Published as: Sirumbal-Zapata, L., Málaga-Chuquitaype, C., Elghazouli, A. (2018).

A three-dimensional plasticity-damage constitutive model for

timber under cyclic loads. Computers and Structures, Vol. 195, pp. 47–63. https://doi.org/10.1016/j.compstruc.2017.09.010
BibTeX:
@article{sirumbal2018three,
title={A three-dimensional plasticity-damage constitutive model for timber under cyclic loads},
author={Sirumbal-Zapata, Luis F and M{\'a}laga-Chuquitaype, Christian and Elghazouli, Ahmed},
journal={Computers \& Structures},
volume={195},
pages={47--63},
year={2018},
}

A three-dimensional plasticity-damage constitutive model for

timber under cyclic loads

Luis F. Sirumbal-Zapataa , Christian Málaga-Chuquitaypea,⇤, Ahmed Y. Elghazoulia

a Department of Civil and Environmental Engineering, Imperial College London, UK

Abstract
The performance of timber structures is governed by the nonlinear response at their
connections, where high deformation levels and stress concentrations are developed, par-
ticularly when subjected to load reversals. To date, no constitutive model for wood
under cyclic load exists which is able to incorporate its most important failure modes
while considering plastic deformations and cyclic sti↵ness and strength degradation si-
multaneously. This paper presents the formulation and implementation of a plasticity-
damage model with these characteristics within a continuum mechanics approach. The
theoretical framework of both plasticity and damage models is described, and a detailed
derivation of the constitutive equations required for their computational implementation
and coupling as well as the return mapping and iterative algorithms for their integra-
tion are presented. The damage evolution process is handled by two independent scalar
variables for tension and compression. A general orthotropic plasticity yield surface with
isotropic hardening is employed to incorporate timber plastic flow in compression. A
closed-form expression for the plasticity-damage consistent tangent operator is derived.
It is demonstrated that the proposed constitutive model captures all the key character-
istics required for an accurate modelling of timber under large deformation levels until
failure.
Keywords: Timber, Plasticity, Continuum damage mechanics, Orthotropy, Cyclic
loading, Numerical algorithm

1. Introduction

The structural performance of timber structures is governed by the nonlinear response

in the connection zones, where high deformation levels and stress concentrations are de-
veloped around the fasteners (nails, dowels or bolts). These zones are also susceptible
5 to significant load-reversals during the service life of structures, especially during ex-
treme scenarios such as earthquakes or strong winds. For this reason, performance-based
assessment of timber structures requires the implementation of a material constitutive

⇤ Correspondingauthor
Email address: c.malaga@imperial.ac.uk ( Christian Málaga-Chuquitaype)

Preprint submitted to Computers and Structures August 24, 2017

 model capable of simulating the nonlinear behaviour of timber under large deformation
levels and load reversals until failure.
10

Timber failure modes can broadly be defined as ductile failure due to compression
stresses and brittle failure due to the combination of shear and tension stresses. There-
fore, in addition to the anisotropy of timber, the numerical modelling of this material
involves the challenge of reproducing completely di↵erent failure modes and nonlinear
15 responses for tensile and compressive stress regimes. Although approximate phenomeno-
logical modelling has been previously attempted [1, 2, 3, 4], a rigorous three-dimensional
material constitutive model which is able to account for cyclic actions is lacking. Building
upon previous developments available for other quasi-brittle materials such as concrete
[5, 6, 7], this paper advances a consistent and detailed 3D plasticity-damage material
20 constitutive model for wood which is able to simulate its key failure modes.

The plasticity component of the model simulates the ductile nonlinear behaviour and
permanent deformation of timber under compressive stresses. Besides, modelling of brit-
tle shear and tensile failure is based on Continuum Damage Mechanics (CDM) theory
25 which, although not suitable for explicit crack representation, allows the monitoring of
damage evolution and the identification of potential rupture zones through a smeared
continuous approach. The coupling of plasticity and damage models is particularly at-
tractive for materials with an inelastic behaviour characterized by the simultaneous oc-
currence of plastic flow and cracks formation. Moreover, plasticity-damage models are
30 able to reproduce the material sti↵ness degradation characteristic of cyclic loading prob-
lems subjected to extensive stress redistribution more accurately [7, 8]. Nevertheless,
in order to obtain reliable results, the thermodynamic consistency of the model should
be verified to ensure that energy is dissipated and to avoid the introduction of spurious
energy into the system [6].
35

In recent years, a few attempts have been made to develop a material constitutive
model for timber subjected to monotonic loading that can deal with both ductile and
brittle failure modes. However, to the authors’ knowledge, no research has been carried
out on timber constitutive models tailored to reproduce the cyclic response of wood. And
40 while most of the available timber models for monotonic loading employ plasticity theory
for failure in compression, two di↵erent approaches have been typically followed for shear
and tension related failure modes. The first approach [9, 10] is based on nonlinear frac-
ture theory for the development of cohesive zone models, and aims to simulate explicitly
the formation and growth of cracks. The major disadvantages of this method are the
45 complexity of the formulation and the practical difficulties associated with the determi-
nation of appropriate values for the input parameters. Furthermore, its implementation
is generally feasible only for discrete crack modelling, which requires an a priori defini-
tion of the cracking path severely limiting its applicability to simulate most timber joints
(commonly dowelled connections) subject to cyclic loading where the crack locations are
50 not known in advance. The second approach, and the one followed herein, is based on
CDM theory [11, 12, 13] and represents a practical alternative for the definition of a
plasticity-damage model for the analysis of timber structures subjected to load reversals.

Among the various existing plasticity models for anisotropic materials, Hill [14] and
2
 55 Ho↵man [15] have been frequently employed for modelling timber failure under mono-
tonic loads. For example, Kharouf et al. [16] employed Hill’s criterion to develop a 2D
elasto-plastic orthotropic model with anisotropic hardening, capable of simulating the
biaxial (perpendicular and parallel to the grain) behaviour of timber under compression
stresses only. Plastic softening and hardening was defined in the directions parallel and
60 perpendicular to the grain, respectively. Brittle failure modes (tension or shear) were not
considered by the authors and their model was employed in the analysis of the monotonic
response of timber bolted connections. Later, Xu et al. [12] developed a timber model
based on a combination of anisotropic plasticity with hardening for compression, and a
simplified continuous damage model for shear and tension. Hill’s yield criterion was used
65 for plasticity, whereas a modified version of the same criterion, considering tension and
shear stresses only, was used for the definition of the onset of damage. A simplification of
the damage evolution law was introduced through a direct reduction of the elastic mod-
ulus in the three material orthogonal directions. This plasticity-damage model was used
to study the behaviour of timber-steel dowelled joints subjected to monotonic tension
70 loads only. Previously, the same model had been used to study the embedding strength
of Glulam dowelled connections [17]. The model prediction accuracy achieved in both
studies was satisfactory. However, its lack of continuous damage evolution laws and
loading-unloading conditions render this model unsuitable for load reversal simulation.

75 On the other hand, Sandhaas et al. [13] implemented a constitutive model for wood
under monotonic loads based exclusively on CDM. The accuracy of this damage model
was evaluated against experimental results of timber specimens subjected to monotonic
tension, compression and dowel embedment. The model takes into account eight types
of brittle and ductile failure modes, each of them associated with a di↵erent failure crite-
80 rion. In spite of not including plasticity, the explicit definition of linear softening (tension)
and perfectly plastic (compression) damage evolution laws, as a function of an internal
threshold variable controlling the size of the damage surface, constitutes an important
contribution for the numerical modelling of wood. More recently, in order to study the
brittle failure modes of dowelled timber-steel connections subjected to monotonic tension
85 loading, a new timber model was developed by Khelifa et al. [11] within the framework
of plasticity coupled with CDM. This refined constitutive model incorporates the e↵ects
of orthotropic elasticity, anisotropic plasticity with isotropic hardening, isotropic ductile
damage, and large plastic deformations. Nonetheless, the model does not consider di↵er-
ent input strength parameters for tension and compression failure. Conversely, only one
90 set of average strength parameters are required to be calibrated for each particular prob-
lem. Moreover, only one Hill’s surface is used simultaneously as plasticity and damage
criteria, making both inelastic behaviours totally dependent on one another. All these
characteristics make this model also unsuitable for the study of the nonlinear response
of timber subjected to cyclic loading.
95

This paper describes the implementation of a 3D material constitutive model for wood
that is capable of reproducing its cyclic response and failure modes, through the cou-
pling of a general orthotropic plasticity model with isotropic hardening, and an isotropic
continuous damage model. To this end, the next section discusses the nonlinear ex-
100 perimental response and failure characteristics of wood under compression, tension and
shear stresses. This is followed by a presentation of the theoretical basis and constitu-
3
 tive equations of the elasticity and damage parts of the proposed model, respectively.
The general orthotropic plasticity formulation is developed and the computational as-
pects of the model implementation are discussed, including the coupled plasticity-damage
105 algorithm and the derivation of the algorithmic consistent tangent sti↵ness matrix. Sub-
sequently, the thermodynamic consistency of the model is verified and the ability of
the model to simulate the uniaxial cyclic response of timber in the directions parallel
and perpendicular to the grain is demonstrated. Finally, the experimental response of
a timber-steel dowelled connection subjected to cyclic loading is employed in order to
110 validate the proposed model, and general conclusions are outlined.

2. Background on the nonlinear behaviour of wood

Wood is an anisotropic material with di↵erent types of failure modes in shear, ten-
sion and compression depending on the loading direction relative to its grain alignment.
This means that the mechanical properties of wood (e.g. elastic moduli, shear moduli,
115 Poissons ratio and strength) vary with the directions and signs (i.e. whether in tension
or compression). In general, the three most failure modes are [12]: (i)ductile failure due
to compression parallel to the grain, (ii) ductile failure due to compression perpendic-
ular to the grain, and (iii) brittle failure due to shear parallel to the grain and tension
perpendicular to the grain.
120

Fig. 1 shows the compressive and tensile stress-strain curves of Scandinavian spruce
with a mean density of 430 kg/m3 obtained experimentally by Karagiannis et al. [3].
It can be appreciated from this figure that, for the direction parallel to the grain (Fig. 1a),
the behaviour in compression is approximately linear elastic until the compressive strength
125 is reached at around 40 MPa. After this point, a minor stress drop is produced followed by
a plastic plateau. Alternatively, the compressive stress-strain relationship in the direction
perpendicular to the grain (Fig. 1b) shows plastic behaviour with moderate hardening.
It is important to note that the compressive strength in the direction perpendicular to
the grain is less than 10% of the strength in the direction parallel to the grain. In the
130 case of tension (Figs. 1c and 1d), an initial linear elastic response is followed by a brittle
failure in both directions. Therefore, the post-elastic behaviour of wood in tension is
markedly di↵erent than that in compression due to absence of plastic deformation and
the sudden loss of strength at failure. Finally, even though the shear stress-strain curve
in the direction parallel to the grain is nonlinear, a brittle failure is also observed in this
135 case (Fig. 2). In light of the above discussion, timber failure modes can be summarized as
ductile elastic-plastic failure with large deformation for compressive stresses, and elastic
brittle failure for the interaction of shear and tensile stresses. Based on this conclusion,
a physically consistent constitutive model is proposed to study the nonlinear response of
3D timber structures subject to cyclic earthquake loading.

