Glass Struct. Eng. (2016) 1:19–37
DOI 10.1007/s40940-016-0015-4
CHALLENGING GLASS PAPER
Finite element analysis of timber-glass walls
Boštjan Ber · Miroslav Premrov ·
Andrej Štrukelj
Received: 29 December 2015 / Accepted: 15 March 2016 / Published online: 1 April 2016
© Springer International Publishing Switzerland 2016
Abstract The paper presents a finite element analysis (FEA) of experimentally investigated timber-glass
composite walls. So far a series of mechanical tests
with in-plane monotonous static and quasi-static cyclic
loading were carried out on timber-glass walls with the
goal to analyse their general behaviour and to experience failure mechanisms that occur. Timber-glass walls
consisted of glass panes, adhesively bonded to the timber frame with different types of adhesives, i.e. silicone,
polyurethane and epoxy. FEA was limited to timberglass walls subjected only to monotonous static loading. A commercial finite element code Ansys was used
to perform the simulation of experiments. As the types
of adhesive are decisive for the behaviour of timberglass walls, additional experiments on bulk adhesives
are briefly discussed. The main aim of the performed
FEA was to build as accurate as possible mathematical
model of timber-glass walls to achieve the correct general behaviour, failure mechanisms and to confirm the
composite action in case of using flexible adhesives,
e.g. silicone and polyurethane. Geometrically nonlinear finite element method was used including solid
B. Ber (B)
Kager hiša d.o.o., Ptuj, Slovenia
e-mail: bostjan.ber@gmail.com
B. Ber · M. Premrov · A. Štrukelj
Faculty of Civil Engineering, Transportation Engineering and
Architecture, University of Maribor, Ptuj, Slovenia
hexahedral finite elements. Some variations of material models and mesh size were performed to calibrate
the numerical model with the results of experimental
investigations in order to reach the desired response.
The above described experimental and FE investigations present a starting point for a comprehensive parametric study.
Keywords FEA · Nonlinear FE method · Timberglass composites · Adhesives · In-plane load
1 Introduction
Timber-frame wall elements (i.e. panel walls) represent a basic structural element of the prefabricated timber structure. Panel walls consist of sheathing boards,
mechanically fastened to a timber frame by means of
staples, nails or self-tapping screws. They have the ability to transmit both vertical (e.g. snow and dead load)
and horizontal loads, caused by gusts of wind or an
earthquake.
In a contemporary lightweight timber-frame building a series of fixed glazing is present due to lighting
conditions and for solar heat gains during the heating
season, which contribute significantly to the energy
efficiency of the building, (Žegarac Leskovar et al.
2013). However, a glazed part of building’s envelope is
unusable in terms of structural integrity. Since the contribution of stiffness and load-bearing capacity of glass
surfaces has been neglected so far, additional problems,
123
20
e.g. structural irregularity and uneven distribution of
horizontal force can occur.
For this matter, a possibility to integrate insulating
glass units (IGU) into the timber structure with thoughtful details has been investigated.
It should be noted that the current state of the art
in the field of timber structures is very conservative in
relation to the use of glass in structural purposes, mainly
due to the lack of investigations and guidelines for the
design of glass and timber-glass composite structures.
A starting point for the development of the European
standard for the design of glass structures represents the
report of the European Commission entitled “Guidance
for European Structural Design of Glass Components„
(Feldmann et al. 2014), in which all previous investigations on glass elements and structures are collected.
However, it states (pp. 66 and 152) that there are no
rules or even guidelines for using shear wall panels as
primary load-bearing elements. Only requirements for
the secondary elements, wherein wall panels loaded
perpendicular to its plane are listed in (Feldmann et al.
2014).
A special motivation to find missing answers on
the latter subject encouraged us to carry out an extensive investigation on timber-glass composite wall elements. The main goal was to achieve higher overall
stiffness of the building and more even distribution of
a horizontal force on panel walls. The assumed loadbearing capacity and stiffness of timber-glass wall elements were established by the use of different assembly concepts, whereby the variation of the adhesive,
bond line conditions and outer dimensions of timberglass walls were made. Among the adhesives, suitable
for joining timber and glass, we limited our choice to
silicone, polyurethane and epoxy. An extensive smallscale laboratory tests were carried out on those as
well.
2 Methods
Timber-glass walls were investigated experimentally
and numerically as well. In following sections the
experimental part of investigation is briefly introduced
to indicate the manner of mechanical testing and to
give a technical description of the specimens. Special
attention is intended to materials and their constitutive
models, which is crucial for further presented numerical study.
123
B. Ber et al.
2.1 Experimental study
Experimental investigations on timber-glass walls were
divided into two parts. First part includes mechanical racking tests of timber-glass walls while a second
part involves shake-table tests of one- and two-story
timber-glass building setups. The latter can be found
in Ber et al. (2015) and is not further discussed in this
work.
To carry out the mechanical tests, two groups
of timber-glass walls were formed, namely TG and
TGWE. Figure 1 presents specimens of the first testing group TG, which have common outer dimensions
and boundary conditions. TG-DS concept consists of
two 6.0 mm thick glass plates, which are adhesively
bonded onto a timber frame with a two-component
silicone. TG-S, TG-P and TG-E concepts, where a
single 10 mm thick thermally toughened glass pane
is placed in the mid-plane of the timber frame, differ from each other depending on the type of adhesive, namely two-component silicone, one-component
polyurethane and two-component epoxy. Similar selection of adhesives and joint types was used by Niedermaier (2005) for timber-glass and Huveners (2009) for
steel-glass connections. To determine the strength and
stiffness characteristics of TG walls, they were fixed to
the testing rig and subjected to a standardized racking
test according to (EN 594 1996, 2011). An increasing
monotonic load (Fh ) represents e.g. wind or seismic
load.
Figure 1 shows the composition and appearance of
TG type walls as well as a detailed description of elements and their dimensions. A more extensive description of the experimental study on TG walls is available
in Ber et al. (2014).
