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

net/publication/360128166

Modeling stiffness of connections and non-structural elements for dynamic

response of taller glulam timber frame buildings

Article in Engineering Structures · June 2022

DOI: 10.1016/j.engstruct.2022.114209

CITATIONS READS

27 11,412

4 authors:

Saule Tulebekova Kjell Arne Malo

Norwegian University of Science and Technology Norwegian University of Science and Technology
12 PUBLICATIONS 58 CITATIONS 90 PUBLICATIONS 1,978 CITATIONS

SEE PROFILE SEE PROFILE

Anders Rønnquist Petter Nåvik

Norwegian University of Science and Technology Norwegian University of Science and Technology
153 PUBLICATIONS 2,437 CITATIONS 41 PUBLICATIONS 639 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Saule Tulebekova on 25 April 2022.

The user has requested enhancement of the downloaded file.

 Engineering Structures 261 (2022) 114209

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Modeling stiffness of connections and non-structural elements for dynamic

response of taller glulam timber frame buildings
Saule Tulebekova a ,∗, Kjell Arne Malo a , Anders Rønnquist a , Petter Nåvik a,b
a
Department of Structural Engineering, Norwegian University of Science and Technology (NTNU), Richard Birkelands vei 1A, Trondheim, 7491, Norway
b
SWECO Norge AS, Drammensveien 260, Oslo, 0212, Norway

ARTICLE INFO ABSTRACT

Keywords: Currently, there is limited knowledge of the dynamic response of taller glue laminated (glulam) timber
Taller timber buildings buildings due to ambient vibrations. Based on previous studies, glulam frame connections, as well as non-
Glulam connections structural elements (external timber walls and internal plasterboard partitions) can have a significant impact
Finite element modeling
on the global stiffness properties, and there is a lack of knowledge in modeling and investigation of their
Dynamic identification
impact on the serviceability level building dynamics. In this paper, a numerical modeling approach with the
Model updating
use of ‘‘connection-zones’’ suitable for analyzing the taller glulam timber frame buildings serviceability level
response is presented. The ‘‘connection-zones’’ are generalized beam and shell elements, whose geometry and
properties depend on the structural elements that are being connected. By introducing ‘‘connection-zones’’,
the stiffness in the connections can be estimated as modified stiffness with respect to the connected structural
elements. This approach allows for the assessment of the impact of both glulam connection stiffness and non-
structural element stiffness on the dynamic building response due to service loading. The results of ambient
vibration measurements of an 18-storey glulam timber frame building, currently the tallest timber building in
the world, are reported and used for validation of the developed numerical model with ‘‘connection-zones’’.
Based on model updating, the stiffness values for glulam connections are presented and the impact of non-
structural elements is assessed. The updating procedure showed that the axial stiffness of diagonal connections
is the governing parameter, while the rotational stiffness of the beam connections does not have a considerable
impact on the dynamic response of the glulam frame type of building. Based on modal updating, connections
exhibit a semi-rigid behavior. The impact of non-structural elements on the mode shapes of the building is
observed. The obtained values can serve as a practical reference for engineers in their prediction models of
taller glulam timber frame buildings serviceability level response.

1. Introduction The stiffness, mass, and damping properties and their distributions in
buildings are the factors that affect the vibration response of taller
Rising awareness of the importance of sustainability in the general glulam timber frame buildings. Estimation of stiffness in glulam timber
public has led professionals in structural engineering to incorporate the frame buildings can be challenging and depends on the structural
use of environmentally friendly materials in their design. Therefore, the system used [4]. Experimental results from ambient vibration mea-
focus on timber, which is known for its natural origin and negligible
surements show that the eigenfrequencies obtained from the numerical
carbon footprint compared to widely used steel and concrete, is under-
modeling can be systematically underestimated due to the assumption
standable. Currently, timber as a structural material is being extensively
utilized in the construction of buildings [1]. of pinned connections in glulam timber frames [5,6].
In general, the lateral loading on taller glulam timber frame build- The past studies on taller glulam timber frame buildings emphasize
ings is governed by wind-induced vibrations [2]. The intrinsic proper- the fundamental importance of connections in timber structures under
ties of wood, including low density and stiffness, make taller timber serviceability level dynamic loading [7]. In glulam timber frame build-
buildings susceptible to horizontal excitation under wind loading [3]. ings, the dowel-type connections make a considerable contribution to

∗ Corresponding author.
E-mail address: [email protected] (S. Tulebekova).

https://doi.org/10.1016/j.engstruct.2022.114209
Received 30 October 2021; Received in revised form 19 February 2022; Accepted 27 March 2022
Available online 22 April 2022
0141-0296/© 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
S. Tulebekova et al. Engineering Structures 261 (2022) 114209

