Development of a Non-Destructive Evaluation System for Mass

Timber Panels

Author
Faircloth, Adam D

Published
2022-06-01

Thesis Type
Thesis (Masters)

School
School of Eng & Built Env

DOI

10.25904/1912/4538

Rights statement
The author owns the copyright in this thesis, unless stated otherwise.

Downloaded from
http://hdl.handle.net/10072/415279

Griffith Research Online

https://research-repository.griffith.edu.au
Development of a Non-Destructive Evaluation
System for Mass Timber Panels

Adam D. Faircloth

BEng. (Mechatronics)

School of Engineering and Built Environment

Griffith University

Submitted in fulfillment of the requirements of the degree of

Master of Philosophy

February 2022
 Abstract
The use of timber in construction originated with timber framing methods then transitioning to
mass timber construction in the last decade. Locally in Australia the impacts of the COVID-19
pandemic have also outlined the societal demand to move from concrete and steel structures to
more sustainable and environmentally friendly buildings. Reported by the Wood Solutions
(2021), timber sales had seen a drastic increase compared to previous years putting pressure on
processors to meet the demand set by the current construction boom. Mass timber construction
uses intelligent design practices to layer, laminate and/ or fasten timber products to create a
light-weight and environmentally friendly structure. With the continued increase in demand for
these mass timber products, manufacturers of these materials will be met with pressure to
ensure high levels of quality are being maintained.

Of the serviceability requirements, the mechanical properties are one of the key design
parameters considered. While these properties are commonly quantified through static test
methods, this testing provides an accurate representation of the material tested. These static
experiments are time consuming, costly to conduct and destructive. Non-destructive evaluation
(NDE) techniques are often equally precise in determining the material properties in a rapid,
and accurate alternative to static testing for beams and boards. These rapid evaluation
techniques have been adopted by industry as a time saving quality assurance (QA) method in
board processing. Several international attempts have been made to develop an NDE system
for use by mass panel manufacturers as a QA tool. Although no such system has focused on
the industry implementation requirements.

This study aims to investigate the experimental parameters involved in NDE of mass timber
panel systems. The system developed through this study will focus on the experimental
requirements of a QA system that is to be used in conjunction with a theoretical model. The
model has been developed in collaboration with researchers from the Department of
Agriculture and Fisheries (DAF) and Centre for International Research Agricultural
Development and (CIRAD). The information presented in this thesis can be separated into four
main parts. Part one (section 3) presents the signal specific section of the research. This section
involves the program development using LabVIEW to create a signal acquisition and post-
processing system for conducting the dynamic vibration experiments. As part of the initial
stages of the research, gaining an understanding for the applicable methods of digital signal
processing (DSP) were identified as an important factor. The resolution of the acquired signal
ii
is a major contributor to the NDE testing approaches accuracy. The acquisition of these signals
are a result of the hardware, software, and experimental method. The signal measurement is an
important phase of the research.

Part two of this thesis reports on the experimental setup requirements for conducting the
dynamic vibration tests (section 4). A series of boundary conditions for supporting the mass
timber panels were selected for investigation. These setups were evaluated for their
repeatability in regard to measured vibrational response, accuracy when compared with an
orthotropic model of a thin plate with similar properties to the test material, and the practical
considerations for industries appetite towards the nominated conditions (in-line application).
Part three of the thesis investigates the effects of the finite element analysis (FEA) structure
used to simulate a CLT panel (section 5). This component of the study was to investigate cross
laminated timber (CLT) panels to evaluate whether there are inaccuracies in assuming the
product responds to impulse as a solid plate element. CLT is a layered panel consisting of
laminated boards bonded face to face with no bonding between boards edgewise, therefore
introducing discontinuities or gaps. This section of the study compares a number of modelling
parameters against experimental modal analysis results to identify potential accuracy increases
to be had in considering a higher degree of detail in the structure and design of the FEA.

Part four of this thesis presents experimental and theoretical test data obtained from imposing
the testing method expanded on through previous sections on a series of mass timber panels
(section 6). Materials were sourced with a focus to evaluate a range of sizes and different
thicknesses to provide a robust proof of concept testing system for measuring mechanical
performance of mass timber panel products. After vibratory testing of the large scale panels
static testing was conducted to evaluate the products elastic moduli (MoE) and shear (G)
properties to validate the NDE system.

iii
 Declaration of Originality
This work has not previously been submitted for a degree or diploma in any university. To the
best of my knowledge and belief, this thesis contains no material previously published or
written by another person except where due reference is made in the thesis itself.

iv
 Acknowledgement of Papers included in this Thesis
Included in this thesis is a paper in section 4 which is co-authored with other researchers. My
contribution to the co-authored paper is outlined at the front of the relevant section. The
reference for this paper including all authors, is:

Section 4: A. Faircloth, L. Brancheriau, H. Karampour, S. So, H. Bailleres, C. Kumar, (2021).