Prefabricated wall elements with fixed insulating
glass units have been in use for quite a while, however
never before for structural reason. In particular they can
be found in timber-frame construction (Kolb 2008). In
general, two different types of timber-glass wall elements (TGWE) appear in such structures. In the first
type, an IGU is of approximately square dimensions,
while the second type has two narrow IGUs separated
by an additional timber stud. In this way, TGWE specimens were designed, as seen in Fig. 2.
In this concept, a groove between a three-layer IGU
and a timber frame is filled with a one-component
polyurethane adhesive over the entire circumference. A
minor step was thus added to the technological process
Finite element analysis of timber-glass walls
21
1250 mm - timber frame
Fh
girder
90/80mm
5
glass pane
t = 10 mm
6
10
10
Detail - A
Fh,r
0.5
15
20
two glass panes
t = 6,0 mm
2C Epoxy
70
2C Silicone
1C PUR
2C Silicone
6
Timber frame
20
7
70
A
90
90
Timber frame
10
Timber frame
Thermally
toughened glass
2x 6,0 mm (TG-DS)
10 mm (TG-S/P/E)
fi16
6
90
12
stud
90/90mm
6
90
2640 mm - timber frame
TG-DS
TG-S/P/E
TG-E (single glass pane)
TG-S/P (single glass pane)
90
TG-DS (double glazing)
90
girder
90/80mm
glass pane
t = 10 mm
Detail - A
Detail - A
Fc
Ft
Fig. 1 Static system and assembly details of TG walls
Fv/3
Fh
girder 80/280 mm
Fv/3
Detail - A
160
Fv/3
girder 80/280 mm
TGWE1
Timber frame
160/160 mm
TGWE2
160
240
50
20
200
IGU - float
0.99 m × 2.03 m
stud 160/160 mm
IGU - float
0.99 m × 2.03 m
stud 160/160 mm
stud 160/160 mm
IGU - float
2.11 m × 2.03 m
stud 160/160 mm
stud 160/160 mm
A
1C PUR
15 5
A
140
Fh
Fv/2
28
Fv/2
girder 120/160 mm
girder 120/160 mm
Ft
12
IGU (6-16-6-16-6)
Fh,r
Fh,r
Fc
Ft
Fc
Fig. 2 Static system and assembly details of TGWE walls
of prefabrication to obtain a load-bearing composite
shear wall. TGWE walls were subjected to racking
tests according to (EN 594 2011) and quasi-static cyclic
tests according to (ISO 16670 2003) in order to analyse
the wall’s ductility, strength and stiffness degradation.
TGWE walls were subjected to a combination of vertical load, Fv , (e.g. load of a roof and floor structure) and horizontal load, Fh , (e.g. wind or seismic
load).
Dimensions of TGWE walls with a structural assembly detail is shown in Fig. 2. While the boundary conditions are schematically presented, a detailed description can be found in Ber et al. (2015).
2.2 Material models
Wood is an organic substance and thus exposed to
external factors during growth, which is reflected in
its non-homogeneity and anisotropy. The latter feature of growing differs between three mutually perpendicular directions, i.e. longitudinal, radial and tangential (Green et al. 1999). Therefore a linear-elastic
orthotropic material model was used to describe the
wood properly, whereby we ignored the difference
between tangential and radial directions. Consequently
six independent elastic constants are sufficient to
describe elastic properties of wood, which are listed in
123
22
B. Ber et al.
Table 1 Elastic constants and mechanical properties of wood
Material
Behaviour E0(X)
(MPa)
Spruce C24
fat,0,k
fac,0,k
ρm
E90(Y) E90(Z) νXY νYZ
νXZ
GXY
(MPa) (MPa) (–)
(–)
(MPa) (MPa) (MPa) (MPa) (MPa) (MPa) (kg/m3 )
(–)
GYZ
GXZ
fm,k
Linear
11,000 660
660
0.42 0.245 0.42 690
33
690
24
14
21
420
Spruce GL24h Linear
11,600 700
700
720
35
720
24
16.5
24
456
a
Values for longitudinal direction parallel to the grain
Table 2 Mechanical properties of glass, steel, concrete and aluminum
Material
Behaviour
E (MPa)
ν
G (MPa)
ft (MPa)
fc (MPa)
ρ(kg/m3 )
Thermally toughened glass
Linear
70,000
0.2
29,167
120
500
2500
Float glass
Linear
Steel
Linear
200,000
0.3
76,923
250
45
250
7850
Concrete
Linear
31,000
0.2
12,917
–
–
2500
Material
Behaviour
E (MPa)
ν
G (MPa)
fy (MPa)
Et (MPa)
ρ(kg/m3 )
Aluminum
Bilinear
71,000
0.33
26,692
280
500
2770
Table 1. Mechanical properties for sawn (C24) and laminated timber (GL24h) were obtained from standards
(EN 338 2003; EN 1194 2000), respectively. Absent
values of Poisson’s coefficients (νij ) and shear moduli
(Gij ) were taken from Green et al. (1999).
The behaviour of glass under increasing load is linearly elastic up to an instantaneous brittle failure. In
case of TG type walls, thermally toughened glass was
used (EN 12150-1 2000), while insulating glass units
(IGU) in case of TGWE type walls were composed of
three float glass panes (EN 572-1 2004). Mechanical
properties of glass are listed in Table 2.
For other materials, that appear in the mathematical models (steel, concrete and aluminum), the default
material properties from the extensive material library
of the commercial finite element code Ansys were
used (Table 2). All mechanical connectors (i.e. screws,
threaded rods, pins), support stirrups, angle brackets
and hold-downs were made of steel. Reinforced concrete was used as the foundation for TGWE walls. For
steel and concrete we assumed a linear elastic behaviour, therefore an isotropic elastic material model was
used. For aluminum spacers between glass panes of
the IGU, a bilinear elasto-plastic model with isotropic
hardening was chosen.
