Journal of the European Ceramic Society 38 (2018) 2994–3003

Contents lists available at ScienceDirect

Journal of the European Ceramic Society

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

Original Article

Modeling complex crack paths in ceramic laminates: A novel variational T

framework combining the phase field method of fracture and the cohesive
zone model
⁎
V. Carolloa, , J. Reinosob,a, M. Paggia
a
IMT School for Advanced Studies Lucca, Piazza San Francesco 19, 55100 Lucca, Italy
b
Elasticity and Strength of Materials Group, School of Engineering, Universidad de Sevilla, Camino de los Descubrimientos s/n, 41092 Seville, Spain

A R T I C L E I N F O A B S T R A C T

Keywords: The competition between crack penetration in the layers and cohesive delamination along interfaces is herein
Phase field model of fracture investigated in reference to laminate ceramics, with special attention to the occurrence of crack deflection and
Cohesive zone model crack branching. These phenomena are simulated according to a recent variational approach coupling the phase
Crack deflection field model for brittle fracture in the laminae and the cohesive zone model for quasi-brittle interfaces. It is shown
Crack branching
that the proposed variational approach is particularly suitable for the prediction of complex crack paths in-
Laminates
volving crack branching, crack deflection and cohesive delamination. The effect of different interface properties
on the predicted crack path tortuosity is investigated and the ability of the method to simulate fracture in layered
ceramics is proven in relation to experimental data taken from the literature.

1. Introduction competition between crack penetration in the layers and delamination

along the existing interfaces. Another approach to foster complex crack
Ceramic materials are largely used in technological applications, paths is to introduce porous layers between the ceramic ones [7].
especially with the aim of achieving a desired resistance to severe wear However, the drawback of porous materials is their low wear resistance.
or corrosive phenomena at high temperatures. However, the main Alternatively, tough interfaces with preexisting defects could be in-
drawback of ceramics regards their brittleness and, to increase their serted among the layers [8]. The tough interface can be made of a
toughness, laminates are often used alternating ceramic and metallic ceramic material which guarantees the resistance to wear and corro-
layers. For instance, in [1,2], Al/SiC and Al/TiN laminates have been sion, also at high temperatures. The defects in the tough interfaces
explored and tested. The metallic layers make the composite able to guarantee the development of crack deflection with a consequent
withstand higher deformations by means of the development of plas- toughening of the material.
ticity at several locations within the specimen, and therefore increasing Considering the technological strategies herein described, the in-
the overall toughness of the laminate. The same toughening process has troduction of an interface with tailored properties clearly emerges as a
been achieved by alternating ceramic layers with polymeric layers in strategy to create a complex crack pattern and consequently enhance
[3]. The main drawback of these solutions is that metals and polymers the apparent material toughness. In this article, we investigate these
loose their mechanical properties at high temperatures and have a low possibilities by examining and modelling the interaction between crack
wear resistance. A possible way to enhance the toughness of ceramics is penetration in the layers and cohesive delamination at the interfaces
to introduce quasi-brittle interfaces [4,5]. Then, a stack of ceramic between materials with different elastic and fracture properties. This
layers alternated by thin layers of a very brittle ceramic is a possible problem has been mathematically addressed in many publications using
effective technological solution. Such brittle layers act as a quasi-brittle different methodologies, cf. [9–11]. Differing from previous investiga-
interfaces which make the crack path very complex, thus increasing the tions, this problem is herein analysed by means of the numerical
overall material toughness by acting on the crack tortuosity. This me- method published in [12], which has been recently extended to a 3D
chanism has been firstly theorized in the framework of linear elastic finite deformation framework in [13] and applied to fracture of ani-
fracture mechanics (LEFM) by the so called Cook–Gordon mechanism sotropic polycrystalline Silicon in [14]. The novelty of the current ap-
[6]. It is the result of crack branching and crack deflection typical of the proach relies upon the innovative variational framework combining the

⁎
Corresponding author.
E-mail address: [email protected] (V. Carollo).

https://doi.org/10.1016/j.jeurceramsoc.2018.01.035
Received 30 November 2017; Received in revised form 21 January 2018; Accepted 22 January 2018
Available online 02 February 2018
0955-2219/ © 2018 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/).
V. Carollo et al. Journal of the European Ceramic Society 38 (2018) 2994–3003

