PROGRESSIVE DAMAGE MODELLING IN CROSS-LAMINATED

TIMBER STRUCTURES BY COMPUTATIONAL HOMOGENISATION

AND COHESIVE ZONE MODELS
Erick I. Saavedra Flores 1, Karin Saavedra 2, Jorge Hinojosa 2, Raj Das 3
ABSTRACT: In this paper we investigate the damage process in cross-laminated timber (CLT) structures by a
computational homogenisation approach enriched with cohesive zone models. In order to predict the undamaged
structural response, four spatial scales are interlinked within a multi-scale finite element modelling scheme. To simulate
the cracking process in the material, a cohesive zone model (CZM) is adopted at the homogenised macroscopic scale.
This double approach allows us to model successfully the progressive damage process in CLT plates subject to threepoint bending.
KEYWORDS: Cross-laminated timber, homogenisation, cohesive zone model, damage.

1 INTRODUCTION 123
Over the last few years, CLT structures have been
increasingly spreading in the construction sector, mainly
in Europe and North America. CLT panels are made up of
several layers of boards stacked crosswise and glued
together on their faces. Among the main advantages of
CLT, we can highlight its fast and efficient on-site
installation, its favourable seismic performance and its
excellent strength. As CLT panels are light-weight
structural elements with high stiffness and strength to
bending, compression and shear, they are an economically
competitive building system when compared to traditional
options and therefore, are a suitable candidate for some
applications in buildings which currently use concrete,
masonry and steel [1].
One important issue in the design of CLT structures which
still requires further investigation is the progressive
damage that may occur across the thickness of CLT
layers, or between adjacent layers [2]. In particular, the
design of CLT floor systems with low span-to-depth ratios
is often governed by the rolling shear capacity of CLT
plates, and therefore, its full understanding is of
paramount importance to provide the required structural
strength and prevent damage in CLT structures. This type
of damage can also be potentially present in CLT walls
subject simultaneously to eccentric compressive loads and
out-of-plane bending or shear loads.
Rolling shear failure consists of inter-fibre cracking due
to shear strains in the plane perpendicular to the
longitudinal axis of wood fibres. Figure 1 shows a typical

Departamento de Ingeniera en Obras Civiles, Universidad de

Santiago de Chile. Av. Ecuador 3659, Estacin Central,
Santiago, Chile.
2 Departamento de Tecnologas Industriales, Universidad de
Talca, Los Niches km 1, Curic, Chile.

rolling shear failure observed in the central layer of a CLT

plate subject to bending [3].

Figure 1: Typical rolling shear failure observed in the

central layer of a CLT plate subject to bending [3].

Despite the increasing interest in this subject, the

complete understanding of the mechanical properties of
wood, and particularly its failure mechanisms in CLT
structures, is still an issue which remains open at present.
In an attempt to shed more light on the simulation of
progressive damage in CLT structures, we present in this
paper a modelling strategy which combines two
approaches. Firstly, we make use of a homogenisationbased multi-scale modelling framework to determine the
undamaged mechanical properties of wood. Secondly, we
enrich this modelling procedure by using cohesive
interfaces at the homogenised macroscopic structural
scale in order to capture inter-fibre and interlaminar
cracking in a CLT plate.

Department of Mechanical Engineering, University of

Auckland, Auckland 1010, New Zealand.

2 THEORETICAL FRAMEWORK
Over the last two decades, homogenisation-based multiscale computational modelling techniques have been
developed to predict accurately the mechanical response
of
heterogeneous
materials
with
hierarchical
microstructures [4]. Successful use of this methodology in
the context of wood materials and structures can be found
in [5-9]. In this constitutive theory, each material scale is
associated with a microstructure whose most statistically
relevant features are incorporated within a representative
volume element (RVE). This RVE is assumed to have a
(microscopic) characteristic length much smaller than the
macro-continuum, and at the same time, a size large
enough to capture the microscopic heterogeneities in an
averaged sense.
In this theory it is also assumed that the macroscopic or
homogenised strain tensor component ij at any arbitrary
point of the macroscopic continuum is the volume average
of the microscopic strain tensor component ij over the
domain of the RVE.
Similarly, the macroscopic or homogenised stress tensor
component ij is assumed to be the volume average of the
microscopic stress tensor component ij over .
In addition, we can introduce a convenient decomposition
in the total displacements field over the RVE domain as a
sum of a linear displacement component ui and a
displacement fluctuation i which represents local
variations about the linear displacement component and
does not contribute to the macroscopic scale strain.
By taking into account the Hill-Mandel Principle of
Macro-homogeneity [10, 11], which establishes that the
macroscopic stress power must equal the volume average
of the microscopic stress power over , the virtual work
equation for the RVE can be reduced to
1