Additional mechanical tests were performed to
determine the structural performance of adhesives for
timber-glass walls when subjected to short-term sta-
123
tic loads. The adhesives tested were a two-part silicone Ködiglaze S; a polyurethane Ködiglaze P and a
two part epoxy Körapox 558 all produced by Kömmerling. The aim of uniaxial tensile tests was to determine
the visco-elastic and elasto-plastic properties independently of each other (Overend et al. 2011). Two types
of specimen’s geometry were used, namely S3a (DIN
53504 2010) and 1B (EN 527-2 2012). Uniaxial tensile tests, performed on Zwick Z030 testing machine
were divided into two interconnected parts. First the
visco-elastic properties were obtained followed by a
discrete loading strategy which was applied to establish
the elasto-plastic properties. It was necessary to undertake these investigations on the adhesives in order to
establish some of the fundamental mechanical properties, which were not available from the manufacturer.
However, some of the mechanical properties required
to characterize the stiffness of the bulk material were
available in the adhesive’s data sheets provided by the
manufacturer. Those are listed in the first portion of
Table 3.
To study the visco-elastic response of the chosen
adhesives, a series of uniaxial tension stress-relaxation
tests were performed. Specimens were first subjected
to instantaneous uniform tension (at a strain rate of
50 mm/min) until a predefined extension was reached.
The extension was then held constant and the stress
relaxation as a function of time was recorded. A decay
Finite element analysis of timber-glass walls
23
Table 3 Mechanical properties given by the manufacturer (first part of the table) and evaluated visco-elastic properties of the adhesives
(second part of the table)
Material
Silicone
Behaviour E
Nonlinear
G
ν
ft
εmax ρ
fv
(MPa) (MPa) (–)
(MPa) (MPa) (%)
(kg/m3 )
TMIN TMAX Gν
td
β
(◦ C)
(◦ C)
(s)
(1/s)
(MPa)
2.8
0.93
0.5
2.1
2.1
230
1370
−40
+150
0.351 100
0.0026
Polyurethane Nonlinear
1
1.3
0.49
2
2
450
1170
−30
+70
0.454 290
0.0016
Epoxy
2800
960∗
0.4∗ –0.46∗ 28∗
1500
−30
+120
Nonlinear
20–22 /
135.4
3600
0.0022
* Values estimated by the manufacturer
Table 4 Evaluated elasto-plastic properties of the adhesives
Adhesive
Time independent elasto-plastic
stress–strain polynomial
R2 -value Mean failure
stress
σmean (MPa)
Mean failure
strain
εmean
Specimen’s geometry S3a (L0 = 50 mm; Lgauge = 10 mm)
Silicone
σ = 1.6856 ε3 − 1.2884 ε2 + 1.9703 ε + 0.0098
0.9995
5.065
1.411
Polyurethane
σ = 2.6085 ε4 − 3.5412 ε3 − 0.0408 ε2 + 3.7315 ε + 0.01
0.9985
4.012
1.229
0.9999
1.397
0.0035
Epoxy
2E + 07 ε3 − 197510 ε2
σ=
≤ ε ≤ 0.00485
+ 624.69 ε + 0.5383 for 0.00021
Specimen’s geometry 1B (L0 = 150 mm; Lgauge = 50 mm)
Silicone
σ = 2.5384 ε3 − 1.3669 ε2 + 1.8454 ε − 0.014
0.9834
2.542
0.937
Polyurethane
σ = 2.1977 ε4 − 3.3422 ε3 + 0.8279 ε2 + 2.1857 ε − 0.0003
0.9779
2.650
1.204
Epoxy
σ = 692508 ε3 − 22633 ε2 + 478 ε + 0.0302
0.9739
4.130
0.014
time td , was reached when the stress fell to the constant
value. Further the stress vs. time curve was converted
into shear modulus vs. time curve, from which the
visco-elastic shear modulus G ν was obtained. Viscoelastic properties for each adhesive are summarised in
Table 3. For this test we used only specimens with S3a
geometry, prepared by the manufacturer.
To establish the elasto-plastic properties of the chosen adhesives, specimens were loaded in 5–10 increments up to failure. In each increment the load was kept
constant for a period td , which we defined in the preceding visco-elastic tests. An expression for the timeindependent elasto-plastic stress–strain relationship for
each type of the adhesive was obtained by polynomial
curve fitting of the discrete stabilised points. Elastoplastic properties were evaluated for all three chosen
adhesives and separately for each specimen’s geometry. The results are summarised in Table 4.
Finite element code Ansys contains an extensive
library of material models, which can simulate different behaviour of the same material. By using available
material characteristics of the adhesives, we chose two
different hyperelastic models and elasto-plastic material model with isotropic hardening. Table 5 shows the
selected material models of adhesives with constants
(hyperelastic models) and descriptions. Designation
of the FE model is composed of adhesive specimen’s
geometry followed by a manner of load application in
uniaxial tensile test wherein “M” denotes a monotonically increasing load and “EP” denotes cyclic loading
protocol.
Hyperelasticity is the property of rubber-like materials that exhibit elastic response at high strains. Such
a behaviour is also observed when testing and working with the silicone. To determine the hyperelastic
response of the silicone we used Arruda–Boyce strain
energy function, which is useful for the level of strains
up to the value of 300 % (Ansys 2014).
The used polyurethane adhesive, which material
properties are similar to those of silicone (Table 3), also
shows hyperelastic behaviour. However, polyurethane
exhibits twice the elongation of the silicone, therefore
Ogden strain energy function was used. Compared to
other, Ogden strain energy function provides the best
123
24
B. Ber et al.
Table 5 Selected material models for each type of adhesive
Adhesive
FE model designation
Material model (Ansys designation)
Silicone
S3a (M)
Initial shear modulus µ
Limiting network stretch λ L
Incompressibility parameter d1
(MPa)
(–)
(1/MPa)
0.7066
2.90E+11
0.2143
Hyperelastic—Arruda–Boyce (TB HYPER BOYCE)
S3a (EP)
Elasto-plastic model with isotropic hardening (TB MISO)
1B (EP)
Hyperelastic—Ogden (TB HYPER OGDEN)
Polyurethane
S3a (M)
S3a (EP)
Material constant µ1
Material constant α1
Incompressibility parameter d1
(MPa)
(–)
(1/MPa)
0.9126
14,486
0.3
Elasto-plastic model with isotropic hardening (TB MISO)
1B (EP)
Epoxy
S3a (M)
Uniaxial tensile test data (TB EXPE UNIAXIAL)
S3a (EP)
Elasto-plastic model with isotropic hardening (TB MISO)
Fig. 3 The display of calculated hyperelastic relations for a Arruda–Boyce and b Ogden strain energy function
response at higher levels of strains (up to 700 %). Based
on the entered nominal values of σ − ε uniaxial tensile
test, the material constants were calculated (Table 5)
and the remaining hyperelastic relations (i.e. biaxial
and shear) were estimated as shown in Fig. 3.
