triaxial stress
fields is available, but there is indirect evidence in the low ratio of compressive to
tensile strength for ice (HAW~ES and MELLOR, 1971). The Griffith failure criterion
essentially predicts when cracks will initiate, not when structural failure will
occur, and there is considerable evidence to show that in uniaxial tests on rocks,
cracks start to form when the load is 50 ~o to 75 ~ of the ultimate value reached
at final failure. For nearly all rocks the ratio of uniaxial compressive to tensile
strength is greater than 8, which implies that the Griffith criterion will still be

Eng. Geol., 5 (1971) 173-225

MEASUREMENT OF TENSILE STRENGTH BY DIAMETRAL COMPRESSION

195

applicable, as is borne out by experimental data. In the case of ice the writers have
observed the development of cracks under uniaxial compressive stress as low as
60 Y/ooof the ultimate collapse stress, which implies a Griffith ratio of uniaxial
compressive to tensile strength of around 2.5. Even assuming that one principal
stress had no influence on the other in causing failure, which is extremely unlikely,
such a low ratio means that failure in an ice disc under diametral compression
is always caused by the compressive stress component. To attempt to analyze
the results of such a test in terms of uniaxial tension is therefore quite futile.

Validity of ring tests. From the available evidence (see also ADDINALLand HACKETT,
1964, 1965; JAEGER and HOSKINS, 1966, HUDSON, 1969) it appears that ring tests
alone cannot provide a useful measure of the uniaxial tensile strength of rocks.
When analyzed in accordance with existing theory, results of ring tests give tensile
strengths which are, in general, far higher than the uniaxial values. Calculated
ring tensile strength is a strong function of p for a given external specimen size,
although it might be more accurate to say that calculated strength is a function of
absolute hole size for a given rock type (AODINALL and HACKETT, 1964, 1965).
There remains the possibility that the ring test might provide a self-consistent index of tensile strength. It is certainly true that ring tests give reproducible
results with simple loading techniques, and that the calculated tensile strength
tends to a limiting asymptotic value for high values of p (p > 0.5). However, there
can be little reasonable doubt that ring tests with values of p less than 0.4 are
likely to produce confusing results, as the following observations show.
Fig.11 gives experimental values of the stress concentration factor K as a
function of p for Indiana limestone and Barre granite; the values differ for the two
rock types. In a program of low temperature tests, parallel sets of ring tests and
Brazil tests were made on Berea sandstone and Indiana limestone; the results
indicated that a consistent ratio of Brazil failure load to ring failure load was not
well maintained (Fig.17), and the observed ratios differed from those obtained
in the tests reported here.
It is concluded that self-consistent results from ring tests on rock can only
be obtained when p is greater than about 0.5, and absolute specimen dimensions
are appropriate to the grain size of the test material (see pp.210ff). The tensile
strength calculated from such a test will approximate the modulus of rupture rather
than the uniaxial tensile strength, since the ring test effectively becomes a beam test
as p ~ 1; this value is generally appreciably larger than the uniaxial tensile strength.
As was the case for Brazil tests, ring test results for ice are anomalous by
comparison with those for rocks. Calculated tensile strength decreases with increasing p as it does for rocks, hut it is smaller than the uniaxial tensile strength.
However, the asymptotic value of calculated strength for high values ofp is comparable to the Brazil strength, as is the case for rocks. Thus ring and Brazil tests for
ice are internally consistent, showing the same relation to each other as do ring
Eng. Geol., 5 (1971) 173-225

196

M. MELLOR AND I. HAWKF_.,S

2400

/
2000

'

I
/
/
/
/

,~/
/
/
/

[]

[]/

/,2~

//

///
/

2////
o./~

800]Unfrozen ^ / " /

Ipecimen$ / /

~i~/:

6erea Sandstone

L] /f/'

Tndianalimestone

1117
0

/
/

i.~ 1200

'

.~

K~'6.Z

1600

~,Dry

400

800

1200

P/TRt, Ring, p=O.OS5(Ibf/inz)

1600

Fig.17. Correlation of failure loads for a solid disc and a ring with p = 0.085. The line
K/Ko = 6.2 gives the simple theoretical relation. K/Ko = 1.25 is the experimental correlation
given by Fig.12 for Indiana limestone, K/Ko is the experimental correlation given by Fig.12 for

Bcrea sandstone. The plotted points give later experimental values for tests on frozen and unfrozen specimens over a range of teml~ratures.

and Brazil tests for rock. The anomalous behavior appears when diametral compression results are compared with uniaxial tensile strength. No beam tests were
made, but results for similar types of ice (BuTKOVICH, 1959; F~NIO~NS~IN, 1959;
PeSCHANSKII, 1967) show that the modulus of rupture is closer to the uniaxial
tensile strength than to the Brazil tensile strength.

Conflicts between theory and experiment for ring tests. There are a number of
factors which may, and probably do, contribute to the conflict between theory
and experiment.
(1) Non-linearity of stress and strain. It has been shown in a separate study
(GARmPY et al., 1971) that the rocks tested in this work all have-non linear stress/
strain characteristics and the quasi-elastic moduli are, in general, different in
tension and compression. Departures from linearity tend to become more serious
as strain rate decreases, and time-dependent internal cracking leads to an apparent
decrease of strength with decreasing strain rate.
Eng. Geol., 5 (1971) 173-225

MEASUREMENT OF TENSILE STRENGTH BY DIAMETRAL COMPRESSION

197

In the ring test it seems clear that inelastic straining will precede failure,
so that at the point of failure (where &r/& = 0 in uniaxial tests) there will be an
effect akin to plastic yielding at points of highest stress. For a detailed quantitative
assessment of these effects it would be necessary to undertake iterative non-linear
analyses, but in the absence of such analytical results there is still ample reason
to believe that peak stress concentrations will be relieved to some extent by quasiplastic yielding. This stress relaxation will tend to become more effective as strain
rate decreases, giving apparent increase of calculated strength, but the effects
of time-dependent internal cracking will offset this to some extent. The loading
rate experiments on ice (Fig.28) tend to support this interpretation.
In brief, a ring of non-linear material cannot develop critical stress concentrations as high as those predicted by linear theory.
(2) Inherent cracks and stress-induced cracks. In the simple theory for
the elastic annulus there is an implicit assumption that any internal defects in
the material, which create their own stress concentrations, are very much smaller
than the dimensions of the test specimen. For typical polycrystalline rocks this
is unrealistic; the stress field at the critical points will almost certainly be perturbed
by inherent defects of the material. Furthermore, there is a possibility that the
failure crack might form slowly, altering the stress field in the process.
To investigate the effect of cracks, some experiments were made on Lucite
which, unlike most rocks, has no gross inherent cracks. Two sets of Lucite specimens were prepared, some of them simple annuli and others annuli with cracks
induced at the critical points. The specimens were 4 inches in outside diameter,
1 inch thick, with a central hole of 0.125 inch diameter. This gave a small value of
p (p = 0.0312), and a restricted field of influence for the hole. The pre-cracked
specimens were prepared in the following way: a 4 inches square Lucite slab with
a 0.125 inch diameter central hole was loaded in uniaxial compression along one
pair of opposite edges until tensile cracks (visible length 0.07-0.09 inch, theoretical
length 0.15 inch) formed at the hole, and a 4 inches diameter annulus was then
machined from the slab. Provided that load was increased slowly, pre-formed
cracks could be made to grow slowly in response to increasing load when the load
was about 40 % of that necessary to break an uncracked annulus. With uncracked
annuli it was not possible to form and h o l d limited cracks; as soon as a crack
formed it propagated instantaneously to the platen contact zones. Thus for Lucite
the load needed to propagate a crack was substantially smaller than the load
needed to initiate a crack, and there was no question of the crack relieving the
stress concentration of the hole and permitting greater load to be carried. The
stress concentration at the crack tip was apparently greater than the stress concentration produced simply by the hole.
Returning to the consideration of rocks, JAEGER and HOSKINS (1966) found
that it was possible to induce cracks at the critical points of marble rings without
the cracks propagating, and they surmised t h a t stress concentrations might be
Eng. Geol., 5 (1971) 173-225