140 3. Orthotropic linear-elastic behaviour of timber

Before yielding (compression) or failure (tension and shear) is produced, the strain-
stress constitutive equation of wood is defined as:

✏ = Ce : (1)
4
 45
Upper bound 5
40
4.5
35 Upper bound
4
30 3.5
25
Stress [MPa]

Lower bound
3

Stress [MPa]
20 2.5 Lower bound

15 Individual Specimen
2
10 1.5 Individual Specimen
Envelope
5 1 Envelope
0.5
0
0 0.02 0.04 0.06 0.08 0.1 0
0 0.05 0.1 0.15 0.2
Strain [mm/mm]
Strain [mm/mm]

(a) Compression paralell to grain (b) Compression perpendicular to grain

40 0.7
Upper bound
35 0.6
30 0.5
25
Stress [MPa]

0.4
Stress [MPa]

Lower bound
20
0.3
15
0.2 Individual Specimen
10
0.1 Envelope
5
0
0
0 0.0002 0.0004 0.0006 0.0008 0.001 0.0012 0.0014
0 0.001 0.002 0.003 0.004 0.005
Strain [mm/mm] Strain [mm/mm]

(c) Tension paralell to grain (d) Tension perpendicular to grain

Figure 1: Experimental compressive and tensile stress-strain relationships of Scandinavian Spruce [3].

6
Upper
5 bound
Shear Stress [MPa]

4
Lower bound
3
Individual Specimen
2
Envelope
1

0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
Shear Strain [mm/mm]

Figure 2: Experimental shear stress-strain relationships of Scandinavian Spruce [3].

5
 where C e is the fourth-order orthotropic linear-elastic compliance tensor, defined as a
6 by 6 matrix in Voigt’s notation such that:
2 1 ⌫Y X ⌫ZX 3
EX EY EZ 0 0 0
6 ⌫XY 1 ⌫ZY
0 0 0 7
6 7
6 ⌫EX EY EZ
7
6 XZ ⌫Y Z 1
0 0 0 7
6 E E E 7
Ce = 6 X Y Z
1 7 (2)
6 0 0 0 0 0 7
6 GXY 7
6 1 7
4 0 0 0 0 GY Z 0 5
1
0 0 0 0 0 GZX

145 Accordingly, the inverse of C e is the fourth-order orthotropic linear-elastic sti↵ness tensor
D e , defined as a 6 by 6 matrix in Voigt’s notation as:

2 EX (1 ⌫Y Z ⌫ZY ) EX (⌫Y X +⌫ZX ⌫Y Z ) EX (⌫ZX +⌫Y X ⌫ZY )

3
0 0 0
6 7
6 EY (⌫XY +⌫ZY ⌫XZ ) EY (1 ⌫XZ ⌫ZX ) EY (⌫ZY +⌫XY ⌫ZX )
0 0 0 7
6 7
6 7
6 EZ (⌫XZ +⌫Y Z ⌫XY ) EZ (⌫Y Z +⌫XZ ⌫Y X ) EZ (1 ⌫XY ⌫Y X )
0 0 0 7
D =6
e
6
7
7
6 0 0 0 GXY 0 0 7
6 7
6 0 0 0 0 GY Z 0 7
4 5
0 0 0 0 0 GZX
(3)

=1 ⌫XY ⌫Y X ⌫Y Z ⌫ZY ⌫XZ ⌫ZX 2⌫Y X ⌫ZY ⌫XZ (4)

The orthotropic directions X, Y and Z correspond to the longitudinal, transverse
and radial timber local axes, respectively. It is assumed that the longitudinal axis runs
parallel to the grain of the timber material, while the transverse and radial axes lay in the
150 cross-section plan and act in the direction perpendicular to the grain. Notice that due
to timber anisotropy the values of the elastic moduli, the shear moduli and the Poissons
ratios are di↵erent for each of the three orthogonal axes or planes. However, due to the
symmetry property of C e and D e , the following conditions hold:
⌫XY ⌫Y X ⌫Y Z ⌫ZY ⌫XZ ⌫ZX
= ; = ; = (5)
EX EY EY EZ EX EZ
Consequently, in order to define the linear-elastic orthotropic behaviour of timber, a
155 total of nine mechanical material parameters are required: three elastic moduli, three
shear moduli and three Poissons ratios. If transverse isotropy is assumed the number of
independent parameters can be reduced to five (EX , EZ , GZX , ⌫XZ , ⌫ZY ), by means of
the following additional set of relationships:
EZ
EY = EZ ; GXY = GZX ; ⌫XZ = ⌫XY ; GY Z = (6)
2(1 + ⌫ZY )

6
 4. Strain-based isotropic damage model for timber

160 Brittle failure due to tension and shear stresses generates voids and micro cracks in
the timber matrix which not only lead to a sudden reduction of the material strength,
but also cause a gradual degradation of its mechanical properties, including its sti↵ness.
When further loads are applied, the micro-cracks grow and their coalescence produce
macro-cracks zones and irreversible damage [11, 18]. CDM, based on the Thermody-
165 namics of Irreversible Processes theory, has been widely used for modelling the nonlinear
behaviour of di↵erent brittle materials, like concrete, rock [19], and more recently, tim-
ber [13]. CDM strain-based damage models rely on the concept of E↵ective Stress and
the hypothesis of Strain Equivalence. The former is defined as the stress acting in the
reduced undamaged net surface area of the material, without considering the portion
170 of area taken by the micro-cracks and voids. Taking into account that the total force
acting in the material body is constant, the magnitude of the e↵ective stress acting in
the reduced undamaged area is higher than the magnitude of the Cauchy stress acting
over the total nominal surface area. On the other hand, the hypothesis of strain equiv-
alence states that the strain associated with the Cauchy stress in the damaged state is
175 equivalent to the strain associated with the e↵ective stress in the undamaged state [19].

The e↵ective stress tensor, ¯ , is transformed into the Cauchy stress tensor, , by
means of the fourth-order tensor M , which is a function of the damage tensor, D [20]:

= M (D) : ¯ (7)
Anisotropic damage is considered by assigning di↵erent values to the damage vari-
180 ables components of D, which can be defined as a second-order tensor, or more generally,
as a fourth-order tensor. However, in spite of being an anisotropic material, there are
two main disadvantages of including anisotropic damage in the timber material model,
a physical and a numerical one. First, the evolution laws for the damage variables of
the stresses terms in each orthotropic direction are not known and are difficult to obtain
185 through experimental tests. Second, the strain equivalence hypothesis is not valid for
anisotropic damage, and therefore, it is not possible to obtain a mechanically consistent
anisotropic damage tensor without losing the symmetry of either the Cauchy or the ef-
fective stress tensors. Given that, by definition, the Cauchy stress tensor is symmetric,
a non-symmetric e↵ective stress tensor would need to be utilised [21]. The reasons for
190 keeping the symmetry of the e↵ective stress tensor are explained below.

In light of above discussion, isotropic damage is considered herein as the best option
for the development of a mechanically consistent timber damage model. To this end, one
of the most common and successful techniques for modelling isotropic damage consists
195 in replacing the tensor M in Eq. 7 by a scalar expression [6, 7, 22, 23], which takes the
form:

= (1 !)¯ (8)
where ! is the scalar damage variable. The damage process starts with the fulfilment of
the damage criterion condition. Once this criterion is met, the value of the damage vari-
able increases gradually and monotonically from 0 (undamaged state) to 1 (total damage
7
200 state). It is important to mention that the scalar damage model defined in Eq. 8 implies
that the degradation ratios of the elastic and shear moduli terms in Eqs. 2 and 3 assume
a value of (1 !). Conversely, it is assumed that no degradation occurs in the Poissons
ratios, remaining constant throughout the analysis. This is an inherent characteristic of
all scalar damage models [20] which is accepted for two reasons: first, it avoids the nu-
205 merical complexities of anisotropic damage models which have a documented restriction
on their applicability in engineering practice [24]; second, it evades the set of assumptions
on the Poisson ratios degradation rules necessary to keep the symmetry of the damage
operator M that lack physical or experimental justification when a complex anisotropic
damage approach is followed [25]. Thus, the relationships between the mechanical prop-
210 erties of the virgin and the damaged materials for the elastic moduli, the shear moduli
and the Poissons ratio are:

E d = (1 !)E 0 ; Gd = (1 !)G0 ; ⌫d = ⌫0 (9)

4.1. Spectral decomposition of the e↵ective tensor

A timber material constitutive model for cyclic loading should be capable of repro-
ducing the di↵erent inelastic responses of wood in tension and compression as described
215 in Section 1. This does not only include the di↵erent strengths and failure modes, but
also the di↵erent post-elastic response and sti↵ness degradation. The latter is particu-
larly important in order to adequately capture the unloading-sti↵ness degradation and
the sti↵ness recovery after load reversal when passing from tension to compression stress
states and viceversa [7, 25, 26]. For this reason, a split of the e↵ective stress tensor into
220 a tensile (¯ + ) and a compressive (¯ ) components is performed such that [27]:
3
X
¯+ = h¯i ipi ⌦ pi (10)
i=1

¯ =¯ ¯+ (11)
where pi and ¯i are the eigenvectors and eigenvalues of the e↵ective stress tensor, re-
spectively. In this way, it becomes possible to manage independent damage mechanisms
for tension and compression [5] and also to model indirectly the opening and closing
of cracks when the material is subjected to reversible loads [22]. In order to obtain the
225 principal values of the e↵ective stress tensor (¯i ) and their associated eigenvectors (pi ), a
spectral decomposition is performed. This justifies the selection of a symmetric e↵ective
stress tensor, and thus, of an isotropic damage model. The Macaulay brackets operator
h·i in Eq. 10 returns the positive values and sets the negative ones to zero. Thus, only the
eigenvectors associated to positive (tensile) principal stresses are retained. Finally, the
230 scalar isotropic damage relationship between the Cauchy and the e↵ective stress tensors
in Eq. 8 can be reformulated in terms of the tensile and compressive components as:

= (1 ! + )¯ + + (1 ! )¯ (12)

8
 4.2. Tensile and compressive damage criteria functions
The initiation of the damage evolution process (when the damage variable ! ± starts
its gradual growth from 0 to 1) is determined by the damage criteria function defined as:

fd± (¯ ± , r± ) = ⌧¯± r± (13)

235 This function is expressed in terms of an e↵ective stress-based norm, called equivalent
⌧ ± ), and a threshold variable (r± ) which controls the size of the damage surface.
stress (¯
Notice that Eq. 13 encompasses two damage functions with di↵erent equivalent stresses
and threshold variables for each independent tension and compression damage processes.
Due to the anisotropy, complex stresses interaction and failure characteristics of wood,
240 a multidimensional damage function based on Hill’s criterion [14] is adopted herein.
Hence, the associated equivalent stress of the Hill’s damage criterion is defined in Voigt’s
notation as:
r
± 1 ±T ± ±
⌧¯ = ¯ H ¯ (14)
2
where the equivalent stress, ⌧¯± , is defined in terms of the corresponding e↵ective stress
tensor, ¯ ± and the Hill’s strength parameters matrix H ± for both tension and com-
245 pression. The definitions of H ± and its strength parameters coefficients ↵ij ±
in terms
of the wood normal (tension or compression) and shear strengths for each of the three
±
orthotropic directions (fX , fY± , fZ± , fXY , fY Z , fZX ) are given as:
2 ± ± ± 3
↵11 ↵12 ↵13 0 0 0
6 ± ± ± 7
6 ↵21 ↵22 ↵23 0 0 0 7
6 7
6 ↵± ↵ ±
↵ ±
0 0 0 7
6 31 32 33 7
H± = 6 7 (15)
6 0 0 0 ↵44 0 0 7
6 7
6 7
4 0 0 0 0 ↵55 0 5
0 0 0 0 0 ↵66
8 ± 2 ± 2 ± 2
> ↵11 = 2 ; ↵22 = 2 ; ↵33 = 2
>
> (fX± ) (fY± ) (fZ± )
>
> 2 2
>
> ↵44 = (fXY )2
; ↵55 = (fY Z )2
; ↵66 = (f 2 )2
>
> ✓ ◆ ZX
>
>
< ↵± ±
= ↵21 = 1
+ 1 1
12 ± 2 2 2
✓ ( fX ) (fY± ) (fZ± ) ◆ (16)
>
>
>
> ±
↵13 = ±
↵31 = 1
+ 1 1
>
> ± 2
(fZ± )
2
(fY± ) ◆
2
>
>
> ✓ ( fX )
>
> ± ±
: ↵23 = ↵32 = 1
2 + 1
2
1
2
(fY± ) (fZ± ) (fX± )