The large strain plasticity formulation, which was
used to describe the elasto-plastic response of adhesives, takes into account the von Mises criterion of plastification, combined with the assumption of isotropic
hardening. Material behaviour was described by entering the data (Table 4) in the form of “true stress—
logarithmic strain”. The finite element code Ansys pre-
123
sumes that the slope of the curve is equal to zero after
the last pair of data is entered.
By entering visco-elastic constants (G ν and β) the
viscous properties of the material are taken into account
implicitly through the evaluation of the strain-energy
constants, while combining with elasto-plastic material
models is not possible.
2.3 Constructing numerical models and simulation of
mechanical tests
The idea of the finite element method (FEM) is a division of a complex model into smaller, manageable
Finite element analysis of timber-glass walls
25
Fig. 4 Used finite elements: a SOLID95, b HYPER86 and c CONTA174 (Ansys 2014)
Fig. 5 Mesh details of TG wall’s elements
pieces. One of the first finite element analyses of adhesively bonded joint was performed by Adams and Peppiatt back in 1974 (Silva and Campilho 2012).
The purpose of the analysis was to develop accurate numerical models of various experimentally investigated timber-glass wall elements. The main focus was
laid on the global behaviour and their response under
the influence of external forces. An academic version of
the commercial finite element code Ansys R14.5 was
used to build and calculate finite element models of
TG walls, which were calibrated to measured values of
displacements from mechanical tests.
The finite element mesh of timber-glass walls consisted of solid volume elements (SOLID95), defined
with 20 nodes and three degrees of freedom per node.
Solid element has the ability to achieve large strains
and it tolerates irregular shape. For calculating hyperelastic materials (i.e. adhesives), the preferred choice
was 8-node solid finite element (HYPER86) designed
for high strains and based on a hybrid formulation with
constant pressure. This type of element can be used
to model compressible and incompressible materials.
In each step of the calculation an update of the model
geometry is done, due to the ability of achieving large
deformations. Touching surfaces between the structural
elements were defined using surface-to-surface nonlinear contact elements (CONTA174) with three degrees
of freedom per node. Contact elements adopt geometric
characteristics of connected solid elements’ surfaces.
Above mentioned finite elements are shown on Fig. 4.
Finite element mesh for all six test groups of TG
and TGWE wall elements is shown on Figs. 5 and 6.
For structural elements with the regular shape (i.e. timber frame, glass and adhesive joint) the use of hexahedral solid elements was required. The other elements of irregular shape (e.g. screws, angle brackets,
hold-downs, steel rods, etc.) were meshed with tetrahedrons.
123
26
B. Ber et al.
Fig. 6 Mesh details of TGWE wall’s elements
Fig. 7 Boundary conditions of TG and TGWE walls with details of the mounting support (D1 and D2)
Due to relatively different dimensions of the structural elements (see Figs. 5, 6) a manual selective
mesh sizing was performed to adjust the mesh quality separate for each member of the timber-glass
model.
3D-models of timber-glass walls were positioned
and supported in the same manner as in a laboratory.
Figure 7 shows the disposition of the simulation, supports and direction of the input load.
Tensile loaded support of the TG type walls (Fig. 7a)
is composed of two 8.0 mm thick steel plates, fastened
to a timber frame with three M16 bolts. Steel plates are
then hinged to a steel tube with a diameter of 55 mm,
which is unmovable. Compressive support was modeled by combining various flat steel elements to achieve
123
the appropriate external dimensions and thickness of
the bedding. Steel elements representing the compressive support were restrained as shown in Fig. 7a. Protection against the lateral torsional buckling was simulated
with a roller support, allowing only in-plane displacements at the free end of the TG model where the vertical
load Fh , was applied.
In case of TGWE models, a simplified steel angles
(type WKR-285) and 12 × 300 mm threaded rods were
used to simulate the connection with the RC foundation slab, which was fixed along the entire bottom
surface (Fig. 7b). However, all translational displacements were released between TGWE model and RC
foundation slab to simulate the actual boundary condition in this area. The model was secured against
Finite element analysis of timber-glass walls
tilting with the introduction of a sliding support, which
allowed only vertical and horizontal in-plane displacement of the model (Fig. 7b). Connections between the
individual components of an assembly were defined
by linear (i.e. bonded) and nonlinear (i.e. frictionless, rough and no separation) types of contact elements.
TGWE models were loaded in vertical and horizontal directions. The horizontal load Fh , was applied to
the upper edge of the wall element, as shown in Fig. 7b.
The load increment was simulated with steps of 10
kN/s, until the value of 70 kN (TGWE2) or 80 kN
(TGWE1) was reached. The total vertical load of 60
kN was applied to the model in the first second of simulation. Vertical load was applied to the model in form
of a point load to every column (Fig. 7b).
3 Results and discussion
The following section presents the results of numerical analysis where we focus on parameters, which
we observed at the experimental investigations. These
parameters are:
• Failure mechanisms,
• In-plane displacement (w and v) as a function of
in-plane load Fh ,
• Stress–strain state in the glass pane.
3.1 Failure mechanisms of TG and TGWE models
The deformed shape of the numerical model is the first
indicator of appropriate boundary conditions depending on the loading conditions (Silva and Campilho
2012). When analyzing a numerical model TG-DS,
an increased deformation of the adhesive joint was
observed at approximately 16 kN of vertical load. The
value of principal stress in the silicone adhesive joint at
this loading stage was approximately 2.7 MPa, which
is already above its plastic limit (Tables 3, 4). The
same occurred with mechanically tested specimens STO1/O2/O3, where in the range of 16–17 kN the deformations of the adhesive joint started to grow large.