the structure stiffness [8]. The estimation of such connection stiffness GL30c in accordance with EN 14080:2013 [17]. The average beam
can, however, be challenging. In a number of studies, it has been cross-section is 400 mm by 500 mm, the corner column cross-section
shown that the static dowel connection stiffness, which is calculated is 600 mm by 1500 mm, and the average diagonal cross-section is
in accordance with Eurocode 5 [9], is different from the stiffness 600 mm by 750 mm. The glulam truss members are interconnected
calculated from the in-service cyclic tests [8,10]. The common practice with combinations of 10 mm thick slotted-in steel plates and 12 mm
in modeling the dynamic response of glulam timber frame buildings diameter dowels [18]. The steel grade for plates is S355 in accordance
is to assign pinned conditions to the connections between glulam with EN 10025-2, which corresponds to 355 MPa yield strength [19].
elements, [5,6,11,12]. The steel grade for dowels is EN 1.4418 in accordance with EN 10088,
The impact of non-structural elements on the serviceability level and has a measured yield strength of 755 MPa [20]. Cross-laminated
response of tall buildings has been acknowledged in both steel and timber (CLT) is used in the elevator and staircase shafts, but is not
reinforced concrete buildings [13]. The previous studies on ambient designed to be part of the horizontal load carrying system. The average
vibration measurements of timber buildings suggested that contribu- thickness of CLT panels is 200 mm. The CLT panels consist of 5 layers
tion from the stiffness of non-structural elements, such as internal of C24 spruce boards in accordance with EN 338 [21]. The first ten
plasterboard partition walls and exterior wall cladding (brickwork, floors are made of ‘‘Trä8’’ system — prefabricated timber decks, which
glass, and timber), affects the serviceability level dynamic response were developed by Moelven [22]. ‘‘Trä8’’ system is composed of Kerto-
of the building [14,15]. One study where operational modal testing Q® structural laminated veneer lumber (Kerto-Q LVL [23]) beams and
of a 6-storey timber frame building was performed, showed that the a top flange, which is topped with 36-mm thick acoustic panel and
addition of internal plasterboards had considerably stiffened the whole 50-mm thick concrete screed (Fig. 3). Stiffening elements made of
building in the translational directions [16]. These studies indicate the Kerto-S® structural laminated veneer lumber (Kerto-S LVL [23]) were
importance of assessment of the effect of non-structural elements on used for stiffening in the perpendicular to span direction. The bottom
the dynamic response of buildings. flanges are made of glulam timber. The wooden decks average span
The main objective of this paper is to present an alternative numer- is 7.5 m, average width is 2.6 m and total thickness is 360 mm. The
ical modeling approach for connections and non-structural elements upper six floors are made of 300 mm thick concrete slabs to increase
(internal plasterboard partition walls and external timber walls), which the self-weight of the structure in order to meet the serviceability
can be used for the investigation of the serviceability level dynamic requirements. The facade of the building is comprised of prefabricated
response of taller glulam buildings. The information at hand for struc- wooden panels. The truss-type glulam pergola is built on top of the
tural engineering of serviceability dynamics of buildings is basically structure for architectural appearance. All structural timber elements
mean values of stiffness and mass of the individual components con- have been provided by Moelven, a Scandinavian producer of structural
stituting the building. The structural interactions of the components timber. The summary of material properties for structural timber are
are dependent on the layout and are usually much more uncertain shown in Table 1.
than the properties of the components. The current practice is to The building foundation consists of a set of circular steel piles which
either design the connections as pin elements or to assign a specified are approximately 30 m long, have a diameter of 400 mm and thickness
stiffness value. The latter can be a cumbersome process since the of 12.5 mm. The pile head is a 2000-mm thick concrete slab at the
stiffness estimates depend on the connection layout, which may vary base of the building. The piles are in contact with the bedrock at an
from floor to floor. The ‘‘connection-zones’’ directly relate the stiffness approximate depth of 30 m. Above, the soil consists of sand, clay, and
of connection to the stiffness of the glulam timber element that is moraine. Based on foundation conditions, the boundary conditions as
being connected. This approach allows for an automated numerical pinned supports under the column ends above the concrete foundation
investigation of the timber connections under ambient vibrations and a were chosen for modeling.
better interpretation of the estimated connection stiffness. The paper is Wind loading was determined to be the dominating load in the
written in the following order. First, the experimental results based on design combinations. The calculated wind speed was 22 m/s corre-
ambient vibration measurements of an 18-storey glulam timber frame sponding to the static wind pressure of 1.12 kN/m2 in accordance with
building (Mjøstårnet, Norway) are presented. Then, in the numerical EN 1991-1-4 [9].
modeling section, the ‘‘connection-zones’’ approach for modeling con-
nections in glulam and non-structural elements is presented, and overall 2.2. Experimental setup
numerical modeling of the Mjøstårnet building is described. After that,
a parametric study is conducted to explore the effect of stiffness of Experimental dynamic characterization of the Mjøstårnet building
connections and non-structural elements on the dynamic response of was performed using ambient vibration measurements. Two different
the building. Finally, a model updating study is conducted to obtain setups have been used: setup on the roof and setup inside the building.
the values for parameters of interest. The goal of this study is to present In both setups, a set of 3 triaxial accelerometers along with a data
values for stiffness of connections and non-structural elements (internal acquisition system were used for experimental testing. In setup on the
plasterboard partitions and external timber walls) for practitioners to roof, the accelerometers have been mounted on the pergola truss at
use as a reference when developing numerical models of taller glulam the top of the building as shown in Fig. 4. This setup allowed for
buildings for serviceability level vibration response prediction. capturing the eigenfrequencies and in-plane movement, but did not
allow for determining the mode shapes along the height of the building.
2. Ambient vibrations testing Therefore, the setup has been moved inside the building and the
accelerometers have been placed at the different levels. In the second
2.1. Building description setup, accelerometers were placed on 3 different floors, 10th, 16th,
and 18th, aligned vertically in order to capture higher translational
The test building is named Mjøstårnet (the Mjosa Tower, Fig. 1) vibration modes (Fig. 5(a)). The locations of the accelerometers were
and is located in Brumunddal, Norway. The total architectural height based on modal analysis results from the numerical model of the build-
of the building (including the truss work on the top) is 85.4 m and ing. The accelerometers were attached with a magnet to a metal plate,
the total height including the spire is 88.8 m. The plan dimensions are which was mounted directly on the glulam beam by screws (Fig. 5(c)).
36.3 m by 15.7 m, see also Fig. 5(b). The summary of the structural Since the experimental setup is being used for long-term measurements,
system of the building is shown in Fig. 2. The load-carrying system the placement of accelerometers was limited to locations that are not
of the building consists entirely of timber: glue laminated timber (glu- obstructing the service of the building and are not accessible to the
lam) beams, columns and diagonals. The strength grade for glulam is public and residents. The building is in a fully operational mode, which

2
S. Tulebekova et al. Engineering Structures 261 (2022) 114209

Fig. 1. Mjøstårnet building (dimensions in m).

Fig. 2. Building structural system.

Table 1
Orthotropic material properties.
Material 𝜌 [kg∕m3 ] 𝐸1 [MPa] 𝐸2 [MPa] 𝐸3 [MPa] 𝐺12 [MPa] 𝐺13 [MPa] 𝐺23 [MPa]
Glulam GL30c 430 13 000 300 300 650 650 91.5
CLT, 5 layers 420 6 960 4 650 300 650 650 650
Kerto-Q® LVL 510 10 500 2 200 130 820 430 22
Kerto-S® LVL 510 13 800 450 130 600 600 11

limited the options for placement of the setup significantly. As seen Since the fire safety system requires that the space be isolated from the
from the building plan (Fig. 5(b)), the corners of the building, which rest of the building, it was not possible to extend the wiring system to
have the highest displacement, were not accessible to the general public other locations on the same floor. The location of the accelerometers in
and the setup could not be installed there. After discussion with the the fire staircase is shown on the plan view of the building (Fig. 5(b)).
building maintenance company, the staircase for the fire escape, which The triaxial accelerometers have high sensitivity of 2000 mV/g and a
is used only for emergency cases, was chosen for mounting the setup. lower frequency range between 0 Hz and 1000 Hz [25]. The sampling

3
S. Tulebekova et al. Engineering Structures 261 (2022) 114209

The calculated poles are then plotted on the stabilization diagram

which allows distinguishing between the spurious poles and true poles.
Spurious poles usually appear due to inaccuracies in measurements,
non-stationarity of data, etc. The procedure for determining the true
poles, i.e. the poles containing the eigenfrequency and eigenvalue of
the physical system, is through assigning the tolerance to the frequency
and damping results. The true poles then appear on the stabilization
diagram as the aligned poles which are constant along the specific
frequency. The DD-SSI method has been applied successfully on various
civil engineering structures [29–31].

2.4. Results
Fig. 3. Graphical illustration of Trä8 floor system [24].

In this study, the stability criteria for frequency and damping were
1% and 5% respectively and the chosen order number was 100. Fig. 6
shows the stabilization diagram of the combined signals from the
accelerometers in the setup inside the building. The aligned vertical
dots show the identified stable frequencies (6 in total). In addition,
the power spectral densities in two orthogonal in-plane directions were
plotted as gray curves. The left vertical axis represents the model
order and the right vertical axis represents the magnitude of the power
spectral density. The stabilization diagram from the accelerometer data
on the roof showed similar identified frequencies to the ones presented
in Fig. 6. The mode shapes from both setups can be seen in Fig. 7 and
Fig. 4. Accelerometer setup on the roof.
Fig. 8. The first two modes are translational modes in two orthogonal
directions: 1st translational mode at 0.493 Hz in the short direction
(Fig. 1(b)), 2nd translational mode at 0.529 Hz in the long direction
rate for data was 400 Hz. The recorded data was resampled to 20 Hz (Fig. 1(a)). The third mode is the torsional mode at 0.813 Hz as seen
using an antialiasing filter in Matlab 2020b [26]. This experimental from Fig. 8. Mode 4 is the second bending mode at 1.918 Hz in the long
setup has been running continuously since December 15, 2020, and direction, and mode 5 is the second bending mode at 2.146 Hz in the
recording the measurements to a server (as of February 2022). Based short direction. The 6th stable frequency at 2.215 Hz is captured in a
on the wind data from the nearest weather station, a 3 hr acceleration long direction, but the mode shape is not possible to determine due to
time series from 5th April 2021 with a mean wind speed of 9.8 m/s its higher-order and lack of accelerometers. Damping factor values for
were chosen for further analysis. modes 1–6 are 1.5%, 2.3%, 2.2%, 1.2%, 1.7% and 2.0% respectively.