= 2 ( + ) ,

As we are interested in modelling cracking in the material,

we adopt a CZM at the macroscopic structural scale. CZM
is the simplest model that allows to describe in full a
fracture process (i.e. initiation and propagation of the
crack). The basic idea of the CZM can be described as a
zero thickness interface transferring tractions which are
related to the displacement jump of the interface by the
meaning of a softening function. This evolution law can
be written in terms of a damageable stiffness operator. At
the beginning, the interface stiffness has no damage (that
is, a value k0). Then, the stiffness decreases with respect
to the displacement jump (up to a value k) and becomes
zero at some critical displacement jump. It is possible to
use a damage variable d to represent the stiffness of the
interface, with d ranging from 0 (healthy interface point)
to 1 (completely damaged interface point). The area under
the entire stress-displacement jump curve is the energy
per unit area necessary to separate completely the
interface at a given point.
The area under the entire stress-displacement jump curve
is the energy Gf (J/m2) necessary to completely separate
the interface at a given point. This energy coincides with
the critical energy release rate, i.e. the mechanical energy
released on the interface during all the decohesion
process, in the case of a steady-state crack propagation
and when the dissipative phenomena are really confined
on the interface.
Nowadays, it is firmly stablished that the quasibrittle
failure of wood, concrete or rocks could be described with
reasonable accuracy using the bilinear approximation, as
shown in Figure 2(a).

(1)

with ij representing the Cartesian components of the

kinematically admissible virtual strains field, i a
component of the virtual displacements vector, both at the
RVE level, and yi a local (RVE) coordinate.
In this investigation, a set of kinematical constraints is
prescribed upon the selected RVE. This set coincides with
the widely used Periodic boundary displacement
fluctuations model, typically associated with the
modelling of periodic media [4]. The kinematical
constraint consists of prescribing identical displacement
fluctuations for each pair of opposite points {y+,y-} on the
RVE boundary domain, such that (y+) = (y-).
Complex materials like timber show a dual behaviour.
That is, for very large compressive strains typically found
in regions near fastener connections, the material shows
high levels of ductility, hardening and energy dissipation.
In sharp contrast to this behaviour, timber shows an
initially linear elastic behaviour when tensile deformation
mechanisms are predominant in the material, followed by
a progressive softening damage after a critical stress. In
fact, the strength of timber elements is normally governed
by the formation of cohesive cracks due to the separation
of wood fibres.

Figure 2: Stress-displacement jump curves in CZMs

assumed in this work [2].
In this law, it is accepted that microcraking and crackbridging take place in the fracture process zone (FPZ) and
precede the propagation of the main crack. Therefore, in
this work this bilinear approximation is adopted to model
the stress-displacement jump curves found in wood.
Since the interlaminar adhesive used to manufacture the
CLT specimens for our experimental validations (that is,
the Emulsion Polymer Isocyanate) has not yet been
identified in the literature, we propose to adopt a power
softening law as shown in Figure 2(b) to simulate the
delamination process in CLT plates.
Due to the large size of the discretised problem, over a
million of degrees of freedom, we propose to perform
parallel computations using a mixed domain
decomposition method which is particularly useful to deal
with cohesive zone models.

