A new multi-axial failure criterion for concrete

M. Como
Department of Civil Engineering, University of Rome Tor Vergata, Rome, Italy

R. Luciano
DiMSAT, University of Cassino, Cassino (Frosinone), Italy

ABSTRACT: In this paper, a new failure criterion for the concrete is developed. Cement, sand, aggregates
and water are mixed together and, after the setting, the cement paste binds aggregates and hardens to form
concrete. Experimentally, pores of different sizes are diffused in the hardened cement paste and micro stresses
arise in the neighborhood of the small pores when the concrete is loaded. The main idea of the proposed failure criterion is based on the assumption that the macroscopic concrete failure occurs when the maximum tensile stress, around the small pores reaches the local hardened cement paste tensile strength. The proposed
model explains many experimental phenomena as, for example, the different compression and tensile
strengths, the different crack geometries in the tension and compression failures and the experimental failure
contours under biaxial stress.
1 INRODUCTION
The research of a failure criterion for concrete under
multi-axial stresses is a very important task because
of its numerous civil engineering applications.
Nowadays, several concrete failure tests under multiaxial stresses are available and many theoretical
failure criteria have been proposed. These criteria,
on the other hand, have been formulated, as a rule,
by modifying failure conditions concerning other
materials, (e.g. the Coulomb criterion) to the concrete. Among all the criteria available in literature,
the Rankine criterion (Van Mier, 1997, Jirsek &
Bazant, 2001, McClintock & Argon, 1966) seems to
have a physical basis for the concrete. According to
this criterion, in fact, concrete fails in brittle manner
as soon as the maximum principal stress exceeds the
concrete tensile strength. It is well known, on the
other hand, that the strict application of the Rankine
criterion could imply that for a compressed concrete
specimen failure could never occur because no tensile stress develops. On the contrary, brittle failures
are ob-served also in compressive regimes. Tensile
stresses, in fact, can arise at microscopic level and
produce failures. The concrete is a multiscale material consisting of aggregates embedded in a matrix
of binder, the hardened cement paste (Van Mier,
1997). Further, the hardened cement paste is weakened by a capillary porosity, produced by evaporation of the water in excess. Diffused pores are visible with the electronic microscope: they have
irregular shape and are spread up among the needles

of calcium silicate hydrates, the structure of the

hardened cement paste. When concrete is loaded,
high stress concentrations take place at microscopic
level around these small cavities: they produce micro cracks that propagate across the hardened cement paste till they become visible at macroscopic
level.

Figure 1. Multiscale micro-geometry of the concrete.

Many contributions are available in literature

which use Micromechanics to estimate the mechanical properties of the concrete e.g. the elastic moduli
(Yang & Huang, 1996). In line with this approach,
ere, two steps homogenization technique is used to
evaluate the stresses supported by the cement paste
when the concrete is loaded (Fig.1). In the first step,
the global moduli of the mortar are evaluated. In the
second step these moduli are used to estimate the
behaviour of the concrete made by the mortar and
the gravel. Combining these two results, obtained at
two different scales, we can evaluate the average
stresses in the hardened cement paste. These mean
stresses represent the asymptotic stress state for the

local stresses acting around the small pores diffused

in the hardened cement paste. These local stresses
are responsible of the failure in the concrete: at microscopic level collapse occurs when the maximum
tensile stress around the pores reaches the tensile
strength of the hardened cement paste.
In the above exposed framework a biaxial failure
contour is analytically obtained for concrete. The
comparison of this theoretical criterion with several
experimental results seems to be satisfying.
2 2 STRESS CONCENTRATIONS AROUND
THE PORES DIFFUSED IN THE HARDENED
CEMENT PASTE
Concrete can be considered a material homogeneous
and isotropic at average level. Mean stresses in the
various components can be obtained by using the so
called mixtures law and the elasticity equations.
(Nemat-Nasser & Hori, 1999). In this formulation
strong simplifications can be obtained by assuming
the same average Poisson ratios for the components
of the concrete, as approximately it does occur. In
this case, as stated beforehand, we consider a two
steps homogenization procedure: in the first step, the
mortar is made by sand and hardened cement paste
while, in the second step, the concrete is composed
by mortar and gravel. Thus we get:

xp = K 2 xc ;

yp = K 2 yc ;

zp = K 2 zc

E p (1 2 2 ) Es2 Em Es Es E p + Em E p
=
1 s
Em
D

K 2m ,c =

(3a)

