Engineering Fracture Mechanics Vol. 60, No. 1, pp.

95±107, 1998
# 1998 Elsevier Science Ltd. All rights reserved
Printed in Great Britain
PII: S0013-7944(97)00094-5 0013-7944/98 $19.00 + 0.00

MIXED MODE FRACTURE OF CEMENTITIOUS

BIMATERIAL INTERFACES; PART II: NUMERICAL
SIMULATION
JAN. CÏERVENKA$
CÏervenka Consulting, Prague, Czechoslovakia

J. M. CHANDRA KISHEN$
Indian Institute of Science, Bangalore, India

VICTOR E. SAOUMA%
Dept. of Civil Environmental and Architectural Engineering, University of Colorado, Boulder, CO
80309-0428, U.S.A.

AbstractÐOn the basis of experimental investigations carried out on large scale simulated rock/concrete
interfaces, fracture properties are extracted from a ®nite element analysis. Gc and GF are extracted from
linear elastic fracture mechanics (LEFM) and nonlinear fracture mechanics (NLFM), respectively.
It is shown that the fracture energy Gc for bimaterial interfaces remains constant as long as the
crack lies within the interface, and an accurate crack pro®le is obtained from the NLFM analysis. #
1998 Elsevier Science Ltd. All rights reserved

KeywordsÐinterface cracks, ®nite element analysis, concrete.

1. INTRODUCTION
IN A companion paper [1], a series of large scale experiments on simulated rock/concrete inter-
faces subjected to mixed mode loading was reported. Whereas important qualitative conclusions
were drawn from these tests, it was not possible to directly extract fracture properties without
use of an appropriate numerical simulation. This will be accomplished in the present paper in
which two models are investigated: a linear elastic one which accounts for the dissimilarity of
the materials across the interface, and a nonlinear elastic one which is based on an extension of
Hillerborg's model [2].

2. LINEAR ELASTIC FRACTURE MECHANICS ANALYSIS

2.1. Theory
The linear elastic fracture mechanics theory for cracks between two dissimilar materials is
fairly well understood. Hence, we merely summarize the established theory ®rst, and then pro-
ceed with the numerical simulation.
Assuming a linear elastic model, a theoretically in®nite stress exists at the tip of the crack,
and we neglect the formation of microcracks which would otherwise have relieved the stress
®eld. If we also account for the dissimilarity between the two materials adjacent to the crack,
then it is well established that an oscillatory type singularity [3] exists at the crack tip. This oscil-
latory singularity stems from the presence of a logarithmic term in the normal stress
equation [4]:
…K1 ‡ iK2 †rie
syy ‡ itxy ˆ p …1†
…2pr†

$Formerly at University of Colorado, Boulder, CO 80309-0428, U.S.A.

%Author to whom all correspondence should be addressed.

95
96 J. CÏERVENKA et al.
p
where i ˆ ÿ1, K1 and K2 are components of the complex stress intensity factor K = K1+iK2.$
The expressions for these bimaterial stress intensity factors were derived by Rice and Sih [5]:
s‰cos…e log 2a† ‡ 2e sin…e log 2a†Š ‡ ft‰sin…e log 2a† ÿ 2e cos…e log 2a†Šg p
K1 ˆ a, …2†
cosh pe

t‰cos…e log 2a† ‡ 2e sin…e log 2a†Š ÿ fs‰sin…e log 2a† ÿ 2e cos…e log 2a†Šg p
K2 ˆ a: …3†
cosh pe
In the above equations, e is the oscillation index given by
 
