Sei sulla pagina 1di 19

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/286126488

A 3D constitutive wood model using the concepts of continuum damage


mechanics

Article · January 2012

CITATIONS READS

2 453

3 authors, including:

Jan-Willem van de Kuilen


Technische Universität München
70 PUBLICATIONS   471 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Entwicklung eines Bausystems für Parkhäuser in Buchen-Furnierschichtholz View project

Glued-in rods View project

All content following this page was uploaded by Jan-Willem van de Kuilen on 11 December 2015.

The user has requested enhancement of the downloaded file.


European Congress on Computational Methods
in Applied Sciences and Engineering (ECCOMAS 2012)
J. Eberhardsteiner et.al. (eds.)
Vienna, Austria, September 10-14, 2012

A 3D CONSTITUTIVE WOOD MODEL USING THE CONCEPTS OF


CONTINUUM DAMAGE MECHANICS
Carmen Sandhaas1, Jan-Willem G. van de Kuilen2, and Hans J. Blass3
1
Delft University of Technology
Faculty of Civil Engineering
Delft, the Netherlands
As of 01.10.2011 at Karlsruhe Institute of Technology
sandhaas@kit.edu
2
Holzforschung München, Technical University Munich,
Delft University of Technology
vandekuilen@wzw.tum.de
3
Karlsruhe Institute of Technology
Timber Structures and Building Construction
blass@kit.edu

Keywords: 3D constitutive model wood, continuum damage mechanics (CDM), finite ele-
ment method

Abstract. Wood is heterogeneous, highly anisotropic and shows ductile behaviour in com-
pression and brittle behaviour in tension and shear. Timber joints with their simultaneous
ductile and brittle failure modes still present a major challenge for modelling. A 3D material
model for wood based on continuum damage mechanics was developed and implemented via
a subroutine into a standard FE framework. Eight stress-based failure criteria were derived
in order to formulate piecewise defined failure surfaces. The damage development was con-
trolled by nine damage variables that were inserted in the damage operator. Mesh dependen-
cy issues connected with softening continuum approaches were addressed. Embedment and
joint tests using three different wood species (spruce, beech, azobé) were carried out whose
results were compared to modelling outcomes. The failure modes could be identified and the
general shape of the load-displacement curves agreed with the experimental outcomes.
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

1 INTRODUCTION
Wood and timber joints are difficult to model. Apart from their heterogeneity, material-
specific issues lead to numerical problems: strong anisotropy with different strengths in ten-
sion and compression and ductile and brittle failure modes occurring simultaneously.
Existing approaches are mostly based on specific approaches for different problem classes.
Brittle problems can be modelled with fracture mechanics approaches within a continuum
framework [i.e. 1, 2] or discrete lattice models [i.e. 3, 4]. For ductile problems, classical flow
theory of plasticity in combination with the Hill criterion [i.e. 5], the Hoffman criterion [i.e. 6]
or even the Tsai-Wu criterion [i.e. 7] is generally used. To overcome the need of separate ap-
proaches for different problem classes, multi-surface plasticity models have been developed
[8-10]. Recent approaches use micromechanical modelling techniques where representative
volume elements are derived starting at chemical level and scaling up the hierarchical levels
of wood [11].
However, multi-surface plasticity models and continuum micromechanical models are not
readily available for timber engineers nor are they easy to handle as often, no clearly defined
material properties are implemented. Therefore, within the framework of continuum damage
mechanics (CDM), a general approach combining the above mentioned issues in one single
3D material model was developed. This developed constitutive model was used to predict the
results of embedment and joint tests.

2 DEFINITIONS
Material directions and arrays must be clearly defined before introducing the material
model. Both are given in Equation (1) and Figure 1. The known indices in timber engineering
are used, e.g. index v = longitudinal shear, index 90 = direction perpendicular-to-grain.
σ   L  R  T  LR  LT  RT    0  90 R  90T  vR  vT  roll 
T T
(1)

Tangential Tangential
Z, T, 3 Z, T, 3

σ33
σ31 σ32
σ13 σ23

σ11 σ12 σ21 σ22


Longitudinal Radial Longitudinal Radial
X, L, 1 Y, R, 2 X, L, 1 Y, R, 2

Figure 1: Definition of material direction and stress components

As already stated, wood is an anisotropic material. As shown in Figure 1, three material di-
rections can be distinguished in order to reduce the complexity of the material model: longi-
tudinal (L), radial (R) and tangential (T). The developed material model is thus orthotropic.
However, the same material properties were used for the radial and the tangential direction
resulting in a smeared direction perpendicular-to-grain. The longitudinal direction corre-
sponds to the direction parallel-to-grain. This is a similar build-up to fibre composites with a
matrix (= ‘perpendicular-to-grain’) and main fibre direction (= ‘parallel-to-grain’) where e.g.
fibre rupture in tension is a brittle failure mode and matrix failure in compression is ductile.
These analogies motivated among others the choice of the CDM framework for wood material
modelling as CDM methods are widespread approaches in composites modelling [i.e. 12, 13].