198

M. MELLORAND I. HAWKES

relieved by formation of non-propagating cracks. The present writers were also


able to form cracks which did not propagate immediately, but once these cracks
were formed the load could not be increased any further without the specimen
failing completely. Thus it was concluded that these stress-induced cracks did
not provide e f f e c t i v e stress relaxation.
The visible stress-induced cracks just described appear to be formed by
growth and coalescence of inherent cracks. Since the latter are associated with
pores and grain boundaries, there is reason to suspect that it may not be possible
for a steep stress gradient to be maintained within about one grain diameter of the
hole wall of the annulus. The maximum tensile stress may well be reached near
the tips of the cracks which are interesected by the hole wall. Some support for
this line of reasoning is provided by ADDINALLand HACKETT(1964, 1965), who
found that calculated ring tensile strength varied with absolute, rather than
relative, hole size. HUDSON (1969) also found that the critical hole size, below
which there is no significant stress-concentrating effect, was related to the grain
size of the material.
This suggested stress-relieving role of inherent cracks actually represents
a physical interpretation of the phenomenological explanation based on quasiplastic yielding, which was put forward in the discussion of non-linearity.
(3) Effect of quasi-plastic yielding at the critical points. In the foregoing
it has been suggested that there may be, surrounding the critical points of the
annulus, a zone of material which undergoes quasi-plastic yielding prior to failure
of the annulus. If it is postulated that the loading diameter of the ring is stressed
elastically up to the boundary of this hypothetical plastic zone, then a comparison
of theoretical and experimental stress concentration factors with the elastic stress
distribution (Fig.2) gives a thickness I for the conceptual plastic zone. In Fig.18
(mm)
3-

0.10

Limestone
(0 2-1 O) Groin si~e,

mm

008
2 -

3O

0 06

-2o A(oep)"

_004

'"(TR)
002

I0
"Stress . . . . . . .

[ . . . . . . . . Io . . . . . . . .

0.2

0.4

0.6

Fig.18. Thickness'qf the quasi-plastic zone I as a function of p for three rock types. Also
shown is the approximate radial stress gradient at the critical point, as given by linear elastic
theory.

Eng. Geol.,

5 (1971) 173-225

MEASUREMENT OF TENSILE STRENGTH BY DIAMETRAL COMPRESSION

199

values of I computed in this way for three rocks are given as a function of p. The
absolute values for I are about an order of magnitude larger than the maximum
pore size of each rock (as determined by mercury intrusion); they are comparable
with the grain size of the granite and with the maximum grain size of the limestone,
but are several times greater than the maximum grain size of the sandstone. For
each rock type, I rises to a maximum at p ~ 0.3.
Thus, while/is largely independent of the grain size of the rock, there appears
to be a significant relationship between l and p. When the theoretical radial stress
gradient at the critical point is plotted against p, it is found that stress gradient
is a minimum at p ~ 0.3, i.e., the maximum value of/coincides with the minimum
value of stress gradient. In other words, "plastic yielding", or crack growth,
extends a greater distance from the critical point as the stress gradient decreases.
From this it might be concluded that there is stable growth or propagation of
cracks down a steep stress gradient and unstable, or abrupt, propagation through
a gentle stress gradient.
(4) Effect of stressed volume. According to the statistical theory of strength,
some departure from the uniaxial strength value might be expected to be caused
by the stress gradient in the ring and by the relatively small volume of the ring
specimen which is subjected to the peak stress, even if the test were otherwise
to behave exactly according to simple theory. These two effects can be discussed
together, since there is an implicit relationship between critically stressed volume
and stress gradient at the critical point.
ADDINALL and HACKETT (1964) plotted computed strength against stress
gradient for their experimental data, and found no simple relationship. A similar
result is obtained when the present data are handled in the same way. This shows
that the broad discrepancy between theory and experiment cannot be related
directly to the stress gradient, but it does not provide a sensitive test of the predictions of statistical strength theory.
There can, however, be little doubt about the reality of "volume effects".
In Fig.26 volumes of the test pieces stressed to various percentages of the maximum
tensile stress are shown for different values of p. These plots were obtained using
the equations derived by NEVEL (1969; see Appendix). The contact stresses at the
loading points have been ignored and the percentage values refer only to the
principal tensile stress, irrespective of its orientation. Where there are no tensile
stresses the values are shown as < 10 %. Fig.33 is a plot of the percentage volume
of the test specimen that is subject to a tensile stress lying within 10 % of the peak
value, as a function of p. For comparative purposes the total volume of the specimen is assumed to be that of the solid disc. For values of p below 0.1, the volume
of the specimen stressed to within 10% of the tensile peak stress is so small that
it is virtually insignificant, i.e., it has no more effect than a naturally occurring pore
or crack. We would therefore expect a disc with a very small hole to act largely as
a solid disc, as is borne out by experimental data.
Eng. Geol., 5 (1971) 173-225

200

M. MELLOR AND I. HAWKES

The previous discussion, which shows a correlation between stress gradient


and the extent of pre-failure yielding or cracking, provides some experimental
justification for the theoretical prediction that measured strength will be influenced
by stress gradient.
(5) Conclusions. The exaggerated estimates of tensile strength of rock given
by the ring test are tentatively attributed to quasi-plastic yielding of the test material
at the critical points. With a steep stress gradient, peak stress will be relieved by
stable crack growth, and structural failure of the ring, which is associated with
unstable crack growth, does not immediately ensue. Size of the hole relative to the
size of controlling structural defects in the material is important, since a very small
hole can become completely ineffective as a stress raiser.
Some writers have put forward the view that measured ring tensile strength
actually reflects an intrinsic property of the material. While this is literally true;
it tends to be misleading where test results are intended for practical application.
From a practical standpoint it seems more reasonable to regard the ring test as
a confusing and unacceptable test for typical rocks.
PRACTICAL CONSIDERATIONS

Brazil test versus ring test

From the foregoing considerations it appears that the Brazil test is preferable
to the ring test. The case for the Brazil test may be summarized as follows:
(1) When properly conducted, the Brazil test gives a good approximation
to the uniaxial tensile strength for "Griffith-type" materials. The ring test gives,
at best, an approximation to the modulus of rupture.
(2) The Brazil test stresses a relatively large volume of material to the critical
level, whereas the ring test confines peak stress to a small volume. There are no
steep stress gradients in the critical zone in the Brazil test, but stress gradients are
very steep in the critical zones of rings.
(3) The Brazil test appears to be relatively insensitive to inelastic behavior
and non-linearity, but the ring test is highly sensitive.
(4) The Brazil specimen is easy to prepare to acceptable tolerances. Preparation of the ring specimen involves more than twice the amount of work
needed for the solid disc.
The only remaining question is whether the reproducibility of the Brazil
test can match that of the ring test, and the tests described in the next section
give answer in the affirmative. Thus it is concluded that the Brazil test is preferable
to the ring test. Having established the general validity of the Brazil test, it remains
to develop sound practical techniques for preparation of specimens and conduct
of the test.

Eng. Geol., 5 (1971) 173-225

201

MEASUREMENTOF TENSILESTRENGTHBY DIAMETRALCOMPRESSION

Reproducibility of ring and Brazil tests.


To check on the reproducibility of the Brazil test and the ring test for
different values of p, a set of tests was made on air-dry Indiana limestone. External
diameter of the discs and rings was 2.125 inches (NX), thickness was 1.0 inch,
and all specimens were prepared by careful drilling and sawing, using precision
diamond tools.
The specimens were loaded by flat steel platens in a jig designed to center
the specimen. The upper platen was loaded through a steel ball, so that it was free
to align itself with the specimen. Platen cushions of smooth cardboard, 0.025
inch thick, were used, following the recommendation of COLBACK (1966). The
platen cushions were rubbed with soft pencil at the end of each test; the imprint
gave the width of the contact strip and provided a check on uniformity of loading
across the thickness of the specimen (the imprint tapers for non-uniform loading).
Load was applied by a 300,000-1bf Riehle testing machine, which was effectively
"stiff" for the small loads required. Head speed of the machine was 1.45 inches/
min. Ten identical ring specimens were tested for each value of p, and sixteen
identical Brazil specimens were tested.
Results of the tests are shown in Fig.19, which indicates that the standard
deviation of 2P/ndt is proportional to 2P/ndt for values of p above 0.1. The
percentage standard deviations are shown in Fig.20. They compare favorably
with those found in good quality uniaxial compression tests. The results indicate
that the reproducibility of ring tests is better than that of Brazil tests. This is not
surprising since, in the ring test, failure is promoted by an introduced stress raiser,
and also the contact stresses are lower than those in the Brazil test.
60

Dry I n d i a n a L i m e s t o n e
Diameter: 2.125 inches
Thickness: 1.0 inch

p-O

/
g

In

o.0294/d'

40

o~oz~/

o
-1
O.161D.~/~

20
o
to

6o311~73~8
~0'~

200

400

P/lrR t

600

8oo

Iooo

(Ibf/in:')

Fig.19. Standard deviation of ring and Brazil tests on air-dry Indiana limestone using
simple testing methods.