2.3. Theoretical background 3. Numerical modeling

The data was analyzed using the operational modal analysis (OMA) A numerical model of the building was developed in Abaqus CAE
technique. The family of stochastic subspace identification (SSI) tech- 2017 [32]. Fig. 9 shows the developed FE model of the building with
niques is a common approach to deal with output-only measurements the mesh geometry. The sensitivity study on the mesh size showed
and is explained in detail in the literature [27]. In the SSI techniques, an insignificant impact on the output natural frequencies. Therefore,
a mathematical model with certain parameters adjusted to fit the raw a larger mesh size was chosen to reduce the computational time and
acceleration time series data is developed and calibrated. In this study, facilitate the optimization process. In Fig. 9(a) the entire assembly
the data-driven stochastic subspace identification technique (DD-SSI) mesh is shown, which includes external walls, glulam frame mem-
approach was used to obtain the modal frequencies, mode shapes, bers, CLT shafts, floor elements, and internal partitions. Fig. 9(b)
and damping ratios [27,28]. In DD-SSI, the technique is performed shows the glulam frame model without the external walls and in-
directly on the measured response data, without pre-processing it. In ternal partitions. ‘‘Connection-zones’’ are added as generalized beam
SSI techniques, the dynamic system is assumed to be described by the elements to represent connections between the glulam frame elements.
discrete stochastic state–space model as shown below: ‘‘Connection-zones’’ are added as shell elements to represent connec-
{𝑧𝑘+1 } = [𝐴]{𝑧𝑘 } + {𝑤𝑘 } (1) tions between non-structural elements/CLT shaft and glulam frame, as
well as connections between floor elements. The chart showing the
stages adding ‘‘connection-zones’’ in glulam timber frame elements and
{𝑦𝑘 } = [𝐶]{𝑧𝑘 } + {𝑣𝑘 } (2)
non-structural elements is shown in Fig. 10. In the following sections,
where {𝑧𝑘+1 } is a state–space vector, which holds the current state of each part of the numerical model of the building is described in detail.
the system, {𝑦𝑘 } is a measured output at a specified sampling rate,
{𝑤𝑘 } and {𝑣𝑘 } are system noise and measurement noise respectively, 3.1. Structural elements modeling
𝑘 is a discrete time step, [𝐴] is a state matrix and [𝐶] is an output
matrix. The modal parameters are then extracted from the identified The glulam frame elements were modeled with one-dimensional
matrices [𝐴] and [𝐶] [29]. The important task is to properly identify the Timoshenko elastic beam elements with an orthotropic material model.
model with a reasonable number of parameters. This is performed in Concrete floors were modeled as 4-node shell elements with the linear
the state–space model by choosing the model order, i.e. the dimension elastic isotropic model. Architectural truss work, (Fig. 1), at the top of
of the A-matrix [28]. The state–space model of order 𝑖 is then used to the roof was modeled as added distributed mass on the roof slab. The
identify the eigenvalue 𝜇𝑖 , from which the corresponding pole 𝜆𝑖 can balconies at levels 12–17 were modeled as added masses on glulam
be obtained at a sampling period 𝑡𝑠 . Hence, the modal frequency and beams at the locations of balconies.
damping ratios, 𝑓𝑖 and 𝜁𝑖 , can be calculated for each pole as follows: The prefabricated timber deck Trä8 is a composite structure and due
𝑙𝑛(𝜇𝑖 ) 𝐼𝑚(|𝜆𝑖 |) 𝐼𝑚(|𝜆𝑖 |) to its complexity, a single shell element representing the global defor-
𝜆𝑖 = 𝑓𝑖 = 𝜁𝑖 = (3) mation model of the floor was chosen for analysis. For this purpose, a
𝑡𝑠 2𝜋 |𝜆𝑖 |

4
S. Tulebekova et al. Engineering Structures 261 (2022) 114209

Fig. 5. Experimental setup.

Fig. 6. Stabilization diagram of the combined signals (vertically aligned dots) and power spectral densities in two orthogonal in-plane directions (gray lines).

5
S. Tulebekova et al. Engineering Structures 261 (2022) 114209

Table 2
Simplified timber deck model properties after optimization.
Model 𝐸1 [MPa] 𝐸2 [MPa] 𝐸3 [MPa] 𝐺12 [MPa] 𝐺13 [MPa] 𝐺23 [MPa] Objective function
Initial 1300 300 300 600 600 50 7.47e−6
Optimized 4800 120 990 600 600 50 5.71e−9

Table 3
Live load estimates.
Occupational type Actual load [kN∕m2 ] Code (SLS) [kN∕m2 ]
Office 0.16 0.9
Hotel 0.21 0.6
Apartment 0.33 0.6
Rooftop deck 0.16 1.2

in Table 3. The actual live load estimates were applied to the building
model.

3.3. Glulam frame connections

Fig. 7. Identified mode shapes from merged accelerometer data sets.
In engineering practice, it is common to assume a pinned connection
between timber elements due to a lack of accurate information about
the connection [34]. In reality, the connection between the timber
elements is neither pinned, nor rigid, but rather semi-rigid, i.e. it
partially transfers the action. Therefore, it is convenient to directly
relate the stiffness properties of the connection to the main timber
elements. Manually calculating and assigning stiffness properties to
the connections and conducting sensitivity study on them can be a
cumbersome task due to the large scale of the model and large variety
of cross-sections. Keeping that in mind, modeling of connections in this
study was performed by introducing the so-called ‘‘connection-zones’’.
Fig. 8. Mode shapes from setup on the roof. The stiffness properties of these ‘‘connection-zones’’ are directly related
to the connected glulam elements, which facilitates the parametric
modeling process. These zones are separate elements with generalized
detailed model of the composite timber deck was developed in Abaqus properties which allow to arbitrarily assign cross-sectional area and mo-
as shown in Fig. 11. The pressure load was applied in three orthogonal ment of inertia. The approach of representing the axial stiffness of the
directions and the resulting deflections were then used as objective connection with ‘‘connection-zones’’ was previously described in the
functions to optimize the elastic constants of an orthotropic shell study by [4]. In this study, the axial and rotational stiffness are assumed
element, shear properties were kept the same. This shell element was to be linearly dependent on cross-sectional area and the second moment
then used as a simplified model of the composite deck in the building of inertia respectively. Thus, reduction factors can be introduced, which
model to reduce computational time and facilitate model updating. The account for the reduced cross-sectional area and reduced moment of
resulting material properties after optimization are shown in Table 2 inertia in the connections. A dimensionless reduction factor equals the
and the thickness of the shell element was 240 mm. ratio between the reduced property of the ‘‘connection-zone’’ and the
The material properties used for wooden elements in the numer- connected element, e.g. 𝑘𝑟𝑒𝑑,𝐴 = 𝐴𝑟𝑒𝑑 ∕𝐴𝑚𝑎𝑖𝑛 or 𝑘𝑟𝑒𝑑,𝐼 = 𝐼𝑟𝑒𝑑 ∕𝐼𝑚𝑎𝑖𝑛 .
ical modeling are given in Table 1. The concrete with a density of The glulam frame in the current study consists of a set of beams,
2400 kg/m3 , an elastic modulus of 34 GPa, and a Poisson ratio of 0.2 columns, and diagonals, which are connected to each other with dowels
was used for modeling. and slotted-in steel plates (Fig. 12(a)). The ‘‘connection-zones’’ were
assigned at the endpoints of beams and diagonals, (Fig. 12(b)). The
3.2. Modeling mass ends of ‘‘connection-zones’’ are tied to the connected elements at their
centerlines. The ‘‘main’’ element in this context is either the glulam
Building structural system consists mainly of prefabricated timber beam element which is connected to the column with the dowel con-
elements with known material and geometric properties. Thus, it is nection or the glulam diagonal element, which is connected to the
assumed that it is possible to estimate the dead load distribution glulam column, and beam elements with the dowel connection. The
accurately. Estimates of added mass, which include the mass of non- length of ‘‘connection-zones’’ is assumed to be equal to the height of
structural elements and live load contribution were made. The added the main element (beam or diagonal). The cross-sectional area and
mass was applied as a uniformly distributed load on the floor elements. moment of inertia are related to the connected structural element
The mass estimates of external walls and internal partitions were taken through the reduction factor 𝑘𝑟𝑒𝑑 . Material properties of the connected
from the technical report of the Norwegian research organization [33]. glulam section are assigned to the ‘‘connection-zone’’ element and the
The weight of the external wall is 1 kN/m2 and the weight of internal density is modified according to the change in geometry to keep the
partitions is 0.5 kN/m2 . According to the Eurocode 1 [9], the specified mass unaltered.
live load values shall be based on the occupation type of the floor
and a fraction of this live load shall be assigned as structural load 3.4. Non-structural elements
for serviceability limit state (SLS) (Table 3). However, these values
significantly overestimate the actual live load in the current building. External walls and internal partitions were modeled for the purpose
Based on the architectural layout of different occupational floors, the of investigating their contribution to the global dynamic behavior of the
actual live loads were investigated and calculated and results are shown building. External walls in the building of interest are composite and

6
S. Tulebekova et al. Engineering Structures 261 (2022) 114209

Fig. 9. Numerical model of the building with mesh geometry.