3 WOOD MICROMECHANICS
At the scale of some few nanometres, wood contains three
fundamental phases: cellulose, hemicellulose and lignin.
These three fundamental constituents form the wood cellwall composite material whose basic unit building block
is called microfibril. This unit can be considered as a
periodic block of rectangular cross-section with infinite
length. This composite comprises reinforcing microfibres which are oriented mainly in a single direction (in
almost the whole cell-walls volume) and are embedded
periodically in a softer matrix. The reinforcing microfibre is made up of periodic alternations of crystalline and
amorphous cellulose fractions. The degree of cellulose
crystallinity is defined as the volume fraction of the
crystalline portion of cellulose with respect to the total
volume of cellulose. The high stiffness of the cellulose is
due to its crystalline fraction whereas its flexibility is
provided by the amorphous part. Only its non-crystalline
fraction may absorb moisture and consequently, change
its mechanical properties.
The matrix of the cell-wall composite is made up of
hemicellulose and lignin polymers. Contrary to crystalline
cellulose, hemicellulose is a polymer with little strength,
built up of sugar units. Its structure is partially random,
with mechanical properties highly sensitive to moisture
changes, softening with increasing water content.
Lignin is an amorphous and hydrophobic polymer and its
main purpose is to cement the individual wood fibres
together and to provide inter-fibre shear strength. It is the
most hydrophobic component in the cell-wall, with
relatively stable mechanical properties under moisture
changes. Figure 3 shows a schematic representation of the
cross section of the microfibril and its fundamental
constituents.

At the scale of some micrometres, the material can be

represented by a periodic arrangement of long slender
tubular fibres, oriented nearly parallel to the axis of the
stem. The cross sections of each wood fibre is (normally)
hexagonal. In softwoods, wood fibres can be divided into
early-wood and late-wood. The early-wood fibres are
characterised by large diameters and thin cell-walls,
whereas late-wood fibres are composed of narrow
diameters with much thicker cell-walls. Figure 4 shows a
scanning electron micrograph of wood at this scale.

Figure 4: Digital photograph of wood microfibers at the

scale of some micrometres [16].

At the scale of some few millimetres, wood is represented

by growth rings, typically found in the cross section cut
through the trunk of a tree. Each growth ring exhibits two
colour regions associated with early-wood (light color)
and late-wood (dark color) fibres, as shown in Figure 5.

Figure 3: Schematic representation of the cross section of

the microfibril and its main constituents at the scale of a
few nanometres [9].

The specific orientation of the microfibrils with respect to

the wood fibres axis is called the microfibril angle, MFA.
Depending on the proportions of constituents and MFA,
wood cell-wall can be divided in different layers. In all
these layers, the microfibril takes different orientations
throughout the cell-wall, generating a mechanical locking
effect and consequently, an increase in the overall
stiffness in the cell. Among all these, S2 layer is the
thickest and most influential factor in the mechanical
behaviour of wood cells. It comprises about 8090% of
the total volume of cell-wall and concentrates a high
content of cellulose of about 50 % in weight. For further
information about cell-wall layers and their functions
from a structural point of view, we refer, for instance, to
[12-15].

Figure 5: Growth rings at the scale of millimetres.

The periodic repetition of these growth rings gives rise to

the macroscopic or structural level, which can be
represented by lumbers. These are then used to
manufacture CLT plates.

4 COMPUTATIONAL MODELLING
In this section, we describe the multi-scale finite element
modelling of CLT. The type of wood chosen for this study
is radiata pine, which has several applications in building
and engineering structures.
As commented in the previous Section 3, three material
scales (at the level of nanometres, micrometres and
millimetres) form the base material for the analysis of the
CLT plate at the structural scale. Therefore, four spatial
scales are interlinked within the present multi-scale
modelling framework in order to predict the structural
behaviour.
A typical finite element mesh of the RVE associated with
the microfibril scale is shown in Figure 6.

The corresponding finite element mesh adopted for the

modelling of the RVE associated with the latewood fibres
is shown in Figure 7.

Figure 7: Finite element mesh adopted for the modelling

of latewood fibres [7].

Figure 6: Finite element mesh of the microfibril RVE [7]

(For the sake of clarity, only one half of the geometry is
shown here)
The mesh consists of 4495 nodes and 3960 hexahedral
elements. We select the eight-noded SOLID185 element
type in ANSYS. The mechanical properties of each of the
basic constituents considered in our study are summarised
in the following Table 1.