phase field method of fracture and the cohesive zone model. In parti- Cauchy stress tensor. The body forces are represented by the function
cular, the role played by the internal characteristic length scales of the fv: Ω → ℝndim . The composite is characterized by quasi-brittle inter-
two approaches is rigorously analysed in order to understand their ef- faces, Γi, and a crack in the layer represented as an internal dis-
fect on the resulting crack path and its tortuosity as a way to enhance continuity, Γb (see Fig. 1(a)). A generic point in the bulk of the body is
the overall composite toughness. denoted by the vector of its Cartesian coordinates x, while a generic
The manuscript is organized as follows. Section 2 summarizes the point on the interface Γi is denoted by the vector xc.
principal features of the variational framework herein employed. In The free energy functional which governs the mechanics of the body
Section 3, the numerical method is applied to predict the crack path in Ω is defined as [15,16]:
laminates. Special attention is devoted to examining the effect of tough
and quasi-brittle interfaces and reproducing also experimental results Π(u , Γ) = ΠΩ (u , Γ) + ΠΓ (Γ) = ∫Ω∖Γ ψe (ε) dΩ + ∫Γ Gc dΓ, (2)
related to a ceramic laminate taken from literature. Finally, the main
where ψ (ε) is the elastic energy density, ε is the strain field, and Gc is
e
conclusions of the current investigation are drawn in Section 4.
the fracture energy.
The main idea to couple the phase field approach for brittle fracture
2. Variational model
and the cohesive zone model is to split the fracture energy function Gc in
two parts. One part (Gcb ) describes fracture in the layers and it is
In this section, the coupled phase field and cohesive zone model modelled by the phase field approach. The second part (G i ) describes
formulation developed in [12] is resumed. We first present the funda- the cohesive fracture of the interfaces and it is modelled by the cohesive
mental hypothesis of the current coupling approach in Section 2.1. zone approach. Then, the free energy functional in Eq. (2) can be re-
Later, the phase field method for brittle fracture formulation and the written as:
cohesive zone model compatible with the phase field model are out-
lined in Sections 2.2 and 2.3, respectively. Finally, the corresponding Π(u , Γb, Γ)
i = ΠΩ + ΠΓb + ΠΓi = ∫Ω∖Γ ψe (ε) dΩ + ∫Γ b
Gcb (u , d) dΓ
finite element formulation within the infinitesimal deformation setting
is derived in Section 2.4. + ∫Γ G i (g,h, d) dΓ,
i (3)

2.1. Fundamental hypothesis where g denotes the vector of displacement discontinuities at the in-
terface, h is an history parameter as in [17] to avoid re-healing of the
The formulation herein presented is developed in the general material and the non uniqueness of the solution, and d is the phase field
Euclidean space of dimension ndim under infinitesimal deformation degradation variable which will be further detailed in the next section.
setting. Let us consider a body Ω ∈ ℝndim with a generic shape, where
the boundaries of the body are denoted by ∂Ω ∈ ℝndim − 1 (Fig. 1). Ki- 2.2. Phase field approach for brittle fracture
nematic and traction boundary conditions can be respectively pre-
scribed on the disjointed parts of the boundaries ∂Ωu and ∂Ωt (with The phase field approach for brittle fracture [15,18] is a variational
∂Ωt ∪ ∂Ωu = ∂Ω and ∂Ωt ∩ ∂Ωu =∅). Then, the prescribed displace- approach which considers a crack as a diffuse damage instead of a sharp
ments and tractions are denoted by: discontinuity (Fig. 1(a)). Within this framework, the potential energy of
the bulk is formulated as follows:
u = u on ∂Ωu and t = σ ·n on ∂Ωt , (1)
where n is the outward normal unit vector to the body, and σ is the
Πb (u , d) = ∫Ω ψ (ε, d) dΩ + ∫Ω Gcb γ (d, ∇x d) dΩ, (4)

Fig. 1. Schematic representation of an arbitrary body with a discontinuity in the domain and an interface: (a) Left: discrete discontinuity in the domain. Right: smeared discontinuity in
the domain based on the phase field concept. (b) Diffusive crack modeling solution for the one-dimensional crack problem.

2995
V. Carollo et al. Journal of the European Ceramic Society 38 (2018) 2994–3003

where ψ (ε, d) is the elastic energy stored in the bulk, the symbol ∇x• n g
⎧ kn g , if 0 <
gn
< 1; tg
⎧ kt g , if 0 <
gt
< 1;
nc gnc tc gtc
denotes the spatial gradient operator, and γ (d , ∇x d) is the so called σ= τ=
gn gt
crack density functional which reads [15]: ⎨ 0, if g ≥ 1, ⎨ 0, if g ≥ 1.
⎩ nc ⎩ tc (9)
1 l
γ (d , ∇x d) = d2 + |∇x d|2 where σ and τ are the Mode I and Mode II tractions, respectively, g is the
2l 2 (5)
relative displacement, and the subscript n and t refers to opening and
where l is the phase field internal length parameter governing the sliding, respectively. The stiffness of the cohesive relation, k, depends
sharpness of the crack according to the equation presented in Fig. 1(b) on damage d according to the formulae:
for the mono-dimensional case.
The elastic energy in Eq. (4) takes the following form: gnc,0 2 gtc,0 2