Fig. 10. Sequence of adding ‘‘connection-zones’’ to the building model.

Fig. 11. Model of the composite timber deck.

consist of several layers. The inner layer starts with the 13-mm gypsum External walls were modeled as simplified 4-node elastic shell el-
plasterboard, followed by the 198-mm isolation made of mineral wool ements in order to investigate their stiffness contribution to the dy-
and 13-mm wind barrier made of gypsum plasterboard. The outer layer namics of the whole structure. Similar to the ‘‘connection-zones’’ in
glulam frames described earlier in this chapter, separate shell elements
consists of the external timber cladding followed by timber framing
were assigned also between the external wall and the glulam element to
work with 73-mm air gap. The external wall is attached to the Kerto-
represent the ‘‘connection-zone’’ (Fig. 13). The in-plane stiffness of the
Q timber panel with screws and angles. The Kerto-Q panel is in turn connection represented by the shell element is assumed to be directly
supported by the composite timber floor which transfers the load to related to its thickness. The variation of stiffness in the shell connection
the load-carrying glulam beams. is performed by introducing the reduction factor, 𝑘𝑟𝑒𝑑 , which represents

7
S. Tulebekova et al. Engineering Structures 261 (2022) 114209

Fig. 12. Modeling of glulam frame with ‘‘connection-zone’’ beam elements.

reduction in thickness of the ‘‘connection-zone’’ element with respect to

the main shell element, 𝑡𝑟𝑒𝑑 ∕𝑡𝑚𝑎𝑖𝑛 . In this context, the ‘‘main shell’’ ele-
ment is an external wall element (or a partition wall element described
later in the section), which is connected to the adjacent structural
timber elements such as glulam frame and floors. The length of the shell
‘‘connection-zone’’ element is assigned to be equal to the thickness of
the external wall element. The ‘‘connection-zone’’ elements are tied to
the adjacent elements at the centerlines (Fig. 13).
Based on the composition of the external wall, only gypsum plas-
terboard can be considered as a material with stiffness contribution.
However, the numerical implementation of such small ‘‘connection-
zone’’ elements in comparison to their length is not feasible, since it
requires much finer mesh, which in turn leads to significantly increased
computational time. Thus, the lowest value for plasterboard thickness
in the numerical implementation was limited to 100 mm, which is
around 10 times higher than the actual single gypsum plasterboard
thickness. Since the thickness of the gypsum plasterboard in the nu-
merical model increased in comparison to actual thickness, the values Fig. 13. Modeling of the external wall with ‘‘connection-zone’’ shell elements.
for the stiffness should be reduced accordingly to represent the actual
panel behavior. Therefore, FEM models of the external wall consisting
of double-layer 13-mm gypsum plasterboard and 100-mm plasterboard for modeling the thickness of internal partition elements was 100 mm,
were developed and an optimization study with a deflection as objec- which is around 2 times higher than the actual gypsum plasterboard
tive function was conducted. The elastic modulus of a single 13-mm thickness. Thus, a similar optimization technique to the one used for
gypsum plasterboard was taken from the literature and is equal to 140 external walls was implemented. The resulting values of elastic and
MPa [35], the shear modulus was taken as 70 MPa. Then, the optimiza- shear moduli after optimization were 72.8 MPa and 36.4 MPa for an
tion study was conducted on the 100-mm plasterboard to determine isotropic model of the 100-mm gypsum plasterboard.
the stiffness properties of the panel with the objective function being The elevator and staircase shafts in the building are made of CLT
shear deflection values. Gypsum plasterboards were modeled with 4- panels stacked on top of each other. The shafts are not intended to carry
node elastic isotropic shell elements. The resulting elastic and shear any lateral load and were not considered as part of the load-carrying
moduli of the 100-mm plasterboard corresponding to the double layer system during the design process. The CLT panels are connected to the
gypsum plasterboard from the external wall after optimization were floors at each level by means of steel brackets and screws. ‘‘Connection-
36.4 MPa and 18.2 MPa respectively. zone’’ shell elements have been added between the CLT shaft elements
Internal partition walls in the building are made of two layers of and a sensitivity study has been conducted, where the thickness of
13-mm gypsum plasterboards on each side of the wall, separated by a the ‘‘connection-zone’’ element was varied in the range 0.1–1.0 while
100-mm gap for steel framing studs. Internal partitions are structurally keeping all other parameters fixed. The change in the outputs of
connected to the floors elements. Similar to the external walls, the interest, i.e. eigenfrequencies and MAC values (explained in the further
in-plane stiffness contribution is assumed to come from the gypsum sections) was negligible, with maximum change reaching 0.4%. Thus,
plasterboards. Therefore, the modeling approach described in the pre- the study on the impact of the stiffness contribution from CLT shafts
vious paragraph was applied. The ‘‘connection-zone’’ elements were was not included in the present parametric study. The fixed value for
introduced at the top and bottom of the partition wall elements at the the thickness of the ‘‘connection-zone’’ between the CLT shafts and floor
location where they are attached to the floor elements. The lower limit elements (𝑡𝐶𝐿𝑇 = 0.5) was used.