1 1ÿb
eˆ ln …4†
2p 1‡b
where b is one of Dunder's elastic mismatch parameters [6], which for plane strain is given by
m1 …1 ÿ 22 † ÿ m2 …1 ÿ 21 †
bˆ …5†
2‰m1 …1 ÿ 2 † ‡ m2 …1 ÿ 1 †Š
in which m,  and a are the shear modulus, Poisson ratio and crack length, respectively, and sub-
scripts 1 and 2 refer to the materials above and below the interface, respectively. We note that b
and e vanish when the materials above and below the interface are identical. When e $0, eq. (1)
shows that the stresses oscillate heavily as the crack tip is approached (r 4 0). Furthermore, the
relative proportion of interfacial normal and shear stresses varies slowly with distance from the
crack tip because of the factor rie. Thus K1 and K2 cannot be decoupled to represent the intensi-
ties of interfacial normal and shear stresses as in homogeneous fracture.
The crack ¯ank displacements for plane strain are given by [4, 7]
4 …1=E 1 ‡ 1=E 2 †…K1 ‡ iK2 † p ie
dy ‡ idx ˆ p rr …6†
2p …1 ‡ 2ie† cosh…pe†
where dy and dx are the opening and sliding displacements of two initially coincident points on
the crack surfaces behind the crack tip. It may be noted from eq. (6) that crack face interpene-
tration is implied when e$ 0, as e is an oscillatory term as shown in eq. (4). The zone of con-
tact, however, is generally exceedingly small compared to the crack tip plastic zone and may
therefore be neglected[7].
The energy release rate, G, for extension of the crack along the interface for plane strain is
given by [4]
…1=E 1 ‡ 1=E 2 †…K 21 ‡ K 22 †
Gˆ : …7†
2 cosh2 …pe†

The complications associated with nonzero e and b may be circumvented by the approach
to bimaterial interface crack problems proposed by He and Hutchinson [8]. They have proposed
to systematically take b = 0 in the analysis of fracture specimens and subsequent application of
experimental data in failure predictions. Usually e is very small and the e€ect of nonzero b is of
secondary importance.
In this investigation, a variation of Stern's contour integral [9], in which the oscillatory
singularity is accounted for, was adopted. Hence, given the stress intensity factors K1 and K2
from a ®nite element analysis, the fracture energy is then determined from eq. (7).

2.2. Analysis
Initially, the specimen is numerically calibrated to obtain the e€ective modulus of elasticity.
The e€ective modulus of elasticity is determined using the relation[10]
C 0n
Eeff ˆ …8†
C 0exp

$Note that for interface cracks it is customary to replace KI and KII by K1 and K2.
 Fracture of bimaterial interfaces; Part II 97

where C 0n is the initial numerical compliance of the notched specimen without crack and C 0exp is
the initial compliance of the experimental splitting load vs crack mouth opening displacement
(CMOD) curve. The criterion used for crack initiation is K1c=0. The horizontal load that is
used in the experiment is applied and kept constant throughout the analysis. The analysis then
proceeded by applying the vertical load (within the experimental range) and incrementally
increasing the crack length along the interface in such a manner that the numerically obtained
CMOD matched with the experimental one. When the CMOD and the vertical load matched
the experimental one, then the mode I and II stress intensity factors were extracted. The Stern
contour integral method [9] was used to extract the bimaterial stress intensity factors. The energy
release rate in turn was computed from eq. (7).
The results of these analyses are presented in the form of plots showing the variation of
stress intensity factors and the energy release rate with crack length, Fig. 1 [11]. Each of the
curves was characterized by three distinct parts: (1) pre-peak; (2) post-peak; and (3) kinked
crack. The pre-peak region of the curve shows the crack length and corresponding SIFs and
energy release rate within the interface before the peak load is reached. This gradual increase
of the computed SIF and G is equivalent to the formation of a fracture process zone which
in this case is approximately equal to 0.08 m. The post-peak solid line indicates the crack
length within the interface after the peak vertical load is reached. This indicates the portion
when steady crack propagation took place within the interface since the energy release rate is
nearly constant. After this constant region, there is an abrupt change in the slope of the
energy release curve. This abrupt change was interpreted as implying that kinking into the
adjacent material should have taken place since such a curved path requires less energy than
that for a colinear interface propagation. Hence, this part of the curve is di€erently labeled.
As seen from this ®gure, the stress intensity factor K1 increases until the peak vertical load is
reached and thereafter decreases, but K2 keeps increasing. The energy release, G, thus
obtained for steady crack propagation within the interface is about 70 N/m. The crack kinks
into the ``rock'' after travelling a distance of 0.3 m from the notch tip as compared to 0.37 m
in the experiment.
Figure 2 shows the results of Test M5. In this test, greased steel plates were introduced in
the ``rock'' side. We observe that in this case the crack remained within the interface only for a
very short distance, and then it branched at the ®rst plate as was experimentally observed. The
energy required for crack propagation in the interface is consistent with the previous test and is
about 74 N/m. Table 1 illustrates the results of other tests.