2
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

3 CONTINUUM DAMAGE MECHANICS


CDM is a nonlinear elastic approach where the nonlinear behaviour is obtained by modify-
ing the stiffness matrix D or its inverse, the compliance matrix C. CDM can be implemented
in an incremental-iterative FE framework. The stress increments are calculated from strain
increments via a variable stiffness matrix. Therefore and as opposed to classical plasticity, the
unloading in damage mechanics is following the secant stiffness and not following the elastic
stiffness, see Figure 2. No plastic deformations can be modelled.
In CDM, a damage variable d, 0 ≤ d ≤ 1, is determined and inserted into the fundamental
Hooke equation as follows:
 ij  (1  d ) Dijkl  kl (2)

If d = 0, no damage is present; if d = 1, the material has failed. This principle of CDM as a


nonlinear elastic approach is visualised in Figure 2. However, anisotropic damage is observed
for wood which means that several damage variables must be defined to represent the 3D be-
haviour of wood.

Figure 2: Basic idea of CDM

Therefore, in order to develop a 3D material model for wood, three major mathematical
definitions need to be established:
 Failure surfaces to identify damage initiation;
 The post-elastic behaviour when 0 ≤ d ≤ 1;
 A constitutive model linking the stresses to the strains.

4 DAMAGE INITIATION
Classical theory of plasticity is generally based on single-surface failure criteria that are
not able to identify specific failure modes as the material starts yielding in all material direc-
tions once the failure criterion is exceeded. An example used in timber research is the already
mentioned Tsai-Wu [14] criterion. Therefore, in order to recognise failure modes, the single-
surface has been subdivided; different failure criteria or damage initiation functions have been
assigned to stress components that are valid only for the respective stress-strain quadrants.
This method is a well-known approach used in fibre composites [12, 13, 15]. For a complete
3D description of wood as an orthotropic material, eight stress-based failure criteria have been
defined with fx being material strengths in tension (index t), compression (index c) and shear
(indices v and roll) in parallel (index 0) and perpendicular (index 90) directions:
 L ≥ 0 - Criterion I: Failure in tension parallel-to-grain is a brittle failure mode of wood
which is caused by tensile stresses L parallel-to-grain. It is assumed that other stress
components do not influence the tension strength parallel-to-grain. Maximum stress cri-
terion:

3
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

L
Ft ,0 ( )  1 (3)
ft ,0

 L < 0 - Criterion II: Failure in compression parallel-to-grain is a ductile failure mode of


wood which is caused by compression stresses L parallel-to-grain. Maximum stress cri-
terion:

 L
Fc ,0 ( )  1 (4)
f c ,0

The transverse tension modes and shear modes have to be combined as for instance split-
ting parallel to the LR-plane can be caused by tension perpendicular-to-grain (mode I), shear
(mode II) or a combination of both (mixed mode). It is not possible to define separate failure
modes for each stress component as degradation of one component also leads to degradation
of the other components. This means that damage due to longitudinal shear also leads to dam-
age in tension perpendicular-to-grain even though the actual tension stress component per-
pendicular-to-grain may still be lower than the transverse tension strength.
 R/T ≥ 0 - Criteria III / IV: Failure in tension perpendicular-to-grain with splitting in LT-
plane resp. in LR-plane is a brittle failure mode of wood which is caused by tensile
stresses R/T perpendicular-to-grain, longitudinal shear stresses LR/LT and/or rolling shear
stresses RT. Quadratic criterion:
 R2 /T  LT
2
 RT
2
Ft ,90 R /T ( )  2
 / LR
 1 (5)
f t ,90 f v2 2
f roll

 R/T < 0 - Criteria V-VIII: Two failure modes “pure transverse compression” and “shear”,
both occurring under compression perpendicular-to-grain, are distinguished. Failure in
compression perpendicular-to-grain is a ductile failure mode of wood which is caused
only by compression stresses R/T perpendicular-to-grain. However, brittle shear failure
can also occur if for instance the compression load is applied with an angle to the grain
creating high shear stress components. Therefore, failure criterion for high shear stresses
under simultaneous compression perpendicular-to-grain must be defined as well:
 R /T
Fc ,90 R /T ( )  1 (6)
f c ,90

 LT
2
 RT
2
FvR /T ( )  / LR
 1 (7)
f v2 2
f roll

5 DAMAGE EVOLUTION
After having identified the onset of failure by means of Equations (3) to (7), the post-
elastic mechanical behaviour and necessary damage variables need to be defined. After an
extensive literature study on available experimental results [16], two main post-elastic behav-
iour types were identified:
 ductile behaviour in compression parallel and perpendicular to the grain, Figure 3a.;
 softening behaviour in tension and shear where the shear is independent of the sign, Fig-
ure 3b.

4
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

a. b.
Figure 3: Stress-strain graph, a. elastic perfectly plastic behaviour, b. softening behaviour

Exceeding the compression strength (fc,0 and fc,90) triggers ductile stress-strain behaviour
where the damage variable follows an elastic perfectly plastic law, Figure 3a.:
1
d(κ )  1  (8)
κ
The failure criteria for tension and shear instead lead to a linear softening stress-strain rela-
tionship as illustrated in Figure 3b.:
1  2 2G f E 
d(κ )  1   f max   (9)
f 2
max  2G f E  κ 
where fmax = maximum strength, E = modulus of elasticity, Gf = fracture energy.