ψ (ε, d) = g (d) ψ+e (ε) + ψ−e (ε), (6a)

kn = kn,0 ( ), gnc
kt = kt ,0 ( ).
gtc (10)

where k0 and g0 are, respectively, the stiffness and critical relative

λ
ψ+e (ε) = ( tr[ε] + )2 + μtr[ε+2 ], displacements at d = 0 .
2 (6b)
In this formulation, among the different potential alternatives, we
λ have adopted the hypothesis of keeping constant the critical energy
ψ−e (ε) = ( tr[ε] − )2 + μtr[ε−2 ], release rate of the interface, Gci , for many reasons. Firstly, the model is
2 (6c)
formulated starting from the Griffith energy balance criterion, then in
where λ and μ are the Lamé constants, tr[•] denotes the trace operator,
Eq.(3) we set up a clear and explicit split between the dissipated energy
ϵ+ and ϵ− denote the positive and negative counterpart of the strain
due to the bulk fracture and to the interface delamination. Corre-
tensor, respectively, and g (d) is a degradation function: spondingly, Gci becomes the material parameter that governs the in-
g (d) = (1 − d)2 + K . (7) terface failure. This choice endows a clear characterization of the in-
terface fracture energy. In fact, among the four parameters entering the
In Eq. (6a), the elastic energy has been split in its positive and ne- cohesive law, Gci can be obtained in a straightforward manner from
gative counterparts according to the formulation in [19,20]. The posi- experimental tests [22].
tive counterpart of the elastic energy is produced by the tensile stresses Finally, the mixed mode failure criterion proposed in [23] is here-
while the negative counterpart is produced by compression. Positive with considered to trigger the interface failure:
and negative stresses can be computed by the derivative of the elastic
2 2
energy with respect to the strain tensor. Then, the introduction of the i i
⎛⎜ G I ⎞⎟ + ⎜⎛ GII ⎟⎞ = 1,
positive and negative split of the strain tensor leads to: G i
G i
⎝ IC ⎠ ⎝ IIC ⎠ (11)
∂ψˆ
σ: = = g (d) σ+ + σ−; with σ± = λ ( tr[ε] ± ) 1 + 2με±, where G Ii and GIIi are the dissipated fracture energies which take the
∂ε (8)
form:
Then, since the degradation function in Eq. (7) affects only the
2
gnc,0 2
gtc,0
positive counterpart, damage can only develop when the material is 1
G Ii (d) = 2 nt ,0 gn2 [(1 − d) g
1
GIIi (d) = 2 kt ,0 gt2 [(1 − d) g
2, 2.
under tension, avoiding damage growth in compression. nc,0 + dgnc,1 ] tc,0 + dgtc,1 ]

(12)
2.3. Cohesive zone model compatible with the phase field The critical fracture energies i
GIC and i
GIIC are:

In this section, the classical linear cohesive zone model with tension i
GIC 21
= 2 gnc,0 i
kn,0, GIIC 2
= 2 gtc,0 kt ,0.
1
(13)
cut-off [21] is particularized in order to take into account the effect of
the bulk damage d. First of all, the cohesive counterpart of the fracture
energy in Eq. (3) is decomposed in the sum of the Mode I and Mode II
fracture energies, GI and GII , respectively. Based on the formulation 2.4. Finite element formulation
outlined in [12], the critical crack opening displacement (gc) depends
on the bulk damage d according to the linear relation The finite element formulation of the previous fracture mechanics
gc (d) = (1 − d) gc,0 + dgc,1, where gc,0 = gc (d = 0) and gc,1 = gc (d = 1) . models is herein derived. First of all, the weak form of the free energy
Then, the cohesive traction vs. relative displacement laws for Mode I functional in Eq. (3) is deduced using the standard Galerkin procedure.
and Mode II take the form shown in Fig. 2, and are described by the Then, the variation of the bulk energy functional (Eq. (4)) with respect
following equations: to the displacements u and the phase field variable d takes the form:

Fig. 2. Schematic representation of the cohesive zone model coupled with the phase field variable for brittle fracture in the bulk. (a) Mode I CZM traction σ vs. gn. (b) Mode II CZM
traction τ vs. gt.