Fig. 1. Test M1: variation of stress intensity factors and energy release rate with crack length.
98 J. CÏERVENKA et al.

Fig. 2. Test M5: variation of stress intensity factors and energy release rate with crack length.

2.3. Observations
The following observations can be made from the results of the linear analyses.
1. From the stress intensity factors and energy release rate vs crack length plots we observe that
cracking starts well before the maximum vertical load is reached. This phenomenon is mainly
responsible for the nonlinear behavior in the ascending branch of the vertical load vs crack
mouth opening displacement curve of the pre-peak region.
2. The average energy release rate or the energy required for crack propagation within the inter-
face is found to be 51 N/m as summarized in Table 1.
3. When the crack propagates at a steady rate within the interface, the energy release rate
remains almost constant and this constant is achieved only after reaching the peak vertical
load as seen in the fracture energy vs crack length plots. This implies that steady crack
propagation takes place only after the peak load is reached.
4. Since the energy release rate remains constant for crack propagation within the interface, the
energy criterion seems indeed to be a more appropriate criterion for crack propagation in dis-
similar materials, whereas a stress intensity factor based one may not be reliable. Hence, for
a crack between two cementitius materials, the fracture energy can be used as a material
property.
5. The ascending part of the energy release rate G is an indication of early formation of loca-
lized damage and its full understanding can only be achieved in the quanti®cation of the frac-
ture process zone size through a nonlinear fracture mechanics analysis.

Table 1. Fracture toughness and fracture energy for crack propagation along bimaterial interface
from linear elastic analysis
p p p p
K1c ‰MPa m …psi in†Š K2c ‰MPa m …psi in†Š Gc [N/m (lbf/in)]

Test M1 0.638 (566.45) 0.622 (552.37) 70 (0.393)

Test M2 0.682 (605.58) 0.460 (408.52) 63 (0.354)
Test M3 0.806 (715.37) 0.435 (386.31) 38 (0.210)
Test M4 1.394 (1236.72) 0.802 (711.87) 58 (0.320)
1.233 (1094.33) 0.445 (394.73) 74 (0.416)
Mean value 0.948 (841.69) 0.553 (490.76) 51 (0.279)

Standard deviation 0.299 (265.4) 0.142 (126.16) 16.13 (0.088)

 Fracture of bimaterial interfaces; Part II 99

Fig. 3. Interface idealization and notations.

3. NONLINEAR FRACTURE MECHANICS

3.1. Model development

Contrary to the LEFM case, a nonlinear fracture mechanics model for a discrete crack
between two homogeneous cementitious materials is not yet available. Hence, we ®rst extend
Hillerborg's model [2] to account for mixed mode loading, and discuss its implementation within
the context of a discrete crack model in a ®nite element code [12].
In the present model, the rock±concrete contact is idealized as an interface between two dis-
similar materials with zero thickness. Thus, the objective is to de®ne relationships between nor-
mal and tangential stresses with opening and sliding displacements. The notation used in the
interface model is illustrated in Fig. 3. The major premises upon which the model is developed
are as follows.
1. Shear strength depends on the normal stress.
2. Softening is present in both shear and tension.
3. There is a residual shear strength due to the friction along the interface, which depends on
the compressive normal stress.
4. Reduction in strength, i.e. softening, is caused by crack formation.
5. There is a zero normal and shear sti€ness when the interface is totally destroyed.
6. Under compressive normal stresses neither the shear nor the normal sti€nesses decrease to
zero. In addition, should a compressive stress be introduced in the normal direction following
a full crack opening, two faces of the interface come into contact, and both tangential and
normal sti€nesses become nonzero.
7. Irreversible relative displacements are caused by broken segments of the interface material
and by friction between the two crack surfaces.
8. Roughness of the interface causes opening displacements (i.e. dilatancy) when subjected to
sliding displacements.
9. The dilatancy vanishes with increasing sliding or opening displacements.
Figure 4 illustrates the probable character of the fracturing process along an interface.
In the proposed model the strength of an interface is described by a failure function:
F ˆ …t21 ‡ t22 † ÿ 2c tan…ff †…st ÿ s† ÿ tan2 …ff †…s2 ÿ s2t † ˆ 0 …9†
where:
. c is the cohesion;
. ff is the angle of friction;
. st is the tensile strength of the interface;
. t1 and t2 are the two tangential components of the interface traction vector;
. s is the normal traction component.
100 J. CÏERVENKA et al.