As can be seen in Equations (8) and (9), the development of the damage variables is not
depending on the strains , but on state variables  instead that keep track of the loading histo-
ry [12]. The damage parameters d cannot be a function of the strains . For instance, once
failure criterion III (= Ft,90R, splitting parallel to LT-plane) is activated due to high longitudi-
nal shear stresses LT, the stiffness values of all other contributing stress components (R and
RT) must degrade as well although the resp. uniaxial stresses R and RT may still be below
the uniaxial strengths ft,90 and froll. In such a case, a dependency of the damage parameters dt,90
and droll on the strains R and RT may not trigger the evolution of damage in these secondary
directions. A dependency on the state variables t,90 and roll will enforce the evolution of the
damage parameters dt,90 and droll instead. The physical reason behind the enforcing of damage
although the uniaxial strengths may not be reached is obvious: once a wooden piece has failed
in shear parallel to the LT-plane, it will not be able to take any loads in tension perpendicular
to the LT-plane.
The failure criteria F can be reformulated introducing state variables  that keep track of
the loading history:
Θ(F, κ )  F  κ  0 (10)
This principle is analogous to the flow theory of plasticity. The “yield” surfaces F are the
damage initiation functions or failure criteria. As in classical plasticity and following [12], the
Kuhn-Tucker conditions must hold:
Θ0 κ  0  0
κΘ (11)
Furthermore, it is required that the damage variables can only grow which follows also
from the second Kuhn-Tucker condition.
d  0 (12)

5
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

The final implementation of the history parameters as state variables is carried out as
shown in Equation (20):


κ t  max 1, max F incr 
incr  0,t
 (13)

As the next step, the number of damage variables that control the post-elastic behaviour is
established to nine:
 dt,0 = damage in tension parallel-to-grain;
 dc,0 = damage in compression parallel-to-grain;
 dt,90R = damage in tension perpendicular-to-grain, radial direction (LT-plane);
 dc,90R = damage in compression perpendicular-to-grain, radial direction;
 dt,90T = damage in tension perpendicular-to-grain, tangential direction (LR-plane);
 dc,90T = damage in compression perpendicular-to-grain, tangential direction;
 dvR = damage in longitudinal shear, LT-plane;
 dvT = damage in longitudinal shear, LR-plane;
 droll = damage in rolling shear, RT-plane.
A transgression of the damage initiation functions I to VIII (Equations (3) to (7)) activates
conjugated damage variables d where the shear damage variables can develop under trans-
verse tension and compression and have to be superposed:
 Criterion I: dt,0
 Criterion II: dc,0
 Criteria III and IV: dt,90R/T, dvR/T and droll
 Criteria V to VIII: dc,90R/T, dvR/T and droll
The two damage variables dt,0 and dc,0 for instance are activated by the same normal stress
component, L parallel-to-grain. However, the damage variables differ depending on the sign
of the stress component. A distinction in tension (index t) and compression (index c) can easi-
ly be made as shown exemplarily for damage parallel-to-grain in Equation (14) using the Ma-
caulay operator as defined in Equation (15). The two damage variables for a normal stress
component can be combined. Damage variables dt,0 and dc,0 can thus be expressed as damage
variable d0.
L  L
d 0  dt ,0  d c ,0 (14)
L L

a :
a  a  (15)
2

6 CONSTITUTIVE MODEL
CDM describes nonlinear material behaviour, especially softening behaviour, as caused by
voids, defects or microcracks which reduce the area or volume of the material that can trans-
mit forces. The effective stress ef is the stress acting on the non-damaged material. A simple
relationship between effective stresses ef and nominal stresses  is shown in Equation (16)
with M being the damage operator which is composed by nine damage parameters d. As de-
fined in Equation (14), the distinction in tension and compression is made by the Macaulay
operator. Therefore, entries d0, d90R and d90T define two damage variables each. The used ar-
ray order was shown in Equation (1).

6
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

σ ef  Mσ
d   d0 d roll 
T
d90 R d90T d vR d vT (16)
 1 1 1 1 1 1 
  if i j
M ij   1  d 0 1  d90 R 1  d90T 1  d vR 1  d vT 1  d roll 

 0 if i j
The relationships of Equation (16) show a simple way to model nonlinear behaviour by
manipulating the elastic stiffness in Hooke’s law. It is obvious that the introduction of ortho-
tropic damage parameters d ranging from 0 to 1 is able to straightforwardly reduce the elastic
stresses in order to represent effective stresses acting on the remaining intact volume of the
material.
As the effective stress ef is the stress acting on the (remaining) undamaged material, it can
also be formulated as shown in Equation (17) for a Hookean material with Del the elastic
stiffness matrix. Or, in the inverse formulation with the elastic compliance matrix Cel:
σ ef  Del ε or ε  Cel σ ef (17)
Equation (17) can be reformulated for a damaged elastic compliance matrix C dam (see also
[13]):
ε  Cel σ ef  Cel Mσ  Cdamσ (18)
With C dam according to Equation (19):
   
1
  21  31 0 0 0 
 (1 d ) E E22 E33 
 0 11 
 12 1  
   32 0 0 0 
 E11 (1 d90 R ) E22 E33 
 
 13  1 
 E  23 0 0 0 
dam (1 d90T ) E33
C  

11 E22


(19)
1
 0 0 0 0 0 
 (1 dvR ) G12 
 
 1 
0 0 0 0 0
 (1 dvT ) G13 
 
 1 
 0 0 0 0 0 
 (1 d roll ) G23 

In finite element programmes, the inverse of the compliance matrix, the stiffness matrix D
is needed in order to update the stresses:

Ddam   Cdam 
1
 σ  Ddamε (20)

If the compliance matrix is positive definite and the damage variables d are less than 1, this
inverse exists.
In the damaged compliance matrix, damage effects and interactions have to be represented.
For instance, damage in compression parallel-to-grain could be assumed to lead to a decrease
of the stiffness in tension parallel-to-grain during a subsequent loading cycle. Furthermore,
assumptions need to be taken about the effects of damage parallel-to-grain on the strength
perpendicular-to-grain.
In the developed material model, no damage interactions were modelled. For instance, it is
assumed that damage in tension perpendicular-to-grain does not lead to a degradation of the
properties parallel-to-grain.