Figure 8 shows the deformations of the timber frame
and adhesive joint of the numerical model compared
with actual deformations, observed in the mechanical
tests.
In the test models TG-S and TG-P we observe
increased deformation (i.e. opening) of the timber
27
frame’s corner from 8.0 to 13 kN of vertical load, which
is presented in Fig. 9a. The default value of the friction
coefficient, which was applied to the contact between
screw and timber has proved to be suitable. Values of
numerical results were ranging from 4.0 to 5.0 mm,
which are close to values of displacements obtained
from the mechanical tests. Those values of displacements were not measured explicitly, however, close to
yielding point they amounted approximately 5.0–7.0
mm.
Figure 9b shows the failure mechanism of a timber
frame corner on TG-E model, where deformations of
the vertical timber element perpendicular to the grain
are visible, due to the compression of the horizontal
timber element. Deformation of the numerical model
at 40 kN of vertical load were 2.0 mm, while the deformation of mechanically tested walls at the same vertical load ranged from 3.0 to 4.0 mm. The reason for
this difference can be attributed to inhomogeneities of
wood, which cannot be modelled exactly. The latter
could be simulated with a more detailed discussion of
orthotropy, i.e. to differ between the radial and tangential direction. However, in our case this was not possible because timber-glass walls had randomly oriented
timber elements.
Characteristic failure mechanisms of TGWE models are shown on Fig. 10. At 70 kN of horizontal load,
a shift of 2.7 mm between the stud and the girder was
observed (Fig. 10a, b, d). Measurement of the optical
system GOM during the mechanical investigations on
the specimen TGWE2-3, at the same value of horizontal load, showed the displacement of 2.4 mm. The
difference of 0.3 mm, between the numerical and measured value, confirms a proper response of the timber frame corner joint. The numerical model was also
showing displacements and deformations in lower tensile loaded corner (Fig. 10d, e), where a horizontal shift
of the joint and uplifting of the timber frame corner off
the foundation can be seen. Although uplifting of the
timber frame’s lower corner is restricted (or rather limited) with a pair of steel hold-downs and two threaded
rods (Fig. 10e), a total vertical displacement of the FE
model equals to 7.0 mm. However, this is a fairly good
result according to measured values ranging from 8.0 to
9.0 mm.
In Fig. 10(c, f), the locations of maximum stresses
according to von Mises criterion of plastification are
shown. Areas of maximum stresses in the adhesive joint
coincide with the initiation of the failure mechanisms,
123
28
B. Ber et al.
Fig. 8 Failure mechanism of TG-DS test group
Fig. 9 Matching failure mechanisms of numerical models and experimentally investigated TG-S/P/E walls
shown in Fig. 10(a, b, d). A detailed analysis of this
mechanism is not specifically considered, whereas
the aim of the investigation is to discuss the general
response of the TG systems.
123
3.2 In-plane displacement of TG and TGWE models
Diagrams of vertical displacement w, as a function of
vertical load Fh , for numerical models as well as for
Finite element analysis of timber-glass walls
29
Fig. 10 Failure mechanisms (a, b, d, e) and the state of stress in the adhesive joint (c, f) in TGWE1 and TGWE2 models
mechanically tested TG walls are presented in Fig. 11.
On the left side of Fig. 11(a, c, e), curves of different
numerical models are shown, together with the curve
of measured mean values. Diagrams on the right side
(Fig. 11b, d, f) show the curve of the selected numerical
model (best fit) together with measured master curves
of the individual experimentally investigated specimen.
For test group TG-DS, only the numerical results
using a hyperelastic material model of the silicone
adhesive are shown (Fig. 11a, b), due to numerical difficulties of elasto-plastic material models. Even though
load-displacement curve of the numerical model does
not have the exact shape as the curve of mechanically
tested walls, it still possess an approximately equal initial stiffness. Moreover, in the range from 12 to 16 kN
the numerical model exhibits comparable vertical displacements with all specimens (ST-O1/O2/O3) as seen
from Fig. 11a, b.
In test groups TG-S and TG-P, different material
models of adhesives were used (see Table 5). However, the solution converged in all cases. Figure 11(c, e)
shows diagrams of different versions of these models,
where a slightly higher stiffness of hyperelastic models
S3a (M) can be seen. In both groups (TG-S and TG-P),
a mesh density variation (2.5 and 5.0 mm) with hyperelastic material models of adhesives was made. With
this alternation 15 % (TG-S) to 20 % (TG-S) lower
value of vertical displacement and more than 10 %
higher stiffness was achieved with the models, where
a finer mesh of finite elements were used. Although,
in both cases (TG-S and TG-P) the most comparable
solutions were obtained with a variation of the numerical model, in which an elasto-plastic material models
of the adhesive S3a (EP) and 1B (EP) were used. The
curve 1B (EP) of the TG-S model (Fig. 11c, d) almost
perfectly fits the master curves of ST-S1 and ST-S2, up
to the value of 8.0 kN. A plastic hinge is then formed
at 25 mm of vertical displacement, indicating the failure of the model. The results of the numerical model
TG-P S3a (EP) and 1B (EP) have converged up to 8.0
kN of vertical load, when they became unstable. The
curve TG-P 1B (EP), shown in Fig. 11(e, f), fits best
up to 4.0 kN of vertical load and even coincides with
the master curve of ST-P3 wall. Numerical Fh − w
curves of models TG-DS, TG-S and TG-P show an
increasing rigidity. There is no noticeable plastification
comparing to mechanical tests, where a plastic limit
can be clearly observed. Models even exhibit bi-linear
123
30
B. Ber et al.
(b) 20
20
TG-DS
18
18
16
16
14
14
Vertical load - Fh [kN]
Vertical load - Fh [kN]
(a)
12
10
8
TG-DS
12
10
8
6
6
4
4
TG-DS (Exp.)
2
2
TG-DS S3a (M)
TG-DS S3a (M)
0
0
0
15 20 25 30 35 40
Vertical displacement - w [mm]
45
50
0
(d) 13
13
TG-S
12
Vertical load - Fh [kN]
10
11
11
10
10
9
9
8
7
6
5
TG-S (Exp.)