Eng. Geol., 5 (1971) 173-225

202

M. MELLORAND I. HAWKF_,8
1

Dry Indiana Limestone


Diameter: 2.125 inches
Thickness: 1.0 inch
Head speed: 1.45 in/rain

8c
o

5>

F l a t S t e e l P l a t e n s with
0.025 inch thick cardboard cushions
0

Improved
technique
us(ng
special jig

i
0.1

l
0.2

t
0.3

I
0.4

I
0.5

I
0.6

Fig.20, Percentage standard deviation of ring and Brazil tests on air-dry Indiana limestone
using simple testing methods, and standard deviation of Brazil test using curved-jaw loading jig.

An attempt was made to improve the reproducibility of the Brazil test.


A new loading jig, the design of which is described later, was made, and operation
of the testing machine was modified so that load was released immediately after
primary fracture. The new loading jig was intended to reduce contact stresses;
thick platen cushions were discarded, but specimens were wrapped with masking
tape. The load release technique was developed after it had been found that the
two halves of a split sample could sometimes carry a higher load than that required
to cause primary cracking. In testing work subsequent to that described here,
load-displacement records were obtained, so that the load causing primary fracture
became easier to identify.
With the improved technique, standard deviation for Brazil tests on Indiana
limestone dropped from 6.3 % to 4.5 %. With the technique finally adopted there
was a further small improvement, since the load-displacement records gave a
more precise value for primary failure load than did the testing machine dial.
It was eventually concluded that a standard deviation of 4 % was readily attainable
in the Brazil test.

Contact s t r e s s e s
If Brazil and ring tests are to be valid, the applied load must be confined
to a narrow strip so as to approximate a uniform line load, but the contact stresses
must not be so high that they cause premature cracking.
Eng. Geol., 5 (1971) 173-225

MEASUREMENTOF TENSILESTRENGTHBY DIAMETRALCOMPRESSION

203

Contact between plane and cylindrical surfaces. Simple theory for discs and rings
assumes line loading at the boundary, although true line loading, which would
imply infinite contact stress, can never be realized in practice. The width o f the
contact area when a fiat platen and a cylindrical sample are pressed together can
be found from elastic theory (e.g., TIMOSHENKOand GOODIER, 1951).

2a=4

[ PR ~ 1 - v~ 1 ~ \
E, + T ~

v~ll '/2
]]

(16)

or:
a

=2(

P ~1/2 [ 1 - v z p + 1 - V s 2 ] 1/2

\~T!

I Ep

(17)

E, ]

where 2a is the width of the contact strip, P is applied load, R is sample radius, t
is sample thickness, and the subscripts p and s refer to platen and sample respectively.
For contact between a steel platen and a sample of rock the width of the
contact strip can be calculated, assuming that the rock is linearly elastic:
with E s = 5. 10 6 lbf/sq, inch, Vs = 0.2, 2a = 1.9. l 0 - 3 (PR/rrt) ~ inch
(P in lbf).
with Es = 10.106 lbf/sq, inch, vs = 0.2, 2a = 1.4.10 -3 (PR/rct)~ inch
(P in lbf).
For an NX diameter Brazil sample with E~ = 5. 10 6 lbf/sq, inch and a
failure load Pit = 3000 lbf/inch, the contact width at failure would be 0.061 inch.
For a similar sample with E s = 10.106 lbf/sq, inch and a failure load Pit = 8000
lbf/inch, the contact width at failure would be 0.073 inch. Thus the contact width
is rather insensitive to rock type; it is approximately 0.07 inch for NX samples or,
more generally, 2a ~ R/15. In angular terms, this is a 2 arc of contact.

Fig.21. Coordinate axes for analysis of contact stresses.

Stresses in the contact zone. The distribution of normal pressure across the width
of the contact strip is semi-elliptic for elastic bodies (TIMOSrmNKO and GOODmR,
1951). I f p is the normal pressure at distance x from the center line of the contact
strip (Fig.21) and P,, is the maximum pressure along the center line of the strip:
P = P m [1 --(x/a)2] *

(18)
Eng. Geol., 5 (1971) 173-225

204

M. MELLORAND I. HAWKF~

The value Ofpm is given by:

Pm=

t~

~E-~~sJ

(19)

A complete analysis o f stress for the elastic half-space loaded over a rectangular strip o f finite length with a semi-elliptic distribution of pressure was
made by KUNERT (1961). His numerical results for a contact strip with t/2a = 10
are summarized in Fig.22. Denoting the tangential or circumferential stress as
a~ and the axial stress (parallel to the disc axis) as a z, stresses at the surface o f
contact, which are the m a x i m u m stresses, can be summed up as follows for the
case when vp = vs = 0.3.
-tO

"10

L---

0.6

or,/,

08

0"K/Lbm)y=O 0 5

04

Pm/Y:o.05

0.6

08

x/Q ~10
0.5

_J
IO

o
0~5

I0

Fig.22. Contact stresses for contact between plane and cylindrical surfaces with 2a/t =
1/10 and semi-elliptical distribution of stress in the x-direction. (After KUr,reRT, 1961).
Mid-section o f contact strip:
center line p = Pro, trx = - 0 . 9 8 Pro, 'z = - 0 . 6 2 Pm
edges o f contact strip p = 0, a x = + 0.020 Pro, trz = -- 0.020 Pm
End zones (within t/4 o f the ends):
center line trx decreases to - 0 . 8 Pm at the end; az increases to - 0 . 8 Pm at
the end; p = P m
edges o f contact strip a~ rises to a peak value o f + 0.066 Pm at 0.03 t from
the end, before falling to zero at the end; trz rises to a peak value o f - 0 . 0 6 6 Pm at
0.03 t from the end, before falling to zero at the end; p = 0 (tensile stress is taken
as positive).
Referring back to the examples o f rock specimens previously quoted, the
values of pro for direct contact between elastic rock and fiat steel are:
E s = 5 . 1 0 6 lbf/sq, inch P m = 2.12" 103 (P/~tRt) lbf/sq, inch 0bf/inch
units)
Es = 1 0 . 1 0 6 lbf/sq, inch P m = 2.82" 103 (P/nRt) ~ lbf/sq, inch (lbf/inch
units)

Eng. Geol., 5 (1971) 173-225

MEASUREMENTOF TENSILESTRENGTHBY DIAMETRALCOMPRESSION

205

Taking the same failure load as in the previous example (3000 and 8000 lbf/inch
respectively), the values of pro at failure would be:
Es = 5 106 lbf/sq, inch Pm = 45,000 lbf/sq, inch
E s = 10.106 lbf/sq, inch Pm = 97,500 lbf/sq, inch
These values are some three to four times greater than the Uniaxial compressive strength for the kinds of rocks represented by the modulus figures, and therefore it seems inevitable that platen cracking of some kind will occur in a Brazil
test when there is direct contact between rock and platen. Although theoretically
the contact stresses in ring tests ought to be at least six times smaller than those
in corresponding Brazil tests, they are actually likely to be no more than four
times smaller for p ~< 0.3 (Fig.12). This still leaves them prone to platen cracking
by direct crushing.
Looking at the tensile stresses, the tangential tensile stresses along the edge
of the contact strip in the mid-section of a Brazil specimen amount to 855 and
1,850 lbf/sq, inch for the low and high modulus rocks respectively; these values
are about the same as the tensile strengths of the rocks represented, so that cracking
is likely. Ring specimens would not be likely to crack in the mid-section under
their smaller tangential tensile stresses. In the end zones of Brazil samples the
tangential tensile stress rises to 2,920 and 6,340 lbf/sq, inch for the low and highmodulus rock types respectively. These values are greatly in excess of the probable
tensile strengths of the rocks represented, and even when reduced by appropriate
factors for ring tests they may still exceed the tensile strength of the rock for values
of p < 0.3. J. A. Hooper (personal communication) 1 has recently carried out a
detailed study into the failure of glass cylinders in diametral compression, and
he concludes that the tensile stresses at the edge of the loading platen invariably
cause failure in this material.
It is concluded from the foregoing that direct contact between steel platens
and rock specimens is unacceptable, except for ring tests with p > 0.3.