2996
V. Carollo et al. Journal of the European Ceramic Society 38 (2018) 2994–3003

δΠb (u , δu , d, δd)= ∫ σ : δε dΩ − ∫ 2(1 − d)δdψe (ε) dΩ+

Ω Ω +
1
∫Ω Gcb l ⎡⎣ l2 dδ d + ∇x d·∇x (δ d) ⎤⎦ dΩ
+ δΠb,ext (u , δu) ∀ δu , δd, (14)

where δ u ∈ Vu
= {δ u | u = u on ∂Ωu , u ∈ H1}
is the vector of the dis-
placement test function, and δ d ∈ Vd = {δ d | δ d = 0 on Γb, d ∈ H 0} is
the damage test function. The contribution of the external forces in the
variation of the bulk energy functional is defined as follow:

δ Πb,ext (u , δ u) = ∫∂Ω t·δ u d∂Ω + ∫Ω fv·δ u dΩ ∀ δ u, δ d. (15)

Finally, the variation of the interface energy functional ΠΓi in Eq. (3)
is defined as:
i i
δ ΠΓi (u , δ u , d , δ d) = ∫Γ ⎛ ∂G ∂(uu, d) δ u + ∂G ∂(ud , d) δ d⎞ dΓ
i
⎜ ⎟ ∀ δ u, δ d.
⎝ ⎠
(16)
The phase field model has been implemented within a 4-node iso-
parametric finite element. The cohesive zone model has been im-
plemented using a 4-node interface finite element. The detailed finite
element implementation in the software FEAP [24] and all the related
operators can be found in [12] and are omitted here for the sake of
brevity.

3. Simulation of complex crack paths in layered ceramics

The prediction of the crack path in ceramic laminates is a very

challenging issue. Various factors such as the elastic mismatch of the
constituent materials and the properties of the interfaces [23,25] make Fig. 3. Specimen geometry.
the crack propagating along tortuous paths which are very difficult to
be simulated using previous numerical methods. The present approach Table 1
resolves most of their drawbacks primarily associated to remeshing or Geometry and material/interface parameters.
corner-case problems.
Geometry parameters
L 6 mm Specimen length
3.1. Complex crack paths for laminates under tensile loading: analysis of ly 0.25 mm Layer thickness
failure patterns with quasi-brittle or tough interfaces h 0.005 mm Interface thickness

Mechanical parameters material 1

In this section we study the effect of the interface toughness on the E1 70,000 MPa Material 1 Young modulus
resulting crack path. The case of a single-edge notched bi-material la- v1 0.34 Material 1 Poisson ratio
G1 0.025 N/mm Material 1 fracture energy
minate under tension (Fig. 3) is herein considered. The laminate has
l1 0.0075 mm Material 1 phase field length scale parameter
been modelled using the phase field finite elements for the bulk and the
interface finite elements compatible with the phase field between each Mechanical parameters material 2
E2 300,000 MPa Material 2 Young modulus
layer. The materials which compose the laminate are: a soft material 1 v2 0.14 Material 2 Poisson ratio
with high fracture toughness; a stiff material 2 with low fracture G2 0.005 N/mm Material 2 fracture energy
toughness. The material parameters are collected in Table 1. l2 0.0075 mm Material 2 phase field length scale parameter
Three cases have been examined: (1) laminate with perfectly Mechanical parameters interface
bonded layers; (2) laminate with tough interfaces; (3) laminate with k0 2000 MPa/mm Initial stiffness of interface
brittle interfaces. Note that the toughness of the interfaces is always σc,0, τc,0 100 MPa Initial peak stress of the tough interface
Gci 2.5 N/mm Critical energy release rate for tough interface
larger than the toughness of the materials composing the laminae, to
simulate configurations consistent with technological solutions for σc,0, τc,0 1 MPa Initial peak stress of the brittle interface
Gci 0.025 N/mm Critical energy release rate for brittle interface
ceramics laminates.
σc,0/σc, τc,0/ 1 Ratio between the initial and final value of the
In the first simulation, no interface elements are introduced in order τc peak stress of the interface
to simulate fully bonded layers. The phase field internal length l is set
very small for both layers (see Table 1), to reproduce LEFM predictions
as discussed in [26]. The results of the simulations are shown in Fig. 4. specimen.
The first layer that shows crack nucleation is the second one (material In the second simulation, we introduce interface elements between
2). In such a layer, two parallel cracks are predicted to propagate si- each layer. The parameters used for the interface are
multaneously (Fig. 4(a)) and can be considered as two branches of the σc,0 = τc,0 = 100 MPa and k0 = 2000 MPa/mm to model a stiff quasi-
initial notch in the layer 1. After increasing the applied load, each crack brittle interface. As expected, the evolution of the predicted crack path
in the second layer further branches in the next brittle layer (Fig. 4(b)). is very different from that of the previous simulation. The first layer
The same process continues for the next brittle layer but only two where cracks nucleate is the second one (Fig. 5(a)) with the same
branches are now developed (Fig. 4(c)). At this stage, the cracks in the pattern as in the first simulation. Immediately after that, delamination
brittle layers start connecting through the material 1. Finally, failure of along the interface between the first and the second layer takes place
the specimen is achieved (Fig. 4(d)) and it is the result of a complex (Fig. 5(b)). Then, the first layer is cracked by the propagation from the
crack path mostly localized in the mid-span cross-section of the

2997
V. Carollo et al. Journal of the European Ceramic Society 38 (2018) 2994–3003

Fig. 5. Crack evolution in the simulation with tough interface (σc,0 = 100 MPa).
Fig. 4. Crack evolution in the simulation with fully bonded layers.