15 20 25 30 35 40
Vertical displacement - w [mm]
45
7
6
5
ST-S1
3
ST-S2
2
2
ST-S3
1
TG-S 1B (EP)
TG-S S3a (EP)
TG-S 1B (EP)
50
8
4
0
0
0
5
10
15 20 25 30 35 40 45
Vertical displacement - w [mm]
50
55
0
60
(f)
15
TG-P
14
13
12
12
11
11
10
10
8
7
6
TG-P (Exp.)
5
TG-P 1B (EP)
60
6
5
3
0
55
7
3
TG-P S3a (EP)
50
8
4
1
15 20 25 30 35 40 45
Vertical displacement - w [mm]
9
4
2
10
TG-P
14
9
5
15
13
Vertical load - Fh [kN]
Vertical load - Fh [kN]
10
3
4
1
(e)
5
TG-S
12
Vertical load - Fh [kN]
(c)
5
ST-P1
ST-P2
2
ST-P3
1
TG-P 1B (EP)
0
0
5
10 15 20 25 30 35 40 45 50 55 60 65
Vertical displacement - w [mm]
0
5
10 15 20 25 30 35 40 45 50 55 60 65
Vertical displacement - w [mm]
Fig. 11 Diagrams of vertical displacement (w) as a function of vertical load (Fh ) for numerical models and experimentally investigated
TG-DS (a, b), TG-S (c, d) and TG-P (e, f) types of walls
123
Finite element analysis of timber-glass walls
31
45
45
TG-E
40
40
35
35
Vertical load - Fh [kN]
Vertical load - Fh [kN]
TG-E
30
25
20
15
30
25
20
ST-E1
ST-E2
ST-E3
ST-E4 (2)
ST-E5 (2)
ST-E6 (2)
TG-E 1B (EP)
15
TG-E (Exp.)
10
10
TG-E(2) (Exp.)
TG-E S3a (M)
5
5
TG-E 1B (EP)
0
0
0
5
10
15 20 25 30 35 40
Vertical displacement - w [mm]
45
50
(a)
0
5
10
15 20 25 30 35 40
Vertical displacement - w [mm]
45
50
(b)
Fig. 12 Diagrams of vertical displacement (w) as a function of vertical load (Fh ) for numerical models and experimentally investigated
TG-E type walls
hardening with increasing tangential stiffness. The
solution can be found in more complex constitutive
models of wood focusing on elastoplasticity and viscoelasticity, proposed by Borst (2012).
When performing non-linear analyzes of TG type
walls with flexible adhesive joints (i.e. silicone and
polyurethane), we often deal with the problem of convergence. In numerical analysis for each load step a
calculation is done. If for the given step or increment
of load the calculation does not lead to a solution, a
procedure is repeated using smaller intermediate steps.
Numerical model does not converge to a solution in
case where at reduced load increment the calculation
is not possible (Silva and Campilho 2012).
A calculation of two numerical models of TG-E
was conducted, namely TG-E S3a (M) and 1B (EP),
(see Table 5). Slightly better results were obtained
with the elasto-plastic material model of epoxy, 1B
(EP), as seen in Fig. 12a. The Fh − w diagram in
Fig. 12a shows a perfect matching of both numerical models with the experiment up to 30 kN of the
vertical load. In the same diagram, the curve of the
test group TG-E (2) shows a little higher stiffness.
A cause for this is probably the fact that TG-E (2)
test group was subjected to an older version of the
(EN 594 1996) testing protocol, which prescribes three
cycles of loading, which could influence on the stiffness
rate.
From 25 to 40 kN, the F-w curve of the model
TG-E 1B (EP) does not show signs of plastification or any stiffness degradation. To simulate this
effect, it would be sensible to use more advance
anisotropic material model of wood. To use this kind
of advanced model, extensive mechanical tests should
be performed, which is unfortunately beyond the scope
of this investigation. However, for an exact prediction
of TG walls behaviour this step is necessary in the
future.
For each type of timber-glass wall element (TGWE1
and TGWE2) we conducted two numerical models
depending on the material model of polyurethane adhesive where S3a (M) represents a hyperelastic and the
S3a (EP) elasto-plastic material model.
Figure 13 represents the diagrams of horizontal displacement v, at the top of the wall, depending on the
horizontal load Fh , for numerical and physical models of TGWE. Once again, a slightly higher stiffness
of the numerical models S3a (M) is observed. Similar occurred with the model TG-P S3a (M) where the
difference between the responses when using different
123
32
B. Ber et al.
90
90
TGWE1
80
80
70
70
Horizontal load - Fh [kN]
Horizontal load - Fh [kN]
TGWE1
60
50
40
30
TGWE1-2 (Exp.)
20
60
50
40
TGWE1-2 (Exp.)
30
TGWE1-1 (1C) - Exp.
20
TGWE1-3 (1C) - Exp.
TGWE1 S3a (M)
10
TGWE1 S3a (M)
10
TGWE1 S3a (EP)
TGWE1 S3a (EP)
0
0
0
5
10
15
20
25
30
35
40
45
0
10
15
20
25
30
35
Horizontal displacement - v [mm]
(a)
(b)
40
45
80
80
TGWE2
TGWE2
70
70
60
60
50
40
30
TGWE2-3 (Exp.)
20
TGWE2 S3a (M)
10
Horizontal load - Fh [kN]
Horizontal load - Fh [kN]
5
Horizontal displacement -v [mm]
50
40
TGWE2-3 (Exp.)
30
TGWE2-1 (1C) - Exp.
20
TGWE2-2 (1C) - Exp.