4.3. Tensile and compressive damage evolution laws

The variation of the damage variable, ! ± , as a function of the threshold variable, r± ,
250 is determined by the damage evolution law defined as:

! ± = gd± r± (17)
9
 The mathematical expression of this monotonically increasing function is related to the
post-elastic stress behaviour (e.g. softening, perfect plasticity, hardening) by means of
Eq. 8. This relationship is more clearly expressed for a uniaxial stress state in terms of
the initial threshold variable, r0± , and the uniaxial strength, fmax
±
, as:

r± ±
±
= fmax 1 gd± r± (18)
r0±
255

The damage evolution law proposed by De Borst et al. [8], capable of modelling
di↵erent types of exponential softening, is assumed herein for simulating the tensile
post-elastic behaviour of wood. This law is expressed as [8]:

r0+ ⇣ b(r + r0+ )

⌘
gd+ r+ = 1 1 n + ne (19)
r+
260 where n is a calibration parameter and b is a constant dependent on the material me-
chanical properties and the finite element mesh size. The influence of n in the damage
evolution of the Cauchy tensile stress, + (Eq. 18), and the tensile damage variable,
! + (Eq. 19), is reproduced in Figs. 3a and 3b, respectively. Fig. 3b shows that n has
almost no influence on the growth rate of the damage variable. However, for n smaller or
265 equal to 1, an asymptotically exponential softening of the Cauchy stress is appreciated
(Fig. 3a). The lower the value of n, the higher the residual stress obtained even for
widespread damage levels. On the other hand, for values n > 1 there is no asymptotic
behaviour and the stress rapidly decays to zero. This resembles the wood behaviour
observed previously in Fig. 1 and justifies its choice.
270

The value of the constant b in Eq.19 depends on the energy dissipated by the material
during post-elastic deformation from the moment in which the maximum strength is
reached until a total failure is produced. The amount of energy dissipated is graphically
represented by the area under the softening curve in Fig. 3a, and is directly proportional
275 to the fracture energy density (Gf ), a mechanical property of wood which varies with
the loading direction and the type of failure. In light of the experimental response
discussed above, tensile failure perpendicular to the grain is assumed as the most frequent
brittle failure for wood. Therefore, the fracture energy density corresponding to Mode
I failure in the direction perpendicular to the grain (Gf,Y Z ) is adopted. Moreover,
280 in order to satisfy the requirement of mesh objectivity, Gf,Y Z needs to be normalized
by the characteristic length, lch , which is a parameter related to the size of the finite
element model discretization [5]. Hence, b can be expressed in terms of the wood material
properties in the direction perpendicular to the grain, through a parameter H defined
as:
2
fY+ Z ⇥ lch
H= (20)
2 ⇥ Gf,Y Z ⇥ EY Z

285 where fY+ Z and EY Z are the tensile strength and the modulus of elasticity, respectively,
in the direction perpendicular to the grain. Accordingly, depending of the value of n, the
constant b is defined as:
10
 0.5
Cauchy stress [Mpa]

n = 0.8 - Gf = 0.5
0.4
n = 1.0 - Gf = 0.5
0.3
n = 1.2 - Gf = 0.5
0.2

0.1

0
0 0.01 0.02 0.03 0.04
Total strain [mm/mm]
(a) Cauchy stress, +.

1
Damage variable

0.8
n = 0.8 - Gf = 0.5
0.6 n = 1.0 - Gf = 0.5
0.4 n = 1.2 - Gf = 0.5

0.2

0
0 5 10 15 20 25 30
Treshold variable
(b) Damage variable, ! + .

Figure 3: Damage evolution law for tensile softening.

11
 8 ⇥ ⇤
< For n A n 1 2H
1 b= r0+
1 + (n 1) ⇥ ln n ; A= 1 H
A H
(21)
: For n < 1 b= [(↵n + n 1) ⇥ ln(↵) + 2n(1 ↵)] ; A=
r0+ 1 n(1 ↵)H

where ↵ is a constant parameter. A detailed discussion on energy dissipation for tensile

brittle failure and the derivation of Eqs. 20 and 21 is o↵ered in Appendix A.
290

In the case of compressive post-elastic behaviour, the definition of the damage evolu-
tion law is straightforward. The typical ductile compressive failure of wood (Fig. 1) can
be accurately represented by a bilinear hardening behaviour. Nevertheless, perfectly plas-
tic linear damage evolution law has been frequently employed for timber [13, 28]. Based
295 on this perfectly plastic linear damage evolution function, a slightly modified expres-
sion with two additional calibration parameters which allow the definition of nonlinear
hardening post-elastic behaviour is proposed herein. Therefore, gd (r ) can be defined
as:
✓ ◆m
r0
gd r = ⇥ 1 (22)
r
where and m are the calibration parameters. The parameter m, which by definition
300 must be higher or equal to one, diminish the growth rate of the damage variable and
determines the order of the uniaxial post-elastic stress-strain curve. Fig. 4 shows the
influence of in the evolution of the compressive damage variable ! and the Cauchy
stress for a linear hardening post-elastic behaviour (m = 1). The slope of the linear
hardening diminishes with the increment of up to a maximum value of 1, for which a
305 perfect plastic behaviour is obtained (Fig. 4a). Moreover, Fig. 4b shows that diminishes
the maximum value of ! (maximum level of damage) in compression from 1 to .
Based on these considerations, and m should be calibrated with the experimental
monotonic (hardening slope) and cyclic (maximum sti↵ness degradation) tests on wood
specimens under compressive loads. It is noteworthy that wood does not show softening
310 behaviour in compression (Fig. 1). Therefore, in this case there is no need to incorporate
a fracture energy density property in the damage evolution law, nor a characteristic
length parameter to satisfy the mesh objectivity requirement.

4.4. Damage constitutive equations

The damage model is completely defined by the damage criteria function, Eq. 13,
315 the damage evolution law, Eq. 17, and the loading-unloading (Kuhn-Tucker) conditions
defined as:

fd±  0; ṙ± 0; ṙ± fd± = 0 (23)

Thus, from Eq. 23, in addition to define the start of the damage evolution process, the
damage function accomplishes two complementary tasks. First, it identifies the loading
and unloading states, and second, it defines the activation and deactivation of the damage
320 variable growth [26]. Finally, from Eqs. 13 and 23, it can be concluded that the value of
the damage threshold variables for a time t can be defined as:

r± = max r0± , max(0xt) ⌧¯x± (24)

12
 25
Cauchy stress [MPa]

20 = 0.8 - m = 1
= 0.9 - m = 1
15 = 1.0 - m = 1

10

0
0 0.05 0.1 0.15 0.2 0.25
Total strain [mm/mm]
(a) Cauchy stress, .

1
Damage variable

0.8

0.6
= 0.8 - m = 1
0.4 = 0.9 - m = 1
= 1.0 - m = 1
0.2

0
0 5 10 15 20 25 30
Treshold variable
(b) Damage variable, ! .

Figure 4: Damage evolution law for compressive hardening.

13
 5. General orthotropic plasticity with isotropic hardening model for timber

Two alternative approaches are possible for the coupling of plasticity and damage in
light of the definition of two parallel stress spaces (nominal and e↵ective) as presented
325 above, depending if the plasticity part is formulated in the nominal Cauchy stress space
or the e↵ective stress space. From a physical point of view, plasticity in compression
occurs in the material matrix, between voids and cracks, leading to local hardening be-
havior [29]. This means that the e↵ective stress tensor, which is defined to act only
over the undamaged material matrix, o↵ers a more consistent stress space for plasticity
330 formulation [19].

Also, numerical considerations are important when defining the plasticity stress space.
In order to obtain a unique solution (Local Uniqueness Condition), plasticity algorithms
based on nominal stresses require the introduction of strong hardening [6]. However,
335 Fig. 1 shows that slight hardening plasticity is the characteristic behaviour of timber
under compressive stresses. Therefore, a nominal stress based plasticity model, which
requires the introduction of strong hardening in order to be numerically stable, is not a
suitable alternative for wood. By contrast, hardening plasticity is not a requirement of
Local Uniqueness in the e↵ective stress space [6]. In light of this, the proposed model
340 employs an e↵ective stress space formulation. This configures a second important reason
for ensuring the symmetry of the e↵ective stress tensor, by coupling the plasticity model
with a scalar isotropic damage model (see previous Section). Therefore, based on the
characteristic strain tensor decomposition of plasticity theory, the relationship between
the e↵ective stress, ¯ , the total strain, ✏, the elastic strain, ✏e , and the plastic strain, ✏p ,
345 is defined as:

¯ = D e : ✏e = D e : (✏ ✏p ) (25)

5.1. General orthotropic yield function

The three components of the plasticity model are the yield function, the plasticity flow
rule and the hardening variable evolution law. The general orthotropic yield criterion
proposed by Oller et al. [30] in combination with with isotropic hardening is adopted
350 herein for the compressive component of the e↵ective stress tensor (¯ ) such that:
1 T T
¯ P (k)¯ + ¯ q(k) ¯y2 (k)
fp = (26)
2
where the reference yield stress, ¯y , the mapping matrix P and the mapping vector q are
functions of the isotropic plastic hardening variable, k (it should be noted that hereafter,
all the equations are formulated in Voigt’s notation, except when indicated otherwise):

2 3 2 3
" # 1 4 5 7 0 0
⌦ 0 6 7 6 7
P = ¯y2 (k) ;⌦ = 4 4 2 6 5; =4 0 8 0 5 (27)
0
5 6 3 0 0 9

⇥ ⇤T
q = ¯y2 (k) 10 11 12 0 0 0 (28)
14
355

The coefficients i (i = 1, ..., 12) are defined in terms of timber anisotropic yield
±
stresses fX , fY± , fZ± , fXY , fXY , fZX and the parameters XY , XZ , XY , X , Y and
Z such that:

8 2 2 2
>
> 1 = +
fX (k)·fX (k)
, 2 = fY+ (k)·fY (k)
, 3 = +
fZ (k)·fZ (k)
,
>
>
>
>
>
< 4 =p +
XY
, 5 =p XZ
, 6 =p YZ
,
fX (k)·fX (k)·fY+ (k)·fY (k) +
fX +
(k)·fX (k)·fZ (k)·fZ (k) fY+ (k)·fY (k)·fZ
+
(k)·fZ (k)

>
> 7 = 2
, 8 = 2
, 9 = 2
2 (k) ,
>
>
2
fXY (k) fY2 Z (k) fZX
>
>
>
: =p 2 X
, =p 2 Y
, =p 2 Z
10 + 11 12
fX (k)·fX (k) fY+ (k)·fY (k) +
fZ (k)·fZ (k)
(29)
The general formulation of Eq. 26 allows the use of several classic (Hill, Ho↵man, Tsai-
360 Wu) and other more recently proposed orthotropic yield functions through the definition
of the parameters and [30].