8
S. Tulebekova et al. Engineering Structures 261 (2022) 114209

4. Model updating The connections between the non-structural elements and the main
structure are given in terms of the thickness of the connection (Table 4).
Sensitivity-based model updating based on ambient vibration data 𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 is the thickness of ‘‘connection-zone’’ elements which are lo-
has been proven to be effective in their application on real structures cated at the top and bottom ends of the external wall panels where they
in structural engineering [36]. The numerical model of the building are connected to glulam beams. The range of values for 𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 is given
presented in this study was scripted in Python [37], and the properties as a percentage of the thickness of the external wall and represents a
of interest were parametrized in order to investigate their sensitivity reduction in the in-plane stiffness of the ‘‘connection-zone’’ elements.
and conduct model updating using the results from the experimental 𝑡𝑝𝑎𝑟𝑡𝑖𝑡𝑖𝑜𝑛 is the thickness of the ‘‘connection-zone’’ elements which are
setup. located at the top and bottom ends of the partition wall elements, where
they are connected to the floor elements. The range of values for 𝑡𝑝𝑎𝑟𝑡𝑖𝑡𝑖𝑜𝑛
4.1. Parameter selection is given as a percentage of the thickness of the internal partition walls
and represents a reduction in the in-plane stiffness of the ‘‘connection-
The dynamic modal response of the numerical model of the building zone’’ elements. Both 𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 and 𝑡𝑝𝑎𝑟𝑡𝑖𝑡𝑖𝑜𝑛 represent the reduction in
depends on both mass and stiffness properties. Optimization of both stiffness contribution to the main structure.
mass and stiffness can lead to an underdetermined problem with an
infinite number of solutions as mentioned in [38]. The mass properties 4.2. Sensitivity analysis
are not investigated in the current study since it is assumed that density
and geometric properties of the prefabricated elements can be well Sensitivity analysis was performed to assess the effect of the pa-
estimated in comparison to the connection stiffness properties. The rameters of interest on eigenfrequencies and modal assurance criterion
glulam bending modulus of elasticity (MOE) and density (𝜌) have a (MAC, explained below) values. The range for each parameter was set
coefficient of variance values (𝐶𝑂𝑉 ) of 0.13 and 0.1 respectively [39]. as a percentage value of the property of the main element, Table 4. The
In addition, the actual live load estimates were calculated based on the lowest value represents the negligible stiffness transfer, whereas the
architectural drawings and site visits, and these values are applied to highest value represents the full stiffness transfer, i.e. rigid connection.
the numerical model. The stiffness properties in the numerical model MAC is the measure of the correlation between the two sets of modes
include the stiffness of main load-carrying elements (glulam frame and is calculated as shown below [40]:
and slabs), connections between those elements, and potentially non- | |
|{𝜑𝐴 }𝑇𝑟 {𝜑𝑋 }𝑞 |2
structural elements. The stiffness parameters of the timber structural
𝑀𝐴𝐶(𝑟, 𝑞) = ( | |
)( ) (4)
elements, such as Young’s modulus, are taken from the specification {𝜑𝐴 }𝑇𝑟 {𝜑𝐴 }𝑟 {𝜑𝑋 }𝑇𝑟 {𝜑𝑋 }𝑞
of the manufacturer and the highly layered composition of the glulam
elements ensures the mean value properties of the wooden elements. where {𝜑𝑋 }𝑞 is the experiment modal vector for mode 𝑞, {𝜑𝐴 }𝑟 is
On the other hand, the stiffness of connections in timber lacks accurate the analytical modal vector for mode 𝑟. MAC takes values between 0
representation. The current practices on the prediction of connection (no similarity between modes) and 1 (high similarity between modes).
stiffness do not give accurate values for the stiffness of the connection. When several modes are compared, the result is a MAC matrix where
Additionally, the effect of non-structural elements is generally excluded diagonal elements have values of 1 in ideal situation.
but might contribute significantly in low-level dynamics. Therefore, The Latin Hypercube Sampling (LHS) method was adopted for the
stiffness of connections in the glulam frame as well as connections in current sensitivity study. LHS is a Monte-Carlo type technique, which
the non-structural elements were chosen for parametric analysis. allows for exploring the entire parameter range with minimized compu-
Based on the results from the ambient vibration measurements, it tational demand [41]. In LHS, the parameter design space is uniformly
can be seen that the first two bending mode frequencies, as well as the divided with the same number of divisions 𝑁 for all factors. The pa-
two second bending mode frequencies, are very close to each other. rameter levels are then randomly combined to create a Latin Hypercube
This might be due to the fact that the span of diagonals, which are design matrix with 𝑁 points. In this study, the sensitivity analysis with
part of the lateral load resisting system, is similar in both directions LHS was performed using the Design of Experiments (DOE) component
(Fig. 5(a)). Therefore, the connection stiffness was studied separately of ISight, a tool for automated simulations which can be integrated into
in two directions of the building plane in order to investigate their Abaqus CAE [42]. The number of samples, 𝑁=100 was selected for
effect on the dynamic response of the building. Table 4 shows the list analysis and a uniform distribution was chosen for each parameter, 𝐾.
of parameters of interest. Each of those parameters is a ‘‘connection- The 𝑁𝑥𝐾 matrix is generated from the randomly sampled values and
zone’’ element in the numerical model. ‘‘Long’’ and ‘‘short’’ subscripts based on this matrix simulations explore the sensitivities of the input
stand for ‘‘connection-zone’’ elements in the long and short direction parameters and evaluate their significance by calculating the partial
of the building respectively (Fig. 9(b)). 𝐼𝑠ℎ𝑜𝑟𝑡 and 𝐼𝑙𝑜𝑛𝑔 are moments rank correlation coefficients (𝑃 𝑅𝐶𝐶). 𝑃 𝑅𝐶𝐶 is the measure of the
of inertia of the ‘‘connection-zone’’ elements at the end of a beam linear relationship between the parameter of interest and the output.
element, where it is connected to the column. The range in values for Sensitivity analysis was performed on a total of 6 input parameters
𝐼𝑠ℎ𝑜𝑟𝑡 and 𝐼𝑙𝑜𝑛𝑔 is given as a percentage of the main element and is with respect to 9 outputs (5 natural frequencies and 4 MAC values).
representative of reduction in the rotational stiffness of ‘‘connection- The summary of the sensitivity study is shown in Fig. 14. For better
zone’’ elements, 𝑘𝑟𝑒𝑑 . The range of values for 𝐴𝑙𝑜𝑛𝑔 and 𝐴𝑠ℎ𝑜𝑟𝑡 is given representations, the absolute values of partial rank correlation coeffi-
as a percentage of the connected element and give the reduction in cients of each parameter were plotted both against eigenfrequencies
axial stiffness of ‘‘connection-zone’’ elements, 𝑘𝑟𝑒𝑑 . Similarly, 𝐴𝑙𝑜𝑛𝑔 and and MAC values. Based on the sensitivity ranks for eigenfrequencies,
𝐴𝑠ℎ𝑜𝑟𝑡 represent the ‘‘connection-zone’’ in the truss diagonals at the parameters 𝐴𝑙𝑜𝑛𝑔 and 𝐴𝑠ℎ𝑜𝑟𝑡 have the largest contributions to most
points where they are connected to the beams and columns (Fig. 9(b)). frequencies (Fig. 14(a)). Parameter 𝐼𝑠ℎ𝑜𝑟𝑡 has a significant effect on
A preliminary parameter study showed that the moment of inertia in natural frequencies as well. Natural frequencies from mode 2, 4 and
diagonals and axial stiffness of the beams have no significant effect on 5 have some contribution from 𝐼𝑙𝑜𝑛𝑔 , 𝑡𝑝𝑎𝑟𝑡𝑖𝑡𝑖𝑜𝑛 and 𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 . Most of the
the eigenfrequencies of the building. This is consistent with the truss- MAC values, Fig. 14(b), have the dominant contribution from 𝐴𝑠ℎ𝑜𝑟𝑡 .
type structural system of the building, where diagonals were designed The first two MAC values are affected by 𝐴𝑙𝑜𝑛𝑔 , whereas the last two
to carry the horizontal loads, which leads to insignificant axial stiffness MAC values are affected by 𝐴𝑠ℎ𝑜𝑟𝑡 and 𝐼𝑠ℎ𝑜𝑟𝑡 . Most of the MAC values
contribution from beams. Therefore, only moments of inertia of beam have minor contributions from 𝐼𝑙𝑜𝑛𝑔 , 𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 and 𝑡𝑝𝑎𝑟𝑡𝑖𝑡𝑖𝑜𝑛 . Mode 4 is
connections and areas of the diagonal connections are studied in the affected considerably by 𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 . Based on the results of the sensitivity
glulam frame. analysis, all 6 parameters from Table 4 were chosen for model updating.

9
S. Tulebekova et al. Engineering Structures 261 (2022) 114209

Table 4
Parameters for sensitivity analysis and model updating.
‘‘Connection-zone’’ Description Range of 𝑘𝑟𝑒𝑑
parameter [% of main element]
𝐼𝑠ℎ𝑜𝑟𝑡 Moment of inertia of beams in long direction 1–100
𝐼𝑙𝑜𝑛𝑔 Moment of inertia of beams in short direction 1–100
𝐴𝑠ℎ𝑜𝑟𝑡 Cross-sectional area of diagonals in short direction 1–100
𝐴𝑙𝑜𝑛𝑔 Cross-sectional area of diagonals in long direction 1–100
𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 Thickness of connection between external wall and beam 1–100
𝑡𝑝𝑎𝑟𝑡𝑖𝑡𝑖𝑜𝑛 Thickness of connection between internal partition and floors 1–100

Fig. 14. Absolute partial rank correlation coefficients.