‘Experimental Modal Analysis of Appropriate Boundary Conditions for the Evaluation of
Cross-Laminated Timber Panels for an in-line Approach’. Forest Products Journal 71(2): 161-
170. Doi:10.13073/FPJ-D-20-00062

v
 List of Publications
Journal Papers:

A. Faircloth, L. Brancheriau, H. Karampour, S. So, H. Bailleres, C. Kumar, (2021).

‘Experimental Modal Analysis of Appropriate Boundary Conditions for the Evaluation of
Cross-Laminated Timber Panels for an in-line Approach’. Forest Products Journal 71(2): 161-
170. Doi:10.13073/FPJ-D-20-00062

Conference Papers:

A. Faircloth, H. Karampour, L. Brancheriau, H. Bailleres, S. So, C. Kumar. ‘Vibratory

Evaluation of Material Damping Performance of Cross-Laminated Timber Panels’. Paper
presented in 2021 at 10th Australasian Congress on Applied Mechanics (ACAM 10, 2021),
Adelaide, Australia. (Submitted 3rd October 2021, Under Review).

vi
 Acknowledgements
I would like to express my gratitude towards Griffith University and my principal supervisors,
Dr. Stephen So and Associate Professor Hassan Karampour for continuously striving to ensure
I achieve my best from this research. Their help throughout this process has been limitless.
Since this higher degree undertaking the connections I have made through my supervisors will
be future contacts I hope to utilise in my growing career. I would like to show my appreciation
to my external supervisors, Dr. Chandan Kumar, Dr. Loic Brancheriau, and Dr. Henri Bailleres.
Their knowledge of the industry for which this research is based around has greatly impacted
the destination and focus of this study; the success of this research would not be possible
without their mammoth input.

For approaching me with this opportunity, I am deeply indebted to my workplace; the

Queensland Government Department of Agriculture and Fisheries (DAF). Their constant belief
in my abilities and push for me to do my best makes me strive to achieve great things. The
research project under which this study has taken place forms part of the Australian Research
Councils (ARC) Centre for Future Timber Structures (CFTS) Hub. It is important I
acknowledge the funding component of this research for without whom, this project would not
be possible. The CFTS Hub serves an important purpose in shining a light on timbers’ wide
range of applications in the construction industry and I am proud to be contributing towards
that awareness.

Furthermore, the success of this project was supported by one person in particular, without
whom the achievements and activities of this research would not be possible. Mr. Eric Littee,
research assistance, DAF technician, colleague and great friend to many was a key and
unquantifiable asset to this project and will be greatly missed. The authors would like to pay
special tribute and acknowledgement to Eric Littee for his dedication and friendship.

vii
 Table of Contents
Abstract ................................................................................................................................................... ii

Declaration of Originality ...................................................................................................................... iv

Acknowledgement of Papers included in this Thesis ............................................................................. v

List of Publications ................................................................................................................................ vi

Acknowledgements ............................................................................................................................... vii

List of Figures ........................................................................................................................................ xi

List of Tables ........................................................................................................................................xiv

1 Introduction ..................................................................................................................................... 1

1.1 General .................................................................................................................................... 1

1.2 Timber Panel Products in Construction .................................................................................. 1

1.3 Quality Assurance Methodologies .......................................................................................... 3

1.4 Non-Destructive Evaluation Techniques ................................................................................ 4

2 Literature Review............................................................................................................................ 7

2.1 Mass timber in Construction ................................................................................................... 7

2.2 Non-Destructive Evaluation Techniques ................................................................................ 9

2.3 Boundary Conditions ............................................................................................................ 14

2.4 Theoretical Background Information .................................................................................... 16

2.4.1 Introduction to Beam Theory ........................................................................................ 16

2.4.2 Plate Theory .................................................................................................................. 18

2.4.3 Evaluation of CLT ........................................................................................................ 22

2.5 Modal Theory........................................................................................................................ 23

viii
 2.5.1 Experimental Modal Analysis ....................................................................................... 25

2.6 Standardised Static Validation Methods ............................................................................... 29

2.7 Summary of Reviewed Literature ......................................................................................... 30

2.8 Research Questions ............................................................................................................... 32

2.8.1 Development of a Non-Destructive Evaluation System for CLT ................................. 32

2.8.2 Assessment of Appropriate Boundary Conditions ........................................................ 32

2.8.3 Modelling Approaches to Identify Potential Inaccuracies ............................................ 32

2.9 Research Objectives .............................................................................................................. 32

2.9.1 NDE system Development ............................................................................................ 32

2.9.2 BC Evaluation ............................................................................................................... 33

2.9.3 Modelling of CLT ......................................................................................................... 33

2.10 Thesis Outline ....................................................................................................................... 33

3 Acquisition System and Testing Method ...................................................................................... 35

3.1 Software Development.......................................................................................................... 35

3.2 Sensor Calibration ................................................................................................................. 38

3.3 Analysis Method ................................................................................................................... 40

4 Experimental Modal Analysis of Appropriate Boundary Conditions for the Evaluation of Cross-

Laminated Timber Panels for an In-Line Approach ............................................................................. 43

4.1 Abstract ................................................................................................................................. 44

4.2 Introduction ........................................................................................................................... 44

4.3 Materials and Methods .......................................................................................................... 47

4.3.1 Test Specimens ............................................................................................................. 47

ix
 4.3.2 Experimental Protocol................................................................................................... 48