σc,0 = τc,0 = 1 MPa and k0 = 2000 MPa/mm. The evolution of the crack
notch (Fig. 5(c)). Continuing with the simulation, cracking proceeds in
path is again quite different from the previous cases (Fig. 6). Thus, in
the material 2 layers together with the development of delamination at
the current case, first, delamination is predicted to occur between the
interfaces (Fig. 5(d)). At failure, delamination makes the crack pattern
first and the second layer (Fig. 6(a)). Then, the crack starts propagating
distributed along the whole specimen, as a primary difference from the
from the notch until it impinges onto the delaminated interface
results of the first simulation. Another important aspect is that the
(Fig. 6(b)). Subsequently, branching is predicted to take place in the
majority of the material 1 layers are not cracked, apart from the first
second layer (Fig. 6(c)), and each branched crack is developed starting
layer containing the notch.
from the points where delamination was arrested. Furthermore, crack
In the third simulation we introduce a more brittle interface, setting

2998
V. Carollo et al. Journal of the European Ceramic Society 38 (2018) 2994–3003

softening. This is due to the gradual crack propagation within the

specimen. The curves corresponding to the second and third simula-
tions show a completely different crack pattern and the nonlinear ef-
fects are much more pronounced in the load–displacement curve due to
the occurrence of delamination. The development of cracking and de-
lamination events cause multiple drops in the resistant force.
The correspondence between crack/delamination events and the
drops in the force can be examined closely in Fig. 8. In these graphs, the
force–displacement curves are plotted together with other two quan-
tities, the total crack propagation length and the portion of crack pro-
pagation length in just the bulk. Both quantities are normalized with
respect to the layer thickness, ly. These plots show that, in general,
delamination and cracking occur simultaneously. This is in agreement
with the patterns in Figs. 5 and 6, where in some cases delamination
triggers crack formation in the layers or vice-versa. Another important
consideration is that the delamination length in the simulation with
quasi-brittle interfaces is bigger than the in simulation with tougher
interfaces.
Moreover, we compare the current predictions with the theoretical
results presented in [6], which complied with the so-called Cook–-
Gordon mechanism. According to that theory, the introduction of a
brittle interface should increase the apparent material strength. The
results reported in Fig. 7 seem to challenge this theory, since a brittle
interface reduces the apparent material strength as compared a tougher
one.
This contradiction can be explained by noting that the assumptions
of the Cook–Gordon model are not fully satisfied in the present com-
putational setting. One of the most relevant differences regards the fact
that in the Cook–Gordon model the material is considered homo-
geneous and linear elastic until the condition of fracture of the interface
is achieved. After delamination, discontinuities are considered along
the interface and, consequently, the interface starts interfering with the
linear elastic fracture mechanics crack tip stress distribution. In the
present model, on the other hand, interfaces are introduced as an ad-
ditional compliant material from the very beginning of the simulation.
In [12] it has been shown that the behaviour of such a system depends
on the ratio between the interface process zone size and the bulk energy
dissipation zone size (lCZM/l). Depending on this ratio, the apparent
strength of the overall material ranges between the strengths ruled by
the following limit models: a model with prefect bonded interfaces and
finite l; a model with elastic bulk material and finite lCZM. In our si-
mulations, the ratio lCZM/l is kept constant. In fact, the bulk properties
do not change (then l is constant), and lCZM is also constant since k is
fixed. This latter aspect is a consequence of dimensional analysis con-
siderations in [12] from which we deduced that
2
lCZM ∝ EGci/ σmax = E /(2k ) . The consequence of having a constant ratio
lCZM/l is that delamination is triggered earlier in the presence of a brittle
interface. Due to this, we have also longer delamination paths for a
brittle interface.
These conclusions lead to the second important difference with re-
spect to the Cook–Gordon model, where the load is supposed to be
applied as a remote tensile stress at infinity. The assumption of infinite
plane plays an important role, since there is no constrain on the size of
the delamination path. As a result, a brittle interface could lead to a
longer delamination with a consequent delayed crack propagation, in-
Fig. 6. Crack evolution in the simulation with brittle interface (σc,0 = 1 MPa). creasing the elastic energy release rate. The specimens that we simu-
lated are not long enough to be considered as infinite. Due to that, when
penetration triggers delamination along the second interface. Finally, the interface is brittle, delamination reaches the boundaries of the
the sample failure is the result of a sudden delamination along all the specimen causing the failure of the whole interface. This is also what
interfaces and cracks in the brittle layers (Fig. 6(d)). Again, the material accelerates the final failure of the specimen. As a result, for a brittle
1 layers have been preserved from cracking. interface, the apparent strength of the specimen results lower than for a
The force–displacement curves of the above three simulations are more ductile one.
shown in Fig. 7. All of them present an initial nonlinearity due to the Finally, there are other assumptions made in the Cook–Gordon
formation of damage in the bulk. After that, the curve of the first si- model that are not in line with the hypotheses of the present approach.
mulation starts loosing the load-bearing capacity slowly with a smooth For instance, Cook and Gordon neglected the effect of the vertical stress
σy in their interface fracture criterion. They made this simplification