4.3. Model updating pool to produce an even higher quality offspring. In Multi-Island GA,
there are several mating pools in which high-quality individual solu-
The purpose of the model updating is to minimize the difference tions are generated, and periodically these pools exchange a portion of
between the numerical model and the experimental modal response. their population in a process called migration. In this study the number
The objective function for model updating included the modal pa- of islands (mating pools) was set to 10, the sub-population size and the
rameters: 5 lowermost eigenfrequencies (𝑓1 − 𝑓5 ) and MAC values number of generations were set to 10, the rate of migration was set to
for the 1st and 2nd bending modes in two translational directions 0.01.
(𝑀𝐴𝐶1 , 𝑀𝐴𝐶2 , 𝑀𝐴𝐶4 , 𝑀𝐴𝐶5 ). MAC value for torsional mode (𝑀𝐴𝐶3 )
was not included in model updating due to the lack of sensors to 4.4. Model updating results
accurately estimate the mode shape. The objective functions used in
the model updating are given as follows: Model updating has been performed using four different sets of
( )
∑𝑛
𝑓𝑖,𝑒𝑥𝑝 − 𝑓𝑖,𝑛𝑢𝑚 2 parameters. In Set 1, all of the parameters from Table 4 were used
𝐹𝑓 𝑟𝑒𝑞 = 𝛾𝑖 (5) for model updating. In Set 2, parameters 𝐼𝑠ℎ𝑜𝑟𝑡 and 𝐼𝑙𝑜𝑛𝑔 were com-
𝑖=1
𝑓𝑖,𝑒𝑥𝑝
bined into one parameter 𝐼𝑏𝑒𝑎𝑚 , and parameters 𝐴𝑠ℎ𝑜𝑟𝑡 and 𝐴𝑙𝑜𝑛𝑔 were
∑
𝑛
( )
𝐹𝑀𝐴𝐶 = 𝛾𝑖 1 − 𝑑𝑖𝑎𝑔(𝑀𝐴𝐶(𝜙𝑖,𝑒𝑥𝑝 , 𝜙𝑖,𝑛𝑢𝑚 )) (6) combined into one parameter 𝐴𝑑𝑖𝑎𝑔𝑜𝑛𝑎𝑙 . Parameters 𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 and 𝑡𝑝𝑎𝑟𝑡𝑖𝑡𝑖𝑜𝑛
𝑖=1 were included into the second set as well. The purpose of generalizing
where 𝐹 is an objective function, (∗)𝑖,𝑒𝑥𝑝 is the experimental output stiffness parameters into 𝐼𝑏𝑒𝑎𝑚 and 𝐴𝑑𝑖𝑎𝑔𝑜𝑛𝑎𝑙 is to compare with the
of mode 𝑖, (∗)𝑖,𝑛𝑢𝑚 is the numerical output of mode 𝑖, 𝜙 is the mode results of the first set updating and observe whether the generalized
shape vector, 𝛾𝑖 is the weight factor, which was set to 1 for 𝑓1 − 𝑓3 and axial and rotational stiffness parameters will be able to predict the
𝑀𝐴𝐶1 − 𝑀𝐴𝐶2 values, the higher modes weight factors were set to building dynamic response. Set 3 includes parameters 𝐼𝑠ℎ𝑜𝑟𝑡 and 𝐼𝑙𝑜𝑛𝑔 for
0.5. The higher weight factor was used for lower modes due to the fact, beam rotational stiffness and parameters 𝐴𝑠ℎ𝑜𝑟𝑡 and 𝐴𝑙𝑜𝑛𝑔 for diagonal
that they are better identified in the OMA analysis and the mode shapes axial stiffness. Parameters 𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 and 𝑡𝑝𝑎𝑟𝑡𝑖𝑡𝑖𝑜𝑛 are excluded from model
have the simple cantilever form, while higher modes have a higher level updating in Set 3. Set 4 includes only two parameters: 𝐼𝑏𝑒𝑎𝑚 and
of uncertainty. 𝐴𝑑𝑖𝑎𝑔𝑜𝑛𝑎𝑙 . Parameters 𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 and 𝑡𝑝𝑎𝑟𝑡𝑖𝑡𝑖𝑜𝑛 were kept constant and were
Model updating was performed using the ISight tool which can be excluded from model updating in Set 4.
integrated into the Abaqus CAE, where the numerical model of the The results of the FE model updating with Set 1 parameters are
building was created. The Multi-Island Genetic Algorithm was chosen shown in Tables 5 and 6. The range for parameters in model updating
for optimization [43]. Genetic Algorithms (GAs) are optimization al- was chosen the same as in the sensitivity analysis, confer Table 4.
gorithms based on natural selection and genetics concepts. GAs work Approximately 1000 iterations were performed in model updating to
on the population of possible solutions, where each individual solution explore the entire design space. The initial values in Table 6 were
is assessed based on the quality of the result (how far it is from the chosen to represent the starting assumption of pinned connections
target values). The mating pool is then generated from the high-quality in the glulam frame and negligible stiffness contribution from non-
individual solutions, and parent combinations are selected from this structural elements. Table 5 shows the natural frequencies and MAC

10
S. Tulebekova et al. Engineering Structures 261 (2022) 114209

Table 5
Comparison between experimental and numerical modal properties (Set 1).
Mode Initial model Updated model (Set 1) Experimental
𝑓 [Hz] Difference 𝑀𝐴𝐶 𝑓 [Hz] Difference 𝑀𝐴𝐶 𝑓 [Hz]
1 0.506 2.64% 0.82 0.511 3.65% 0.93 0.493
2 0.509 −3.78% 0.88 0.518 −2.08% 0.97 0.529
3 0.820 0.86% – 0.817 0.49% – 0.813
4 1.915 −0.16% 0.25 1.954 1.88% 0.85 1.918
5 1.957 −8.81% 0.38 1.972 −8.11% 0.97 2.146