Reduction of contact stresses. The estimates of contact stress made in the previous
section may somewhat exaggerate actual peak stresses, since few rocks are linearly
elastic up to failure, but nevertheless they establish approximate magnitudes.
Reduction of contact stress is clearly necessary, at least in the Brazil test, and the
figures already arrived at suggest that t h e reduction should be by a factor of 5.
There are two ways o f reducing contact stresses: the first is to increase the contact
area; the second is to alter the distribution o f stress so as to reduce stress concentrations.
In order to reduce stresses solely by increasing contact area, it would be
necessary to increase the contact width by at least a factor of 5, from R/15 to R/3,
t Hooper, J. A., The Failure of Glass Cylinders in Diametral Compression. Unpublished report
for Ove Arup and Partners, London, April 1970.
Eng. Geol., 5 (1971) 173-225

206

M. MELLOR AND I. HAWKE$

but there is a possibility that this might alter the stress conditions in the critical
test zone of the sample. However, photoelastic observations suggest that stress
distribution in the critical zones of discs and annuli is virtually unaffected by
distribution of boundary loads over contact arcs up to 15 (COLBACK, 1966; also
unpublished observations by the writers). Furthermore, theoretical solutions based
on the assumption that contact pressure is distributed uniformly across the contact
arc confirm this. HONDROS'S (1959) solutions indicate that the simple Brazil
formula for principal tensile stress at the center of the sample is in error by less
than 2 ~o for contact arcs up to 15 wide, while studies by JAEGER and HOSKINS
(1966) indicate that the ring stress concentration factors for line loads differ from
those for loads distributed over 15 o arcs only by some 2 9/0 for typical values of p.
From this evidence it is concluded that 15 , i.e., 2a ~ R/4, is an acceptable upper
limit for contact width. There are other factors involved in the choice of contact
angle, as will be discussed in connection with fracture initiation.
Once it has been decided that distribution of the contact load is permissible,
there remains the problem of how to achieve this distribution in practice. One
deceptively obvious solution, which has been adopted b y a number of investigators (e.g., JAEGER and HOSRINS, 1966) is to groove the platens so that they
make direct contact over an arc of the sample's periphery. Although this arrange-

Fig.23. Improved version of curved-jaw loading jig. The upper segment has only one
effective degree of freedom, viz. freedom to rotate about the x-axis of Fig.21. Load is applied
through a properly designed spherical seat.

Eng. Geol., 5 (1971) 173-225

MEASUREMENT OF TENSILE STRENGTH BY D I ~ T R A L

COMPRESSION

207

ment was found unsatisfactory by COLBACK(1966), the writers built a testing jig
with reversible platens which were grooved on one side to the radius of the specimen, so as to give initial contact over 10 . The grooved platen was tested and
found to be unsatisfactory: under load, the edges Of the platen grooves bit into
the rock specimen and promoted platen cracking.
The testing jig finally designed had curved "platens", or jaws. Its design
was based partly on the need for a very simple self-centering jig which could be
loaded easily at very low temperatures. The jig (Fig.23) consists of a pair of jaws,
the contact surfaces of which are circular arcs of radius larger than the specimen
radius. The jaw radius was chosen so as to give a contact arc of approximately
10 (2a ~ R/6) at failure; with this width errors in the simple formula are negligible for practical purposes. The required radius was calculated from the more
general form of eq.16, which gives the width of the contact strip for contact between two cylindrical surfaces (TIMOSHENKOand GOODIER, 1951):
2a=4[

P
nt

RsR' [ 1 - v 2
Rs + Rj \ ~

1 - v2~] ~
+ ~ ]

(20)

where 2a is contact width, P is applied load, t is sample thickness, R~ is sample


radius, Rj is jaw radius, and subscripts s and J on the elastic constants denote
specimen and jaw respectively. Taking R~ = 1.063 inches, P/t = 8000 lbf/inch,
E, = 6.106 lbf/sq, inch, v~ = 0.2, Ej = 30- 106 lbf/sq, inch, v, = 0.3, the required
radius of the jaw R~ is 1.44 inches (concave). The jig was actually built with R, =
1.50 inches.
While this jig greatly improves the contact conditions, it does not lower
the contact stresses sufficiently to permit direct steel-rock contact. Tests showed
that when the jig was used with direct steel-rock contact there was sometimes
spalling at the face of Brazil specimens. This spalling apparently initiated in the
ends of the contact zone, approximately 0.05 inch in from the face of the 1 inch
thick disc, which accords reasonably well with the theoretical prediction of maximum tensile stress at a distance 0.03 t from the end of a contact strip of comparable
proportions. Spalling was more frequent in sandstone and limestone than in
granite, as might be expected, since the ratio of Pit to E, and hence the contact
width, is smaller for sandstone and limestone. In this connection it is interesting
to note that J. A. Hooper (personal communication), by chamfering the periphery
of glass cylinders, was able to increase the load before failure by 50 ~ compared
to unchamfered cylinders.
Reluctance to adopt a wider contact than 10 was partly intuitive, based
on the observation that the resistance mobilized by a specimen after initial fracture
may exceed the load required to cause initial fracture. It was felt that a jaw of
smaller radius would tend to grip the specimen and mask the primary failure when
continuous records of load and deformationwere not available. Thus it was
necessary to supplement the effects of increased contact area with platen cushioning,
Eng. Geol., 5 (1971) 173-225

208

M. MELLOR AND I. HAWKES

The second broad way of lowering peak contact stress is to reduce stress
concentrations and spread the load by permitting plastic yielding or otherwise
arranging inelastic contact. This can be achieved by making false platens of low
yield point metal (e.g., aluminum, copper, brass), but a cheaper and more convenient expedient is to use cardboard "platen cushions", which both spread the
load and to some extent relieve stress concentrations. COLBACK (1966) carried
out tests and recommended the use of cardboard 0.025 inch thick with flat steel
platens. This recommendation was followed by the present writers for initial
work with flat platens, and it was found to be quite satisfactory for 2.125-inches
diameter specimens. One advantage of this method is that the imprint of the specimen contact can be developed by rubbing with a soft pencil. This immediately
gives the actual contact width and shows whether or not the load was distributed
uniformly across the thickness of the specimen: a tapered imprint indicates improper distribution of load. Fig.24 shows the results of some imprint measurements
on 0.025 inch thick card. Comparing the effective contact widths with those
calculated for direct steel-rock contact, it appears that the cushions spread the
load over an area sufficient to avoid platen cracking, even without amelioration
of stress concentrations. Tests with 0.009 inch thick card showed that this thickness
was inadequate for 2.125-inches diameter specimens, but satisfactory for 1-inch
diameter specimens. It is suggested that the thickness of cardboard platen cushions
should be about 0.01 D, where D is specimen diameter.
i
Specimen

diameter:2125

//

inches

-03

//
Flat Steel
with 0025
cardboard

Platens
i n c h thick

cushions/

t/

~_._~/ /
/

~
E

/&

//

.=

//

-0.2

//

/i

//

//

c
o
u

/
/,/
/
/

./

g~
OI

Direct
Platen
Contact
E s = 5x I0 6 Ibl/in 2 ~

800

I0
16

P/t

I
0

i
2400

3200

(Ibflin)

Fig.24. Platen contact width as a function of load per unit thickness for flat platens with
cardboard cushions 0.025 inch thick. The calculated contact width for direct contact between
steel and rock is shown for comparison.

Eng. Geol., 5 (1971) 173-225

MEASUREMENTOF TENSILESTRENGTHBY DIAMETRALCOMPRESSION

209

With the curved-jaw testing jig ease of loading and unloading was a prime
consideration, so the specimens were wrapped around the periphery with a single
layer of double-thickness masking tape. This avoided the need for separate insertion of platen cushions and also prevented failed samples from disintegrating.
With taped samples the spalling problem was eliminated and there was no evidence
of platen cracking. The contact width was effectively increased by the tape, but
since the testing machine was effectively "stiff" and chart records of load-deformation were obtained, there were no problems from secondary cracking.