2999
V. Carollo et al. Journal of the European Ceramic Society 38 (2018) 2994–3003

Fig. 7. Force–displacement curve of the three simulations.

Fig. 8. Force–displacement curve compared with the crack and delamination length for the case of σmax = 100 MPa (left) and σmax = 1 MPa (right) normalized by the layer thickness ly.
The letter refer to the corresponding image in Fig. 5 for the case of σmax = 100 MPa and to Fig. 6 for the case of σmax = 1 MPa.

Fig. 9. (a) Si3N4/BN structure of the microlaminate; (b) 4-point bending experimental test geometry. Image (a) reprints from [4].

because their system is an homogeneous material with an interface, ceramics. The laminate is structured with layers of Si3N4 of thickness
while here we have a laminate with different elastic properties for the between 40 μm and 60 μm, alternated by layers of BN of variable
laminae. This means that σy is not constant across the laminae and there thickness between 2 μm and 10 μm (Fig. 9(a)). The BN layers act as
are jumps in correspondence of the interfaces. We cannot say quanti- quasi-brittle interface between the Si3N4 layers.
tatively how much is the effect of the variation of σy on the interface The 4-point bending test geometry is shown in Fig. 9(b). The di-
delamination and on the overall strength. Nevertheless, we have good mensions of the specimen are: total span L = 5 mm, thickness T =
motivations to believe that σy cannot be neglected and could be another 3 mm, width W = 4 mm. The outer and inner span of the 4-point
important source for possible discrepancies with respect to the bending test are, respectively, S1 = 4 mm and S2 = 2 mm. This geo-
Cook–Gordon model. metry is discretized with phase field finite elements for the Si3N4 layers,
representing the bulk material, and cohesive interface finite elements
compatible with phase field for the BN layers. Due to the variable
3.2. Crack propagation in Si3N4/BN micro-laminate thickness of the Si3N4 layers, in our simulation the thickness associated
to the layers of this material is set equal to 40, 50 or 60 μm. The as-
In this section, we reproduce the experimental results in [4] con- signment of the Si3N4 layer thickness is randomly chosen according to a
cerning the 4-point bending test of a silicon nitride/boron nitride uniform distribution. The interface thickness, on the other hand, is set
(Si3N4/BN) micro-laminate. Both constituent materials are brittle

3000
V. Carollo et al. Journal of the European Ceramic Society 38 (2018) 2994–3003

Fig. 10. Numerically predicted vs. experimental force–displacement curves.

Fig. 11. Fracture propagation and delamination resulting from the numerical simulation. For each subfigure there is the damage contour plot on top and the x-displacement contour plot
on bottom.

3001
V. Carollo et al. Journal of the European Ceramic Society 38 (2018) 2994–3003

Fig. 12. (a) Experimental crack pattern; (b) numerical predictions of the crack pattern (magnified).
Image (a) adapted from [4].