4.3.3 Vibration Assessment ................................................................................................... 50

4.3.4 Finite Element Analysis ................................................................................................ 51

4.4 Results and Discussion ......................................................................................................... 53

4.5 Conclusions ........................................................................................................................... 59

5 Mode Shape Regeneration ............................................................................................................ 60

5.1 Background ........................................................................................................................... 60

5.2 Materials and Methods .......................................................................................................... 60

5.2.1 Test Specimens ............................................................................................................. 60

5.2.2 Experimental Protocol................................................................................................... 61

5.2.3 Theoretical Techniques ................................................................................................. 62

5.2.4 ANSYS FEA ................................................................................................................. 62

5.2.5 MAC Analysis .............................................................................................................. 64

5.3 Results and Discussion ......................................................................................................... 65

5.4 Conclusion ............................................................................................................................ 73

6 Evaluation of NDE System ........................................................................................................... 75

6.1 Background ........................................................................................................................... 75

6.2 Materials and Methods .......................................................................................................... 75

6.2.1 Test Specimens ............................................................................................................. 75

6.2.2 Dynamic Assessment Protocol...................................................................................... 76

6.2.3 Optimisation Algorithm ................................................................................................ 78

6.2.4 Static Evaluation Protocol ............................................................................................. 79

x
 6.3 Results and Discussion ......................................................................................................... 82

6.3.1 Robustness of the NDE Optimisation Procedure .......................................................... 82

6.3.2 Experimental Testing .................................................................................................... 83

6.4 Conclusion ............................................................................................................................ 92

7 Overall Conclusions ...................................................................................................................... 93

7.1 Thesis Findings ..................................................................................................................... 94

7.2 Recommendations for Future Work...................................................................................... 96

8 Bibliography ................................................................................................................................. 97

List of Figures
Figure 1-1: Image of 25 King Street, mezzanine floor. .......................................................................... 2

Figure 1-2: CLT cross layering orientation (Middleton, 2017). ............................................................. 3

Figure 1-3: (a) time domain, (b) frequency spectrum. ............................................................................ 5

Figure 2-1: example constructions of (left) –NLT, (centre) – GLT, and (right) – CLT. ........................ 8

Figure 2-2: CLT elastic and shear modes and sections. ........................................................................ 11

Figure 2-3: Shows difference between planar section being both perpendicular and not to the deflected

beam (FEA, 2019)................................................................................................................................. 18

Figure 2-4: Geometric definition of a rectangular plate with moments and forces (Lekhnitskii, 1968).

.............................................................................................................................................................. 19

Figure 2-5: First mode example. ........................................................................................................... 24

Figure 2-6: Second mode example........................................................................................................ 25

Figure 2-7: ‘VibraPann System’ diagram of modal testing and modal analysis for full sized wood-

based panels. ........................................................................................................................................ 27

Figure 2-8: mode shapes of thin isotropic plate (F. Bos et al., 2003). .................................................. 27

xi
Figure 2-9: Comparison of two models through the MAC criterion (Vacher et al., 2010)................... 29

Figure 2-10: Flow chart of systematic approach to NDE system development. ................................... 34

Figure 3-1: Signal duration example. .................................................................................................... 36

Figure 3-2: LPF example. ....................................................................................................................... 37

Figure 3-3: Flow chart of signal processing and test diagram. ............................................................. 38

Figure 3-4: Experimental setup for pendulum test of accelerometer. ................................................... 39

Figure 3-5: Pendulum test results for (top) the phase angle of the pendulum in radial axis, (b) velocity

of pendulum, and (c) the measured acceleration (orange) and predicted response (blue). ................... 39

Figure 3-6: Four mode shapes showing the bending and torsional modes.......................................... 40

Figure 3-7: Free-Free boundary condition supports; recommended impact and sensor point. ............. 40

Figure 3-8: Experimental test setup for preliminary samples ............................................................... 41

Figure 3-9: Static bending test used for evaluating plywood panel EN789 (2004). ............................. 42

Figure 4-1: (top) Experimental setup of a) SSSS, b) SFSF (shown) and FSFS BCs, c) Airbags (FFFF),

d) Suspended on flat (FFFF), e) Suspended on edge (FFFF), (bottom) diagram of the BCs shown in a)

through e). ............................................................................................................................................. 49

Figure 4-2: Impulse location and stress wave propagation through a CLT panel................................ 50

Figure 4-3: First four mode shapes for the 4 boundary conditions (a) a freely supported system

(FFFF), (b) simply supported widths and free lengths (SFSF), (c) free widths and simply supported

lengths (FSFS), and (d) all sides simply supported (SSSS). ................................................................. 52

Figure 4-4: BC supports and recommended impact locations (a) – FFFF, (b) – SFSF, (c) – FSFS, (d) -

SSSS...................................................................................................................................................... 53

Figure 4-5: (top) – FFFF variation comparison, (bottom) – SFSF, FSFS, and SSSS comparison. ...... 54

Figure 4-6: Comparative results between simulation and experimental resonance frequencies for the 6