In agreement with the observed experimental behaviour (Fig. 1) and the material
model hypothesis adopted, the yield function is defined herein only for the compressive
365 part of the e↵ective stress tensor. Thus, the proposed model considers a damage-only
behaviour in tension with no plastic response. However, the incorporation of the tensile
+
yield stresses (fX , fY+ , and fZ+ ) is required for the definition of pressure-dependent yield
criteria [31]. To this end, linear hardening functions are assumed for each of the timber
yield stresses and the reference yield stress in terms of the isotropic hardening variable
370 k and the hardening moduli (h± ± ±
X , hY , hZ , hXY , hY Z , hZX , and h) such that:

( ± ±
fX = fX,0 + h±
X · k, fY± = fY,0
±
+ h±
Y · k, fZ± = fZ,0
±
+ h±
Z · k,
(30)
fXY = fXY,0 + hXY · k, fY Z = fY Z,0 + hY Z · k, fZX = fZX,0 + hZX · k

¯y = ¯y,0 + h · k (31)
And if proportional hardening is assumed in all directions for numerical analysis simpli-
fication [31]:

h±
X h±
Y h±
Z hXY hY Z hZX h
± = ± = ± = = = = (32)
fX fY fZ fXY fY Z fZX ¯y
which, combined with Eqs. 30 and 31 leads to:
± ± ±
fX,0 fY,0 fZ,0 fXY,0 fY Z,0 fZX,0 ¯y,0
± = = = = = = (33)
fX fY± fZ± fXY fY Z fZX ¯y
Finally, by replacing Eqs. 33 and 29 in Eqs. 27 and 28 for an initial reference yield stress
375 equal to one (¯y,0 = 1):

15
 2 3 2 3
" # 1,0 4,0 5,0 7,0 0 0
⌦ 0 6 7 6 7
P = ;⌦ = 4 4,0 2,0 6,0 5; =4 0 8,0 0 5 (34)
0
5,0 6,0 3,0 0 0 9,0

⇥ ⇤T
q = ¯y (k) 10,0 11,0 12,0 0 0 0 (35)
which renders P independent from the reference yield stress while q becomes a linear
(not quadratic) function of the reference yield stress. Moreover, the condition in Eq. 33
allows to keep the definition of the coefficients i,0 (as per Eq. 29), but now in terms
of the initial instead of the current yield stresses. This means that the coefficients i,0
380 become constants and independent from the hardening variable k. Accordingly, based
on the new definition of the mapping matrix and vector, the general orthotropic yield
function with isotropic proportional hardening becomes:
1 T T
fp = ¯ P¯ +¯ q(k) ¯y2 (k) (36)
2
5.2. Plastic flow rule and strain-hardening law
The flow rule for associated plasticity is given by:

✏˙p = ˙ n ¯ , k (37)

✓ ◆T
@fp ( ¯ , k)
n= =P¯ + q(k) (38)
@¯
385 where ˙ is the plastic multiplier and n is the tensor normal to the yield surface. Likewise,
the normalized strain-hardening evolution law is defined as:
p
k̇ = ˙ nT T n (39)
where T is defined as:

T = diag [1, 1, 1, 1/2, 1/2, 1/2] (40)

Finally, to complete the formulation of the plasticity model, the following loading-
unloading conditions are assumed:

fp  0; ˙ 0; ˙ fp = 0 (41)

390 5.3. Hardening modulus

The hardening modulus can be expressed as (from Eqs. 30 and 31):
8 ±
> dfX ± dfY± ±
±
dfZ ±
>
< dk = hX , dk = hY , dk = hZ ,
dfXY dfY Z dfZX (42)
= hXY , dk = hY Z , = hZX ,
>
> dk dk
: d¯y
dk = h
16
 Therefore, the hardening modulus is an input parameter of the plasticity model which
can be directly obtained from the uniaxial stress-strain diagram of the material in any
of the orthotropic directions. In the case of wood it is appropriate to define the value
395 of the hardening modulus using the compressive stress-strain curve in the direction per-
pendicular to the grain (Y or Z axes). And by noting that, in line with the plasticity
consistency condition, plastic flow can take place only if the value of the yield function is
equal to zero and remains constant for at least one instant of time (fp = 0 ^ f˙p = 0) [32],
the following plasticity condition can be obtained from Eq. 36 for a uniaxial compressive
400 stress state in the Z direction:
✓ ◆
˙ d¯y d¯y
fp = 3,0 · ¯Z · ¯˙ Z + 12,0 · ¯˙ Z · ¯y + ¯Z · k̇ 2¯y k̇ = 0 (43)
dk dk

By recalling the definition of hardening modulus (Eq.42), and re-arranging for the
particular case of uniaxial plasticity, the following expression for the hardening modulus
405 can be obtained:
⇣ ⌘
¯Z
12,0 + 3,0 ¯y @ ¯Z
h= ⇣ ⌘ · p (44)
12,0 · ¯Z
¯y 2 @✏Z

where:
q
2
¯Z 12,0 ( 12,0 ) +2 3,0
= (45)
¯y 3,0

And since ¯Z is negative and the reference yield stress is always ¯y 1, the ratio ¯¯Zy
corresponds to the negative root of the yield function (fp = 0). The substitution of the
coefficients 3,0 and 12,0 (Eq. 29) in Eqs. 44 and 45 leads to:
1 @ ¯Z
h= p (46)
fZ,0 @✏Z
410 Finally, the following expression can be used to calculate the derivative of the uniaxial
e↵ective stress with respect to the plastic strain in terms of the elastic (EZ ) and plastic
tangent (TZ ) moduli:
@ ¯Z EZ · T Z
p = (47)
@✏Z EZ T Z

5.4. Integration of the stress-strain rate equation

The backward Euler method is employed herein to integrate the stress-strain rate
415 equations. To this end, the finite increment counterpart of the plasticity flow rule in
Eq. 37 is expressed as:

✏p = n ¯ ,k (48)

17
 Correspondingly, Eq. 25 is also reformulated for the compressive component of the e↵ec-
tive stress as:

¯ = De ✏ De n ¯ , k (49)

420

Before the load increment is applied, the total strain increment, ✏ , the initial
e↵ective stress, ¯ j , and the hardening modulus, h, are known. The addition of ¯ j to
both sides of Eq. 49 makes it possible to obtain the final e↵ective stress, ¯ j+1 , as an
elastic predictor-plastic corrector process in which ¯ e is the elastic trial e↵ective stress
425 vector:
e
¯ j+1 = ¯ e j+1 D n ¯ j+1 , kj+1 ; ¯ e = ¯ j + D e ✏j+1 (50)
Substituting Eq. 38 into Eq. 50:

h 1
i 1
1
¯ j+1 = A ( j+1 ) De ¯e j+1 q(kj+1 ) ; A = De + j+1 P (51)

Besides, two additional equations can be derived from the hardening law (Eq. 39) and
the yield function (Eq. 36) such that:
q
kj+1 = j+1 nT ¯ j+1 , kj+1 T ¯ j+1 , kj+1 n ¯ j+1 , kj+1 (52)

1 T T
fp = ¯ P ¯ j+1 + ¯ j+1 q (kj+1 ) ¯y2 (kj+1 ) (53)
2 j+1
And the system of three equations (Eqs. 51 to 53) with three unknowns (¯ j+1 , kj+1
430 and j+1 ) is completely defined. This system can be reduced to one system of two
equations with two unknowns ( kj+1 and j+1 ):

8 p
>
> F = kj+1 j+1 nT ( j+1 , kj+1 ) T ( j+1 , kj+1 ) n ( j+1 , kj+1 ) = 0
>
<
1 T T
> fp = ¯ j+1 ( j+1 , kj+1 ) P ¯ j+1 ( j+1 , kj+1 ) + ¯ j+1 ( j+1 , kj+1 ) q ( kj+1 )
>
> 2
:
¯y2 ( kj+1 ) = 0
(54)
which is based on the following relationship between the final hardening variable and its
increment:

kj+i = kj + kj+1 (55)

The solution of Eq. 54 can be obtained by means of an iterative method like Newton-
435 Raphson as outlined in Appendix B.

18
 6. Computational implementation

6.1. Numerical considerations for plasticity and damage coupling under load reversals
Some numerical aspects need to be accounted for when formulating a coupled damage-
plasticity model such as the well-known mesh-dependency numerical problem of finite el-
440 ement approximations with softening inelastic behaviour [29]. This issue is not relevant
for wood in compression since hardening plasticity and hardening damage evolution are
considered. However, for wood in tension, the development of a fracture process zone
produces softening [33] which is expressed mathematically by the damage evolution law
in Eq. 19. In order to avoid mesh-dependency, a crack band regularization technique [34]
445 is employed. Accordingly, the softening damage evolution law was adjusted through the
incorporation of the characteristic length (lch ) in Eq. 20, which is assumed to be equal
to the cubic root of the volume of the smallest finite element of the model mesh.

Additionally, the damage criterion functions formulated in terms of the total strain
450 only (without considering the influence of plastic strains) can generate spurious high ten-
sile strengths. This artificial rise of the tensile strength delays the start of the damage
process when a tension reversal loading is applied after the occurrence of a plastic flow
in compression [6]. This is the reason why the damage function equivalent stress defined
in Eq. 14 is expressed in terms of the e↵ective stress tensor, which is a function of the
455 elastic strain or, analogously, of both the total and plastic strains [7]. This approach is
essential for an appropriate definition of the damage function for cyclic loading appli-
cations. Furthermore, the coupling of damage and plasticity formulations is proposed
to occur exclusively in the compressive component of the e↵ective stress space. Con-
versely, the tensile component of the e↵ective stress is governed by a pure damage model
460 (no plasticity), and as a consequence, there is no need of coupling between damage and
plasticity. This is a particularly important and original characteristic of the proposed
model, which is grounded on the typical post-elastic behaviour of wood in both tensile
and compressive stress states.

465 Since there is no experimental evidence of an equivalent behaviour to the Bauschinger

e↵ect in wood subjected to cyclic loading, there is no reason for incorporating a kinematic
hardening rule in the plasticity part of the model. Besides, since plasticity is valid only for
the compressive component of the stress tensor, the e↵ect of such a kinematic hardening
would be diminished. Nevertheless, the di↵erent evolution of the elastic boundaries in
470 tension and compression is governed by the two damage variables. Likewise, the influence
of the loading in one direction on the elasticity boundary in the other direction is defined
by the concepts of structural damage, characteristic of tensile failure, and constitutive
damage, characteristic of compressive failure [22]. The structural damage in tension does
not a↵ect the sti↵ness degradation or the strength of the material when a reversed load
475 is applied in compression (sti↵ness recovery). In contrast, the constitutive damage or
crushing does reduce the tensile capacity of the material when a reverse loading from
compression to tension is applied.