Fig. 4. Interface fracture.

The shape of the failure function in the two-dimensional case is shown in Fig. 5, and corre-
sponds to the failure criterion ®rst proposed by Carol et al. [13]. The general three-dimensional
failure function is obtained by mere rotation around the s-axis.
The evolution of the failure function is based on a softening parameter uie€ which is the
norm of the inelastic displacement vector ui. The inelastic displacement vector is obtained by de-
composition of the displacement vector u into an elastic part ue and an inelastic part ui. The
inelastic part can subsequently be decomposed into plastic (i.e. irreversible) displacements up
and fracturing displacements uf. The plastic displacements are assumed to be caused by friction
between crack surfaces and the fracturing displacements by the formation of microcracks:
F ˆ F…c, st , ff †, c ˆ c…uieff †, st ˆ st …uieff †

u ˆ u e ‡ u i , ui ˆ up ‡ uf

i2 1=2
uieff ˆ kui k ˆ …ui2 i2
x ‡ uy ‡ uz † : …10†

In this work both linear and bilinear relationships are used for c(uie€) and st(uie€):
  9
uieff >
ieff
c…u † ˆ c0 1 ÿ 8u <wc >
ieff >
>
wc >
>
=
ieff
c…u † ˆ 0 8u rwc linear for cohesion
ieff
>
>
>
>
2G IIa
F >
>
wc ˆ ;
c0

s1c ÿ c0 9
c…uieff † ˆ c0 ‡ uieff 8uieff <w1c >
>
w1c >
>
>
>
 ieff
 >
>
u ÿ w1c >
ieff
c…u † ˆ sc 1 ÿ 8uieff 2 hw1c , wc i >
=
wc ÿ w1c bilinear for cohesion …11†
>
>
c…u ieff
†ˆ0 8u ieff
> wc >
>
>
>
>
>
2G IIa ÿ …s1c ‡ c0 †w1c >
>
wc ˆ F ;
s1c
 Fracture of bimaterial interfaces; Part II 101

Fig. 5. Failure function.

  9
uieff >
>
st …uieff † ˆ st0 1 ÿ 8uieff <ws >
>
ws >
>
=
st …uieff † ˆ 0 8u rwst linear for tensile strength
ieff
>
>
>
>
2G IF >
>
ws ˆ ;
st0

s1s ÿ st0 9
st …uieff † ˆ st0 ‡ uieff 8uieff <w1s >
>
w1s >
>
>
>
 ieff
 >
>
u ÿ w1s >
st …uieff † ˆ s1s 1 ÿ 8uieff 2 hw1s , ws i >
=
wst ÿ w1s bilinear for tensile strength …12†
>
>
st …uieff † ˆ 0 8uieff > ws >
>
>
>
>
>
2G IF ÿ …s1s ‡ st0 †w1s >
>
ws ˆ ;
s1s
where G IF and G IIa
F are mode I and II fracture energies. s1c, w1c and s1s, w1s are the coordinates
of the breakpoint in the bilinear softening laws for cohesion and tensile strength, respectively.
The critical opening and sliding corresponding to zero cohesion and tensile strength are denoted
by ws and wc, respectively, and they are determined from the condition that the area under the
linear or bilinear softening law must be equal to G IF and G IIa
F , respectively. The signi®cance of
these symbols can best be explained through Fig. 6. It should be noted that G IIa F is not the pure
mode II fracture energy (i.e. the area under a t±ux curve), but rather is the energy dissipated
during a shear test with high con®ning normal stress. This parameter was ®rst introduced by
Carol et al. [13] in their microplane model. This representation seems to be more favorable to
the pure mode II fracture energy G II II
F . The determination of G F would require a pure shear test
without con®nement, which is extremely dicult to perform. Alternatively, a G IIa F test requires a
large normal con®nement, and is therefore easier to accomplish. Furthermore, if G II F is used, the
whole shear±compression region of the interface model would be an extrapolation from the
observed behavior, whereas the second approach represents an interpolation between the upper
bound G IIa
F and the lower bound G F .
I

The residual shear strength is obtained from the failure function by setting both c and st
equal to 0, which corresponds to the ®nal shape of the failure function in Fig. 5 and is given by:

Fig. 6. Bilinear softening laws.