constant and equal to 5 μm. The material and fracture parameters of the field method of fracture and the cohesive zone model approach has
bulk are: Young modulus E = 310 GPa, Poisson ratio ν = 0.27, fracture been applied for modelling complex crack paths in ceramic laminates
energy Gs = 9 μN/μm and phase field internal length scale l = 2 μm. within the context of the finite element method.
The interface parameters are: σc,0 = τc,0 = 32 MPa and k0 = 70 MPa/ In particular, the predictive capabilities of the proposed modelling
μm. tools have been exploited in relation to ceramic laminates. The effect of
The experimental force–displacement curve (Fig. 10) shows an in- tough or quasi-brittle interfaces has been quantified and compared to
itial linear behaviour until the peak load of 475 N is reached. After this the case of fully bonded layers. Based on the numerical predictions, we
point, the load-carrying capacity of the specimen drops down to around have numerically quantified that introducing interfaces is a way to
40% of the peak load. Then, the load continue increasing until a second increase the tortuosity of the crack path increasing the energy dis-
drop is observed when the value of 240 N is reached. After this second sipated before failure. It has been also identified that the peak cohesive
drop, the load-carrying capacity is reduced to 10% of that at peak load. tractions govern the development of crack penetration, crack branching
The specimen maintains this level until final failure. and crack deflection phenomena.
Fig. 11 shows the evolution of the predicted crack pattern. Due to Moreover, the applicability of the current computational method
symmetry of the geometry and of the boundary conditions, only half of has been examined through its application to real laminates structures
the domain has been simulated, using around 380.000 finite element composed by Si3N4/BN layers. The numerical predictions showed that
nodes and a finer mesh in the region where the crack is expected to the proposed framework enabled reproducing the key features of the
propagate. The simulation shows an initial linear behaviour until the crack pattern and the force–displacement curve observed in the ex-
peak load of 475N is reached (Fig. 11(a)). Then, an interface situated in periments, proving the capabilities of the proposed approach to be used
the middle of the specimen thickness fails due to delamination as a tool for the design and characterization of ceramic laminates.
(Fig. 11(b)). After this first delamination event, the specimen continues In light of the previous arguments, this work is expected to provide
gaining load-carrying capacity until the Si3N4 layers start failing. Ac- a suitable modelling framework for more application-based problems
cording to the numerical predictions, the layers that first fail are that at concerning crack pattern predictions in ceramic composites, among
the intrados and the one immediately next to the delaminated interface other alternative multilayer or heterogeneous materials.
(Fig. 11(c)). The crack is predicted to continue its propagation towards Moreover, as a research perspective, a comparison of the proposed
the extrados of the specimen until final failure (Fig. 11(d)). The result of approach with the method provided in the theory of structured de-
the simulation in terms of force–displacement curve shows a good formations [27–29] could be of interest for further investigations, to
agreement with the experimental one (Fig. 10). The initial part of the provide a re-interpretation of the phase field approach to fracture in
curve, the peak load and the drop in load-carrying capacity are very terms of structured deformations.
well predicted, capturing the critical load at which damage events
occur. The final part of the simulation shows a more progressive failure Acknowledgments
evolution until collapse. This is not in perfect agreement with experi-
ments, since we suppose that the second drop in the force noticed in The authors acknowledge funding received from the European
experiments is caused by a second severe delamination event which we Research Council under the European Union's H2020 Programme ERC
were not able to reproduce numerically. One possible source for this Grant Agreement No. 737447 (ERC Proof of Concept 2016
mismatch can be attributed to the interface parameters, since no ex- “Photovoltaic with superior crack resistance” – PHYSIC). JR acknowl-
perimental characterization was available. edges the support of the project funded by the Spanish Ministry of
Another important result of the simulation is the very satisfactory Economy and Competitiveness/FEDERMAT2015-71036-P and the
prediction of the crack pattern features conforming to the experimental Andalusian Government Project of Excellence No. P12-TEP-1050. The
evidences. The experimental image in Fig. 12(a) shows the strong crack authors would like to thank the anonymous reviewer for his useful
deflection and branching developed due to delamination. The same suggestions.
behaviour has been reproduced by the numerical simulation
(Fig. 12(b)) where both phenomena are present in the simulated crack References
pattern.
[1] M. Sebastiani, K.-E. Johanns, E.-G. Herbert, F. Carassiti, G.-M. Pharr, A novel pillar
indentation splitting test for measuring fracture toughness of thin ceramic coatings,
4. Conclusions Philos. Mag. 95 (16–18) (2015) 1928–1944.
[2] N. Li, H. Wang, A. Misra, J. Wang, In situ nanoindentation study of plastic co-
deformation in Al-TiN nanocomposites, Sci. Rep. 4 (2014).
In this work, a novel variational framework combining the phase