7
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

Furthermore, when calculating the inverse of the damaged compliance matrix as given in
Equation (19), non-diagonal, non-zero entries of the form ij(d) containing Poisson’s ratios
and damage parameters will be obtained in the damaged stiffness matrix. By means of the
Poisson’s ratios, the relationship between the normal stresses resp. strains is defined. Subse-
quently, these non-diagonal entries must also be adjusted taking damage into account: ij =
ij(d) [13].
It is safe to assume that with increasing damage also the Poisson’s ratios will degrade.
However, except for [17], no literature is known to the author where the evolution of the Pois-
son’s ratios during a test was measured. Linear degradation as shown in Figure 4 or exponen-
tial degradation could be implemented. Here, the chosen Poisson’s ratios have the value of the
damaged Poisson’s ratios already at the beginning of the modelling, see Figure 4. Therefore,
the normal damage parameters d0 and d90R/T are considered to be decoupled.

Figure 4: Degradation of Poisson’s ratios

7 MESH REGULARISATION
Every continuum mechanics approach with softening will suffer from mesh dependency
problems. The source of mesh dependency is of mathematical nature as with the onset of sof-
tening, the previously well-posed problem has turned into an ill-posed problem. In computa-
tions, this results in a stiffness matrix which is no longer positive definite. The mathematical
solution is a localisation zone of zero width without energy dissipation. The numerical solu-
tion tries to capture this physically inadmissible mathematical solution which yields a local-
ised zone of smallest possible width, i.e. a single element in most cases.
In literature, many different regularisation methods are available [e.g. 18]. Here, the crack
band method as described in [19] was chosen to regularise the mesh. With the crack band
method, the fracture energy is expressed in terms of characteristic finite element length h
where h is a geometrical value in [length] containing information on the element’s aspect ratio:
Gf
gf  (21)
h
In equation (21), the fracture energy Gf needs to be replaced by the characteristic fracture
energy gf if the crack band model is activated. The introduction of a characteristic fracture en-
ergy gf that is correlated to the element size provides the transformation of the fracture energy
Gf into a ‘mesh-dependent’ value. For instance, in a coarse mesh, h will be large and leads
hence to a small characteristic fracture energy gf in comparison to the large gf of a fine mesh
with a small characteristic element length h. This adjustment of the fracture energy consider-
ing the mesh size compensates for the trend of continuum softening models to dissipate as
less energy as possible.
However, the crack band model only works if one failure mode is dominating and if a lo-
calised solution occurs. It is valid if damage develops only in a band of elements and not in all

8
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

elements homogeneously. Therefore, the decision whether to activate the crack band model or
not must be carefully taken for each modelling problem.
Figure 5 shows load-slip graphs of a simple tension beam model where locally, slightly
lower tension strength was assigned in order to trigger mesh dependency. A model with a lo-
calised softening zone was thus created and the crack band model should alleviate mesh de-
pendency. It can be seen that the mesh dependency obtained without regularisation technique
(Figure 5a.) is much weaker when the crack band model is activated, see Figure 5b. [further
information see 16]
25 25
Ux = Uz = 0
10 elements 10 elements
Ux = 0 80 elements 80 elements
640 elements 640 elements
20 20
default material

Ux = Uy = Uz = 0
15 15
Load [N]

Load [N]
weaker material

10 10

Y
Z X
5 5
Ux = 0.5mm

0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
Displacement [mm] Displacement [mm]

a. b.
Figure 5: a. Model without regularisation, b. Crack band method activated

8 VISCOUS STABILISATION
Viscous stabilisation similarly to [12] is used in order to improve the convergence charac-
teristics of the developed material model. A fictitious viscous parameter  is introduced in the
model. Due to the additional viscous component, the stiffness matrix is generally positive def-
inite. Therefore, viscous stabilisation leads to a more robust solution process with less con-
vergence problems.
As viscosity is a time-dependent material parameter, viscous stabilisation must be a func-
tion of the rate of damage. The rate of the damage variable or of the damage threshold (= fail-
ure criteria) may be used. Equation (22) shows the stabilisation as a function of rate of
damage variables:

d 
 d  dV  (22)

In Equation (22), the viscous parameter  defines the rate at which the true damage d and
the stabilised damage dV as defined in Equation (23) approach each other. Equation (22) can
now be discretised in time. Equation (23) shows the discretised equation based on the back-
ward Euler algorithm to insert an artificial viscosity  that is acting on the damage variables.
  dt 
dVt  max 0, dVt 1 , dVt 1  dt  (23)
   dt   dt 
As viscous stabilisation is applied to improve convergence, the energy output must be con-
trolled in order to judge the model performance and reliability, i.e. to control that fictitious
viscosity is not influencing modelling results. To this scope, the total dissipated energy and
the dissipated energy associated with viscous regularisation are calculated as internal state
variables. Both variables do not have any influence on the solution. They are merely used for
energy output, control and post processing.