BCs (top) modes 1 to 6, (bottom) modes 7 to 12. ................................................................................. 55

Figure 4-7: Quality factor vs mode number for BCs. ........................................................................... 56

xii
Figure 4-8: Comparative results for airbag pressure investigation (top) modes 1 to 6, (bottom) modes

7 to 12. .................................................................................................................................................. 58

Figure 4-9: Pressure sensitivity investigation, changes in signal quality with rising pressure. ............ 58

Figure 4-10: Rapid decrease in peak quality, leading to unwanted error in peak identification

(presented in 70, 350 and 700 kPa). ...................................................................................................... 59

Figure 5-1: CLT panel layering and design. ......................................................................................... 61

Figure 5-2: (top view) Highlighted modal testing points on panel. ...................................................... 62

Figure 5-3: Mode shapes obtained through (left) experimental modal analysis and mode shape

reconstruction, (right) FEA of a 3-layer plate CLT composite panel (G10/G8/G10) in ANSYS, and

MAC analysis for each paired mode shape comparison. ...................................................................... 70

Figure 5-4: a) 3D representation of MAC values, and b) 2D representation of MAC values. ............. 71

Figure 5-5: a) 3D representation of MAC values, and b) 2D representation of MAC values. ............. 73

Figure 6-1: Experimental BC setup for vibration assessment of CLT panels....................................... 77

Figure 6-2: (left) impulse signal in time domain, and (right) the result of the frequency response

function (FRF). ..................................................................................................................................... 77

Figure 6-3: Cutting plan applied to CLT panels for test strips. ............................................................ 79

Figure 6-4: 4-point bending test setup in accordance with EN408 (2010) cl. 10 and 11 for measuring

MoE and Shear...................................................................................................................................... 80

Figure 6-5: DIC setup for deformation measurement to determine shear moduli. ............................... 81

Figure 6-6: Robustness error for rank groupings. ................................................................................. 82

Figure 6-7: Ex comparison with static and dynamic assessment for the linear fitted data set. .............. 84

Figure 6-8: Corrected Ex for outliers with (a) a linear response, and (b) a balanced response of the data

set. ......................................................................................................................................................... 85

Figure 6-9: Ey comparison with static and dynamic assessment for the linear fitted data set. .............. 85

Figure 6-10: Corrected Ey for outliers with (a) a linear response, and (b) a balanced response of the

data set. ................................................................................................................................................. 86

xiii
Figure 6-11: Gxy comparison with static and dynamic assessment for the linear fitted data set. .......... 88

Figure 6-12: Gxy percentage difference within sub-strip pieces for static testing and NDE. ................ 88

Figure 6-13: Corrected Gxy for outliers with (a) a linear response, and (b) a balanced response of the

data set. ................................................................................................................................................. 89

Figure 6-14: Example of CLT panel defects noted through testing. ..................................................... 90

Figure 6-15: Values of the minima for the elastic moduli values (a) 3D view and (b) topographic

view....................................................................................................................................................... 91

Figure 6-16: Values of the minima for the transverse shear (a) 3D view and (b) topographic view. ... 91

List of Tables
Table 3-1: Signal acquisition and DSP settings required for large scale trials. .................................... 37

Table 3-2: Results of plywood NDE predicted mechanical properties (on full sheet), and the static

values. ................................................................................................................................................... 42

Table 5-1: Assumed Stiffness compositions. ........................................................................................ 63

Table 5-2: Material properties for 3-layer shell CLT model in FEA (USDA, 2010)............................ 63

Table 5-3: Single layer CLT scenario, material properties obtained from classical laminate theory

(Oregon State University OSLU freeware)........................................................................................... 64

Table 5-4: Results summary of stiffness compositions and experimental modal testing. .................... 65

Table 5-5: Results summary for CLT scenario considerations of gap size and board friction. ............ 71

Table 6-1: Table of CLT specimen constructions, nominal sizes, and densities. ................................. 76

Table 6-2: Robustness percentage difference values. ........................................................................... 83

Table 6-3: Summary of NDE and static results for MoE and G. .......................................................... 87

xiv
1 Introduction
1.1 General

Timber has long been the primary material used in construction for residential and low-rise
structures. Recent advances in the use of engineered wooden products (EWPs) has enabled
their expanded use into multi-rise timber structures. Wood-based mass panels (WBMPs) are
becoming more and more prevalent in the construction industry for their excellent seismic
performance, advantages of prefabrication, cleaner and faster on-site construction timelines,
significant carbon storage, avoidance of fossil-fuel intensive materials, reduced construction
cost and the potential for improved building envelope thermal performance (Wang et al., 2018).
WBMPs are manufactured by combining timber boards and/ or veneers together by means of
adhesives, mechanical fasteners or a combination of the two. This section will explore a series
of relevant introductory sections to establish the narrative on the use of timber in construction
followed by a detailed review of currently published literature.