3002
V. Carollo et al. Journal of the European Ceramic Society 38 (2018) 2994–3003

[3] A. Hiltner, K. Sung, E. Shin, S. Bazhenov, J. Im, E. Baer, Polymer microlayer [16] M.-J. Borden, C.-V. Verhoosel, M.-A. Scott, T.-R. Hughes, C.-M. Landis, A phase-field
composites, MRS Online Proc. Lib. Arch. 255 (1991). description of dynamic brittle fracture, Comput. Methods Appl. Mech. Eng. 217
[4] H. Liu, S.-M. Hsu, Fracture behavior of multilayer silicon nitride/boron nitride (2012) 77–95.
ceramics, J. Am. Ceram. Soc. 79 (1996) 2452–2457. [17] C.-V. Verhoosel, R. de, Borst, A phase-field model for cohesive fracture, Int. J.
[5] J.-X. Zhang, D.-L. Jiang, S.-Y. Qin, Z.-R. Huang, Fracture behavior of laminated SiC Numer. Methods Eng. 96 (2013) 43–62.
composites, Ceram. Int. 30 (2004) 697–703. [18] B. Bourdin, G.-A. Francfort, J.-J. Marigo, The variational approach to fracture, J.
[6] J. Cook, J.-E. Gordon, C.-C. Evans, D.-M. Marsh, A mechanism for the control of Elast. 91 (2008) 5–148.
crack propagation in all-brittle systems, Proc. R. Soc. Lond. A 282 (1391) (1964) [19] V.-A. Lubarda, D. Krajcinovic, S. Mastilovic, Damage model for brittle elastic solids
508–520. with unequal tensile and compressive strengths, Eng. Fract. Mech. 49 (1994)
[7] L. Yong-Ming, P. Wei, L. ShuQin, C. Jian, W. RuiGang, L. JianQiang, Mechanical 681–697.
properties and microstructure of a Si3N4/Ti3SiC2 multilayer composite, Ceram. Int. [20] R.-W. Ogden, Non-linear Elastic Deformations, Courier Corporation, 1997.
28 (2002) 223–226. [21] J.-G. Williams, H. Hadavinia, Analytical solutions for cohesive zone models, J.
[8] S. Bueno, C. Baudin, Design and processing of a ceramic laminate with high Mech. Phys. Solids 50 (2002) 809–825.
toughness and strong interfaces, Compos. A: Appl. Sci. Manuf. 40 (2009) 137–143. [22] P.-G. Charalambides, J. Lund, A.-G. Evans, R.-M. McMeeking, A test specimen for
[9] G. Mishuris, A. Piccolroaz, A. Vellender, Boundary integral formulation for cracks at determining the fracture resistance of bimaterial interfaces, J. Appl. Mech. 56
imperfect interfaces, Q. J. Mech. Appl. Math. 67 (2014) 363–387. (1989) 77–82.
[10] L. Morini, A. Piccolroaz, G. Mishuris, E. Radi, Integral identities for a semi-infinite [23] P.-P. Camanho, C.-G. Davila, M.-F. De Moura, Numerical simulation of mixed-mode
interfacial crack in anisotropic elastic bimaterials, Int. J. Solids Struct. 50 (2013) progressive delamination in composite materials, J. Compos. Mater. 37 (2003)
1437–1448. 1415–1438.
[11] A. Piccolroaz, G. Mishuris, Integral identities for a semi-infinite interfacial crack in [24] O.-C. Zienkiewicz, R.-L. Taylor, The Finite Element Method: Solid Mechanics, vol. 2,
2d and 3d elasticity, J. Elast. 110 (2013) 117–140. Butterworth-Heinemann, 2000.
[12] M. Paggi, J. Reinoso, Revisiting the problem of a crack impinging on an interface: a [25] P.-P. Camanho, C.-G. Dávila, Mixed-mode decohesion finite elements for the si-
modeling framework for the interaction between the phase field approach for brittle mulation of delamination in composite materials. Technical Report NASA/TM-
fracture and the interface cohesive zone model, Comput. Methods Appl. Mech. Eng. 2002-211737, NASA Langley Research Center, Hampton, VA United States, 2002.
321 (2017) 145–172. [26] A. Braides, Approximation of Free-discontinuity Problems. Number 1694, Springer
[13] V. Carollo, J. Reinoso, M. Paggi, A 3D finite strain model for intralayer and inter- Science & Business Media, 1998.
layer crack simulation coupling the phase field approach and cohesive zone model, [27] D.-R. Owen, Elasticity with gradient-disarrangements: a multiscale perspective for
Compos. Struct. 182 (2017) 636–651. strain-gradient theories of elasticity and of plasticity, J. Elast. 127 (2017) 115–150.
[14] M. Paggi, M. Corrado, J. Reinoso, Fracture of solar-grade anisotropic polycrystalline [28] L. Deseri, D.-R. Owen, Stable disarrangement phases arising from expansion/con-
Silicon: a combined phase field-cohesive zone model approach, Comput. Methods traction or from simple shearing of a model granular medium, Int. J. Eng. Sci. 96
Appl. Mech. Eng. 330 (2018) 123–148. (2015) 111–130.
[15] C. Miehe, M. Hofacker, F. Welschinger, A phase field model for rate-independent [29] L. Deseri, D.-R. Owen, Submacroscopic disarrangements induce a unique, additive
crack propagation: robust algorithmic implementation based on operator splits, and universal decomposition of continuum fluxes, J. Elast. 122 (2016) 223–230.
Comput. Methods Appl. Mech. Eng. 199 (2010) 2765–2778.

3003