Platen freedom. In all the tests described here, the loading platens or loading jaws
were free to rotate about two orthogonal horizontal axes (loading direction being
vertical). While it is clearly necessary to have rotational freedom such that load
is distributed uniformly across the thickness of the specimen (about the x-axis
of Fig.21), it is both inconvenient and undesirable to have the platens free to
rotate on the perimeter of the specimen (about the z-axis). The inconvenience
arises from the need to align the platens relative to the specimen and each other
by hand, while the undesirability lies in the perturbation of the stress field caused
by lack of parallelism between platens in the plane of the disc. This last effect
is easy to appreciate when attempts are made to press a disc of ice between smooth
platens that are not perfectly parallel--the disc simply slips sideways.
In the most recent version of the curved-jaw loading jig (Fig.23) the guide
rods and their slide holes have a clearance of 0.004 inch, permitting a rotation
about the x-axis of 4 . 1 0 -3 rad when the test specimen is in place. This allows
the platen to correct for variation in diameter up to 0.004 inch over the 1-inch
thickness of the specimen. Freedom to rotate about the z-axis through a similar
angle is not considered to be significant. Load is transmitted from the testing
machine to the jig through a spherical seat designed in accordance with the principles outlined elsewhere (HAwI<~ and MELLOn, 1970).
Specimen dimensions and size effects
Size and proportions of test specimens are usually chosen to suit the convenience of the individual experimenter, but to facilitate comparison of data it
seems desirable to standardize dimensions, and also to choose dimensions that
will permit meaningful comparison of results with data from uniaxial tests.
Specimen diameter. Specimen diameter should obviously be chosen so as to give
a representative sample of the test material within the critically stressed volume
of the specimen. This probably means that the smallest linear dimension of this
critically stressed volume ought to be at least 10 times the grain diameter of the
rock. In the Brazil test, although the strict Griffith fracture conditions occur only
along the axis of rotational symmetry, it seems more realistic to assume that the
critically stressed area of the circular cross-section has a diameter of the order
Eng. Geol., 5 (1971) 173-225

210

M. MELLORAND 1. HAWKES

of R/2 (see FMRHURST, 1964; also Fig.3). In the ring test, which strictly has only
two critically stressed points in the circular cross-section, it might be assumed
for present purposes that there are two critically stressed areas, each having a
diameter of the order of R/IO (s Rn'PEROER and DAVIDS, 1947; also Fig.2 and
18). This then suggests minimum requirements for Brazil tests of R ~ 20d and
for ring tests of R ,-~ 100d, where d is grain diameter.
Another consideration is that the critically stressed volume should be comparable with the critically stressed volume of a uniaxial test since, to a first approximation, strength is inversely proportional to a power of the stressed volume
(see HAWKES and MELLOR, 1970, for further details). Following the assumptions
made above, the stressed volumes of Brazil and ring specimens are of the order
of R 2 t/5 and R 2 t/130 respectively, where t is specimen thickness. Assuming that
t ,-, R, this means that for stressed volumes directly comparable with those of
typical uniaxial tests (1-10 cubic inches) there is a requirement for R > 1.7 inches
in Brazil tests and R > 5 inches in ring tests. This condition agrees with that
derived from grain size considerations if it is assumed that the test must be valid
for materials with grains up to millimeter-size.
These considerations lead to prohibitive demands in the case of ring tests-outside diameters of 10 inches or more. For Brazil tests they indicate that outside
diameter should be more than 3 inches. However, the range of diameters for test
specimens of rock is normally limited by practical considerations; it tends to be
from 1-3 inches diameter.
A number of Brazil tests were made in order to investigate the influence
of specimen diameter ~over the practically feasible range of sizes. The results
(Fig.25) show that size effect for typical rocks is significant for diameters of 1 inch
and less, but they suggest that measured strength may tend to a steady value for
diameters above about 3 inches.
c

2400

1600

'$
c

"Fndiano
Limeston~

800

--x. . . . . . .

T/Bereo
~onC~i~on
e

-I---

Sondstone

<

i
l

i
2
Disc Diorne$er,inches

Fig.25. Apparent Brazil tensile strcn~h as a function of disc diameter, with thickness

equal to radius.
Eng. Geol., 5 (1971) 173-225

MEASUREMENTOF TENSILESTRENGTHBY DIAMETRALCOMPRESSION

211

Taking into account theoretical considerations, experimental findings,


and practical limitations, it is suggested that N X core size (2.125 inches diameter)
should be taken as the minimum acceptable size for Brazil and ring tests on typical
rock materials.
Specimen thickness. Theoretically the controlling stresses are the same for plane
stress and plane strain, so that there is no explicit restriction on Specimen thickness
and, in fact, there is a choice between testing a thin disc or a long cylinder. In
rock mechanics applications, however, the disc has been favored over the cylinder,
perhaps because of its economy or else because perfectly straight core cannot
usually be guaranteed.
The minimum thickness of a ring or disc is set in ~ibsolute terms by the size
of grains or defect structures in the rock. There should be sufficient thickness to
provide a representative sample, which suggests a minimum thickness of the order
of 10 grain diameters. The thickness should also be sufficient to reduce surface
effect to a relatively small role. The influence of surface grains can be estimated
roughly on the assumption that the influence of a given grain extends to a radius of
approximately two diameters. In the critically stressed regions, the volume of
rock unaffected by grains at the boundary (V) is related to the total volume (Vt) by:

(21)

V/Vt = 1 - 4 d/t

where d is grain diameter and t is specimen thickness. From Fig.26, where V/Vt
is plotted against t/d, it can be seen that a thickness greater than 20 grain diameters
is desirable, while a thickness less than 10 grain diameters is bound to make surface
effects significant. This means, for example, that for a rock with a l-ram grain
size, it ought to be greater than 2 cm, and it should not be less than 1 cm.
It is also desirable to have a critically stressed volume comparable with
that for uniaxial tests, as mentioned already in connection with specimen diameter.
1,0

to

~,

0.8

9
~

0.4

Fig.26.

I
20

V/Vt

1
40

as a function o f

t/dandp/p'

I
80

as a function o

t/R.
Eng. GeoL, 5 (1971) 173-225

212

M. MELLOR AND I. HAWKES

If the previous estimate of critically stressed volume is accepted, the ideal requirement is for t > 5/R 2 in Braziltests and t > 130/R 2 in ring rests (t and R in inches).
The latter requirement obviously cannot be met when R ~ 1 inch. The Brazil
test requirement indicates that, for a 2.125-inches diameter (NX) disc, the thickness
should be not less than about 1 inch.
In relative terms, a lower limit of thickness is set by the need to avoid eccentric loading, misalignmeht and buckling. This limit has not been determined
analytically, but in practice it is difficult to ensure that the plane of the disc or
ring is normal to the platen if t is less than about R/4, especially when platen
cushions are used.
The upper limit of thickness tends to be set by considerations of practicality
and expediency, such as availability of material, ability to ensure uniformity of
loading across the thickness of the specimen, and testing machine size and capacity.
For simple Brazil tests there may be an advantage in testing long cylinders (as
in concrete testing) rather than thin discs. If spalling occurs in the end zones
of the contact strip, a thin disc will inevitably fail completely from the platen
contact, whereas a long cylinder may be able to tolerate spalling without suffering
complete platen failure. I f p and A are contact pressure and contact area respectively before spalling and p' and A ' are contact pressure and contact area after
spalling:
p/p' = A ' / A = 1 - R/t
(22)
assuming that 2a = R / 5 and the end zones are approximately 5 a long (following
KUNERT, 1961; see Fig.22). This suggests (Fig.26) that spalling can probably be
tolerated for t > 10 R.
Experimental data for Brazil specimens of 2.125 inches diameter (Fig.27)
indicate that test results become erratic and anomalous for thicknesses less than
1 inch. For thicknesses greater than 1 inch there is a slow decrease of apparent
strength with increasing thickness.
From the foregoing considerations it is concluded that t = R should be
taken as the minimum acceptable thickness for Brazil specimens which meet the
standards given previously for acceptable diameter.
Volume effects. Once the minimum acceptable dimensions for test specimens
have been met, statistical theory of strength suggests that there should be a slow
decline of apparent strength with increasing specimen volume. A simple relationship expressing the dependence of strength S on volume V is:
Sv2/Sv1 ~. (V1/V2) l[m

(23)

where subscripts 1 and 2 denote two different volumes and m is a characteristic


material constant. For specimens of constant radius by varying thickness, eq.23
can be written as:
S v 2 / S v l = (tl/t2) 1/m

(24)
Eng. Geol., 5 (1971) 173-225

213

MEASUREMENT OF TENSILE STRENGTH BY DIAMETRAL COMPRESSION

and for geometrically similar specimens of varying radius as:

Sv21Svi

(25)

(RIIR2) ~/m

If sufficiently extended, plots such as Fig.25 and 27 ought to give limiting


slopes for large values of R and t which represent the volume effect. However,
if the present limited results are plotted logarithmically it is found that the thickness tests and the diameter tests give quite different values for the constant m.

4000
o

%
o

3000

o
o

"-..~.
o "~'~......~

Granite
""~0" ....~Oven - d r y
2 pts J"
"'" "--

el
I
7

"~

-200

2000

" " .*--....

Granite
Saturated

.-2
N
0

bar
-I00

Limestone
Oven-dry/

1000

J
0
I
0

I
I

I
2

i
O. 5

f
3

I
1.0

I
Disc

8O0

I
4

J
1.5

I
5

I
2.0

I
6

7cm
I
I
2.5 inches

Thickness

600
_Q

'~ 4 0 0

Berea Sandstone
Air-dry

o
m

200

%
I
0

6 cm

i
Z inches

Oisc T h i c k n e s s

Fig.27. Apparent Brazil tensile strength as a function of disc thickness with constant
diameter. A. Results of tests on random-thickness offcuts (high loading rate). B. Results of tests on
specially prepared specimens (low loading rate).