The mesh consists of 3141 nodes and 1840 SOLID185

elements. Those elements coloured in red represent the S2
layer in the cell-wall with a corresponding relative volume
of 80% with respect to the total volume of cell-wall. This
material is obtained through the computational
homogenisation of the microfibril RVE shown in Figure
6. The green colour represents the compound middle
lamella (P + M), comprising a relative volume of 20%.
The MFA of the S2 layer is taken into account during the
homogenisation process.
The finite element mesh used in the modelling of the
earlywood fibres is shown in the following Figure 8. It
consists of 1626 nodes and 672 SOLID185 elements.

Figure 8: Finite element mesh adopted for the modelling

of earlywood fibres [2].

Table 1: Summary of the elastic material properties of the

basic constituents of wood.
This set of data corresponds to dry conditions, i.e. 12%
moisture content.

We note the considerable difference observed in Figure 7

and 8 between the thicknesses of earlywood and latewood
fibres.
The finite element model of the growth ring RVE is
shown in Figure 9

Figure 9: Finite element mesh adopted for the modelling

of the growth ring RVE [2].
The finite element mesh consists of 288 nodes and 165
hexahedral SOLID185 elements. The brown colour shows
the material obtained by the homogenisation of the
latewood RVE shown in Figure 7.The light green colour
represents the portion of material calculated by the
computational homogenisation of the earlywood RVE
shown in Figure 8.
Finally, by homogenisation the growth ring RVE shown
in Figure 9, we obtained the (undamaged) material
properties for the macroscopic structural (or largest) scale.
In this work, this scale consists of a CLT specimen of
three 4-cm-thick layers with a length of 75 cm and a width
of 39 cm. The span length between supports was 60 cm.
We note that the main motivation of choosing these small
dimensions was to produce shear effects in the panel.
ASTM D3737 [17] suggests a span-to-depth ratio near 5
to produce a high percentage of shear failure at a large
extent. The outer layers were made up of timber pieces
oriented in the strong direction of the panel. The central
layer was made of members oriented in the weak
direction. In addition, edgegluing is considered in the
manufacture of the panel.
This CLT plate is modelled (at the macroscopic scale)
using a finite element mesh with 861696 linear wedge
elements and 2.9 millions of DOFs (because of the
symmetry of the problem, only one quarter of the
geometry is considered). Figure 10 shows one quarter of
the 12-cm-thick CLT plate (i.e., three 4-cm-thick layers).

Cohesive interfaces are used to simulate the rolling shear

failure, but a distinction is made between the interlaminar
interfaces (for delamination) and the inner interfaces of
the central layer. The interfaces in green are defined using
a power law and the red ones using a bilinear behaviour.
When a cohesive interface is completely delaminated,
contact conditions are considered to avoid
interpenetration.
The fracture parameters of the Chilean radiata pine and
the interlaminar adhesive have not been reported in
previous works. Therefore, the parameters of both
cohesive laws were approached from the literature for
similar species [18-21].

5 NUMERICAL RESULTS
In this section we present the numerical results obtained
by the present multi-scale computational approach
enriched with macroscopic cohesive laws.
The computational loading programme consists of a
displacement-controlled load at mid-span where the
deflection increases monotonically from zero up to reach
extensive failure.
Figure 1 shows the computational simulation of the
progressive damage process in the CLT. The
corresponding contour plot shows the damage variable d
that ranges from 0, when an interface point is fully
healthy, to 1, when the interface point is completely
damaged. In our simulation, two parallel cracks are
predicted consistently with experiments (not detailed
here), along with some delaminated regions, mainly
between the bottom and central layer. In our numerical
predictions, we note that after a cohesive interface has
become fully damaged, it is converted into a contact
interface as well.

Figure 11: Computational simulation of the progressive

damage process in a CLT plate subject to bending.

Figure 12 shows the deformed shape of the CLT

plate for the same simulation. The contour plot of

the principal stresses are shown here for the same
load level considered in Figure 11.
Figure 10: Finite element mesh adopted for the modelling
of a CLT plate subject to bending (only interfaces are
shown here for the sake of clarity).