Table 6 of the applied timber material [4]. Therefore, the structural behavior of
𝑘𝑟𝑒𝑑 values for Set 1 after model updating. the dowelled connections is subject to uncertainty. Several sensitivity
Parameter Initial Updated studies have been done to explore the effect of both axial and rotational
𝐼𝑠ℎ𝑜𝑟𝑡 1% 65.3% stiffness of glulam frame connections on the vibration response of the
𝐼𝑙𝑜𝑛𝑔 1% 92.7% building [4,6]. The studies on the sensitivity of the rotational stiffness
𝐴𝑠ℎ𝑜𝑟𝑡 100% 33.5%
showed a negligible difference between the pinned connections and
𝐴𝑙𝑜𝑛𝑔 100% 68.3%
𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 1% 94.1% connections with stiffness values based on Eurocode 5 calculations [6].
𝑡𝑝𝑎𝑟𝑡𝑖𝑡𝑖𝑜𝑛 1% 45.8% The Eurocode 5 calculations, however, lack the accurate representation
of the connection stiffness and underestimate the actual value based
on the sub-assembly experimental results [10]. On the other hand, the
rigid connection scenario in the study by [6], showed a considerable
values after model updating along with the initial model results and
increase in fundamental frequencies of the building.
experimental natural frequencies. It can be seen that while there is a Model updating of parameters 𝐼𝑙𝑜𝑛𝑔 and 𝐼𝑠ℎ𝑜𝑟𝑡 in this study showed
slight improvement in natural frequencies, MAC values have improved that the rotational stiffness in glulam connections is neither negligible,
significantly after model updating. In particular, considerable improve- nor rigid, but rather exhibits a semi-rigid behavior. Similarly, the axial
ment in mode shape prediction is observed for modes 1 and 2. The total stiffness parameters 𝐴𝑙𝑜𝑛𝑔 and 𝐴𝑠ℎ𝑜𝑟𝑡 have updated values which are
objective function improved from 0.647 to 0.071 after model updating, less than the initial value of 100% of the main elements, meaning that
resulting in an 89% decrease. The numerical mode shapes after model connection is semi-rigid in the axial direction as well. The sensitivity
updating are shown in Fig. 15. study by [4] showed that variation of axial stiffness in glulam truss has
The updated values of 𝐼𝑠ℎ𝑜𝑟𝑡 and 𝐼𝑙𝑜𝑛𝑔 increased considerably com- an effect on the eigenfrequencies of the structure, which is consistent
pared to the initial values (Table 5). This means that the glulam with the findings in this paper. As seen from Table 6, both 𝐴𝑙𝑜𝑛𝑔 and
connections have considerable rotational stiffness contribution which is 𝐼𝑙𝑜𝑛𝑔 have updated values, which are consistently higher than the values
different from the pin-condition assumption that was selected initially. of 𝐴𝑠ℎ𝑜𝑟𝑡 and 𝐼𝑠ℎ𝑜𝑟𝑡 respectively. The model updating of glulam timber
The value of 𝐼𝑙𝑜𝑛𝑔 updated to 92.7% of the main element, while the frame connections implies that there exists significant variation in the
updated value for 𝐼𝑠ℎ𝑜𝑟𝑡 is 65.3%. The values of 𝐴𝑠ℎ𝑜𝑟𝑡 and 𝐴𝑙𝑜𝑛𝑔 were stiffness, which should be investigated further. The geometry of the
reduced in comparison with the initial values, which means that the building might have a considerable impact on the updated stiffness
initial assumption of full axial load transfer in the connection was not of the connections since it can be seen that the model updating with
sufficient. Similarly to the moments of inertia, updated 𝐴𝑙𝑜𝑛𝑔 value is separate elements for long and short directions achieves better results
larger than the updated 𝐴𝑠ℎ𝑜𝑟𝑡 , 68.3% and 33.5% respectively ( Table 6). in comparison with other sets.
The parameters 𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 and 𝑡𝑝𝑎𝑟𝑡𝑖𝑡𝑖𝑜𝑛 , which represent the non-structural
elements contribution, increased significantly from their initial value. 5.2. Non-structural elements
Tables 7 and 8 show the summary of the model updating for Sets 1–4. In
comparison with the Initial Set, the objective function for Sets 1–4 was The study explored the effect of the external wall and internal parti-
lower after model updating. However, the objective function for Sets tion stiffness on the dynamic response of the building. By modifying the
2–4 was higher than for Set 1, which had an objective function value of connection thickness parameters, 𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 and 𝑡𝑝𝑎𝑟𝑡𝑖𝑡𝑖𝑜𝑛 , the contribution
0.071. As seen from Table 7, the eigenfrequencies achieved after model of non-structural member stiffness was varied from negligible, 𝑡∗ = 1%,
updating with Sets 1–4 were in good accordance with the experimental to a full stiffness transfer, 𝑡∗ = 100%. Based on the sensitivity study, the
results. Model updating with Sets 1–3 achieved acceptable MAC values. external partition stiffness contribution to the natural frequencies was
insignificant since the load-bearing contribution consisted only of two
5. Discussion 13-mm thick gypsum plasterboards and low stiffness. However, there
was a noticeable effect on the MAC values (Fig. 14(b)) for modes 1 and
The discussion of the model updating results in this section is based 4 (Fig. 15). After Set 1 model updating the value of 𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 was 94.1%
on the assumption that the structural masses, live loads, and structural of the main shell element, whereas after Set 2 model updating the value
material stiffness properties were estimated with an acceptable level of 𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 was 88.6%. The actual connection between the external wall
of uncertainty. While the authors have used their best engineering and main structure is, however, not very rigid based on the detailing
judgment in the calculation of structural masses and live loads, the drawings. On the other hand, the stiffness contribution from non-
uncertainty related to these estimations can still be considerable. The structural elements might be considerable at low-level vibrations and
presented numerical modeling approach of the glulam building using further investigation is required on the topic. The internal partitions
‘‘connection-zones’’ helps to generalize the connection properties mak- have a slightly higher stiffness contribution compared to external walls
ing it easier for practicing engineers to use the obtained results as a since they consist of the 2-ply double-sided gypsum plasterboard with
reference for their prediction models. a steel frame. Therefore, non-structural element parameters 𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 and
𝑡𝑝𝑎𝑟𝑡𝑖𝑡𝑖𝑜𝑛 have contributions to mode shapes of the structure and should
5.1. Glulam connections be accounted for when performing modal analysis of the building.
The results of the model updating showed that under ambient vibra-
Dowelled glulam connections exhibit a nonlinear behavior even at tions, the non-structural members have a considerable impact on the
low excitation levels due to several factors including an initial slip of MAC values in the dynamic response of the timber building, which is
dowels inside the drilled holes and inherent cracks and imperfections consistent with the previous studies [14,15].

11
S. Tulebekova et al. Engineering Structures 261 (2022) 114209

Fig. 15. Numerical mode shapes after model updating with the first set.

Table 7
Comparison between experimental and numerical modal properties for all sets.
Mode Set 1 Set 2 Set 3 Set 4 Experim.
𝑓 [Hz] Change [%] MAC 𝑓 [Hz] Change [%] MAC 𝑓 [Hz] Change [%] MAC 𝑓 [Hz] Change [%] MAC 𝑓 [Hz]
1 0.511 3.6 0.93 0.510 3.4 0.88 0.513 4.0 0.93 0.507 2.8 0.73 0.493
2 0.518 −2.0 0.97 0.513 −3.0 0.93 0.519 −1.89 0.97 0.509 −3.7 0.79 0.529
3 0.817 0.4 – 0.826 1.6 – 0.817 0.4 – 0.821 0.98 – 0.813
4 1.954 1.8 0.85 1.961 2.24 0.81 1.966 2.5 0.84 1.954 1.88 0.70 1.918
5 1.972 −8.1 0.97 1.970 −8.2 0.93 1.979 −7.7 0.96 1.959 −8.71 0.85 2.146

6. Conclusions in a parametrized manner. The dynamic properties of the glulam timber

frame building have been achieved successfully using the proposed
This paper presents an approach to model connections in glulam approach for connections modeling. An ambient vibration procedure
frame buildings. ‘‘Connection-zones’’ are introduced in the model and and subsequent system identification of an instrumented 18-storey
represent the rotational and axial stiffness, in this case, of doweled con- glulam building identified 5 vibrational modes, which were used for
nections. This approach allows for the implementation of modeling of model validation and model updating. In summary, the axial stiffness
the connection stiffness influence on the dynamic response of building parameter has the largest impact on both fundamental frequencies and

12
S. Tulebekova et al. Engineering Structures 261 (2022) 114209

Table 8 [6] Lazzarini E, Frison G, Trutalli D, Marchi L, Scotta R. Comfort assessment of