Fig. 5 shows the initial (r± = r0± = 1) elastic boundaries in tension and compression,
480 according to Hill’s damage criteria (Eqs. 13 and 14), and the initial (k = 0) plasticity
surface, according to Ho↵man’s particular case of the general orthotropic yield function
19
 Hoffman Plasticity Yield Surface perp Hoffman Plasticity Yield Surface
Compressive Hill Damage Surface Compressive Hill Damage Surface perp
Tensile Hill Damage Surface Tensile Hill Damage Surface

perp
paral

(a) Plane of normal stresses in the directions (b) Plane of normal stresses in the directions
parallel and perpendicular to the grain. perpendicular to the grain.

Hoffman Plasticity Yield Surface Hoffman Plasticity Yield Surface perp

Compressive Hill Damage Surface shear Compressive Hill Damage Surface
Tensile Hill Damage Surface Tensile Hill Damage Surface

shear
paral

(c) Plane of normal stress in the direction par- (d) Plane of normal stress in the direction per-
allel to the grain and shear stress. pendicular to the grain and shear stress.

Figure 5: Initial compressive and tensile Hill’s damage surface and initial Ho↵man’s plasticity yield
surface.

in Eq. 36. All curves in Fig. 5 represent the intersection of these multidimensional sur-
faces with some two-dimensional planes (shear stresses and normal stresses parallel and
perpendicular to the grain) of the e↵ective stress space. Fig. 5a shows that under biaxial
485 compression in the directions parallel and perpendicular to the grain damage occurs first
than plasticity, while for mixed tension-compression stress states an interaction between
plasticity in compression and damage in tension prevails. Similarly, for biaxial compres-
sion in the plane perpendicular to the grain, Fig. 5b shows that damage occurs first
than plasticity as well. What is more, due to the elliptic paraboloid shape and the pres-
490 sure dependency of the Ho↵man surface, plasticity is unlikely to occur for a hydrostatic
compression. In this plane also an interaction between tensile damage and plasticity can
be appreciated for mixed biaxial tension-compression stress state. Finally, Figs. 5c and
5d show the interaction of shear with normal stresses in both parallel and perpendicular
to the grain directions, respectively. In the same way than in the previous cases, dam-
495 age is activated earlier than plasticity when compression normal stresses interact with
shear stresses. However, the distance between both compressive damage and plasticity
surfaces is considerably smaller than in the case of biaxial normal stress (in the direction
parallel to the grain, Fig. 5c, both surfaces are almost coincident in compression). This
clearly indicates that for this type of plasticity yield function a stress state including
500 shear stresses is more critical than one of normal stresses only. In the tension zone, the
interaction of shear and normal stresses is governed exclusively by the tensile damage
surface.

All the characteristics discussed above are consistent with the expected behaviour of

20
505 wood where compressive behaviour, with or without shear, is governed by the combi-
nation of plasticity and damage, with the latter occurring before the former. It is also
expected that a combination of shear stresses with compression stresses produce a ductile
failure in which plasticity plays the most important role. By contrast, a hydrostatic bi-
axial compression is assumed to produce failure due to crushing, and therefore, damage
510 plays the most important role. On the other hand, any stress state in which tension
normal stresses are present, the tensile damage surface clearly prevails over the other
two. Noticeably, these initial surfaces will expand once the applied stresses magnitudes
reach them and the inelastic behaviour starts.

6.2. Algorithm design

515 The algorithm of the plasticity-damage model for timber is shown in Table 1. First
the total trial e↵ective stress is calculated (Step 1). Then the spectral decomposition of
the total trial e↵ective stress is performed and the eigenvalues and eigenvectors are saved
(Step 2). In the next step the compressive stress is evaluated into the yield function
in order to determine if there is plasticity or not. In case of plasticity, the Newton-
520 Raphson algorithm is executed in order to determine the value of the plastic multiplier
and the hardening variable increment (Step 3). Then, the compressive e↵ective stress is
updated and the increment of the plastic strain vector is determined. Then, both e↵ective
stress components (compressive and tensile) are evaluated in their correspondent damage
functions. If there is damage for any of the stress components, the value of the damage
525 variable is updated using the corresponding damage evolution law (Step 4). The Cauchy
nominal stress is calculated (Step 5) and the total and plastic strains are updated (Step 6).
Finally, the consistent plasticity-damage tangent sti↵ness matrix (Step 7) is calculated
to be delivered as one of the output parameters of the model.
Table 1: Plasticity-damage timber constitutive model algorithm

Initial values: ✏0 = 0; ✏p0 = 0; ¯ ± ± ±

0 = 0; k0 = 0; !0 = 0; r0 = 1

Load step input parameters: ✏j+1 , ✏j , ✏pj , ¯ ± ± ±

j , k j , ! j , rj

1. Calculate the total elastic trial e↵ective stress tensor:

e
¯e = ¯+
j + ¯j + D ✏

2. Spectral decomposition of the total elastic trial e↵ective stress tensor:

2.1. Obtain the eigenvalues and eigenvectors for i = 1, 2, 3
2.2. Calculate the compressive and tensile components:
Eq. 10 ! ¯ +
j+1 Eq. 11 ! ¯ j+1
3. Plasticity verification for the compressive component only:
3.1. Evaluate the yield function

530

21
 T T
Eq. 36 ! fp (¯ j+1 , kj ) = 12 ¯ j+1 P ¯ j+1 + ¯ j+1 q(kj ) ¯y2 (kj ) < 0 ?

3.2. YES ! No plasticity

j+1 = 0; kj+1 = kj ; ✏pj+1 = 0

3.3. NO ! Plasticity algorithm

3.3.1. Assign the compressive elastic trial e↵ective stress tensor

¯ e = ¯ j+1

3.3.2. Initiation of the unknown variables

0 0
j+1 = 0; kj+1 =0

3.3.3. Calculate j+1 and kj+1 : Newton-Raphson algorithm for

iteration i

i i
Eq. 55 ! kj+1 = kj + kj+1

i i
Eq. 31 ! ¯y,j+1 = ¯y,0 + h · kj+1

i
Eq. 35 ! q kj+1
i i i
Eq. 51 ! ¯ j+1 j+1 , kj+1
⇣ i
⌘
Eq. 38 ! nij+1 ¯ j+1 , i
kj+1

bi+1 i
j+1 = bj+1 (J ij+1 ) 1 i
r j+1

where b, r, and J are the unknown variables vector, the residuals

vector, and the Jacobian matrix, respectively (Appendix B).

Do while r i+1
j+1 > tolerance

3.3.4. Calculate the compressive e↵ective stress tensor and the plastic
strain tensor increment:
Eq. 51 ! ¯ j+1 ( j+1 , kj+1 )
Eq. 48 ! ✏pj+1 ¯ j+1 , kj+1
4. Damage verification for the tensile and compressive stress components:
4.1. Calculate the equivalent stress and evaluate the damage function
q
T
Eq. 14 ! ⌧¯ = 12 ¯ ± H ± ¯ ±
±

Eq. 13 ! fd± (¯ ± , r± ) = ⌧¯± r± < 0 ?

22
 4.2. YES ! No damage
±
rj+1 = rj± ; !j+1
±
= !j±
4.3. NO ! Damage algorithm
± ±
rj+1 = ⌧¯j+1
+
Eq. 19 ! !j+1 = gd+ (rj+1
+
)
Eq. 22 ! !j+1 = gd (rj+1 )
5. Calculate the Cauchy stress tensor:
+
Eq. 12 ! j+1 = (1 !j+1 )¯ +
j+1 + (1 !j+1 )¯ j+1

6. Calculate the total strain and plastic strain tensors:

✏j+1 = ✏j + ✏j+1 ; ✏pj+1 = ✏pj + ✏pj+1

7. Calculate the consistent plasticity-damage tangent sti↵ness matrix:

7.1. Calculate the consistent elasto-plastic tangent sti↵ness matrix for the
compressive stress component (D epj+1 )
7.2. Calculate the compressive and tensile fourth-order tensor operators
(Q±
j+1 )
7.3. Calculate the consistent plasticity-damage tangent sti↵ness matrix
(D pd
j+1 )

Output parameters: j+1 , D pd p ± ± ±

j+1 , ✏j+1 , ✏j+1 , ¯ j+1 , kj+1 , !j+1 , rj+1

6.3. Consistent plasticity-damage tangent sti↵ness matrix

The consistent plasticity-damage tangent sti↵ness matrix, D pd , sets the relation be-
535 tween the Cauchy nominal stress rate, , and the total strain rate, ✏. This matrix
is employed in the calculation of the global structural tangent sti↵ness matrix, which
ensures the robustness of the Newton-Raphson method by maintaining its quadratic
convergence rate [35]. The particular characteristics of the consistent tangent sti↵ness
matrix depend on each constitutive model and may be non-symmetric in many cases,
540 which increases the computational time and numerical complexity. For this reason, some
authors recommend to adopt a nonlinear equation solver which employs only the initial
elastic sti↵ness matrix during the entire analysis, particularly when performing seismic
analysis of massive structures [5]. In spite of saving computational time for some specific
applications, the use of a unique tangent sti↵ness matrix is not suitable for problems
545 where high levels of nonlinear post-peak behaviour are expected. In these cases a consis-
tent tangent sti↵ness matrix needs to be derived, not only to attain the high deformation
levels required but also to drastically curtail the number of iterations needed [27]. For
the plasticity-damage model proposed herein the consistent tangent sti↵ness matrix is

23
 obtained from the Cauchy stress, Eq. 12, as a function of the compressive and tensile
550 threshold variables (r± ) and the total strain tensor increments ( ✏± ):
⇥ ⇤ ⇥ ⇤
= 1 ! + r+ ¯+ ✏+ + 1 ! r ¯ ✏ (56)

Likewise, the tangent sti↵ness matrix (in tensor notation) can be defined as:

d @ @r+ @ @r @ @✏+ @ @✏
= +⌦ + ⌦ + + : + : (57)
d✏ @r @✏ @r @✏ @✏ @✏ @✏ @✏
Based on Eqs. 17 and 56, the partial di↵erentiation of the Cauchy stress tensor with
555 respect to the threshold variables leads to:
±
@ @ @! ± ± @gd
= = ¯ (58)
@r± @! ± @r± @r±
while on the basis of Eqs. 13 and 14, the partial di↵erentiation of the threshold variables
with respect to the total strain tensor is:

@r± @r± @ ⌧¯± d¯ ± @✏±

= : : (59)
@✏ @ ⌧¯± @ ¯ ± d✏± @✏
±
It should be noted, from Eq. 24, that @@r⌧¯± can only be equal to 1, when damage occurs,
or 0, when there is no damage. Subsequently, the partial di↵erentiation of the Cauchy
560 stress tensor with respect to the tensile and compressive components of the total strain
tensor leads to:

@ @ d¯ ± d¯ ±
= : = 1 !± I : (60)
@✏± @ ¯ ± d✏± d✏±
where I is the unitary fourth-order tensor.