TGWE2 S3a (M)
10
TGWE2 S3a (EP)
TGWE2 S3a (EP)
0
0
0
5
10 15 20 25 30 35 40 45 50 55 60 65 70
0
5
10 15 20 25 30 35 40 45 50 55 60 65 70
Horizontal displacement - v [mm]
Horizontal displacement - v [mm]
(c)
(d)
Fig. 13 Diagrams of horizontal displacement (v) as a function of horizontal load (Fh ) for numerical models and experimentally
investigated TGWE type walls
material models is much higher, comparing to TGWE
models. Results of both TGWE S3a (M) models show
more realistic initial stiffness in the range between
0kN < Fh < 50kN for TGWE1 and 0kN < Fh <
25kN for TGWE2. From these values of Fh on, S3a
(EP) models more accurately describe the response of
123
TGWE1-2 and TGW2-3. After formation of a plastic hinge at approximately 66 kN of Fh , the TGWE2
S3a (EP) curve is still in a rather good agreement with
the slope of the TGWE2-3 (Exp.) curve. Figure 13b
and d show the Fh − w diagrams, where envelopes of
the first cycle of mechanically tested walls subjected
Finite element analysis of timber-glass walls
33
Fig. 14 Maximum and minimum values of principal stress in glass for models TG-S, TG-P and TG-E
to quasi-static cyclic load are added. In both cases
(TGWE1 and TGWE2) wall elements exhibit a similar response up to 60 kN of Fh , afterwards they fail
at different values of horizontal load (62 kN < Fh <
84 kN). Again a higher flexibility of numerical model
S3a (EP) can be seen, which is now even more pronounced. However, more pronounced is also the difference between maximal horizontal displacements of
S3a (M) and S3a (EP), which is in favor of the latter.
Using a hyper-elastic material model of the adhesive joint seems to predict the initial stiffness better,
but it overestimates the actual load-bearing capacity
of TGWE. On the other side, the elasto-plastic material model of the adhesive joint slightly underestimates
the initial stiffness, but more accurately describes the
global response of the wall element.
3.3 Stress–strain state in the adhesive joint and glass
of TG and TGWE models
In Fig. 14 principal stresses in glass with the local maxima and minima of numerical models TG-S, TG-P and
TG-P-E can be observed. Since the stress state image
of TG-S 1B (EP) and TG-P 1B (EP) is practically the
same, double values are written in Fig. 14a.
In all three cases (TG-S, TG-P and TG-E), maximum
value of the principal stress is located in the vicinity
of supports. The fact that maximum stress at tension
support is only two times higher at four times higher
value of Fh , comparing TG-E 1B (EP) and TG-S 1B
(EP), it is evident stresses are more evenly distributed
over the glass pane. A similar phenomena is observed
when analyzing TG-P 1B (EP). In this case, a five times
lower vertical load yields only three times lower value
of maximum stress at tension support, compared to TGE 1B (EP).
For a better insight into the performance of adhesive
joints, the deformed shape of silicone and polyurethane
together with strains in epoxy adhesive joint are shown
on Fig. 15. The deformed shape of silicone and
polyurethane adhesive joint clearly shows the rotation
of the glass pane in a timber frame (Fig. 15a), causing
tension strains in the upper-right and lower-left corner
and compression strains in the opposite corners. Strains
in epoxy adhesive joint (Fig. 15b) are directly connected with principal stresses in a glass plate (Fig. 14b),
therefore the highest values in both cases are located
at supports. Now it is evident that TG-E due to the
stiff epoxy adhesive joint exhibits a fully composite
action.
During mechanical tests a wall was deformed in
form of a parallelogram, followed by shear failure of
the adhesive joint in the corners. Larger global displacements of the wall caused the outer glass pane
to detach from IGU. This failure mechanism began
123
34
B. Ber et al.
Fig. 15 Deformed model of silicone and PUR adhesive joint (a), and strains in epoxy adhesive joint (b)
Fig. 16 Optical measurements of surface strains by GOM system (left) and FE simulation of the stress state in the glass (right) for
both types of TGWE
to show at Fh ≈ 30 kN in case of TGWE2-3 and at
Fh ≈ 40 kN in case of TGWE1-2. A parallel measurement of strains on the wall’s surface by an optical system GOM confirms this phenomena. Figure 16
123
shows the comparison between the measured major
strains on the wall’s surface (left) and the von Mises
stress state (right) in the first (i.e. outer) glass plate of
IGU.
Finite element analysis of timber-glass walls
35
Fig. 17 State of stress in the second and third glass pane at Fh = 70 kN for TGWE1
Fig. 18 State of stress in the second and third glass pane at Fh = 70 kN for TGWE2
In both cases (TGWE1-2 and TGWE2-3) the limit
state of strains in the adhesive joint is evident. In upper
left and lower right corners, measured major strains
are reaching 4.0 %. In the same corners of numerical
models (TGWE1 and TGWE2), stress in the glass pane
at the surrounding area began to reduce, which indicates
a gradual failure of the adhesive joint.
In continuation of mechanical tests, the second
and third glass pane take over all (vertical and horizontal) load. Figures 17 and 18 present von Mises
stress state in both glass plates for numerical models
TGWE1 and TGWE2 at loading stage; Fh = 70 kN and
Fv = 60 kN.
In the case of TGWE1 (Fig. 17) the observed stress
state is of the same magnitude, wherein the second glass
pane is subjected to a slightly higher load.
In case of TGWE2 (Fig. 18), the stress in the third
pane is slightly higher. The difference in stress magnitude between the second and third glass pane is quite
noticeable (20 % for TGWE2), while in case of TGWE1
123
36
B. Ber et al.
TG-DS S3a (M)
TG-DS (Exp.)
TG-S 1B (EP)
536
TGWE1 S3a (M)
543
TGWE1 S3a (EP)
376
TG-S (Exp.)
373
TG-P 1B (EP)
353
4607
4190
TGWE1-2 (Exp.)
4900
TGWE1-1 (Exp.)
4945
TGWE1-3 (Exp.)
4654
TGWE2 S3a (M)
TG-P (Exp.)
324
TGWE2 S3a (EP)
TG-E 1B (EP)
1570
TG-E (Exp.)
1544
TG-E(2) (Exp.)
2415
TGWE2-3 (Exp.)
2875
TGWE2-1 (Exp.)
2020
0
2717
500
1000
1500
2000
Racking stiffness - R [N/mm]
2948
TGWE2-2 (Exp.)