9
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

9 MATERIAL PROPERTIES
A major issue of all material models is the need of mechanical input parameters such as
stiffness and strength values which usually derive from tests. The developed wood model
needs seventeen properties as listed in Table 1 for three different wood species.
Generally, stiffness values are rather easily assembled with a satisfying reliability whereas
already seemingly easy parameters as uniaxial strength values can be procured only with dif-
ficulty. This is due to two main issues. Firstly, the inherent large scatter of mechanical proper-
ties for wood and secondly, difficulties connected with testing and measuring. For instance,
the uniaxial shear strength can hardly be assessed without triggering stress peaks or secondary
stresses [20]. Furthermore, not always all parameters are measured, e.g. due to Poisson effect,
or the positioning of the measuring instruments is not clear.
Also the fracture energies suffer from a large scatter and the fact that they are usually
measured on small clear wood specimens whereas for a heterogeneous material, it can be
safely assumed that fracture energies are not constant. Moreover, the needed material values
are rather easily procurable for softwoods, but not for hardwoods.
In [16], a thorough literature study has been carried out where the issues around reliable
material parameters are discussed and the chosen values from Table 1 are motivated.

Para- Spruce Beech Azobé Variation A


Units
meter (Picea abies ) (Fagus sylvatica ) (Lophira alata ) Beech
E11 11000 13000 20000 13000
E22 = E33 370 860 1330 860
G12 = G13 690 810 1250 810
G23 50 59 91 59
ft,0 24 41 72 41
MPa
fc,0 36 45 58 45
ft,90 0.7 1.0 1.0 10
fc,90 4.3 14.2 23.2 14.2
fv 6.9 6.9 8.6 10
froll 0.5 0.5 0.6 10
Gf,0 60 100 180 100
Gf,90 0.5 0.71 0.71 50
N/mm
Gf,v 1.2 1.2 1.5 50
Gf,roll 0.6 0.6 0.7 10
 - 0.0001 0.0001 0.0001 0.0001
Table 1: Material properties for spruce, beech and azobé (Variation A Beech needed for later modelling)

10 FLOW DIAGRAM OF SUBROUTINE


The developed material model was programmed in a user subroutine UMAT and inserted in-
to the commercial FE software ABAQUS®.
Figure 6 shows the flow diagram of the developed CDM material subroutine. The subrou-
tine is given in [21].
The chosen simple constitutive law with explicit mechanical properties facilitates under-
standing and inspection of the models. No smeared input values need to introduced and their
effect can be directly translated to the model results.

10
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

Read actual total deformations


at end of increment t
tot t-1 t
kl = kl + kl

Calculate predictive elastic


stress at end of increment t
ef tot
ij = Dijkl· kl

Check failure initiation


ef
F(fmax, material, ij ) ≥1

YES NO

Calculate damage parameters


dij = 0
dij with: 0≤dij<1
and ∂dij/∂t≥0

Update stresses
t
ij = inv(Cijkldam)· kl
tot

Figure 6: Flow diagram of CDM

The chosen CDM approach allows for identification of failure modes and visualisation of
damage. For instance, progressive damage during a loading process is easily visualised in the
post processing. Ductile and brittle failure modes can be combined in a 3D material model
and mixed failures combining ductile and brittle modes occurring for instance under loading
with an angle to the grain can be captured. Also early damage in e.g. embedment tests due to
transverse or shear stress components can be captured while the global load-slip behaviour of
the model is still ductile.

11 MODELLING

11.1 Validation
When developing constitutive laws, the material model needs to be validated on single el-
ement models or on models with few elements. For this purpose, different model validations
were carried out among which compression on a cube with different angles to the grain.
Figure 7 shows modelling results on spruce wood using the developed UMAT together with
the FE model (here with 125 linear brick elements with full integration). The crack band
model was deactivated as no local crack band is expected and therefore, the crack band model
is not valid. As can be observed in Figure 7, compression parallel- and perpendicular-to-grain
showed elastic perfectly plastic behaviour as implemented whereas for 22.5° and 45°, brittle
behaviour due to the influence of brittle shear damage laws could be observed.
Figure 8 shows the overlap of the Hankinson equation [22] with FE results. The load carry-
ing capacity can be satisfyingly predicted with a maximum difference between FE model and
Hankinson equation of 15%.
Figure 7 and Figure 8 underline the prediction capability of the CDM model. Not only load
carrying capacities can be predicted, but also the general stress-strain behaviour with com-
bined ductile and brittle failures. Especially the prediction of the different load-slip behaviour
between the models loaded at an angle to the grain of 22.5°, brittle, and 45°, more ductile, is
good.