The tensile and compressive components of the e↵ective stress tensor are obtained
565 based on elasticity and plasticity formulations, respectively. Therefore, their total deriva-
tives in function of the corresponding tensile and compressive components of the total
strain tensor are precisely the linear-elastic sti↵ness tensor, D e , and the consistent elasto-
plastic tangent sti↵ness tensor, D ep :

d¯ +
= De (61)
d✏+

d¯
= D ep (62)
d✏

570

From Eqs. 57 to 62, the following expression for the consistent plasticity-damage
tangent sti↵ness matrix can be obtained:

24
 
d @r+ @gd+ + @ ⌧¯+ e @✏
+
= 1 !+ I ¯ ⌦ : D :
d✏ @ ⌧¯+ @r+ @ ¯+ @✏
 (63)
@r @gd @ ⌧¯ @✏
+ 1 ! I ¯ ⌦ : D ep :
@ ⌧¯ @r @¯ @✏

6.3.1. Consistent elasto-plastic tangent sti↵ness matrix

The elasto-plastic tangent sti↵ness matrix is obtained consistently with respect to the
575 stress update procedure followed in the corresponding local return mapping algorithm
[36, 37]. Noting that the compressive e↵ective stress (Eq. 51) is a function of the plastic
multiplier increment, , the hardening variable increment, k, and the compressive
total strain tensor increment, ✏ , then:
h 1
i
¯ = A 1 ( ) De ¯ e ✏ q ( k) (64)

The tangent sti↵ness matrix is obtained from the total di↵erentiation of the compressive
580 e↵ective stress tensor, ¯ , with respect to the compressive total strain tensor, ✏ :

d¯ @¯ @¯ @ @¯ @ k
= + + (65)
d✏ @✏ @ @✏ @ k @✏

On the other hand, based on the Newton-Raphson iterative algorithm for plasticity,
the derivative of the plastic multiplier and the hardening variable increments with respect
585 to the compressive total strain can be obtained in terms of the Jacobian matrix, J , as
follows:
" # " @f #
@ p
@✏ @b 1 @r 1 @✏
@ k
= = J = J (66)
@✏
@✏ @✏ @F
@✏

The substitution of Eq. 66 into Eq. 65 leads to the expression for the consistent elasto-
plastic tangent sti↵ness matrix Dep :
" @f #
d¯ @¯ h i p
ep @ ¯ @ ¯ 1 @✏
=D = @ @ k
J (67)
d✏ @✏ @F
@✏

The expressions to calculate the derivatives in Eq. 67 and the coefficients of the Jacobian
590 matrix are presented in Appendix B.

6.3.2. Partial derivatives of the tensile and compressive components of the total strain
tensor with respect to the total strain tensor
In order to obtain the expressions of the derivatives of the tensile and compressive
strain components with respect to the total strain tensor, first the fourth-order tensor
595 operators, Q± , must be defined:
±
e e
¯˙ = Q± : ¯˙ (68)
25
 Q =I Q+ (69)
Notice that the previous expressions are defined to be valid in the elastic stress state
only, before any plasticity correction is performed for the compressive e↵ective stress
component. Therefore, the time derivatives of the elastic e↵ective stresses are defined as
follows:
e
¯˙ = D e : ✏˙ (70)

±
e
¯˙ = D e : ✏˙± (71)
600 From Eq. 10, the total di↵erentiation of the tensile elastic e↵ective stress, in tensor
notation, is:
3
X
+
d¯ e = [H (¯i ) (pi ⌦ pi ) d¯i + h¯i id (pi ⌦ pi )] (72)
i=1

where H(·) is the Heaviside function. The total derivatives of the eigenvalues, ¯i , and
eigenvectors, pi , of the elastic e↵ective stress tensor, ¯ e , can be expressed as [27]:

d¯i = Rii : d¯ e (73)

3
X 
1
d (pi ⌦ pi ) = 2 (Rij ⌦ Rij ) : d¯ e (74)
¯i ¯j
j=1
j 6= i

where the second-order symmetric tensor Rij is defined as:

1
Rij = p ⌦ pj + pj ⌦ pi (75)
2 i
605 Replacing Eqs. 73 and 74 into Eq. 72, and from the definition given in Eq. 68, the tensile
fourth-order tensor operator is obtained [27]:
3
X X3 3
X 
+ h¯i i
Q = [H (¯i ) (Rii ⌦ Rii )] + 2 (Rij ⌦ Rij ) (76)
i=1 i=1
¯ i ¯j
j=1
j 6= i

Finally, after substituting Eqs. 70 and 71 into Eq. 68, the derivative of the tensile and
610 compressive strain tensor components as a function of the total strain tensor is obtained
as:

@✏± 1
= De : Q± : D e (77)
@✏
where Q+ and Q are given by Eqs. 76 and 69, respectively.

26
 6.3.3. Consistent plasticity-damage tangent sti↵ness matrix formulation
The final expression of the consistent plasticity-damage tangent sti↵ness matrix in
615 Voigt’s notation can be obtained by replacing Eq. 77 into Eq. 63:


pd @r+ @gd+ + @ ⌧¯+
D = 1 !+ I ¯ Q+ D e
@ ⌧¯+ @r+ @ ¯+
 (78)
@r @gd @ ⌧¯ ep e 1 e
+ 1 ! I ¯ D (D ) Q D
@ ⌧¯ @r @¯

7. Thermodynamic consistency verification

The verification of the thermodynamic consistency of a constitutive model is a nec-

essary requirement to guard against the danger of spurious energy generation, and is
particularly important for cyclic material models as the one presented herein [6]. The
620 first step is to define the elastic free energy potential in tensor notation as [5]:
1 1
¯ : ✏e = ¯ : (✏ ✏p )
0 = (79)
2 2
Based on Eq. 11, the elastic free energy potential can be expressed as:
1 + 1 1
0 = ¯ +¯ : ✏e = ¯ + : (✏ ✏p ) + ¯ : (✏ ✏p ) = + 0 + 0 (80)
2 2 2
Therefore, under the assumption of isothermal conditions and uniform temperature, the
free energy potential of the plasticity-damage model developed herein can be expressed
in terms of the free (✏) and internal (✏p , ! + , ! ) variables as follows:
+
✏, ✏p , ! + , ! = 1 !+ 0 (✏, ✏p ) + 1 ! 0 (✏, ✏p ) (81)
625

In order to fulfil the second principle of thermodynamics, the total energy introduced
to the system should be higher than the total energy dissipated [5]. This condition is
mathematically expressed by the Clausius-Duheim inequality [19] such that:

˙ = : ✏˙ ˙ 0 (82)
630 The time derivative of the free energy potential is:

˙ = @ : ✏˙ + @ : ✏˙ p + @ !˙ + + @ !˙ (83)
@✏ @✏p @! + @!
And, from Eq. 81, the partial derivatives of the free energy potential in function of the
free and internal variables are:

@ @ + @ 0
= 1 !+ 0
+ 1 ! (84)
@✏ @✏ @✏

@ @ + @ 0
= 1 !+ 0
+ 1 ! (85)
@✏p @✏p @✏p
27
 @ ±
= 0 (86)
@! ±
It can also be shown that [5]:

@ ±0
= ¯± (87)
@✏

@ ± 0
= ¯± (88)
@✏p
The substitution of Eq. 83 into Eq. 82 leads to:
✓ ◆ ✓ ◆
@ @ p @ + @
˙ = : ✏˙ : ✏˙ + !˙ + !˙ 0 (89)
@✏ @✏p @! + @!
635 Given that the condition in Eq. 89 must be fulfilled at all time steps, the first term,
which is a function of the free variable, should always equal zero:
@
= (90)
@✏
Therefore, the substitution of Eqs. 84 and 87 into Eq. 90 confirms the model constitutive
law previously defined in Eq. 12. Moreover, substituting Eqs. 85, 86 and 88 into Eq. 89
allows the energy dissipation condition to be expressed in terms of the plasticity and
640 damage dissipation rates as:

˙ = ˙p + ˙d 0 (91)

⇥ ⇤
˙p = 1 !+ ¯ + + 1 ! ¯ : ✏˙p = : ✏˙p 0 (92)

+ +
˙d = 0!
˙ + 0 !˙ 0 (93)

It can be shown that ± ˙ ± is always positive as

0 is positive [5], while by definition !
well. Therefore, thermodynamic consistency is always ensured for the damage component
645 of energy dissipation rate ˙ d . However, some additional conditions should be met in order
to ensure that the plasticity part of the energy dissipation is positive as well. From Eq. 92
it can be concluded that the scalar multiplication of the e↵ective stress components and
the plastic strain tensor should be non-negative. For this to be true it is sufficient, but
not necessary, that the plastic yield surface, Eq. 36, is convex and the plastic flow rule,
650 Eq. 38, is associated [23, 6]. Both condition are met by the plasticity model developed
herein, hence verifying its thermodynamic consistency.

28
 8. Application: Uniaxial stress-strain cyclic response

The newly proposed timber plasticity-damage model was coded in the finite element
software DIANA [38] by means of a user-supplied subroutine (USS) and the stress-strain
655 response of a timber specimen was obtained. The specimen was subjected to uniaxial
cyclic loading in the directions parallel and perpendicular to the grain. In addition to the
typical orthotropic elasticity parameters, the proposed plasticity-damage model requires
a total of 15 input parameters. The set of material model parameters employed in this
application are provided in Table 2 and is consistent with the assumption of orthotropic
660 behaviour with transverse isotropy.

Table 2: Uniaxial cyclic loading. Timber material plasticity-damage model parameters

Poisson’s ratio ⌫XY = 0.41 ⌫Y Z = 0.37 ⌫XZ = 0.41

Shear moduli [MPa] GXY = 690 GY Z = 125.9 GZX = 690
Young’s moduli [MPa] EX = 9936 EY = 345 EZ = 345
Compressive strength [MPa] fX = 40 fY = 4 fZ = 4
+
Tensile strength [MPa] fX = 20 fY+ = 1 fZ+ = 1
Shear strength [MPa] fXY = 4 fY Z = 4 fZX = 4
Fracture energy density [N/mm] Gf,Y Z = 1
Characteristic length [mm] lch = 333
Exponential softening parameter n=1
Hardening modulus h = 12.9
Compressive hardening constant = 0.85
Compressive hardening power m=1
Orthotropic yield criterion Ho↵man

Figs. 6 and 7 depict the cyclic stress-strain response obtained by means of the new
665 timber material model subjected to uniaxial action in the directions parallel (X) and
perpendicular (Z) to the grain, respectively. The geometry of the one-meter per side
cubic timber specimen modelled with hexahedral solid finite elements (333 by 333 mm)
is also shown in Figs. 6 and 7. It can be observed from these figures that the responses
under tensile and compressive stress states are markedly di↵erent. The tensile inelastic
670 response is characterized by an exponential softening, while the compressive counter-part
features plasticity with minor hardening. Accordingly, permanent plastic deformations
occur in the negative side of the strain axis when the stress passes from compression to
tension. Moreover, the level of sti↵ness degradation in tension (structural damage) is
significantly higher than in compression (constitutive damage). This is clearly appreci-
675 ated by comparing the slopes of the unloading branches in both tensile and compressive
zones. The lower strength and damage-only (no plasticity) characteristic response of
wood in tension accelerates the degradation of the secant unloading sti↵ness. On the
other hand, the higher strength in compression and the inclusion of plasticity generates
a milder sti↵ness degradation with an unloading sti↵ness in-between the elastic and the
680 secant branches. Overall, the proposed model captures the key characteristics of the
cyclic behaviour of timber materials, namely, the tensile softening brittle failure, the
29
 20

10

0
Stress X [MPa]

-0.01 -0.008 -0.006 -0.004 -0.002 0 0.002 0.004

-10

-20

-30

-40

-50
Strain X [mm/mm]

Figure 6: Plasticity-damage model. Uniaxial cyclic loading-unloading stress-strain diagram in the direc-
tion parallel to the grain.