P
P

=
= A( ) zp
z

at the equator Ez:

E
E
RE = 0;
= B( ) zp ;
= C( ) zp
z

(3b)

A(), B(), C() depend on the Poisson ratio of the

hardened cement paste and are given by:

(2)

27 15
3 + 15
, B ( ) =
,
2 (7 5 )
2 (7 5 )
3 15
C ( ) =
2 (7 5 )
A( ) =

K 2 = K 2p ,m K 2m,c
K

P
RR
=0

(1)

where:

p ,m
2

Nasser & Hori, 1999). If such a constraint is not satisfied, i.e. Ec and Em are chosen arbitrarily, the results obtained by using the proposed failure model
can be physically uncorrected since the elastic
moduli are meaningless. Therefore in eqs. (3a) and
(3b), concrete and mortar moduli will be suitably
chosen by using micromechanics and accurate test
results in order to effectively represent the elasticity
of the composite materials. Strong stress concentration occurs around the small cavities spread up inside the hardened cement paste. In order to define
these local stresses can be assumed that the mean
stress inside the hardened cement paste is the asymptotic stress acting at large distance from the single pore. Thus, when the concrete is, for instance,
axially loaded by the uniform stresses z, and the
mean stress zP takes place inside the hardened cement paste, by assuming for the pore a spherical
shape, the local stress around the cavity is
(McClintock & Argon, 1966):
at the pole Pz (see Fig.2):

2
Em (1 2 2 ) Eg Ec Eg Eg Em + Ec Em )
Ec
1 g
D

(2a)
(2b)

(4)

where indicates the local Poisson ratio of the hardened cement paste without pores, larger than the average Poisson ratio .

with:
D = Em2 (1 2 2 ) Eg Em (2 2 4 2 )
+ E (1 2 )
2
g

(2c)

where E is the Young modulus and the indices c, g,

m, s, p indicate concrete, gravel, mortar, sand and
cement paste respectively. The elastic moduli of the
constituents and of the concrete are related; in fact
the elastic modulus of the mortar must be the result
of an homogenization process in which the constituents are the sand and the cement paste. In the same
way, the elastic modulus of the concrete is obtained
from the homogenization of mortar and gravel.
Therefore Ec and Em can be obtained by using a micromechanical models and must satisfy the HashinShtrikman bounds which depend on the geometry
and the volume fraction of the constituents (Nemat-

P
P
z

P
E
y

Figure 2. Local stress around the spherical pore under an asymptotic compressive stress.

Actually, the small pores spread up in the hardened

cement paste have irregular shape. Pores of irregular
shape in plane elasticity can be represented by small
elliptic holes having various orientation. High stress
concentrations occur on the boundaries of elliptic
cavities with the major axis orthogonal to the direction of the applied uniaxial tensile stress. The stress
component y, acting at the end of the major axis, increases strongly while the other ones hold almost the
same (Fig.3).

3 THE FAILURE CONDITION

Failure occurs when the maximum principal tensile
stress around the capillary pores of the hardened
cement paste reaches its tensile strength. This condition, even if of local character, concerns all the cavities continuously diffused in the hardened cement
paste and, when reached, produces the macroscopic
destruction of the binder and, accordingly, the concrete failure.
tz

cz

tz

cz

Figure 4 Uniaxial tension and uniaxial compression

Under increasing tensile stress acting on the conEz

crete,
, acting at the poles Py and Px, i.e. along
the equatorial circle Ez of the cavities, can reach the
tensile strength of the hardened paste and produce
Ez
failure. Cracks are thus orthogonal to the
, i.e.
Figure 3 Stress concentrations around an elliptic cavity