Eng. Geol., 5 (1971) 173-225

214

M. MELLOR AND I. HAWKES

The thickness tests suggest values of m from about 7 to 9, whereas the diameter
tests suggest that m ~ 60 for sandstone and limestone and m ~ 30 for granite.
Loading rate

There are at least three broad principles to be considered in choosing loading


rates for ring and disc tests:
(1) Strength is calculated from test results on the assumption that the
discs or rings are linearly elastic, so loading rates must be fast enough to evoke
elastic response from the specimens.
(2) Most rocks fail by time-dependent development of internal microcracks and, in general, apparent strength increases as loading rate increases or
test duration decreases.
(3) There are practical limitations on loading rates. Typical testing machines
become unreliable at their highest speeds (head speed ~ 2 inches/min, loading
rate ~ 4.105 lbf/min) due largely to inertial effects in the load measuring and
dial readout systems. At very low loading rates (head speed ~ 0.001 inch/min,
loading rate ,-~ 400 lbf/min) test durations become inordinately long--15 min
or more.
Elastic response. Observations show that the quasi-elastic response of rocks be-

comes more nearly linear as loading rate, or strain rate, increases (STOWE and
AINSWORTH, 1968; GREENand PERKINS, 1968). At low loading rates non-linearity
of the stress/strain characteristic is pronounced, particularly at high stresses,
since there is sufficient time for flow and internal cracking to exhibit their effects.
Thus the actual peak stresses at the critical points of disc and ring specimens will
be lower than those predicted by linear elastic theory, and test results calculated
from elastic theory will tend to exaggerate the strength of the rock.
This effect can be seen very clearly in disc and ring tests on ice at high
homologous temperatures (Fig.28). At very low loading rates (machine head
speed ~ 10 -s inches/rain) a disc or annulus of ice may suffer large visco-plastic
strains without breaking (Fig.29). At somewhat higher machine speeds the ice
specimens eventually break after the load-deformation curve has peaked, but it
seems clear that the stress concentrations predicted by elastic theory have been
relieved by visco-plastic flow, since the calculated strengths are double those
measured in careful uniaxial tests on the same material. The machine used by the
writers for disc and ring tests on ice could not be driven fast enough to define an
asymptotic limit for apparent ring tensile strength as a function of loading rate,
but tests on low salinity sea ice by PAIGEand KENNEDY(1967) suggest that apparent
ring tensile strength is tending to a limit at a head speed of 20 inches/min. However
in the present work the limiting ring tensile strength, which appears to converge
with the Brazil tensile strengttf:,(K factor unity), is much smaller than the direct
tensile strength for this kind of ice.
Eng. Geol., 5 (1971) 173-225

215

MEASUREMENT OF TENSILI~ STRENGTH BY DIAMETRAL COMPRESSION


I000

, i,~,1

, Ir~ll

I i'l',---

' I '

Fine-grained Ice
Temp:-?*C

6o

800

600

2KP
~dt

40
/

J-

~'~-

Annulus

bet

.,,~=8 75

(ibf/in 2) 4 0 0

"'t\

200

--{

_~

Disc K=IO

20

-~--__
t -

10-3
k , I,t,[

IILl,l

I
10-3

10-2
I ,ILI,J

;lll~l

iO "1
I il,l,I

10-2

iI,l,l

r
I0-1

I0 o
I , J~lil

II,

i
I
IOOin/min

I0 i cm/min
I

Head S p e e d

Fig.28. Ring tensile strength and Brazil tensile strength as a function of loading rate for
fine-grained polycrystalline ice at --7 C.

Fig.29. Permanent deformation of a 2-inches diameter disc of polycrystalline ice pressed


between flat platens converging at 0.005 cm/min (temperature --7 C.)

Eng. Geol., 5 (1971) 173-225

216

M. MELLOR AND I. HAWKES

For most rocks at normal environmental temperatures this marked transition


from viscous to elastic predominance is likely to occur at strain rates very much
smaller than those imposed by typical testing machines, but nevertheless
the general point remains valid, and loading rates should be sufficiently fast to
suppress creep effects.

Time-dependent fracture. Most rocks fail by progressive growth or development


of numerous internal microcracks over a finite period of time, and consequently
the apparent strength of a uniaxial test specimen increases with increasing loading
rate. The same effect can be expected in ring and Brazil tests. This effect works
in the opposite sense to the visco-elastic effect just described, but it appears that
for common rocks it is the dominant effect for tests at typical loading rates.
PRICE (1966) made tests on 1-inch diameter ring specimens, varying loading rate
over two orders of magnitude, from 102 to 104 kgf/min (2.2.102 to 2.2.104
lbf/min). In general, ring tensile strength increased by 17% to 449/0 as loading
rate was increased over the stated range, but in one set of tests the apparent strength
became erratically lower at the highest loading rates. The writers made a small
number of Brazil tests on 2.125 inches diameter discs of two rock types over a
range of machine speeds from 0.004 ineh/min to 2.8 inch/min (approximate
loading rates 300 lbf/min to 400,000 lbf/min). The results (Fig.30) show a similar
trend to those of Price, and again there is an apparent drop in strength at the highest
loading rates for one of the rock types.
,

[ ,l,i,

i ,i,i,

'

I 'l'l'l

~ 'I'TI

'

I 'l'l
"30

4000

3000

.~00
o
Barre

Granite

o--

bor

2000
-I00

IO00
8erso

Sandstone

&

I ,I,1.1

I0 "4
I

i ,I,l.I

10 .3
i ,I,l,I

10-3

I ,1~1,1

I 0 -z

IJl,t,I

I 0 -z

I0"
Nominal

J ,I.hl

I 0 "l

I tlJiJ

Head

[ I ,I,Lu

I0
I ,I,1.1

I0

I0 e in/rain
I ,I,l,I

IO I

era/rain

Speed

Fig.30. Apparent Brazil tensile strength as a function of loading rate for two rock types.

Machine speeds. A typical speed range for high-capacity hydraulic testing machines
of the older type is from about 0.001 to about 3.0 inches/min. For lower capacity
Eng. Geol., 5 (1971) 173-225

MEASUREMENTOF TENSILESTRENGTHBY DIAMETRALCOMPRESSION

217

screw-drive machines a typical range is from 0.002 to 2.0 inches/min. At the top
end of these ranges, failure of a small ring or disc specimen is virtually instantaneous, and the response of a dial pointer or a pen recorder may be too slow to
give reliable readings. An electrical load cell with oscilloscope readout is then
required. At the highest loading rates disc and ring specimens usually suffer both
primary and secondary cracking and, as is shown in another section, may affect
the test result. At the bottom end of the typical speed range the testing machine
usually operates reliably and it is easy to halt the test after primary fracture without
inducing secondary cracks, providing the machine is adequately "stiff". However,
at the lowest speeds test durations are excessive for a program of multiple testing.

Choice of loading rate. Just as in other forms of testing, it is futile to specify loading
rates until the purpose of the test has been defined. If the disc or ring test is to be
used as an expedient alternative to the uniaxial direct test, then the loading rate
for the test should be related to that used in the reference test. If the disc or ring
test is to be used as a self-consistent index of tensile strength, then there is need
for a rate which will not become a hidden implicit variable in a test program.
The properties of the test material must also be taken into consideration: a low rate
of loading which may be perfectly acceptable for a highly elastic rock may be
wholly unacceptable for materials which creep readily.
On the basis of test experience with a wide range of materials and five
different testing machines, the writers are presently of the opinion that, for common rocks and typical machines, head speeds above 1.5 inches/min are too high
unless oscilloscope readout is provided, while nominal head speeds much below
0.1 inch/min may result in excessive test durations. With simple equipment, reproducibility tends to be best at the lowest speeds, but there are no great problems
in this respect up to a nominal head speed of about 0.5 inch/min. However, rate
sensitivity may be quite high over the practically feasible range of speeds.
For solid discs of Lucite and some resins, available machine speeds were too
low to provide valid Brazil tests. Lucite discs invariably suffered gross deformation
at the platen contact, and failure occurred by separation of crescentic fragments
from outside the contact zone while a large central segment of material directly
between the platen contacts remained intact. The width of the contact strip at
failure was equal to or greater than the specimen radius, and the loading diameter
had a permanent strain of the order of 3 70 after failure.
Available machine speeds were also too small to provide any assurance of
elastic response for rings of polycrystalline ice at temperatures within 10 C of the
melting point.
Load readout system
The object of a diametral compression test is to measure the force required
to form the primary diametral crack in the specimen so that tensile strength can
Eng. GeoL, 5 (1971) 173-225