102 J. CÏERVENKA et al.

Fig. 7. Sti€ness degradation in the equivalent uniaxial case.

t21 ‡ t22 ˆ tan2 …ff †s2 : …13†

Sti€ness degradation is modeled through a damage parameter, D $ h0, 1i, which is a relative
measure of the fractured surface. Thus, D is related to the secant of the normal sti€ness Kns in
the uniaxial case:
Af Kns
Dˆ ˆ1ÿ …14†
A0 Kn0
where Kn0 is the initial normal sti€ness of the interface; A0 and Af are the total interface area
and the fractured area, respectively. It is assumed that the damage parameter D can be deter-
mined by converting the mixed mode problem into an equivalent uniaxial one (Fig. 7). In the
equivalent uniaxial problem the normal inelastic displacement is set equal to uie€. Then, the
secant normal sti€ness can be determined from:
s st …uieff † st …uieff †
Kns ˆ p
ˆ e p f p
ˆ …15†
uÿu u ‡u ‡u ÿu st …u †=Kn0 ‡ …1 ÿ g†uieff
ieff

where g is the ratio of irreversible inelastic normal displacement to the total value of inelastic
displacement. Experimentally, g can be determined from a pure mode I test through:
up
gˆ …16†
ui
where up is the residual displacement after unloading and ui is the inelastic displacement before
unloading (Fig. 7). For concrete, g is usually assumed equal to 0.2 (Dahlblom and Ottosen [14])
or 0.3 (Alvaredo and Wittman [15]). Then, the evolution of the damage parameter D is de®ned
by the formula:
st …uieff †
Dˆ1ÿ …17†
st …uieff † ‡ …1 ÿ g†uieff Kn0
which is obtained by substituting eq. (15) into eq. (14).
The stress±displacement relationship of the interface is expressed as:
s ˆ aE…u ÿ up †; …18†
where: (a) s is the vector of tangential and normal stress at the interface:
s ˆ ft1 , t2 , sgT ; …19†

(b) a is the integrity parameter de®ning the relative active area of the interface, and is re-
lated to the damage parameter D:
jsj ‡ s
aˆ1ÿ D: …20†
2jsj
It should be noted that a can be di€erent from one only if the normal stress s is positive (i.e.
the interface is in tension). In other words, the damage parameter D is activated only if the
interface is in tension. In compression, the crack is assumed to be closed, and there is full con-
 Fracture of bimaterial interfaces; Part II 103

tact between the two crack surfaces. The activation of D is controlled through the fraction
…jsj ‡ s†=2jsj, which is equal to one if s is positive, and is zero otherwise.
(c) E is the elastic sti€ness matrix of the interface:
2 3
Kt0 0 0
E ˆ 4 0 Kt0 0 5: …21†
0 0 Kn0

It should be noted that the o€-diagonal terms in the elastic sti€ness matrix E of the interface
are all equal to zero, which implies that no dilatancy is considered in the elastic range. The dila-
tancy is introduced later after the failure limit has been reached through the iterative solution
process. The dilatancy of the interface is given by dilatancy angle fd, which is again assumed to
be a function of uie€. In the proposed model, a linear relationship is assumed:
 
uieff
fd …uieff † ˆ fd0 1 ÿ 8uieff Rudil
udil

fd …uieff † ˆ 08uieff > udil …22†

where udil is the critical relative displacement after which the interface does not exhibit the dila-
tancy e€ect any more, and fd0 is the initial value of the dilatancy angle.

3.2. Analysis
Based on the preceding model which was implemented in a special purpose ®nite element
code [16], nonlinear analysis on the bimaterial specimens is carried out. The numerical values for
the normal sti€ness (Kn) and tangential sti€ness (Kt) are computed from known values of the
elastic modulus and Poisson ratio of the two materials on either side of the interface, using the
following relations:
Eeff
Kn ˆ …23†
t

Kn
Kt ˆ …24†
2…1 ‡ eff †
where Ee€=0.5(E1+E2), e€=0.5(1+2), E1, E2, 1 and 2 are the elastic modulus and Poisson
ratio of the two materials on either side of the interface, respectively; t denotes the thickness of
the interface, which is assumed to be around 0.1 units in the analyses performed. It should be
noted that the ratio Eeff =Kn 10:1 corresponds to the assumption of interface thickness being
equal to 0.1 times a unit length of the problem, which is quite reasonable. In some analyses, the

Fig. 8. Boundary conditions and ®nite element discretization.