The influence of wood density on the mechanical

behaviour of CLT is also investigated in this paper. Four
densities are considered here. These are 400, 450, 500 and
550 kg/m3. These values have been obtained by modifying
the cell-walls thickness and keeping the rest of the
micromechanical parameters fixed in our model.
Figure 14 shows the numerical prediction of the variation
of the longitudinal Youngs modulus when wood density
increases.
Figure 12: Deformed shape and principal stresses in a
CLT plate subject to bending. Rolling shear cracks are
visible here.
In practice, when manufacturing CLT panels, the edgegluing of lumber pieces (within each layer) is optional. If
edgegluing is not present, the adhesive is only applied to
the opposite sides of adjacent layers, leaving the joint
lines free to expand and contract in response to changes in
temperature and relative humidity [2]. By avoiding edge
gluing, CLT manufacturers can also achieve further
benefits by saving adhesive consumption and improving
production times. Therefore, in this paper we investigate
this effect on the rolling shear capacity.
Figure 13 shows the computational simulation of a CLT
plate when no edge-gluing is present.

Figure 14: Variation of the longitudinal Youngs modulus

when wood density increases.
We note that these results are obtained within the
infinitesimal range of elastic deformations. That is, the
material is undamaged. As expected, a linear trend is
observed. We also note that for radiate pine, the
experimental Youngs modulus and density are 12.7 GPa
and 506 kg/m3, respectively [2], which are in good
agreement with the values predicted by our model.
Figure 15 shows the variation of the critical peak load
when wood density increases. We note that this critical
load is associated with the beginning of rolling shear
cracks in the central layer of the CLT specimen.

Figure 13: Deformed shape and principal stresses in a

CLT plate with non-edge-gluing effect. Rolling shear
cracks and edge separation can be observed here.
The non-edge-gluing effect is modelled by introducing
frictionless contact interfaces between opposite edges of
individual pieces of lumber, allowing us to capture joint
opening/closure effects between adjacent edges, with no
penetration. Here, surfaces are free to slide and separate
without resistance. In our computational simulations, we
note that after a cohesive interface has become fully
damaged, it is converted into a contact interface as well.
When removing edge-gluing effects from the CLT plate,
a reduction of 40% of the critical load is observed.
Nevertheless, further investigation must be carried out in
this context, particularly with experimental tests in order
to validate this numerical prediction.

Figure 15: Variation of the critical peak load (beginning

of rolling shear cracks) when wood density increases.
We observe here that the critical peak load increases
linearly from 142.5 kN to 172.5 kN, approximately, when
wood density increases from 400 to 550 kg/m3. That is,
21% increase in the critical peak load is predicted when
the density increases 37.5%.

6 CONCLUSIONS
In this paper, we have investigated the progressive
damage process in CLT plates by a computational
homogenisation approach in conjunction with a CZM.
Four spatial scales have been interlinked within a multiscale framework in order to predict the undamaged
structural behaviour. The CZM has been adopted at the
macroscopic scale to simulate the formation of
interlaminar and interfibre cracks in the material. The
results obtained in this study demonstrate that the present
approach may represent a promising framework to
describe multiple crack fronts or to describe not wellknown crack paths in complex materials and structures.
We believe that the present multi-scale modelling strategy
combined with CZMs, and with the support of
experimental works, can provide a robust platform for
further investigations on the dissipative response of mass
timber structures with a particular view to clarify features
of the microscopic behaviour, and their influence on the
macroscopic response, which are not well-understood at
present. For example, the investigation of non-linear
buckling combined with delamination in CLT walls
subject to compressive loads is relatively straightforward
and is likely to shed more light into the complex
behaviour of such structures. Finally, we remark that
studies are currently under way to predict numerically inplane shear strength of CLT plates based on the present
framework. This will the subject of a future publication.

ACKNOWLEDGEMENT
E.I. Saavedra Flores acknowledges the financial support
from the Chilean National Commission for Scientific and
Technological Research (CONICYT), FONDECYT
REGULAR research project No 1140245.
K. Saavedra acknowledges the financial support from
CONICYT, FONDECYT Initiation into Research project
No 11130623.