𝑘𝑟𝑒𝑑 values for all sets after model updating. high-rise timber buildings exposed to wind induced vibrations. Struct Design
Parameter Initial Set 1 Set2 Set 3 Set 4 Tall Spec Build 2021.
𝐼𝑠ℎ𝑜𝑟𝑡 1% 65.3% 3.4% 83.1% 1.0% [7] Foster R, Reynolds T, Ramage M. Proposal for defining a tall timber building. J
𝐼𝑙𝑜𝑛𝑔 1% 92.7% 3.4% 51.0% 1.0% Struct Eng 2016;142.
𝐴𝑠ℎ𝑜𝑟𝑡 100% 33.5% 88.1% 34.0% 97.0% [8] Reynolds T, Harris R, Chang W. Stiffness of dowel-type timber connections under
𝐴𝑙𝑜𝑛𝑔 100% 68.3% 88.1% 94.9% 97.0% pre-yield oscillating loads. Eng Struct 2014;65:21–9.
𝑡𝑒𝑥𝑡𝑒𝑟𝑛𝑎𝑙 1% 94.1% 88.6% – – [9] CEN1991-1-1. Eurocode 1: Actions on structures. Part 1-1: General actions.
𝑡𝑝𝑎𝑟𝑡𝑖𝑡𝑖𝑜𝑛 1% 45.8% 82.6% – – Brussels, Belgium: European Committee for Standardization; 2002.
𝑂𝑏𝑗.𝑓 𝑢𝑛𝑐𝑡𝑖𝑜𝑛 0.647 0.071 0.119 0.075 0.360 [10] Solarino F, Giresini L, Chang W, Huang H. Experimental tests on a dowel-type
timber connection and validation of numerical models. Buildings 2017;7.
[11] Johansson M, Linderholt A, Jarnero K, Landel P. Tall timber buildings - A
preliminary study of wind-induced vibrations of a 22-storey building. In: World
conference on timber engineering. Vienna, Austria; 2016.
MAC values, whereas the rotational stiffness parameter has the least
[12] Edskar I, Lidelow H. Dynamic properties of cross-laminated timber and timber
impact. The non-structural parameters accounting for external walls truss-building systems. Eng Struct 2019;186:525–35.
and internal partitions have a considerable effect on the MAC values [13] Li B, Hutchinson G, Duffield C. Contribution of typical non-structural components
of the structure. The assumption of an accurately predicted mass was to the performance of high-rise buildings based on field reconnaissance. J Buil
adopted, which is the main limitation in the current study since there is Apprais 2010;6:129–51.
some uncertainty in relation to the mass estimates. A future study will [14] Mugabo I, Barbosa A, Riggio M. Dynamic characterization and vibration analysis
of a four-story mass timber building. Front Built Environ 2019;5.
include the long-term monitoring of the building in order to evaluate
[15] Worth M, Omenzetter P, Morris H. Ambient and forced vibration testing and
the dependence of the amplitude of the wind-induced vibrations. The finite element model updating of a full-scale post-tensioned laminated veneer
authors believe that the stiffness ratios for ‘‘connection-zones’’ in this lumber building. In: 2012 new zealand society for earthquake engineering
study can serve as a useful reference for practitioners designing glulam (NZSEE) annual conference. 2012.
timber frame buildings. [16] Ellis B, Bougard A. Dynamic testing and stiffness evaluation of a six-storey timber
framed building during construction. Eng Struct 2001;23:1232–42.
[17] 14080 C. EN 14080-2013: Timber structures - glue laminated timber and
CRediT authorship contribution statement glued solid timber - Requirements.. Brussels, Belgium: European Committee for
Standardization; 2013.
Saule Tulebekova: Conceptualization, Methodology, Validation, [18] Abrahamsen R. Mjøstårnet - construction of an 81 m tall timber building. In:
Investigation, Formal analysis, Writing – original draft, Writing – re- 23rd inernational holzbau-forum IHF 2017. 2018.
view & editing, Visualization. Kjell Arne Malo: Conceptualization, [19] 103 CE. NS-EN 10025-2. Hot rolled products of structural steels. Part 2: Technical
delivery conditions for non-alloy structural steels. Brussels, Belgium: European
Methodology, Writing – review & editing, Supervision, Project admin- Committee for Standardization; 2019.
istration, Funding acquisition. Anders Rønnquist: Conceptualization, [20] 10088 C. EN 10088-3-2014: Stainless steels. Part 3: Technical delivery conditions
Methodology, Writing – review & editing, Supervision. Petter Nåvik: for semi-finished products, bars, rods, wire, sections and bright products of
Investigation, Writing – review & editing. corrosion resisting steels for general purposes.. Brussels, Belgium: European
Committee for Standardization; 2014.
[21] 338 C. EN 338-2016: Structural timber - strength classes. Brussels, Belgium:
Declaration of competing interest
European Committee for Standardization; 2016.
[22] Moelven. Flervåningshus Trä8, https://www.moelven.com/se/produkter-och-
The authors declare that they have no known competing finan- tjanster/allt-om-limtra/flervaningshus-tra8/.
cial interests or personal relationships that could have appeared to [23] MetsäWood. Mechanical properties. Kerto LVL. 2021, https://www.
influence the work reported in this paper. metsawood.com/global/Tools/MaterialArchive/MaterialArchive/Kerto-manual-
lvl-mechanical-properties.pdf.
[24] Liven H. Treet i Bergen og Mjøstårnet i Brumunddal. Høyhus i tre - Utfor-
Acknowledgments
dringer. 2017, https://www.treteknisk.no/resources/filer/aktuelt/brd-foredrag/
Mjostarnet-og-andre-hoge-bygg.pdf.
This research has been conducted as part of the Dynamic Response [25] Kistler. MEMS capacitive triaxial accelerometer for low frequency applica-
of Timber Buildings under Service Load (DynaTTB) project. The au- tions, K-beam: Type 8396a2d0. 2021, https://www.kistler.com/en/product/type-
thors are grateful to the ERA-NET Cofund Forest Value and all the 8396a/.
[26] The Math Works I. MATLAB (R2020b). Natick, Massachusetts: The MathWorks
corresponding funding bodies for their assistance and financial support.
Inc.; 2019.
In Norway funding was provided by the Research Council of Norway, [27] Van Overschee P, De Moor B. Subspace identification for linear systems: theory,
grant no. 297513. The authors would also like to thank students Daniel implementation, applications. Leuven, Belgium: Kluwer Academic Publishers;
Hjolman Reed and Lars Håkon Wiig for parametrizing the numerical 1996.
model of timber building in Python. The authors are grateful to Moel- [28] Rainieri C, Fabbrocino G. Operational modal analysis of civil engineering
structures: an introduction and guide for applications. New York, USA: Springer;
ven Limtre, and in particular Rune Abrahamsen; and SWECO Norge and
2014.
in particular Magne Aanstad Bjertnæs for their help and cooperation in [29] Kvåle K, Øiseth O, Rønnquist A. Operational modal analysis of an end-supported
connection to the Mjøstårnet project. pontoon bridge. Eng Struct 2017;148:410–23.
[30] Zhou X, Cao L, Khan I, Li Q. Data inspecting and denoising method for
References data-driven stochastic subspace identification. Shock Vib 2018;2018.
[31] Øiseth O, Rønnquist A, Kvåle K, Sigbjörnsson R. Monitoring wind velocities and
[1] Zubizarreta M, Cuadrado J, Orbe A, Garcia H. Modeling the environmental dynamic response of the hardanger bridge. Springer, Cham; 2014, Dynamics
sustainability of timber structures: A case study. Environ Impact Assess Rev of Civil Structures, Volume 2. Conference Proceedings of the Society for
2019;78. Experimental Mechanics Series.
[2] Zhao X, Zhang B, Kilpatrick T, Sanderson I, Liu D. Numerical analysis on global [32] Simulia. Abaqus FEA software. Dassault systèmes; 2017.
serviceability behaviours of tall glulam frame buildings to the eurocodes and UK [33] SintefByggforsk. Egenlaster for bygningsmaterialer, byggevarer og bygningsdeler.
national annexes. J Civ Eng Const 2021;10:109–22. Byggforskserien 471.031, Trondheim, Norway: SINTEF; 2013.
[3] Ramage M, Burridge H, Busse-Wicher M, Fereday G, Reynolds T, Shah D, et al. [34] Shu Z, Li Z, Yu X, Zhang J, He M. Rotational performance of glulam bolted
The use of timber in construction. Renew Sustain Energy Rev 2017;68:333–59. joints: Experimental investigation and analytical approach. Constr Build Mater
[4] Malo K, Abrahamsen R, Bjertnæs M. Some strucutral design issues of 2019;213:675–95.
the 14-storey timber framed building treet in Norway. Eur J Wood Prod [35] Telue Y, Mahendran M. Behaviour and design of cold-formed steel wall frames
2016;74:407–24. lined with plasterboard on both sides. Eng Struct 2004;26:567–79.
[5] Bjertnaes M, Malo K. Wind-induced motions of ‘‘Treet’’ - A 14-storey timber [36] Kodikara K, Chan T, Nguyen T, Thambiratnam D. Model updating
residential building in Norway. In: World conference on timber engineering. of real structures with ambient vibration data. J Civ Struct Health Monit
Quebec City, Canada; 2014. 2016;6:329–41.

13
S. Tulebekova et al. Engineering Structures 261 (2022) 114209

[37] Reed D, Wiig L. Parametric study of tall timber buildings. Trondheim, Norway: [41] Gomero B. Latin hypercube sampling and partial rank correlation coefficient
Norwegian University of Science and Technology; 2020. analysis applied to an optimal control problem. Knoxville, Tennessee, USA:
[38] Aloisio A, Pasca D, Tomasi R, Fragiacomo M. Dynamic identification and model University of Tennessee; 2012.
updating of an eight-story CLT building. Eng Struct 2020;213. [42] DassaultSystemes. Isight. Automated design exploration and optimization. 2014,
[39] JCSS. JCSS probabilistic model code. Part 3: Resistance models. Section 3.5. https://www.3ds.com/fileadmin/PRODUCTS-SERVICES/SIMULIA.
properties of timber. Joint Committee on Structural Safety; 2006. [43] Chen H, Ooka R, Kato S. Study on optimum design method for pleasant
[40] Pastor M, Binda M, Harcarik T. Modal assurance criterion. Procedia Eng outdoor thermal environment using genetic algorithms and coupled simulation
2012;48:543–8. of convection, radiation and conduction. Build Environ 2006;43:65–71.

14

View publication stats