104 J. CÏERVENKA et al.

thickness of the interface and hence the normal and tangential sti€ness are determined by data
®tting in the elastic region.
Hence, from the nonlinear analyses on the bimaterial specimens we seek to determine the
material parameters for rock/concrete interfaces on the basis of experimental ones.
The ®nite element discretization and the boundary conditions are shown in Fig. 8. Three-
noded triangular plane stress elements are used for simulated rock, concrete, top backing plate
and bottom backing plate. Four-noded interface elements are provided along the bimaterial
interface. The backing plates used for the front and rear side of the specimen in the experiment
are substituted by linear truss elements in the numerical analysis. The con®ning compressive
forces or the horizontal load are applied as tractions on the side steel plates and this is kept con-
stant throughout the analysis.
The analysis is carried out by incrementally increasing the displacement at the top of the
specimen (Fig. 8). Displacement increments of 0.05 in are imposed during the ®rst four or ®ve
increments. Thereafter, smaller increment sizes ranging from 0.01 to 0.001 in are used until the
end of the analysis.
Cracking is assumed to begin when the maximum principal stress reaches the tensile
strength of the interface. The direction of crack propagation is assumed to be along the direc-
tion of the minimum principal stress. Whenever the maximum principal stress reaches the tensile
strength at the crack front, the crack is extended by a small amount. Interface elements are pro-
vided along the new crack and the analysis continued until ®nal failure, i.e. when the crack
extends to the full length of the specimen.
The fracture parameters in the numerical analyses using the interface crack model are
shown in Tables 2 and 3 for interface and kinked cracks (kinked into ``rock''), respectively.
Since a linear softening law for tension and cohesion was assumed for all the analyses, the
values of parameters s1, sw1, c1 and cw1 are all considered to be zero. The values for the angle

Table 2. Fracture parameters for the interface crack

Test M1 M2 M3 M4 M5
7
Normal sti€ness (10 psi) 4.45 4.45 3.42 3.68 9.83
Tangential sti€ness (107 psi) 1.89 1.89 1.45 1.56 4.16
Tensile strength (psi) 120 120 120 225 250
Cohesion (psi) 220 320 320 420 450
Friction angle (deg) 60 51 55 60 60
Dilatancy angle (deg) 45 45 45 40 45
GIF (lb/in) 0.35 0.34 0.35 0.45 0.35
GII
F (lb/in) 1.05 1.05 1.05 1.05 1.05
g 0.15 0.15 0.15 0.15 0.15
uDmax (in) 0.50 0.50 0.50 0.50 0.50
s1 (psi) 0.00 0.00 0.00 0.00 0.00
sw1 (in) 0.00 0.00 0.00 0.00 0.00
c1 (psi) 0.00 0.00 0.00 0.00 0.00
cw1 (in) 0.00 0.00 0.00 0.00 0.00

Table 3. Fracture parameters for kinked crack

Test M1 M2 M3 M4 M5
7
Normal sti€ness (10 psi) 5.55 5.55 3.70 3.85 3.71
Tangential sti€ness (107 psi) 2.35 2.35 1.57 1.63 1.57
Tensile strength (psi) 120 120 120 120 120
Cohesion (psi) 220 220 220 220 220
Friction angle (deg) 60 51 55 60 60
Dilatancy angle (deg) 45 45 45 40 45
GIF (lb/in) 0.35 0.34 0.35 0.45 0.35
GII
F (lb/in) 1.05 1.05 1.05 1.05 1.05
g 0.15 0.15 0.15 0.15 0.15
uDmax (in) 0.50 0.50 0.50 0.50 0.50
s1 (psi) 0.00 0.00 0.00 0.00 0.00
sw1 (in) 0.00 0.00 0.00 0.00 0.00
c1 (psi) 0.00 0.00 0.00 0.00 0.00
cw1 (in) 0.00 0.00 0.00 0.00
 Fracture of bimaterial interfaces; Part II 105