11
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

Further validations were carried out with different mesh sizes and material orientations,
different loadings from reversed cyclic loading to combined loading of normal stresses and
shear. All validations are shown in [16] and confirmed the correct implementation and calcu-
lation of the subroutine.
60

Stick−slip,
all other BC: Ux=0 n = 0.5

50
Ux=Uz=0

Load [kN] 40

0 degrees
90 degrees
30 Steelplates

Ux=Uy=Uz=0
22.5 degrees
Wood,
UMAT
45 degrees
Ux=−20mm
20 Y

Z X

10

0
0 0.5 1 1.5 2 2.5
Displacement [mm]

Figure 7: Results FE model for compression with different angles to the grain, spruce
40

35 Hankinson
Compression strength [MPa]

30 FE 1 element

25

20

15

10

0
0 10 20 30 40 50 60 70 80 90

Angle  to the grain

Figure 8: Superposition of FE results with Hankinson equation, spruce

11.2 Embedment models


Embedment models were developed whose outcomes could be verified with tests taken
from literature [23]. Three different wood species were chosen, spruce, beech and azobé. The
material properties are given in Table 1. The FE model is shown in Figure 9, only a quarter of
an embedment specimen with a 24 mm dowel was modelled. Also here, no localised damage
is expected and the crack band model was deactivated. 3D solids with linear interpolation be-
tween the eight integration points were chosen. Friction between dowel and wood was simu-
lated with the stick-slip model and a friction coefficient of  = 0.5.
Figure 10 shows exemplarily the modelling outcomes of embedment models in comparison
with test results. In Figure 10, damage due to compression parallel-to-grain directly under-
neath the dowel of a spruce specimen can be clearly seen. Also brittle damage due to tension
perpendicular-to-grain and shear is identified. Both failure modes are coupled as defined in
Equation (5). Furthermore, two elements are indicated in Figure 10 whose integration point
results in terms of tension perpendicular-to-grain and longitudinal shear are given in Figure 11.
It can be seen that damage in element a initiated due to high shear stresses and in element b,
damage was triggered by high tension stresses perpendicular-to-grain. Both predictions are
correct if the position of the elements in the model is considered.

12
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

Symmetry plane

wood,
linear elastic

Symmetry plane

Ux=5mm,
displacement for load−slip
Steel
wood,
UMAT
Y

Z X
Ux=Uy=Uz=0,
reaction force
for load−slip

Figure 9: Quarter of embedment model with boundary conditions, materials and default mesh
1
1
1
1
1
1
1
1
1
1
2
element b
0
0

element a

a. b. c.
Figure 10: Embedment of spruce with 24mm dowel, a. damage variable dc,0 in compression parallel-to-grain,
b. test result, c. damage variable dt,90 in tension perpendicular-to-grain
2.5

22 22
6.0
S, S22 PI: TIMBER−1 E: 1662 IP: 7 S, S22 PI: TIMBER−1 E: 1670 IP: 1

12 12
S, S12 PI: TIMBER−1 E: 1662 IP: 7 S, S12 PI: TIMBER−1 E: 1670 IP: 1
5.0
a [MPa] a

a [MPa] b

2.0
element

element

4.0
1.5
element

element
Stress[MPa]

[MPa]

3.0

1.0
Stress
Stress

Stress

2.0

0.5
1.0

0.0 0.0
0.00 0.20 0.40 0.60 0.80 0.00 0.20 0.40 0.60 0.80
Time
Time Time
Time

a. b.
Figure 11: Stress components 22 and 12; a. element a; b. element b (Figure 10)

Figure 12 shows load-slip graphs of azobé specimens in overlap with the modelling result.
The stiffness prediction is good. The reached displacement was too low. However, the source
of the brittle failure observed in Figure 12 at a displacement of less than 1 mm is purely nu-
merical and is not expressing the global splitting of the embedment model. The load drop of
the simulated load-slip graph represents a numerical failure with no physical meaning other
than that the wood underneath the dowel is collapsed completely being thus an artificial sof-
tening. Once the row of elements directly underneath the dowel fails, spurious energy modes
develop in the used elements. The load cannot be transferred to neighbouring elements as they
are already unloading although they may not have reached their ultimate load carrying capaci-
ty yet. Basically, the model fails completely on a local level of collapsed elements whereas
the neighbouring elements still can carry load.

13
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

80

70

Embedment strength [MPa]


60

50

40

30

20

10 model
tests
0
0 1 2 3 4 5
Displacement [mm]

Figure 12: FE result versus test results of embedment tests on azobé with 24mm dowels

Accordingly, there are three main options to improve the model performance and to avoid
local early brittle failures due to spurious energy modes:
 Introduction of a threshold value for the damage variables in order to avoid complete
damage and to keep a residual stiffness. This however is not thought to be a good solu-
tion as an uncontrollable fictitious parameter would be introduced.
 Change resp. increase mechanical properties in order to avoid early complete damage of
elements.
 Use different element formulations or arbitrary Lagrangian-Eulerian (ALE) approaches
[i.e. 24] to control excessive mesh distortion.
Point three is part of future research as then, the material subroutine will be optimised and
solutions will be developed to control excessive element distortion.
The second option however was carried out in order to understand the influence of the ma-
terial parameters. Also considering the rather difficult determination of material properties, it
is thought to be admissible to modify these parameters. Table 1 gives modified material pa-
rameters for beech wood nominated “variation A” where the properties controlling the 3D
brittle behaviour are increased. Figure 13 shows the modelling results for an embedment test
specimen with the default properties for beech and variation A superposed with test results.
The load-slip behaviour applying the default properties shows again too brittle behaviour due
to numerical softening. However, “variation A” reaches higher deformations before failure.
The higher values for strength and fracture energies as defined for “variation A” lead to nu-
merical softening at higher deformations which is more realistic.
As the resulting curve for “variation A” still yields too high strength values, also the com-
pression strength parallel-to-grain is reduced as indicated in Figure 13. Such a reduction is
thought to reduce the load carrying capacity and should therefore lead to a lower plateau.
Modelling outcomes confirmed this. A better agreement between tests and model can thus
be obtained purely by modifying clearly recognisable mechanical properties.
Another important outcome of modelling can also be deduced from Figure 13. The earlier
initiation of damage in the default model with a decrease in stiffness in comparison to “varia-
tion A” is clearly observable in the detail. The global stiffness is thus decreasing although the
model still shows a very clear load increase. This is a realistic prediction.