1.2 Timber Panel Products in Construction

WBMPs rely on the design of the panel system to achieve enhanced mechanical performance
through the combination of single laminates or veneers, combined through either adhesion or
fastener fixing. A selection of WBMPs have been categorised into the following four classes:

 Timber or sawn board-based products: These products are manufactured through the
combination of boards using either a fastener or adhesive. Two relevant examples of
this technique in practice are (i) cross laminated timber (CLT – ‘Monterey apartments’
as a floor and wall element, Kangaroo Point, Brisbane, QLD, Australia) and (ii) nail
laminated timber (NLT – ‘The Hudson’ floor systems, Vancouver).
 Veneer based products: These products are manufactured through the use of peeled
logs into veneers and are then bonded together to form a veneer-based panel. These are
used to create both plywood and laminated veneer lumber (LVL). These products can
then be further bonded together to form combinations of the two or larger dimensions
of smaller panels to form veneer-based mass panels (VBMPs – Mass plywood pavilion,
Portland art museum, Portland, Oregon, USA).
 Strand based products: Manufactured from flaked or chipped wood products these
panels are often seen as a means of reducing waste from board processing and an

1
 appropriate example of the circular economic model timber-based manufacturing can
enhance. Some examples of these include orientated strand board (OSB – commonly
used in framing and cladding of walls and floors under short spans) or particle board.
 A combination of the above: The above methods are all equally suitable for various
applications throughout the construction process and environments, a combination of
these techniques is not uncommon and is not limited to the above listed methods. For
example, a commonly used combination with other building materials would be a
timber concrete composite (TCC – Okanagan College, Penticton, Canada) slab.

Recent constructions locally within Australia witnessed by this researcher include the CLT
office building constructed in late 2018 (25 King Street, Brisbane, Australia), the tallest
engineered-timber office tower world-wide at the time (shown in Figure 1-1). A more recent
local development is the Monterey 10 storey CLT apartment building (9 Lambert Street,
Kangaroo Point, Brisbane). An impressive structure located in central Norway, playfully
named the ‘plyscraper’ Mjøstårnet is an 18-storey timber hotel completed in mid-2019 and is
the tallest timber-framed structure in the world. The proposed hybrid timber tower for the
central Sydney CBD will reach an impressive 40 storey’s tall and provide a template for the
benefits of incorporating timber within multi-storey structures. With many other similar
structures across the globe, the stigma that previously and falsely assumed timber is weaker
than steel and concrete appears to be dissolving.

Figure 1-1: Image of 25 King Street, mezzanine floor.

2
CLT is the most widespread and versatile mass timber product used commonly for ceilings,
floors and wall elements. With panels of different surface finish, size and structural
applications, the materials are designed for the application. Meaning manufacturers are able to
respond to complex designs with ease. CLT has been available overseas since it was developed
in the 1970’s and continues to evolve through emerging construction methodologies with new
capabilities being identified regularly through on-going research and development. Figure 1-2
presents a graphical representation of the construction method of CLT.

Figure 1-2: CLT cross layering orientation (Middleton, 2017).

Bonding in CLT is commonly concentrated to the face adhesion where the edge surfaces are
left unbonded. This approach in regard to layering and bonding gives CLT a consistent
performance in both axial directions. As the primary product in the mass timber construction
arsenal. This study has focused its attention on CLT as the product and manufacturing industry
of interest and how technologies can be employed to maintain the consistency of produced
material.

1.3 Quality Assurance Methodologies

Quality assurance (QA) is a process by which the production process and procedures can be
validated and maintained or alternatively in some cases altered if required. Timber processing
QA practices involve regular sampling of batch manufactured boards and panels for selected
testing (for example static testing to confirm grade). This is a common recommendation by
most material specific design standards as well as a process employed by certification bodies

3
to monitor on-going product performance. QA techniques employed by processors and
manufacturers include an in-line stress testing device for rapid grading of boards, known as a
machine stress grader (MSG) (MSG, Metriguard Inc., Washington, USA). As the elastic
modulus (stiffness) is one of the major governing parameters for the use of timber in a structure,
the MSG conducts a rapid pre-stress test to determine stiffness and assign a grade in accordance
with the relevant design standards. These methods are well established for board QA and are
used regularly throughout the timber industry by processors world-wide. Unlike boards, mass
timber products such as CLT are not made to a set specific dimension and can vary depending
on the application. CLT also has no established rapid QA approach. As such QA methods
currently used by manufacturers include static testing of sections taken from the manufactured
panels to confirm grade and composite theories (such as the shear analogy method - SAM).
The SAM is used to design CLT panels based on the requirements by assigning graded boards
to appropriate positions within the panels.

While both static testing and SAM methods provide relevant information as to the expected
performance of the manufactured CLT, they are not without their limitations. SAM is a pre-
manufacturing method as it does not observe the effects of the various manufacturing stages,
it is highly dependent on the quality of the graded boards. Static testing provides data specific
to the material being tested post-manufacturing. While this is a more representative approach
to QA it is a time consuming, costly and destructive method. The following section describes
alternatives to current streamlined QA techniques and its suitability to CLT.