218

M. MELLORAND I. HAWKES

be calculated. In most laboratories this force is read directly from the dial of a
universal testing machine, but the writers found that the peak force measured
in this way is sometimes too large. Solid discs and rings with small values of p
sometimes suffer primary failure at a certain load, but the two halves of the specimen (which are clamped securely between the platens of the loading machine)
then go on to carry a somewhat higher load than that which caused the primary
failure. There may also be exaggeration of the peak force by intertial effects at
high rates of loading.
To avoid errors of this kind, the writers supplemented the dial readout
system with chart records of load as a function of displacement, and also attempted
to release the load at the first indication of failure. Fig.31 gives a representative
selection o f records for solid discs tested in a high-capacity machine with rapid
I

DISC

Force

Displacement

Force

Displacement

Fig.31. Representative selection of test records for: A. solid disc specimens; and B. ring
specimens 6o = 0.085) of reck. The ordinate is load, the abscissa is crosshead displacement.
Eng. Geol., 5 (1971) 173-225

MEASUREMENT OF TENSILE STRENGTH BY DIAMETRAL COMPRESSION

219

release of load at failure. In some instances the diametral crack occurred and load
was removed before any secondary cracks could form. In other cases the diametral
crack occurred and secondary cracks formed during unloading. In some cases the
primary crack occurred, loading continued, and a higher load was carried by the
cracked sample before final collapse. Sometimes the primary and secondary
cracks occurred at the same load level. Corresponding data are given for ring
tests on annuli with p = 0.085.

Specimen preparation
Disc specimens are usually prepared in the laboratory by coring block
samples and slicing the resulting cores. The easiest way of making ring specimens
from block samples is by drilling and overcoring, and slicing the resulting hollow
cylinders. Alternatively, disc specimens may be pierced with a central hole by
drilling on the lathe.
There are no established tolerances for specimen dimensions in diametral
compression testing. Probably the most critical requirement is that the generator
along which the specimen is loaded should be sensibly free from bumps and waviness. In the absence of any direct theoretical or experimental guidance on this
point, it may be inferred from studies of the uniaxial compression test (HAWKES
and MELLOR, 1970) that bumps up to 0.001 inch can probably be tolerated in
most cases, even when no platen cushions are used. This tolerance can be met
routinely when specimens are cored from block samples with a rigid drill press and
good diamond tools; use of a platen cushion, even a single layer of tape around
the specimen, provides an adequate safety factor. With specimens of adequate
thickness (t ~ R), variations in thickness from specimen to specimen of a few
percent are unimportant as long as the exact thickness is entered into the computation of tensile strength for each individual specimen. However, it is more convenient to have all specimens of uniform thickness to within + 1 ~o so that a single
factor can be applied in computation of results. Within a single specimen it is
desirable that the fiat surfaces should be perfectly plane and parallel, and also
normal to the rotational axis of symmetry. However, in a specimen of adequate
thickness there is no apparent reason to believe that relevant tolerances need be
unduly stringent. It is suggested that end surfaces should be flat to better than 0.01
inch and that parallelism and squareness should be within 0.25 . These seem
realistic goals insofar as they can be attained with a good cutoff machine and
suitable jigs, without calling for surface grinding or lapping.
Bearing in mind that a prime merit of a Brazil test is its simplicity, the following basic procedure for specimen preparation is suggested.
(1) Extract core from oriented block samples using a rigid drill press and
maintaining "clean" machining conditions. Details of tool speeds and contamination problems are discussed elsewhere (HAWKESand MELLOR, 1970). Keep track
of rock orientation by marking the head of each core prior to drilling it out of

Eng. Geol., 5 (1971) 173-225

220

M. MELLOR AND I. HAWKES

the block, and by transferring the head marking along generators after the core
has been extracted (felt tip pens are suitable for this).
(2) Slice the core into discs of the required thickness with a precision cutoff
machine and a diamond blade, using clean water as coolant. Support the core in
fixed vee-blocks on both sides of the cut, using a stop to control the disc thickness.
The vee-bloeks must be set with great care to ensure normal cuts.
(3) Wash the specimens in clean water, dry in an oven at ll0C for 24 h,
and finally bring the specimens to the desired water content. Very low water
contents can be induced by exposure to constant humidity atmospheres, using
saturated aqueous solutions of various salts to control humidity. For standardization of the "air-dry" condition it is suggested that specimens should be equilibrated with an atmosphere of 100 % relative humidity, i.e., an atmosphere saturated
with respect to pure water. Complete saturation of the rock can be induced either
by vacuum technique or by immersion in heated water. Intermediate water contents
can be induced on a trial and error basis by partial drying of saturated specimens
or, in highly porous rocks, by addition of water to each specimen from a vernier
syringe.
These basic procedures must obviously be modified for preparation of rocks
which are soluble or highly moisture-sensitive. If specimens have to be made from
field-drilled core rather than block samples, it may be necessary to dress the cylindrical surface of the core with a lathe and toolpost grinder.

Recommended procedure for Brazil tests


The following notes are offered as an outline of optimum procedure for
the Brazil test:
Specimen preparation. Specimens should be cut and prepared using clean methods.
The cylindrical surface should be free from obvious tool marks, and any bumps
or waves across the thickness of a specimen should not exceed 0.001 inch in height.
End faces should be fiat to within 0.01 inch and, if possible, square and parallel
to within 0.25 . Specimen orientation should be known, and water content should
be either controlled or measured.
Specimen dimensions. It is recommended that specimen diameter should be not
less than NX core size (2.125 inches diameter); greater diameter may be necessary
for coarse-grained material. It is suggested that specimen thickness should be
approximately equal to the specimen radius.
Application of load. Load should be distributed over the contact zones so as to:
(a) avoid platen cracking; and (b) induce unit stress severity in the specimen
mid-section for a wide range of rock types. It is suggested that the loading jaws
should be designed to give air ~ 1/12 for direct rock-steel contact, and that in
Eng. Geol., 5 (1971) 173-225

MEASUREMENT OF TENSILE STRENGTH BY DIAMETRAL COMPRESSION

221

addition the specimens should be wrapped with a single layer of double-weight


masking tape. The curved-jaw loading jig with a spherical seat is recommended.
Loading machine and load readout. The loading machine should be as "stiff"
as possible in order that head travel may be effectively arrested at failure. It is
desirable to have an automatic load-displacement record so as to identify the
force for primary fracture unequivocally.
Loading rate. Loading rate must be determined to some extent by the properties
of the materal (avoidance of excessive creep) and the proposed application of the
test data (structural, fragmentation). For NX-size specimens of typical rocks,
rate sensitivity becomes appreciable at machine speeds 1 above about 0.1 inch/min,
while test durations tend to become excessive for machine speeds much below
0.1 inch/min. For typical rocks a loading rate of approximately 0.25 inch/min
will probably be satisfactory for many practical purposes.
Test results. With careful application of the above technique there should be no
difficulty in achieving a standard deviation of approximately 5 ~o.
CONCLUSION

Indirect tensile tests, which induce non-uniform stress fields controlled


partly by the properties of the test material, can never fully substitute for the direct
uniaxial tensile test. However, the Brazil test is capable of giving a very good
measure of uniaxial tensile strength for Griflith-type materials when it is carefully
performed, with special attention paid to control of contact stresses, to accurate
load readout, and to other factors listed in the foregoing outline of recommended
procedures.
There are some materials for which the standard Brazil test gives consistent
but wholly erroneous results. Ice is one of these materials: it is anomalous in
that its ratio of compressive strength to tensile strength is very low in comparison
with typical rocks and ceramics, and this means that its failure criterion does not
match the ususal tacit assumptions made in the interpretation of the Brazil test.
There are several serious objections to the ring test which appear to completely invalidate it as a practically useful test. Values of tensile strength calculated
from ring tests on Griffith-type materials are greatly in excess of the uniaxial
tensile strength when the hole size is small. Calculated strength tends asymptotically towards the modulus of rupture (flexural strength) as hole size increases in
relation to outside diameter. The test is not self-consistent, as the effective stress
concentration factor for a given specimen size varies with rock type and with
t No-load machine speed.
Eng. GeoL, 5 (1971) 173-225

222

M. MELLOR AND I. HAWKES

condition (e.g. temperature, water conten0. For materials that creep readily,
the ring test is highly sensitive to rate of loading.
APPENDIX
STRESSES IN AN ELASTIC CIRCULAR RING WITH SYMMETRICAL NORMAL SURFACE
LOADING AT OPPOSITE ENDS OF A DIAMETER (AFTER

NEVEL, 1969)

TI

(a) p-O,l

(b) p-0.3

STRESS % OF PEAK V A L U E
A

0-10

10-20
20-30

C
D

30-40

40-50

50-60

60-70

70.80

8O-9O

90-100

( C) p " 0 l ~

Fig.32. Principal tensile stress magnitudes in a ring under diametral compression. Values
expressed as a percentage of the peak value.