At the end of the minor axis the circumferential

stress x changes sign but its intensity almost does
not increase. Thus, taking into account the irregular
shape of the pores diffused in the hardened cement
paste, we assume that the long ellipsoids pores with
the major axis orthogonal to the applied tensile
stress, will exhibit the highest stress concentration
and will be the first to fail. To the authors knowledge, there is no simple analytical solution able to
describe the local stress field in the neighbourhood
of an ellipsoidal cavity. Thus, for sake of simplicity,
we continue to use the field equations describing
stresses occurring along the boundary of the spherical pore but increasing the intensity of the stress
component directed as the applied tension. Thus, introducing the intensity factor kB to take into account
the irregular shape of the cavity, we modify the expression of the local stress field (3a) and (3b) concerning the spherical hole in the following way:
at the pole Pz:
P
RR
=0
z

(3a)

P = P = A( ) zP
z

but, at the equator Ez:

RE = 0;
z

E = k B B( ) zP ; E = C ( ) zP
z

(3b)

orthogonal to the applied tensile stress. Concrete in

uniaxial tension thus fails when
Ez
f rtp,loc =
= k B B ( ) K 2 f rtc

(5)

where f rtp,loc and f rtc indicate the tensile strengths of

the hardened cement paste and concrete respectively. Conversely, in presence of uniform compression z, the tensile strength of the cement paste is
reached at poles Pz of the of the cavities. In this case
the failure condition is:
f rtp,loc = A( ) K 2 f rcc .un

(6)

Cracks will now be parallel to the compression

stress z acting on the concrete.
By tests, in the case of uniform uniaxial compression, cracks turn out parallel to the direction of the
applied compression while, in the case of uniaxial
tension, cracks are orthogonal to the applied tensile
stress. These results are captured by the present
model of concrete failure because it takes into account the different local failures occurring at the
pole or at the equator of the pores diffused in the
cement paste.
At the same value of the stress acting on the concrete, the maximum local tensile stress, produced in
the binder when the concrete is in uniaxial compression, is much lower than the maximum local tensile
stress occurring when the concrete is in uniaxial tension. Thus, the intensity of the stress applied to the

concrete, able to produce compression failure, will

be much higher than the stress intensity required to
produce tensile failure. The proposed failure model
is therefore able to explain why the tensile concrete
strength is much lower than the compression one.
In order to validate the proposed concrete failure
model, on the other hand, it is required that the
maximum local tensile stress, corresponding to the
concrete compression failure, is equal to the maximum local tensile stress occurring in the concrete in
uniaxial tension. Thus we have the condition
A( ) K 2 f rcc .un = k B B( ) K 2 f rtc

(7)

and:

A( ) f rcc .un
kB =
B( ) f rtc

4 BIAXIAL FAILURE CONTOUR FOR THE

CONCRETE
Biaxial stress states are very frequent. The biaxial
stress regimes are of four different kinds: 1. compression-compression (Bcc) 2. compression-traction
(Bct), 3, traction-compression (Btc) 4. tractiontraction (Btt). We will examine the corresponding
possible concrete failures.
4.1 Compression - compression (Bcc)
The uniform vertical compression z is applied on
the concrete perpendicular to the horizontal compression y (Fig.5).

(8)

cz

For normal strength concrete (NC), the uniform

compression strength f rcc .un is lower than the standard cylindrical compression strength and we can
write

f rcc .un = f rcc

(9)

where depends on the used restraint at the specimen boundary (Van Mier, 1997) and can be assumed
equal to 0,850,90. Further, the tensile concrete
strength f rtc can be evaluated in terms of the cylindrical compression strength f rcc with:

f rtc = f rcc

(10)

f rtc = f rcc .un /

(11)

(14)

large distance from the pore in the hardened cement

paste are:

z = cz K 2 y = c y K 2 x = 0

(14)

to which correspond the local stresses:

P = [ A( )cz C ( )c y ]K 2
z

(12)

= [k B B( )cz A( )c y ]K 2
Py

Taking into account the range of values of the

We can also obtain the local tensile strength of

the hardened cement paste, in fact
= k B B ( ) K f

c
2 rt

(13)

and we get, by assuming for the value of the local

Poisson ratio of the hardened cement paste = 0,25,
f rtp,loc 3, 75 f rtc

y = c y z = cz

and the concentration factor kB is about 2,5.