Fig. 9. Test M4: vertical load vs CMOD and crack pattern.

of friction, cohesion and the fracture energies in mode I and II are the ones obtained from the
experiments [1].
The normal and tangential sti€ness are obtained using eqs (23) and (24). These equations
hold good for both the interface crack and the crack kinked into ``rock''.
The results of the nonlinear analysis are shown in the form of vertical load vs CMOD
plots. Figure 9 shows the results of Test M4. As seen from this ®gure, a very good match is
obtained between numerical and experimental curves. The sti€ness and the maximum load are
very well captured. The crack pattern also shows an excellent match.
In Fig. 10, the results of Test M5 are shown. As in the linear analysis, the greased plates
used in the experiment to simulate bedding planes in the rock are modeled by the use of truss el-
ements. The initial sti€ness and the post-peak behavior are well captured. The peak load is
slightly underestimated. This might be caused by the complications involved in the modeling of
the simulated bedding planes for the ``rock''.

3.3. Observations
The results of the numerical simulation call for the following comments.
1. The interface crack model has good capabilities of modeling the interface between two dis-
similar materials. From the results of the analyses, the initial sti€ness, peak load and the
post-peak behavior are well captured in comparison to the experimental ones.
2. In the analyses, cracking took place when the maximum principal stress reached the tensile
strength. The direction of crack propagation was perpendicular to the maximum principal
stress. Using these criteria produced a good match between the experimental and numerical
crack patterns. This implies that fracture in low tensile strength materials seems to be a mode

Fig. 10. Test M5: vertical load vs CMOD and crack pattern.
106 J. CÏERVENKA et al.

Table 4. Summary of fracture parameters

Material type Experimental Numerical
GIF (N/m) GIIa
F (N/m) p LEFM p NLFM
K1c …MPa m† K2c …MPa m† Gc (N/m) GF (N/m)
Intact concrete[18] 182 1.04
Intact limestone[10] 37.4 0.80
Concrete±limestone 38.3 0.28 0.01 2.87 40
Mixed mode test on simulated 187 0.948 0.553 51 61
rock±concrete
Lee et al.[19]
low strength mortar±granite 24±56
high strength mortar±granite 62±97
Yan and Mindess[20]
0.5% steel ®bre-conc. 54±65
1.0% steel ®bre-conc. 71±86
0.5% polyprop-conc. 40±45
1.0% polyprop-conc. 40±45
Stankowski[21]
GIF
GIIa
ˆ 0:1
F

I mechanism. The e€ect of shear is to rotate the direction of principal stresses. This phenom-
enon is also observed for fracture in concrete [17].
3. The in¯uence of boundary conditions on the crack pattern is clearly observed in the numeri-
cal analysis, very similar to the experimental one. This implies that proper boundary con-
ditions must be considered in the numerical analysis, to correctly capture the cracking
patterns.
4. A summary of the fracture energy values obtained in the present work through di€erent
methods and by other investigators is presented in Table 4. As is seen, the fracture energy
values for interfaces fall within a reasonable range.

4. CONCLUSIONS
In this paper, the linear and nonlinear fracture mechanics based analysis performed on
bimaterial specimens are described in detail.
In the linear elastic fracture mechanics analysis, based on existing models, it is shown that
the fracture energy obtained for bimaterial interface remains constant as long as the crack lies
within the interface. Hence the energy based criteria seems to be more favorable for crack
propagation within bimaterial interfaces.
In the nonlinear fracture mechanics analysis, a new model is ®rst developed. Then, when
applied to simulate the experimental results reported in Ref. [1], the complete crack pro®le was
adequately replicated as well as load displacement curves.
The nonlinear analysis in turn yielded a critical energy release rate GF equal to 61 N/m,
whereas the linear elastic one yielded a value of Gc equal to 51 N/m. Given that those values
were obtained from radically di€erent models, and given the complexity of the problem, those
values can be considered quite representative.

AcknowledgementsÐThis research has been funded by the Electric Power Research Institute, Palo Alto, through
Research Contract No. RP-2917-08. The support of its program manager, Mr. Doug Morris, is gratefully acknowledged.