2600
0
2500
(a)
1000 2000 3000 4000 5000
Racking stiffness - R [N/mm]
6000
(b)
Fig. 19 Racking stiffness diagram of numerical models and experimentally investigated TG and TGWE type walls
the difference is only about 4.0 %. The difference in
stress between the left and the right panes in TGWE2
is caused by the direction of the horizontal load.
4 Conclusions
The results of numerical analysis are in good agreement
with the experimental analysis of TG walls. With accurate material models of adhesives, failure mechanisms
are well simulated and calculated values of vertical displacements can be compared to measured ones.
For models with elastic adhesive joints (TG-DS, TGS and TG-P), we managed to capture the behaviour
of experimentally tested wall elements, which was our
main goal. For a better match, especially of the stress–
strain state of the model, further investigations on bulk
adhesives should be carried out to improve material
models of flexible adhesives.
The response of TG-E model is the closest to actual
situation. Vertical displacements and the stress–strain
state of the numerical model are to a great extent
comparable to measured values. The next step here is
to use the elasto-plastic material model of timber to
include irreversible deformations, which are absent at
the moment.
Figure 19 shows a racking stiffness diagram, which
includes selected numerical models and mean values of
the experimentally investigated TG and TGWE walls.
There is a good correlation in all four groups of numerical and physical models, with the negligible difference
123
of stiffness (<2.0 %) in groups TG-DS, TG-S and TGE. However, the difference in group TG-P amounts to
8.9 %.
In case of TGWE, both material models of the
polyurethane adhesive joint present a right choice. With
the use of hyperelastic material model, a closer-toreality response was achieved, especially in the first
part, as seen from diagrams in Figs. 13 and 19b. The
stiffness of both numerical models S3a (M) in comparison with TGWE1-2 and TGWE2-3, differ by less
than 6.0 %. In the second part of the simulation, a more
realistic response is achieved with the use of an elastoplastic material model S3a (EP), which is evident as
the numerical Fh − v curve fits better in this final stage.
According to Figs. 17 and 18, we conclude that the
stress state in the glass panes of IGU is the function of
the IGU’s position in the timber frame. By moving an
IGU towards the mid-plane of the timber frame, a more
even distribution of stress in the glass panes would be
achieved. It would be sensible to carry out a parametric
study of the IGU’s position in a timber frame and to
determine the size of this impact.
Acknowledgments The research support provided by the EU
through the European Social Fund ’Investing in your future’ is
gratefully acknowledged.
References
ANSYS, Inc. Theory Manual. 001369. Twelfth Edition. SAS IP,
Inc. Accessed June 2014
Finite element analysis of timber-glass walls
Ber, B., Šusteršič, I., Premrov, M., Štrukelj, A., Dujič, B.: Testing of timber-glass composite walls. Proc. Instit. Civ. Eng.:
Struct. Build. 168(7), 500–513 (2015). doi:10.1680/stbu.
13.00105
Ber, B., Premrov, M., Štrukelj, A., Kuhta, M.: Experimental
investigations of timber-glass composite wall panels. Constr. & Build. Mater. 66, 235–246 (2014)
da Silva, L., Campilho, R.: Advances in Numerical Modelling of
Adhesive Joints. Springer, Heidelberg (2012)
de Borst, K., et al.: Mechanical characterization of wood: an
integrative approach ranging from nanoscale to structure.
Comput. Struct. (2012). doi:10.1016/j.compstruc.2012.11.
019
Deutsche Norm. DIN 53504.: Testing of rubber—determination
of tensile strength at break, tensile stress at yield, elongation
at break and stress in a tensile test, German Standard. Beuth
Verlag GmbH (2010)
European Committee for Standardization. EN 594:1996: Timber
structures—test methods—racking strength and stiffness of
timber frame wall panels. Brussels (1996)
European Committee for Standardization. EN 1194:2000: Timber structures—glued laminated timber strength classes and
determination of characteristic values. Brussels (2000)
European Committee for Standardization. EN 12150-1:2000:
Glass in building—thermally toughened soda lime silicate
safety glass. Part 1: definition and description. Brussels
(2000)
European Committee for Standardization. EN 338:2003: Structural timber—strength classes. Brussels (2003)
European Committee for Standardization. EN 572-1:2004: Glass
in building—basic soda lime silicate glass products—Part 1:
definitions and general physical and mechanical properties.
Brussels (2004)
37
European Committee for Standardization. EN 594:2011: Timber
structures—test methods—racking strength and stiffness of
timber frame wall panels. Brussels (2011)
European Committee for Standardization. EN ISO 527-2:2012.
Plastics—determination of tensile properties—Part 2: test
conditions for moulding and extrusion plastics. Brussels
(2012)
Feldmann, M., Kasper, R. et al.: JRC Scientific and Policy
Report—Guidance for European Structural Design of Glass
Components. Publications Office of the European Union,
Luxembourg (2014). doi:10.2788/5523
Green, D.W., Winandy, J.E., Kretschmann, D.E.: Mechanical
properties of wood. In: Wood Handbook: wood as an Engineering Material, Chapt. 4.1–4.45. USDA Forest Service
GTR-113 (1999)
Huveners, E.M.P.: Circumferentially adhesive bonded glass
panes for bracing steel frames in facades. Ph.D Thesis. University of Technology Eindhoven, Eindhoven (2009)
International Standard. ISO 16670:2003(E): Timber structures—
joints made with mechanical fasteners—quasi-static
reversed-cyclic test method. Zurich (2003)
Kolb, J.: Systems in Timber Engineering. Birkhäuser Verlag AG,
Basel (2008)
Niedermaier P.: Holz-Glas-Verbundkonstruktionen. Ein Beitrag
zur Aussteifung von filigranen Holztragwerken. Ph.D Thesis. University of Technology Munich, Munich (2005)
Overend, M., Jin, Q., Watson, J.: The selection and performance
of adhesives for a steel-glass connection. Int. J. Adhes.
Adhes. 31, 587–597 (2011)
Žegarac Leskovar V., Premrov M.: Energy-Efficient TimberGlass Houses, (Green Energy and Technology). Springer,
London (2013). doi:10.1007/978-1-4471-5511-9
123