REFERENCES
[1] M. Mohammad, S. Gagnon, B.K. Douglas and L.
Podesto (2012). Introduction to cross laminated
timber. Wood Design Focus 22(2), 3-12.
[2] E.I. Saavedra Flores, K. Saavedra, J. Hinojosa, Y.
Chandra and R. Das. Multi-scale modelling of rolling
shear failure in cross-laminated timber structures by
homogenisation and cohesive zone models.
International Journal of Solids and Structures, 81
(2016), pp. 219232.
[3] Q. Zhou, M. Gong, Y. Hei Chui and M. Mohammad.
Measurement of rolling shear modulus and strength
of cross laminated timber fabricated with black
spruce. Construction and Building Materials 64
(2014) 379-386.
[4] D. Peric, E. A. de Souza Neto, R. A. Feijoo, M.
Partovi, A. J. C. Molina, On micro-to-macro
transitions for multi-scale analysis of non-linear
heterogeneous materials: unified variational basis

and finite element implementation, International

Journal for Numerical Methods in Engineering 87 (1
5) (2011) 149170.
[5] E. I. Saavedra Flores, M. I. Friswell, Multi-scale
finite element model for a new material inspired by
the mechanics and structure of wood cell-walls,
Journal of the Mechanics and Physics of Solids 60 (7)
(2012) 1296 1309.
[6] E. I. Saavedra Flores, M. I. Friswell, Ultrastructural
mechanisms of deformation and failure in wood
under tension, International Journal of Solids and
Structures 50 (13) (2013) 20502060.
[7] E. I. Saavedra Flores, I. Dayyani, R. Ajaj, R. CastroTriguero, F. DiazDelaO, R. Das, P. Gonzlez Soto,
Analysis of cross-laminated timber by computational
homogenisation and experimental validation,
Composite Structures 121 (0) (2015) 386 394.
[8] E.I. Saavedra Flores, F.A. DiazDelaO, R.M. Ajaj,
M.I. Friswell, and G.F. Fernando, Mathematical
modelling of the stochastic mechanical properties of
wood and its extensibility at small scales. Applied
Mathematical Modelling, 38(1516) 2014, pp. 3958
3967.
[9] E.I. Saavedra Flores, E.A. de Souza Neto and C.
Pearce, A large strain computational multi-scale
model for the dissipative behaviour of wood cellwall. Computational Materials Science, 50(3) 2011,
pp. 1202-1211.
[10] R. Hill, A self-consistent mechanics of composite
materials, Journal of the Mechanics and Physics of
Solids 13 (4) (1965) 213222.
[11] J. Mandel, Plasticit Classique et Viscoplasticit.
CISM Lecture Notes, Springer-Verlag, Udine, Italy,
1971.
[12] J. Bodig, B. Jayne, Mechanics of Wood and Wood
Composites. Von Nostrand Reinhold, New York,
1982.
[13] L. Salmn, Micromechanics of the wood cell wall: a
tool for the better understanding of its structure. In:
Navi, P. (Ed.), Proceedings of 1st International
Conference of the European Society for Wood
Mechanics. EPFL, Lausanne, Switzerland, 2001, pp.
385398.
[14] Dinwoodie, J., 1981. Timber Its nature and
behavior. Von Nostrand Reinhold, New York.
[15] Booker, R., Sell, J., 1998. The nanostructure of the
cell wall of softwoods and its functions in a living
tree. European Journal of Wood and Wood Products.
Holz als Roh- und Werkstoff 56 (1).
[16] L.J. Gibson. The hierarchical structure and
mechanics of plant materials. J. R. Soc. Interface
(2012) 9, 27492766.
[17] ASTM American Society for Testing and
Materials, Standard test methods of static tests of
lumber in structural sizes. Designation: D198-09,
USA.
[18] O. Allix, D. Lveque, L. Perret, Identification and
forecast of delamination in composite laminates by
an interlaminar interface model, Composites Science
and Technology 58 (5) (1998) 671678.
[19] N. Dourado, S. Morel, M. de Moura, G. Valentin, J.
Morais, Comparison of fracture properties of two

wood species through cohesive crack simulations,

Composites Part A: Applied Science and
Manufacturing 39 (2) (2008) 415 427.
[20] S. E. Stanzl-Tschegg, D. Keunecke, E. K. Tschegg,
Fracture tolerance of reaction wood (yew and spruce
wood in the TR crack propagation system), Journal
of the Mechanical Behavior of Biomedical Materials
4 (5) (2011) 688 698.
[21] F. Daghia, P. Ladeveze, Identification and validation
of an enhanced mesomodel for laminated composites
within the WWFE-III, Journal of Composite
Materials 47 (20-21) (2013) 26752693