1.4 Non-Destructive Evaluation Techniques

Non-destructive evaluation (NDE) practices broadly refer to the alternative of static testing or
destructive evaluation. This research focuses on the dynamic based NDE techniques
determined through measured vibrations of a material. NDE is widely used through the
construction industry for materials such as steel, concrete and timber. These techniques mainly
include:

 Visual testing (appearance grading or image based NDE),

 Ultrasound testing,
 Acoustic emissions testing,
 Modal testing and transient response testing,

4
The testing methods listed above all consist of the assessment of impulses injected into the
material of interest. Impulse vibration analysis has been widely used to estimate the physical
and mechanical properties of many materials including wood and wood-based composites for
several decades. It is also commonly used for the identification of defects. Vibration analysis
cannot directly determine the strength properties of EWPs. Correlation between the elastic and
strength properties are generally significant for wood-based products allowing for the
development of predictive models of strength on the basis of simple correlations. These results
act as an indication of the properties, they give a good estimation of the mechanical
performance of the specimen, especially for wood-based panels for which these properties are
fairly uniform at this scale. NDE methods and analysis have been widely used to differentiate
many sorts of materials and their internal properties. Timber and EWPs are no exception to this
statement and these techniques have been known to replace traditional evaluation methods such
as visual characterisation or static tests. Most of these techniques are based around simple
physical parameters such as vibration analysis and acoustic measurement. Vibration based
measurement methods have been selected as the optimum testing method for this project, as
detailed in the literature review. As such, reference to ‘NDE’ from this point will refer to the
vibrational measurements of impulse signals created through impacting the material. NDE
consists of the experimental assessment of a materials response to vibrations injected into the
surface of the material through impact. Using a surface mounted device such as an
accelerometer, the output impulse can be observed and recorded, this time-based response is
referred to as the time domain (Figure 1-3 (a)). The vibration based NDE approaches use the
measured materials resonance frequency as an input parameter along with the dimensions and
density of the material.

Figure 1-3: (a) time domain, (b) frequency spectrum.

5
The time signal (Figure 1-3 (a)) represents that of a decaying impulse function, where the signal
begins at a maximum (the point of impact), and decays over time back to the zero point. The
short duration loading that is experienced through the impacting of the panel provides potential
energy to the material that excites a wide frequency range. This impulse signal is formed
through the summation of several sinusoids oscillating at different frequencies, where the
largest amplitude sinusoid correlates to the natural frequency or first harmonic. As is the
operation of the fast Fourier transform (FFT) function, these frequencies are decomposed and
represented as a peak along the frequency axis. The frequency spectrum (example depicted in
Figure 1-3 (b)) can contain information other than the impulse response of the measured
material such as effects of the environment, supporting conditions, and sensor cables. In these
systems the experimental signal obtained from the vibrational analysis of the tested material is
coupled with a robust and accurate theoretical mode and solution to the material being
evaluated. This is a crucial component to an NDE system as the accuracy is dependent on the
appropriate input parameters being selected from the experiments as well as the supports or
boundary conditions (BCs). The theoretical solution allows for peak detection assistance in the
frequency spectrum and avoids selection of the false modes.

The use of NDE techniques is well established within the timber industry with research
expanding into the evaluation of engineered timber and mass timber products. As described
further in the literature review in section 2, attempts have been made at adopting such NDE
techniques to CLT with success. Although no such system has been established within a
manufacturing facility to date and to the knowledge of this researcher. Therefore, the aim of
this study has been to develop a rapid non-destructive QA approach for CLT panels. As part of
this main focus, parameters concerned with the development of a rapid NDE system for CLT
will also be investigated such as efficient supporting conditions, approaches to modelling CLT
and validation through comparative methods.

6
2 Literature Review
2.1 Mass timber in Construction
Cross-laminated timber (CLT) is the most widespread mass timber product and is commonly
used across ceilings, floors and wall applications. CLT is very versatile in its applications, with
panels of different surface finishes and structural properties that can be applied to different
areas of construction. The material is known to be fire-resistant and good for large span
applications. CLT has been available overseas since it was developed in the 1970’s (Xia et al.,
2014), and continues to evolve through emerging construction methodologies with new
capabilities being identified. A recent example in Brisbane was an all-timber office building
constructed by LendLease in late 2018. The 45-metre-tall building is currently the tallest
engineered-timber office tower on Earth. More detail on the production process and quality
control of CLT panels can be found in (Brandner, 2013).

Glued laminated timber (GLT), also known as Glulam is a type of engineered wood product
(EWP) composed of layered boards bonded together with a durable structural adhesive. One of
the main differences between CLT and GLT composite panels is the way in which the boards
are layered. While CLT consists of layers of timber boards orientated perpendicular to the
preceding layer, GLT boards are layered in the same direction; the grain direction of all boards
is orientated in the length direction. The laminates used in GLT are commonly finger jointed
together into continuous lengths, enabling columns and panels to be constructed longer than
the dimensions of the timber they are created from. Similar to CLT one advantage of the GLT
assembly is the minimisation of strength-reducing characteristics such as defects found in solid
wood (knots, checks, etc.) either being absent from the construction (cut away) or restricted to
one board. An example of GLT can be seen in Figure 2-1.