0
-0.025 -0.02 -0.015 -0.01 -0.005 0 0.005
-1
Stress Z [MPa]

-2

-3

-4

-5
Strain Z [mm/mm]

Figure 7: Plasticity-damage model. Uniaxial cyclic loading-unloading stress-strain diagram in the direc-
tion perpendicular to the grain.

sti↵ness cyclic degradation and recovery after load-reversal, the compressive permanent
plastic deformation due to ductile failure, and the interaction of multiaxial stress states.

9. Experimental validation

685 Although a large amount of experimental research has been carried out on the re-
sponse of timber elements and timber connections of various types at the component
level, including studies on their global cyclic behaviour, there is a dearth of information
regarding cyclic experimental results for wood at the local material level. In this sec-
tion, the experimental results obtained by Popovski et al. [39] are employed to validate
690 the newly proposed plasticity-damage model for wood. These tests were conducted on
timber-steel dowelled connections between glulam braces and steel side plates using steel
bolts. In particular, the specimen with two 19mm diameter bolts is employed since its
nonlinear response was governed by that of the wood material.
30
 3

2
Displacement [ y]

-1

-2

-3

Figure 8: Cyclic loading protocol [39].

9.1. Finite element model

695 The glulam brace under consideration had a length of 1.5 m and a cross section of
130 mm x 152 mm; the steel plates were 12 mm thick; the spacing between the two bolts
was 76 mm, while the end distance was 228 mm. The load was applied to the connection
in the axial direction of the brace (parallel to the grain) through a servo-controlled actu-
ator attached to the steel side plates. Fig. 8 shows the cyclic loading protocol in function
700 of the yielding displacement ( y ) which based on the monotonic experimental test was
determined to be equal to 1.65 mm [39]. Further details about the test set-up, speci-
men geometry, material properties and the cyclic testing protocol can be found in [39, 40].

Fig. 9 shows the finite element model developed in DIANA taking into account the
705 symmetry of the specimen. The glulam brace and the steel plate and dowels were mod-
elled using hexahedral solid elements. The plasticity-damage USS described above was
assigned to the timber brace, while a standard Mises plasticity model was assigned to the
steel dowels. Following experimental observation, the steel side plates are expected to
work mainly under the linear elastic range. The interaction between the dowel and the
710 timber surrounding was handled by a bi-dimensional interface element which accounted
for the contact nonlinearity between both components. A displacement-control approach
was followed as represented by the arrows in Fig. 9.

A mesh sensitivity analysis was carried out to determine an appropriate model mesh
715 size. The results obtained were observed to have limited mesh dependency. However,
very small elements in the timber zone around the dowels should be avoided due to
their potential to generate spurious stress concentrations in the contact zones, which can
hinder the stress transfer and lead to unrealistic reductions in the predicted maximum
capacity of the connection. In general, and regardless of the material model employed, it
720 is recommended that the smallest elements in the timber zone be larger than any element
in the dowels, thus avoiding the presence of free nodes.

31
 Z

Z
Y
X
X

(a) 3D view

Z X

Z X

(b) Plan view, boundary conditions (green arrows) and applied load (red arrows)

Y X

Y X

(c) Lateral view, boundary conditions (black arrows) and applied load (red arrows)

Figure 9: Timber-steel dowelled connection finite element model.

32
 The same timber constitutive model parameters presented before in Table 2 are em-
ployed in this simulation, excepting the characteristic length parameter which for this
725 case is lch = 2.65 mm. The values of all the parameters listed in Table 2, except the
exponential softening parameter (n), the plasticity hardening modulus (h), the compres-
sive damage hardening constant ( ) and power (m) parameters, are directly related to
the mechanical properties of the timber material or to the geometric characteristics of
the finite element model. Therefore, only the values of the four previously mentioned
730 parameters need to be assumed or obtained by means of a calibration process. For this
particular case study, a value of n equal to 1 was assumed based on typical material ten-
sile and shear tests in timber specimens (see Figs. 1 and 2). These test show that wood
has no residual strength after brittle failure. Moreover, a value of n higher than 1 was
not considered adequate following the discussion o↵ered in Section 4.3 and illustrated in
735 Fig. 3, given that it would imply the definition of a post-failure ultimate strain which
cannot be recalled from the available experimental data.

On the other hand, in a dowelled connection the interaction of both compressive

hardening parameters (h and ) has a direct influence on the maximum capacity of the
740 connection. In addition, h defines the level of plastic deformation in the timber zone
around the dowels, while limits the maximum level of sti↵ness degradation in compres-
sion. However, the measurement of plastic deformations in laboratory tests of timber
specimens is complex and rare. Similarly, cyclic tests on timber specimens at the material
level to determine the maximum level of sti↵ness degradation in compression are also un-
745 usual. Under these circumstances, it is desirable to calibrate the parameters h and on
the basis of experimental force-displacement relationships of embedment tests in the di-
rections parallel and perpendicular to the grain. For the case study under consideration,
embedment tests were not available, and therefore, the monotonic experimental curves
of the dowelled connection were employed for the calibration of . Previously, the value
750 of h has been determined by means of Eqs. 46 and 47. A plastic tangent moduli equal
to the 10% of the elastic moduli was assumed [17]. Finally, a linear hardening damage
evolution law for compressive post-elastic behaviour was employed [13, 28]. Accordingly,
the value of the power parameter m was defined as 1.

755 Preliminary sensitivity analyses have demonstrated the reliability of the global nu-
merical response when the values of any of these parameters are modified slightly with
the model being most sensitive to variations of the compressive damage hardening con-
stant ( ). Noting that the admissible values for this parameter range between 0 and 1,
a variation of 0.1 was observed to lead to noticeable changes in the numerical response
760 whilst variations in the order of 0.01 of the assumed values had negligible e↵ects.

9.2. Monotonic response

Fig. 10 compares the experimental and numerical results of the timber-steel dowelled
connection subject to tension monotonic loading. Besides the load-deformation response
obtained by means of the new plasticity-damage model proposed herein, the correspond-
765 ing curve for a plasticity-only model has also been included in Fig. 10. In both models
the Ho↵man orthotropic yield surface was employed. It can be noticed from this figure
that the plasticity-damage model is able to reproduce the sti↵ness degradation shown by
the experimental curves, eventually reaching a maximum load plateau and then failure.
33
 160
140
120
100
Load [kN]

80
60 Experimental
40 Plasticity only
20 Plasticity-Damage
0
0 1 2 3 4 5 6 7
Deformation [mm]

Figure 10: Timber-steel dowelled connection subjected to monotonic tension. Experimental [40] and
numerical.

Moreover, the prediction of the maximum load and the ultimate deformation is good. In
770 fact, the numerical curve obtained with the new plasticity-damage model closely resem-
bles one of the three experimental curves. On the other hand, the plasticity-only model
shows a sti↵er response which makes the numerical and experimental load-deformation
curves separate at an early load stage. Consequently, the load-deformation curve shows
no ductile behavior (brittle failure) and a higher failure load than the experimental ones.
775 In general, the overall nonlinear monotonic response of the timber-steel dowelled connec-
tion is accurately predicted by the new plasticity-damage model.

9.3. Cyclic loading

Fig. 11 shows the experimental and numerical load-deformation hysteretic response
of the timber-steel dowelled connection until failure. It can be observed from this figure
780 that the numerical results obtained by means of the new constitutive material model
reproduce fairly the cycle-to-cycle sti↵ness degradation. The loading, unloading and
reloading slopes corresponding to each hysteretic loop are precisely captured by the
model. Furthermore, the maximum load per cycle and the in-cycle strength degradation
are accurately predicted as well. Finally, pinching behaviour and plastic deformations
785 weel described by the numerical response. In general, a good correlation between the
experimental and numerical hysteretic curves is obtained until the failure of the specimen.

10. Conclusions

This paper has presented the theoretical framework, mathematical formulation and
computational implementation of a three-dimensional plasticity-damage constitutive model
790 for timber materials under cyclic loading. The numerical features and the thermodynamic
consistency verification of the algorithm, which was implemented in the framework of
DIANA FE software, have been detailed. Care was taken to ensure that most model
34
 120

80

40
Load [kN]

0
-4 -3 -2 -1 0 1 2 3 4
-40
Experimental
-80
Plasticity-Damage
-120
Deformation [mm]

Figure 11: Timber-steel dowelled connection subjected to reverse cyclic loading. Experimental [39] and
numerical comparison.

parameters have a clear link with the physical material properties of wood, minimizing
the need for numerical calibration.
795

Preliminary numerical results obtained for a dowelled timber-steel connection, and

their comparison against the experimental results, including monotonic and cyclic load-
ing tests, have confirmed that the coupling of orthotropic plasticity in compression with
isotropic damage in both tension and compression stress states, constitutes a promis-
800 ing approach for modelling of timber structures subjected to load reversals. Moreover,
the incorporation of continuous damage evolution is essential to capture timber sti↵ness
degradation and to obtain a good agreement between numerical and experimental results.
It has been demonstrated that all key features of the nonlinear timber behaviour can be
captured by the proposed model, including: tensile softening brittle failure, sti↵ness
805 cyclic degradation and recovery after load-reversal, as well as compressive permanent
plastic deformation due to ductile failure.

The proposed material constitutive model for wood under cyclic loads relies on an
efficient and reliable algorithm which implementation in a widely-used FE software opens
810 the door for realistic simulations to be performed by the broader structural engineering
community. This study represents a fundamental first step towards a faithful estima-
tion of the ultimate capacity of timber structures subjected to extreme loads such as
earthquakes and, as a consequence, a more genuine assessment of their performance.

Acknowledgements

815 The authors wish to thank the support from the Peruvian Government through its
CONCYTEC/CIENCIACTIVA Program.

35
 Appendix A: Energy dissipation for tensile brittle failure

The energy dissipated in the post-elastic region after tensile brittle failure can be
expressed as the fracture energy density (Gf ) of wood normalized by the characteristic
820 length (lch ) of the finite element discretization such that:
Z 1
Gf
= ˙ d+ dt (94)
lch 0

where the tensile component of the damage energy dissipation rate, ˙ d +, is:
+ +
˙ d+ = 0!
˙ (95)

On the other hand, the e↵ective stress for a uniaxial tensile stress state is expressed
825 as:

r+ +
¯+ = fmax (96)
r0+
while the total strain is:

1 r+ +
✏= f (97)
E r0+ max
And since there is no plastic strain for uniaxial tensile stresses, the tensile elastic free
energy potential is:
✓ ◆2
+ 1 r+ +
0 = fmax (98)
2E r0+
The time derivative of the tensile damage variable is defined as:

@gd+ +
!˙ + = ṙ (99)
@r+
830 Therefore:
2
+
fmax 2 @gd+ +
˙ d+ = 2 r+ ṙ (100)
2E r0+ @r+

Finally, the energy dissipation due to tensile failure is obtained through the solution
of the following integral:
2 Z +
Gf +
fmax ru
2 @gd+ +
= 2 r+ dr (101)
lch 2E r0+ r0+ @r+

835 which integration by parts leads to:

36
 2  ✓ ◆ ✓ ◆
Gf f+ 2 +
b( r u r0+ ) 2
= max+ ru+ r0+ (1 n) n ru+ + e +n r0+ + (102)
lch 2Er0 b b
for the particular tensile damage evolution law of Eq. 19.