14
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

60

50
40

Embedment strength [MPa]


40

30 30

20 model
variation A
variation A with f = 40 MPa
10 c,0
variation A with fc,0= 35 MPa
20
tests
0
0 1 2 3 4 5
Displacement [mm]
10
Figure 13: FE result versus test results of embedment tests on beech with 24mm dowels,
material parameters modified – variation A, with detail

11.3 Joint models


Finally, double-shear timber joint models with slotted-in steel plates and dowel-type fas-
teners were developed. In comparison to embedment models, the dowels do not remain
straight in the joint models, but develop plastic hinges instead. An exemplary joint model with
one dowel is shown in Figure 14 (linear 3D solid elements). Figure 15 and Figure 16 show
typical modelling results using the material properties for spruce given in Table 1.

Ux = 20mm

symmetry plane

displacement for
load-slip curve

wood,
linear elastic
wood,
UMAT
steel dowel

symmetry plane Ux = Uy = Uz = 0,
reaction force for
load-slip curve

a. b. c.
Figure 14: Typical model with one dowel, default mesh and boundary conditions;
a. overview, b. symmetry plane in width, c. symmetry plane in thickness

Z Z
X X
Y Y

SDV11
(Avg: 75%) S, S33
+1.560e+00 (Avg: 75%)
+1.000e+00
+9.167e−01 +5.004e+02
+8.333e−01 +4.035e+02
+7.500e−01 +3.065e+02
+6.667e−01 +2.096e+02
+5.833e−01 +1.126e+02
+5.000e−01 +1.566e+01
+4.167e−01 −8.130e+01
+3.333e−01 −1.783e+02
+2.500e−01 −2.752e+02
+1.667e−01 −3.722e+02
+8.333e−02 −4.691e+02
+0.000e+00 −5.661e+02
−1.316e+00 −6.631e+02

a. b.
Figure 15: Spruce joint with one 24 mm high strength steel (hss) dowel at last increment;
a. damage variable dc,0, b. Stress component 33 in axial dowel direction

15
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

200 400

180
350
160
300
140
250
Force [kN]

Force [kN]
120

100 200

80
150
60
100
40 model S24.3 with E =13000MPa
11
50 model S24.3 with E =4000MPa
20 model S24.1 11
tests S24.1 tests S24.3
0 0
0 1 2 3 4 5 0 1 2 3 4 5
Displacement [mm] Displacement [mm]

a. b.
Figure 16: Spruce joints with 24 mm very high strength steel (vhss) dowels, test results and model
a. one dowel, b. three dowels, second model with reduced Young’s modulus parallel-to-grain

In Figure 15a., modelling results in terms of damage variable dc,0 are shown where the fi-
bre crushing underneath the dowel can be observed. Figure 15b. shows the stress component
33 to visualise the dowel bending (material properties hss fy = 563 MPa, fu = 590 MPa,
 = 0.3).
Figure 16a shows the comparison between test results taken from [16] and the model for a
spruce joint with one 24 mm vhss dowel (material properties vhss fy = 1311 MPa,
fu = 1389 MPa,  = 0.3). The numerical results were offset by 0.5 mm to exclude the initial
slip observed in tests. The model prediction is good in terms of load carrying capacity, stiff-
ness and ductility. However, the calculations stopped too early due to the extensively dis-
cussed spurious energy modes and the reached ductility is hence less than the one observed in
tests.
In Figure 16b., the comparison between tests (taken from [16]) and models for joints with
three 24 mm vhss dowels is shown. The model predicted the tests outcomes satisfyingly well,
both in terms of load carrying capacity and shape of the load-slip curve with the saw-tooth
shape indicating damage in the perpendicular and shear directions. The predicted stiffness is
too high. If a longitudinal stiffness value of E11 = 4000 MPa is taken, the stiffness prediction
quality improves. However, the prediction of the load carrying capacity decreases.

12 CONCLUSIONS
A promising constitutive 3D model for the material wood was developed that can identify
failure modes and combine simultaneous ductile and brittle failures within one model. The
mechanical material parameters needed for the constitutive relationship are clearly defined.
The material model runs in a complex FE environment in combination with other material
models and contact formulations.
Modelling results were satisfying in terms of stiffness and load carrying capacity. Artificial
softening caused by spurious energy modes of completely collapsed elements however led to
early brittle failure of models that did not represent physical failure. Furthermore, once un-
loading has started in the most stressed elements, the load is not transferred to neighbouring
elements. These numerical problems need to be solved in order to optimise the subroutine.
As for any other modelling approach, a major issue lies in the determination of the neces-
sary mechanical properties. As the model performance and prediction capacity is highly de-
pendent on these properties, methods to derive reliable values must be developed.