Due to the construction method used for these panels, it is not uncommon for manufactures to
design beams or panels with higher strength rated lamella in areas of high stress, such as the
top or bottom layers of the panels/ beams, and likewise lower strength lamella can be placed in
lower stress areas (core). As Figure 2-1 (right) shows, CLT’s construction employs each layer
to support the other. This effectively limits the number of weak points within the panel in terms
of its strength properties. GLT’s (Figure 2-1 (centre)) lamella arrangement does not have as
great an impact on these parameters, directing most of the strength enhancement towards the
adhesive used.

7
 Figure 2-1: example constructions of (left) –NLT, (centre) – GLT, and (right) – CLT.

In any application of adhesives, the bond between surfaces should be stronger than the material
strength. Therefore, the bond between the laminates needs to be as strong or stronger than the
inherent strength of the timber to ensure a robust product. Nail laminated timber (NLT) panels
get their strength from the fixings that secure the individual laminates together. Unlike some
other wood and wood composite panel constructions, NLT does not require any dedicated
manufacturing facilities. Its construction method consists of stacked timber and mechanically
fastening the individual laminates together with either screws or nails. Due to NLT’s modest
design, manufacturing costs are lower than its other relative panel constructions (such as CLT
and GLT) which makes it a much more cost-effective option. Figure 2-1 shows that there are a
number of differences between NLT and CLT. For example, visual identification that CLT
laminates are not limited in terms of thickness, whereas NLT is essentially dependant of the
length of the nail holding the laminates together. Similar to NLT, the dowelled laminated
timber (DLT) panels are produced from softwood lumber fixed together with hardwood dowels
(commonly between 12-24 mm diameter) to friction fit instead of glues, resins, nails, or metal
fasteners to form beams and panels. Laminated veneer lumber (LVL) is a composite wood
material made from sheets of veneer that are laminated together with their grain orientation in
the same direction (Suzuki et al., 2016). The manufacturing processes of LVL are described in
AS/NZS4357.2 (2006) and the European standard EN14374 (2004). In the strand-based panels

8
such as parallel strand lumber (PSL), laminated strand lumber (LSL), and oriented strand
lumber (OSL), strands are used instead of veneers or boards. They are manufactured by gluing
strands of wood together under pressure that are oriented primarily along the length of the
member with a waterproof adhesive. These mass panels can be combined with other mass panel
or other construction material to form composite panels.

Although the lamellas are stress graded prior to combination, the panels still prove difficult to
effectively categorise into different strength groups in terms of the global panel properties. This
is a common problem for WBMPs, with NDE methods being the ideal analysis technique when
evaluating mechanical properties. Vibration analysis techniques can be widely applied to a
variety of WBMP types. Commonly CLT is manufactured at lengths up to 18 m, widths of up
to 3 m and thicknesses of up to 400 mm (Brandner et al., 2016). Majority of construction
applications and transportation restrictions call for panels 3.2 m in length, 2.8 m in width, and
thicknesses ranging from 75 to 300 mm. The system will be designed around the CLT panels,
their properties, material composition, and dimensions. The size and properties of the tested
panels will be nominal with respect to similar products available of either different construction
strategies or materials. The testing will be conducted on large scale samples manufactured with
Radiata Pine (pinus radiata); the specific panel constructions and sizes will be defined in their
respective sections.

2.2 Non-Destructive Evaluation Techniques

A classical version of the NDE method is referred to as the ‘tap test’ or coin tap method
commonly used in assessments of walls, floors and columns (Cawley et al., 1988). The
hypothesis behind the use of the method is that with a delaminating layer behind the surface
being tapped, there should be an audible or force felt difference in comparison to a solid point
along the structure. While the use of NDE has continued to transform, the premise is still sound
that the method is a rapid, efficient, and cost-effective means of QA. As introduced in section
1 the NDE concept and approach for this project centres around dynamic vibration. Dynamic
vibration refers to the use of transient impulse measurements through the material of interest
to collect the resonance frequencies as well as damping performance.

Morandi et al. (2021) conducted a review of five relevant evaluation methods for acquiring the
required information from CLT panels. Of these five methods, three considered time domain
signals (maximum peak, cross correlation, and kurtosis) and two through the frequency

9
spectrum (phase difference and wave correlation). Morandi et al. (2021) conducted a series of
tests both experimental and theoretical for isotropic, non-homogeneous material and evaluated
the results of the tests through the five evaluation methods.

Vibration impulse evaluation is a common method used through NDE for the determination of
mechanical properties (Zang et al., 2017). These measurements are taken by using a sensing
device for either vibrational or acoustic based detection and coupled with an excitation device
such as an impactor. The sensor, whether it be an accelerometer or microphone measures the
time taken for impulse vibrations to pass from the excitation point to the sensing location.
Bucar et al. (2011) evaluated a group of stress graded boards using both NDE and static
modulus of elasticity (MoE) tests to authenticate the grade assignment. One hundred boards
were evaluated using transient vibration assessment and from which the fundamental frequency
used to calculate the dynamic MoE. Result comparison between the two methods produced a
92% agreement across the sample population. A study by Shirmohammadi et al. (2020)
compared 15 species of small clear timber boards for their suitability as fingerboards within
acoustic guitars. Of the properties evaluated MoE was of interest and determined through a
comparison of NDE and static testing. The NDE evaluation was conducted through a
commercially available acoustic/ vibration tool for beams and boards, BING (CIRAD, France).
The BING system uses digital signal processing (DSP) techniques and acoustic and/or
vibrational sensors to deliver frequency information to determine MoE. The transverse MoE is
determined through classical formulas such as Eq. 2-1 for Bernoulli’s principle and Eq. 2-2 for
Timoshenko’s theory (Brancheriau et al., 2002):

𝜌𝐴𝐿 𝑓
𝐸 = 4π Eq. 2-1
𝐼 𝑃

where 𝜌 is density, 𝐿 is sample length, 𝑓 is the natural frequency of the transverse

vibration, 𝐴 is the cross section on the test specimen, and 𝐼 is the moment of inertia
about the cross section.