Depending on the value of the parameter n, the expressions to calculate the maximum
value of the threshold variable, ru+ , are:
(
For n 1 ru+ = 1b ln nn 1 + r0+
(103)
For n < 1 ru+ = 1b ln (↵) + r0+

840 where ↵ denotes a fraction of the initial slope of the exponential softening curve. Accord-
ingly, for n < 1, ru+ is defined to occur at the point where the slope of the exponential
curve equals ↵ times the initial slope ( ↵  0.01). From this, the following expressions
are obtained for b:
8
> 2[1+(n 1) ln( nn 1 )]
> For n 1
> b= " #
>
> + 2EGf
< r 0 + 2 1
(fmax ) lch
(104)
>
> (↵n+n 1) ln ↵+2n(1 ↵)#
> For n < 1 b = " 2EG
>
>
: r0+ f
2 n(1 ↵)
+
(fmax ) lch

845

Eq. 104 corresponds to Eq.21 which accomplishes the demonstration. Finally, to

ensure that the tensile damage evolution law in Eq.19 corresponds to an exponential
softening function, the parameter b should be positive. This condition defines a restriction
for the maximum characteristic length of the finite element discretization, lch , such that:
8
> 2EG
< For n 1 lch < + f 2
(fmax )
2EGf
(105)
>
: For n < 1 lch < f + 2 n(1
( max ) ↵)

850 Appendix B: Derivatives of the Jacobian matrix for the Newton-Raphson

iterative solution of the plasticity algorithm and the consistent elasto-plastic
tangent sti↵ness matrix
The solution of the system of equations 54 can be obtained by means of the iterative
Newton-Raphson method:
1
bi+1 i
j+1 = bj+1 J ij+1 r ij+1 (106)
855 where i is the iteration number, j is the previous load step, b is the unknown variables
vector, r is the residuals vector, and J is the Jacobian matrix:
" #
i
j+1
bij+1 = i
(107)
kj+1
37
 " #
i i
fp j+1 , kj+1
r ij+1 = i i
(108)
F j+1 , kj+1
2 3
dfp i i dfp i i
d j+1 , kj+1 d k j+1 , kj+1
J ij+1 = 4 dF i i dF i i
5 (109)
d j+1 , kj+1 d k j+1 , kj+1

The expressions to calculate the four total derivatives of the Jacobian matrix are:

dfp @fp @ ¯ @¯
= = nT (110)
d @¯ @ @

dfp @fp @ ¯ @fp @q @fp @ ¯y @¯ T @q

= + + = nT +¯ 2¯y h (111)
d k @¯ @ k @q @ k @ ¯y @ k @ k @ k
✓ ◆ p
dF @F @n @F @F @n @ ¯
= + = nT T n (112)
d @n @ @ @n @¯ @
✓ ◆
dF @F @n @F @F @n @ ¯ @n
= + = + +1 (113)
d k @n @ k @ k @n @¯ @ k @ k
860 While the expressions of the partial derivatives in Eqs. 110 to 113 are defined as follows:

@¯ h ⇣ 1
⌘ i
2
=A P De ¯e q + Aq (114)
@
@¯ 1 @q
= A (115)
@ k @ k
@q ⇥ ⇤T
=h 10,0 11,0 12,0 0 0 0 (116)
@ k
@F
= p nT T (117)
@n nT T n
@n
=P (118)
@¯
@n @q
= (119)
@ k @ k
Similarly, the other partial derivatives required for the calculation of the consistent elasto-
plastic tangent sti↵ness matrix in Eq. 67 are:

@¯ @ ¯ @ ¯e 1
= =A (120)
@✏ @ ¯ e @✏
@fp @fp @ ¯ 1
= = nT P A (121)
@✏ @ ¯ @✏
38
 @F @F @n @ ¯
= = p nT T P A 1 (122)
@✏ @n @ ¯ @✏ nT T n
After convergence, all the derivatives in Eq. 67 employed for the calculation of the consis-
tent elasto-plastic tangent sti↵ness matrix, Dep , should be evaluated again for the given
865 converged solution values.

References
[1] R.O. Foschi,Load-slip characteristics of nails, Wood Science,7 (1974) 69–76.
[2] J.P. Hong, Three-dimensional nonlinear finite element model for singe and multiple dowel-type wood
connections, PhD thesis (2007), Program of Forestry, University of British Columbia, Canada.
870 [3] V. Karagiannis,C. Málaga-Chuquitaype, A.Y. Elghazouli, Modified foundation modelling of dowel
embedment in glulam connections, Construction and Building Materials 102(2) (2016) 1168–1179.
[4] V. Karagiannis,C. Málaga-Chuquitaype, A.Y. Elghazouli, Behaviour of hybrid timber beam-to-
tubular steel column moment connections, Engineering Structures 131 (2017) 243-263.
[5] R. Faria, J. Oliver, M. Cervera, A strain-based plastic viscous-damage model for massive concrete
875 structures, International Journal of Solids and Structures 35(14) (1998) 1533–1558.
[6] P. Grassl ,M. Jirasek M, Damage-plastic model for concrete failure, International Journal of Solids
and Structures 43(22-23) (2006) 7166-7196.
[7] L. Jason, A. Huerta, G. Pijaudier-Cabot, S. Ghavamian, An elastic plastic damage formulation for
concrete: Application to elementary tests and comparison with an isotropic damage model, Computer
880 Methods in Applied Mechanics and Engineering, 195(52) (2006) 7077-7092.
[8] R. De Borst, J. Pamin, M.G.D. Geers, On coupled gradient-dependent plasticity and damage theories
with a view to localization analysis, European Journal of Mechanics, A/Solids 18(6) (1999) 939-962.
[9] B. Franke, P. Quenneville, Numerical Modeling of the Failure Behavior of Dowel Connections in
Wood, Journal of Engineering Mechanics 137(3) (2011) 186-195.
885 [10] C.L Dos Santos, J.J.L. Morais, A.M.P.De Jesus, Mechanical behaviour of wood T-joints, Experi-
mental and numerical investigation, Frattura Ed Integrita Strutturale 31 (2015) 23–37.
[11] M. Khelifa, A. Khennane, M. El Ganaoui, A. Celzard, Numerical damage prediction in dowel con-
nections of wooden structures. Materials and Structures 49(5) (2016) 1829–1840. doi:10.1617/s11527-
015-0615-5.
890 [12] B. Xu, A. Bouchair, P Racher, Appropriate Wood Constitutive Law for Simulation of Nonlinear
Behavior of Timber Joints, J. Mater. Civ. Eng. (ASCE) 26 (2014) 1–7.
[13] C. Sandhaas, J.W. Van de Kuilen, H.J. Blass, Constitutive Model for Wood Based on Continuum
Damage Mechanics, World Conference on Timber Engineering (2012) Auckland, New Zealand.
[14] R. Hill, A theory of the yielding and plastic flow of anisotropic materials. Proc. Roy. Soc., A193
895 (1947) 281–297.
[15] O. Ho↵man O , The brittle strength of orthotropic materials, Journal Composite Materials, 1 (1967)
200–206
[16] N. Kharouf, G. McClure, I. Smith, Postelastic Behavior of Single- and Double-Bolt Timber Con-
nections, Journal of Structural Engineering 131(1) (2005) 188–196.
900 [17] B.H. Xu, A. Bouchar, M. Taazount, P. Racher P, Numerical simulation of embedding strength of
glued laminated timber for dowel-type fasteners, Journal of Wood Science 59(1) (2013) 17–23.
[18] T. Rabczuk, J. Eibl, Modelling dynamic failure of concrete with meshfree methods, International
Journal of Impact Engineering, 32 (2006) 1878-1897.
[19] J.C. Simo, J.W. Ju, Strain- and stress-based continuum damage modelsI. Formulation , Interna-
905 tional Journal of Solids and Structures 23(7) (1987) 821–840.
[20] J.W. Ju, Isotropic and Anisotropic Damage Variables in Continuum Damage Mechanics. Journal
of Engineering Mechanics ASCE 116(12) (1990) 2764–2770.
[21] S. Valliappan, M. Yazdchi, N. Khalili,Seismic analysis of arch dams – a continuum damage mechan-
ics approach, International Journal for Numerical Methods in Engineering 45(11) (1999) 1695–1724.
910 [22] J.A. Paredes, S. Oller, A.H. Barbat, New Tension-Compression Damage Model for Complex Anal-
ysis of Concrete Structures, J. Eng. Mech. 142 (2016) 1–16.
[23] F. Parisio, S. Samat, L. Laloui, Constitutive analysis of shale: A coupled damage plasticity ap-
proach, International Journal of Solids and Structures, 7576 (2015) 88-98.
39
 [24] J.Y. Wu, J. Li, R. Faria, An energy release rate-based plastic-damage model for concrete, Interna-
915 tional Journal of Solids and Structures, 43(3-4) (2006) 583-612.
[25] A. Matzenmiller, J. Lubliner, R.L Taylor, A constitutive model for anisotropic damage in fiber-
composites, Mechanics of Materials, 20(2) (1995) 125–152.
[26] R. Faria, J. Oliver, M. Cervera, Modeling Material Failure in Concrete Structures under Cyclic
Actions, Journal of Structural Engineering,130(12) (2004) 1997–2005.
920 [27] R. Faria, J. Oliver, M. Cervera, On isotropic scalar damage models for the numerical analysis of
concrete structures, CIMNE Monograph, No.198 (2000) Barcelona, Spain.
[28] M. Wang, X. Song, X. Gu, Y. Wu, Mechanical Behavior of Bolted Glulam Beam-To-Column Con-
nections With Slotted-in Steel Plates Under Pure Bending, World Conference on Timber Engineering
(2016) Viena, Austria.
925 [29] R. De Borst, Some recent issues in computational failure mechanics, International Journal for
Numerical Methods in Engineering, 52(12) (2001) 63–95.
[30] S. Oller, E. Car, J. Lubliner, Definition of a general implicit orthotropic yield criterion, Computer
Methods in Applied Mechanics and Engineering, 192(78) (2003) 895-912.
[31] X. Li, P.G. Duxbury, P. Lyons, Considerations for the Application and Numerical Implementation
930 of Strain Hardening with the Ho↵man Yield Criterion, Computers and Structures, 52(4) (1994)
633–644.
[32] R. De Borst, M.A. Crisfield, J.J.C. Remmers, C.V. Verhoosel, Non-linear Finite Element Analysis
of Solids and Structures, Wiley, 2nd edition (2012)
[33] P. Grassl P, On a damage-plasticity approach to model concrete failure, Engineering and Compu-
935 tational Mechanics, 162(EM4) (2009) 221-231.
[34] Z.P. Bazant, B.H. Oh, Crack band theory for fracture of concrete, Materials and Structures, 16(3)
(1983) 155–177.
[35] I. Lapczyk, J.A. Hurtado, Progressive damage modeling in fiber-reinforced materials, Composites
Part A Applied Science Manufacture 38 (2007) 2333-2341.
940 [36] J.C. Simo, R.L. Taylor, Consistent tangent operators for rate-independent elastoplasticity, Com-
puter Methods in Applied Mechanics and Engineering, 48(1) (1985) 101-118.
[37] F. Hashagen, R. De Borst, Enhancement of the Ho↵man yield criterion with an anisotropic hard-
ening model, Computers and Structures, 79 (2001) 637-651.
[38] DIANA, Users Manual - Version 10.1, DIANA FEA BV, Delft, The Netherlands (2016).
945 [39] M. Popovski, H.G. Prion, E. Karacabeyli, Seismic performance of connections in heavy timber
construction, Canadian Journal of Civil Engineering, 29 (2002) 389-399.
[40] M. Popovski, Seismic performance of braced timber frames, PhD Thesis, (2000) Department of
Civil Engineering, The University of Bristish Columbia, Vancouver, Canada.

40