16
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

REFERENCES
[1] M. Schmid: Anwendung der Bruchmechanik auf Verbindungen mit Holz. PhD thesis,
Technische Universität Karlsruhe, 2002.
[2] M. Ballerini, M. Rizzi: A numerical investigation on the splitting strength of beams
loaded perpendicular-to-grain by multiple dowel-type connections. CIB-W18 Meeting
38. Karlsruhe, Germany, Paper 38-7-1, 2005.
[3] F.K. Wittel, G. Dill-Langer, B.H. Kröplin: Modeling of damage evolution in soft-wood
perpendicular to grain by means of a discrete element approach. Computational
Materials Science, 32 (2005), 594-603.
[4] T. Reichert: Development of 3D lattice models for predicting nonlinear timber joint
behaviour. PhD thesis, Napier University Edinburgh, 2009.
[5] A.M.P.G. Dias, J.W.G. van de Kuilen, H.M.P. Cruz, S.M.R. Lopes: Numerical
modelling of the load-deformation behavior of doweled softwood and hardwood joints.
Wood and Fiber Science, 42 (2010), 480-489.
[6] B.H. Xu, M. Taazount, A. Bouchaïr, P. Racher: Numerical 3D finite element modelling
and experimental tests for dowel-type timber joints. Construction and Building
Materials, 23 (2009), 3043-3052.
[7] A. Bouchaïr, A. Vergne: An application of the Tsai criterion as a plastic-flow law for
timber bolted joint modeling. Wood Science and Technology, 30 (1995), 3-19.
[8] M. Fleischmann: Numerische Berechnung von Holzkonstruktionen unter Verwendung
eines realitätsnahen orthotropen elasto-plastischen Werkstoffmodells. PhD thesis,
Technische Universität Wien, 2005.
[9] M. Grosse: Zur numerischen Simulation des physikalisch nichtlinearen Kurzzeit-
tragverhaltens von Nadelholz am Beispiel von Holz-Beton-Verbundkonstruktionen.
PhD thesis, Bauhaus-Universität Weimar, 2005.
[10] J. Schmidt, M. Kaliske: Zur dreidimensionalen Materialmodellierung von Fichtenholz
mittels eines Mehrflächen-Plastizitätsmodells. Holz als Roh- und Werkstoff, 64 (2006),
393-402.
[11] K. Hofstetter, C. Hellmich, J. Eberhardsteiner: Development and experimental
validation of a continuum micromechanics model for the elasticity of wood. European
Journal of Mechanics A/Solids, 24 (2005), 1030-1053.
[12] P. Maimí: Modelización constitutiva y computa-cional del daño y la fractura de
materiales compuestos. PhD thesis, Universitat de Girona, 2006.
[13] A. Matzenmiller, J. Lubliner, R.L. Taylor: A Constitutive Model for Anisotropic Dam-
age in Fiber-Composites. Mechanics of Materials, 20 (1995), 125-152.
[14] S.W. Tsai, E.M. Wu: A general theory of strength for anisotropic materials. Journal of
Composite Materials, 5 (1971), 58-80.
[15] Z. Hashin: Failure criteria for unidirectional fiber composites. Journal of Applied
Mechanics, 47 (1980), 329-33.
[16] C. Sandhaas: Mechanical behaviour of timber joints with slotted-in steel plates. PhD
thesis, University of Technology Delft, in preparation.

17
C. Sandhaas, J.W.G. van de Kuilen, and H.J. Blass

[17] S. Franke: Zur Beschreibung des Tragverhaltens von Holz unter Verwendung eines
photogrammetrischen Messsystems. PhD thesis, Bauhaus-Universität Weimar, 2008.
[18] L.J. Sluys: Wave Propagation, Localisation and Dispersion in Softening Solids. PhD
thesis, Technische Universiteit Delft, 1992.
[19] Z.P. Bažant, B.H. Oh: Crack band theory for fracture of concrete. Matériaux et
Constructions, 16 (1983), 155-177.
[20] D.M. Moses, H.G.L. Prion: Stress and failure analysis of wood composites: a new
model. Composites Part B-Engineering, 35 (2004), 251-261.
[21] C. Sandhaas: 3D Material Model for Wood, Based on Continuum Damage Mechanics.
Stevinrapport 6-11-4, Stevin II Laboratory, University of Technology Delft, 2011.
[22] R. L. Hankinson: Investigation of crushing strength of spruce at varying angles of grain.
U. S. Air Service Information Circular Vol. 3 No. 259, 1921.
[23] C. Sandhaas, J.W.G. van de Kuilen, H.J. Blass, G. Ravenshorst: Embedment tests
parallel-to-grain and ductility in tropical hardwood species. Proceedings WCTE. Riva
del Garda, Italy, 2010.
[24] A. Rodríguez-Ferran, A. Pérez-Foguet, A. Huerta: Arbitrary Lagrangian-Eulerian (ALE)
formulation for hyperelastoplasticity. International Journal for Numerical Methods in
Engineering, 53 (2002), 1831-1851
Finite Element Software package:
ABAQUS® Version 6.8, Dassault Systèmes Simulia Corp., Providence, RI, USA, 2008.

18

View publication stats

Potrebbero piacerti anche