If cz e and c y are compressions, then

P = [ A( )cz k B B ( )c y ]K 2

with f rcc .un / f rtc 8 9 and

A( )
kB =
B ( )

Figure 5. The compression compression regime

with cz c y . The asymptotic stresses acting at

where the constant is about 0,10 (Van Mier, 1997).

Thus we get

p
rt ,loc

cy

(13)

The local strength of the hardened cement paste,

i.e. the tensile strength of the binder not weakened
by the diffused porosity, is much higher than the
tensile concrete strength.

= [C ( )cz A( )c y ]K 2
Py

(15)

P = [k B B( )cz C ( )c y ]K 2
x

P = [C ( )cz k B B( )c y ]K 2
x

Taking into account the sign of the quantities

A(), B() e C(), the higher tensile stress is reached
Pz
by the
at the pole Pz of the pore. On the contrary, with cy dominating, the higher tensile stress is
P
produced by the y at the pole Py (Fig.6). The failure condition will be attained when firstly one of the
following conditions, linear in cy and cz is satisfied:

P

= [ A( )cz C ( )c y ]K 2 = f rtp,loc , cz c y
z

= [ A( )c y C ( )cz ]K 2 = f rtp,loc , c y cz
Py

(16)

cz

(16)
ty

or:
cz +

C ( )
c y = f rcc ,un
A( )

C ( )
cz = f rc ,un
A( )

cy +

(17)

Figure 7. The (Bct) and the (Btc) regimes

At the poles of the cavities diffused in the hardened cement paste we have the stresses

according to cz cy or cy cz.

P = K 2 [ A( )cz + k B B ( )t y ]
z

P = K 2 [ A( )cz + C ( )t y ]
z

yP

z
y

Px

= K 2 [ k B B( )cz + A( )t y ]
Py

Pz
Ey

Pz

Py

Px

Ez

P = K 2 [ k B B( )cz + C ( )t y ]
x

Ex
yP

P = K 2 [ C ( )cz + k B B( )t y ]
x

Figure 6. Local stresses around the pores in the Bcc regime

In Figure 12 the half lines AK and BK are the

failure contours corresponding to eqs.(16) e (16).
As a rule, A()<0 e C()>0 when >0,20. Thus, the
biaxial concrete compression strength is larger than
the uniaxial strength. Particularly, when:
c y = cz = cbiass

(18)

we have the concrete strength under equi-biaxial

compressions
cbiass = f rcc ,un.biass

(19)

with:
f rcc ,un.biass =

(22)

Py

Px
Px
z

= K 2 [ C ( )cz + A( )t y ]

c
rc ,un

1 + C ( ) / A( )

(20)

Thus, with cz dominating and taking into account

the values of the coefficients A(), B() e C(), the
higher local tensile stress is reached at the pole Pz by
the component Pz .(Fig.8). Thus the failure condition is now:
A( )cz + k B B ( )t y = f rtp,loc / K 2

(23)

or
f rcc .un
ty
f rtc

cz = f rcc .un

(24)

represented in Figure 12 by the straight line BP.

Analogously, in the case of traction compression, we have
f rcc .un
tz
f rtc

c y = f rcc .un

(24)

For instance, with = 0,25, eq. (20) gives:

f rcc ,un.biass = 1,125 f rcc ,un

Py

z
y

Px

4.2 Compression-tension (Bct) e tensioncompression(Btc)

Pz
Ey

In this case, the uniform vertical compression z is

applied on the concrete perpendicular to the horizontal tensile stress y (Fig.7) and we have

z = cz y = t y

Pz
Py

Px
Pz

xP

Ez

Ex
yP

(21)

where cz and ty are the intensities of the applied

stresses.