Eng. Geol., 5 (1971) 173-225

MEASUREMENT OF TENSILESTRENGTH BY DIAMETRAL COMPRESSION

0.3

0.10.~~2

0.5

0.6

Vol%ume

0.1

223

0,P5

0.2

0.4

Fig.33. Percentage volume of a diametrally compressed ring expedencing a tensile stress


witl~ 10~/~of the peak value, plotted against dim'neterratio P.
The required general stress solution in plane polar coordinates (r, O) with
symmetry about the line 0 = O, n is:

_~o2 ~o (r)-2
0"~

r~

+ r-~- ~o
_

+ ~ [~
n

- - ~ ' +~'(~oI - ~
0"0 - A2ro
.2 .

+~

B
( ~. o ) - 2.r ~

E[

"-~

, \ -ro/

+ D(2 + n)(1 - n)

(~oI-']

cos

(nO)

~o~

+ Co
[ 32 + 2 1 g ( ~ o ) ] r o

[A(2+n)(l+n)(~o)"

+ Cn(1 + n)\~o/
a,o

(rl

~ ~'+ +, (~o)'+ ~ '

+D(2-

(~)" - B ( 1 -

A(1 + n) 7o

Bn(1

(r]

n) -\ rol

"-2

n ) ( 1 - - n) ~o
n)

~r ] .~ -

ro2
C(1 + n )

( r)

--n--2

n
+ 1)(1 - n)

(~o) +] .sin~n0,
r~

where Ao, Bo, A, B, C, D, ro are constants and n is a positive integer.


For a circular ring loaded as shown in Fig.32, the constants are as follows:

Eng. Geol.,5 (1971) 173-225

224

M. MELLOR AND I. HAWKES


r0 is the o u t e r radius,

Ao =

~___

Pro
2n(1 - a 2)

, Bo -

n(1 - a 2)

,Co =0

Pro sin
_ _ not [ a 2n
nR
not

1 + na 2"-2]

Pro sinnot [ a 2 .

1-na

PF_ 0 _sin/10t [ a2 n _
nR
not

a 4n + n a 2n+2 ]
1 + n

Pro
s i n n o t [ a2 n
_
_

a 4n - n a 2 n - 2 ]

_ _

Pro a 2

IrR

n~

l~-+-n

2n+2]

i -

where
(1

- -

a2~) 2 - n 2 a 2~-2 (1 - a2) 2

ri/ro

Ot, r l , r, 0 are d e f i n e d as i n Fig.32


P is the a p p l i e d force per u n i t t h i c k n e s s o f the ring.
REFERENCES

ADDINALL,E. and HACKETT,P., 1964. Tensile failure in rock-like materials. Symp. Rock Mech.,
6th, Rolla, Mo., 1964: 515-538.
ADDrSALL, E. and HACI~Tr, P., 1965. Rock in tension. A problem in strata control. Civil Eng.
(N.Y.), 60: 1056-1058.
AssuR, A., 1960. Composition of sea ice and its tensile strength. U.S.A. SIPRE, Res. Rept.,
44:49 pp.
BBn~NBAOM, R. and BRODm, I., 1959. Measurement of the tensile strength of brittle materials.
Brit. J. AppL Phys., 10: 281-287.
BROWN, E. T. and TROLLOPE,D. H., 1967. The failure of linear brittle materials under effective
tensile stress. Rock Mech. Eng. Geol., 5: 229-241.
Btrr~ovrrcn, T. R., 1959. Some physical properties of ice from the TUTO tunnel and ramp,
Thule, Greenland. U.S.A. SIPRE, Res. Rept., 47:17 pp.
COLBACK,P. S. B., 1966. An analysis of brittle fracture initiation and propagation in the Brazilian
test. Proc. Congr. Intern. $oc. Rock Mech., 1st, Lisbon, 1966: 385-391.
FAmmr~sT, C., 1964. On the validity of the Brazilian test for brittle materials. Intern. J. Rock
Mech. Mining ScL, 1: 535-546.
FmON, L. N. G., 1924. The stresses in a circular ring. Inst. Civil Engrs., SelectedPapers, 12.
FRANKm~STEn~,G. E., 1959. Strength data on lake ice. U.S.A. SIPRE, Corps Engrs., Tech. Rept.
59.
GAm~Pv, S., HAWK,S, I. and MEta.og, M., 1971. Stress-strain relations for rocks in uniaxial
tension, in preparation.
G~G.
J., 1965. The axial cleavage fracture. Eng. GeoL, 1: 31-72.

Eng. Geol., 5 (1971) 173-225

MEASUREMENT OF TENSILE STRENGTH BY DIAMETRALCOMPRESSION

225

GREEN, S. J. and PERKINS,R. D., 1968. Uniaxial Compression Tests at Strain Rates from lO-4/sec
to 104/sec on Three Geologic Materials. Materials and Structures Laboratory, General
Motors Technical Center, Warren, Mich., 35 pp.
HAWKES,I. and MELLOR, M., 1970. Uniaxial testing in rock mechanics laboratories. Eng. Geol.,
4: 179-285.
HAWKES, I. and MELLOR, M., 1971. Deformation and fracture of ice under uniaxial stress. J.
GlacioL, in press.
Hoens, D. W., 1964. The tensile strength of rocks. Intern. J. Rock Mech. Mining ScL, 1: 385-396.
Honas, D. W., 1965. An assessment of a technique for determining the tensile strength of rock.
Brit. J. AppL Phys., 16: 259-268.
HONDROS, G., 1959. The evaluation of Poisson's ratio and the modules of materials of a low
tensile resistance by the Brazilian (indirect tensile) test with particular reference to concrete.
Australian J. Appl. ScL, 10: 243-268.
HUDSON, J. A., 1969. Tensile strength and the ring test. Intern. J. Rock Mech. Mining ScL, 6:
91-97.
JAEGER, J. C. and HOSKINS,E. R., 1966. Stresses and failure in rings of rock loaded in diametral
tension or compression. Brit. J. AppL Phys., 17: 685-692.
KUNERT,K., 1961. Spannungsverteilungim Halhraum bei elliptischer Flachenpressungsverteilung
iiber einer rechteckigen Druckflache. Forsch. Gebiete Ingenieurw., 27: 165-174.
NELSON, C. W., 1939. Stresses and Displacements in a Hollow Circular Cylinder. Thesis Univ.
Michigan, Ann Arbor, Mich. (Some results given in RrPPERGERand DAVIDS, 1947.)
NEVl~L, D. E., 1969. The ring test, Brazil test and strength of sea ice. U.S. Army Cold Reg. Res.
Eng. Lab., Tech. Note (unpublished).
PAIGE, R. A. and I~NNEDY, R. A., 1967. Strength studies of sea ice. Effect of load rate on ring
tensile strength. U.S. Navy Civil Eng. Lab., Tech. Rept., R545:25 pp.
PESCHANSgaL I. S., 1967. Ledovenie i Ledotekhnika. Gidrometereologicheskoe Izdatel'stvo,
Leningrad, 461 pp.
PRICE, D. G. and K~LL, J. L., 1966. A study of the tensile strength of isotropic rocks. Proc.
Congr. Intern. Soc. Rock Mech.. Ist, Lisbon, 1966: 439--442.
Rn'~ROER, E. A. and DAVIDS, N., 1947. Critical stresses in a circular ring. Trans. Am. Soc.
Civil Engrs., 2308: 619-635.
RL~NICK, A., HUNTER,A. R. and HOLDEN,F. C., 1963. An analysis of the diametral compression
test. Mater. Res. Std., 3: 283-289.
SEAF~n~D,K. J., G~VND, H. and PrNcus, G., 1967. An experimental investigation of the strain
distribution in the split cylinder test. J. Mater., 2: 703-718.
STOWE, R. L. and AINSWORTH,D. L., 1968. Effect of rate of loading on strength and Young's
modulus of elasticity of rock. Proc. Syrup. Rock Mech., lOth, Austin, Texas, 1968, in press.
TIMOSHENKO,S. and GOOD~, J. N., 1951. Theory of elasticity. McGraw-Hill, New York, N.Y.,
372 pp.
WEmULL,W., 1939. A statistical theory of the strength of materials. Proc. Roy. Swed. Inst. Eng.,
151: 1-45.
WEIBULL,W., 1952. A survey of "statistical effects" in the field of material failure. Appl. Mech.
Rev., 5(11): 449-451.

Eng. GeoL, 5 (1971) 173-225

Você também pode gostar