REFERENCES
1. Slowik, V., Chandra Kishen, J. M. and Saouma, V., Mixed mode fracture of cemetitious bimaterial interfaces; Part
I: experimental results. Engineering Fracture Mechanics, 1998 60.
2. Hillerborg, A., ModeÂer, M. and Petersson, P. E., Analysis of crack formation and crack growth in concrete by
means of fracture mechanics and ®nite elements. Cement and Concrete Research, 1976, 6(6), 773±782.
3. Williams, M. L., The stresses around a fault or crack in dissimilar media. Bulletin of the Seismological Society of
America, 1959, 49(2), 199±204.
4. Carlsson, L. A. and Prasad, S., Interfacial fracture of sandwich beams. Engineering Fracture Mechanics, 1993, 44,
581±590.
 Fracture of bimaterial interfaces; Part II 107

5. Rice, J. R. and Sih, G. C., Plane problems of cracks in dissimilar media. Journal of Applied Mechanics, 1965, 418±
423.
6. Dunders, J., Edge-bonded dissimilar orthogonal elastic wedges under normal and shear loading. Journal of Applied
Mechanics, 1969, 36, 650±652.
7. Hutchinson, J. W. and Suo, Z., Mixed mode cracking in layered materials. Advances in Applied Mechanics, 1992, 29,
63±191.
8. He, M. Y. and Hutchinson, J. W., Kinking of a crack out of an interface. Journal of Applied Mechanics, 1989, 56,
270±278.
9. Hong, C.-C. and Stern, M., The computation of stress intensity factors in dissimilar materials. Journal of Elasticity,
1978, 8(1), 21±34.
10. BruÈhwiler, E. and Saouma, V. E., Fracture testing of rock by the wedge splitting test. In Proc. 31st U.S. Symposium
on Rock Mechanics, Golden, CO, June 1990. Balkema, pp. 287±294.
11. Chandra Kishen, J. M., Interface cracks: fracture mechanics studies leading towards safety assessment of dams.
Ph.D. Thesis, University of Colorado, Boulder, CO, June 1996.
12. CÏervenka, J., Discrete crack modeling in concrete structures. Ph.D. Thesis, University of Colorado, Boulder, CO,
December 1994.
13. Carol, I., BazÏant, Z. P. and Prat, P. C., Microplane type constitutive models for distributed damage and localized
cracking in concrete structures. In Proc. Fracture Mechanics of Concrete Structures, Breckenridge, CO, June 1992.
Elsevier, pp. 299±304.
14. Dahlblom, O. and Ottosen, N. S., Smeared crack analysis using a generalized ®ctitious crack model. Journal of
Engineering Mechanics, 1990, 116(1), 55±76.
15. Alvaredo, A. M. and Wittman, F. H., Crack formation due to hygral gradients. In Third Int. Workshop on
Behaviour of Concrete Elements Under Thermal and Hygral Gradients, Weimar, Germany, 1992.
16. Reich, R., CÏervenka, J. and Saouma, V. E., Merlin, a three-dimensional ®nite element program based on a mixed-
iterative solution strategy for problems in elasticity, plasticity, and linear and nonlinear fracture mechanics.
Technical report, EPRI, Palo Alto, 1997, http://civil.colorado.edu/ 0saouma/Merlin.
17. Schlangen, E., Experimental and numerical analysis of fracture processes in concrete. Ph.D. Thesis, Delft University
of Technology, 1993.
18. Slowik, V. and Wittmann, F. H., In¯uence of strain gradient on fracture energy. In Proc. First Int. Conf. on
Fracture Mechanics for Concrete and Concrete Structures, Breckenridge, CO, ed. Z. BazÏant, June 1992. Elsevier, pp.
424±429.
19. Lee, K. M., Buyukozturk, O. and Kitsutaka, Y., The role of interfacial fracture toughness in cracking behavior of
high-strength concrete. In Interface Fracture and Bond, ACI Special Publication, 1995, SP-156, pp. 69±84.
20. Yan, C. and Mindess, S., Fracture mechanics analysis of bond behavior under dynamic loading. In Interface
Fracture and Bond, ACI Special Publication, 1995, SP-156, pp. 107±123.
21. Stankowski, T., Numerical simulation of progressive failure in particle composites. Ph.D. Thesis, University of
Colorado, Boulder, CO, 1990.

(Received 22 November 1996, in final form 5 July 1997, accepted 8 July 1997)