Figure 8. Local stresses around the pores in the (Bct)/(Btc) regimes

The straight lines BP and AQ of Figure 12 represent the failure conditions (24) and (24) for the
compression- tension and tension- compression regimes. Thus, when a small lateral tensile component
is applied, the strength decreases significantly in the
compressive direction. When the two stresses with
opposite sign have the same intensity, as it occurs in
the case of pure shear (Fig.9),

z = or y = or

(25)

with tz dominating on tz, the higher tensile stress is

reached by the component Px and occurs at the
pole Px (Fig.11). The failure condition thus is
k B B ( )t z + C ( )t y = f rtp,loc / K 2

(30)

or,
C ( )t z + k B B ( )t y = f rtp,loc / K 2

(30)

the failure shear stress is:

or =

f rt ,un

(26)

1 + f rt ,un / f rc ,un

yP

z
y

Px

Pz
Ey

zP
Py

Px
zP

xP

Ez

Ex

Figure 9. The pure shear regime

Py

4.3 Traction - traction(Btt)

The mean stresses acting on the concrete are the tensile stresses (Fig.10)

Figure 11 Local stresses around the pores in the (Btt) regime

z = tz

Eqs. (30) and (30) , more clearly, become

y = ty

(27)

tz

ty

Consequently, the mean stresses acting on the

hardened cement paste are
(28)

P = [ A( )t z + k B B( )t y ]K 2
z

P = [ A( )t z + C ( )t y ]K 2
z

= [k B B( )t z + A( )t y ]K 2
Py

Py

P = [k B B( )t z + C ( )t y ]K 2
x

P = [C ( )t z + k B B ( )t y ]K 2
x

ty +

C ( )
t z = f rtc
k B B ( )

t z = t y = tbiass

(31)

(31)

(29)

(32)

Thus we get:
tbiass =

and the local stresses around the pores are

= [C ( )t z + A( )t y ]K 2

C ( )
t y = f rtc
k B B( )

It is relevant the case of the uniform equi-biaxial

tensile stress:

Figure 10. The tension tension regime

z = t z K 2 y = t y K 2

tz +

f rtc
C ( )
1+
k B B( )

(33)

Eqs. (31) e (31) respectively correspond to the

segments QW e PW of Figure 12.
Thus the equi-biaxial tensile strength has to be
lower than the uniaxial tensile strength. Each of the
failure conditions (31), (31), (24), (24), (17), concerning the different stress regimes of the concrete,
is linear in the components y, z and define convex
sub-regions in the y, z plane. The intersections of
all these convex regions define the convex safe region in the biaxial stress components y, z whose
boundary is the biaxial failure contour of the concrete (Fig.12).

REFERENCES

Figure 12. The obtained concrete biaxial failure contour

5 CONCLUSIONS
Figure 13 gives the biaxial failure contour for a concrete obtained experimentally by (Kupfer, 1973).
Other failure contours have similar behaviour (Van
Mier, 1997). It is immediate to recognize the essential good agreement of these test results with the
contours corresponding to the proposed failure
model of the concrete. In the compression compression regime, when the ratio between the two
principal stresses is about 0.5, the agreement is less
satisfying since in this case the test results seem to
be strongly influenced by the restraints used at the
specimen boundary, as shown in Figure 3.84 of (Van
Mier, 1997).

Figure 13. Failure surface of the concrete in biaxial stress conditions obtained by (Kupfer, 1973)

Van Mier, JGM. 1997. Fracture processes of concrete, CRC

Press, Boca Raton.
Jirsek M, Bazant ZP. 2001. Inelastic Analysis of structures,
John Wiley, New York.
McClintock FA, Argon AS. 1966. Mechanical Behaviour of
materials, Addison Wesley Publishing Company, inc Reading, Massachusetts.
Yang CC, Huang R. 1996. Double Inclusion Model for Approximate Elastic Moduli of Concrete Material, Cement
and Concrete Research, 26(1): 83-91.
Nemat-Nasser S, Hori M. 1999. Micromechanics: overall
properties of heterogeneous materials, Elsevier, Amsterdam.
Kupfer, H. 1973. Behaviour of concrete under multiaxial short
term loading, with emphasis on biaxial loading, Deutscher
Ausschuss fur Stahlbeton, Vol.254, Berlin (in German)