𝐸 𝐸 𝐴𝐿 𝑓 𝜌𝐴𝐿 𝑓
− 𝑄𝐹 (𝑚)4𝜋 = 4π [1 + 𝑄𝐹 (𝑚)] Eq. 2-2
𝜌 𝐾𝐺 𝐼 𝑃 𝐼 𝑃

where:

𝐹 (𝑚) = 𝜃 (𝑚) + 6𝜃(𝑚),

10
 𝐹 (𝑚) = 𝜃 (𝑚) − 2𝜃(𝑚), and
tan(𝑚) tanh(𝑚) 𝜋
𝜃(𝑚) = 𝑚 , 𝑤𝑖𝑡ℎ 𝑚 = 𝑃 = (2𝑛 + 1)
tan(𝑚) − tanh(𝑚) 2
the parameters 𝑚, 𝑃 , 𝐹 (𝑚), and 𝐹 (𝑚) are determined for index n referred to the
as the resonance of each mode type.

Based on the intended application and aim of this research surface mounted accelerometers
were the chosen sensor type. The approach implemented by Steiger et al. (2010) has been the
focus method for these experiments where the CLT panels were supported by a selected
boundary condition (BC) and accelerometers attached to the panel surface. Signals were
analysed using the system detailed in section 3 and compared against a theoretical model as
detailed in section 2.4. The method was originally referred to as experimental and theoretical
modal analysis and was expanded on by Steiger et al. (2010) by using the experimental results
as feedback to the theoretical model and provided a solution to the inverse problem. The
experimental modal analysis was conducted by measuring the materials response to vibration
at a series of points across the panels surface using the roving sensor approach. The research
outlined by Steiger et al. (2010) aimed at developing an in-line NDE system for evaluating
CLT. Targeting a freely supported (FFFF) boundary condition the system was effectively able
to identify 2 elastic moduli (Ex and Ey) and three shear moduli (Gxy, Gyz, and Gxz); Figure 2-2
displays these parameters.

Figure 2-2: CLT elastic and shear modes and sections.

To correctly validate the compared theoretical and experimental frequencies, the modal
assurance criterion (MAC) analysis is commonly employed (Steiger et al., 2010). Using MAC

11
analysis, the generated and measured modes for the corresponding frequencies were compared
and changes were made to the elastic and shear values until the MAC agreement was within
95%. The two classical approaches for modal analysis include (i) roving hammer or impactor
approach and (ii) roving sensor approach (Avitabile, 2001). These two methods are identical
in their requirements where the equipment needed involves the test material, the sensor(s), the
impacting device, and the signal processing equipment. The two differ only in their approach
to data collection where one fixes the position of the sensor and changes the location of impulse
(roving impactor) and the other fixes the impacting location and changes the sensor position
(roving sensor). While the two methods are comparatively identical, to extract modal
information of the test material with minimal interpretation as to the location specific signal
velocity, a roving sensor approach has been adopted in the thesis and the experiments detailed
in section 5.

Similarly to Steiger et al. (2010) and Zhou et al. (2016) implemented modal analysis to
evaluated wood-based panels for the determination of mechanical properties. Using this
method across two 5.25 x 2.15 x 0.1 m CLT panels, five 2.44 x 1.22 x 0.011 m oriented strand
board (OSB) panels, and five 2.46 x 1.24 x 0.015 m medium density fibreboard (MDF) panels
the researchers were able to identify the elastic and shear moduli of the tested panels. Based on
the dimensions of the materials through this study, all panels can be considered as ‘thin’ where
the length to thickness (Lx/h) ratio is > 20. Through an investigation into the BCs required, it
was identified that they have a large influence on the accuracy of the obtained modal results.
This is especially correct when relying on the comparison between an experimental and
theoretical approach. Zhou et al. (2016) used an iterative approach where the theoretical
models’ initial values for elastic and shear moduli were increased or decreased by
approximately 0.2%. This was continued until the error between the compared frequency
values from the experimental approach were below 5%. These results were validated against
static testing and an agreement of 6.4% (major direction (Ex)), 3.8% (minor direction (Ey)), and
4.6% (in-plane shear modulus (Gxy)) was found.

Following this study, Zhou et al. (2020) expanded on their previous research by continuing into
the evaluation of 3 CLT panels (two 3-layer 3.24 x 1.07 x 0.105 m, and one 5-layer 4.06 x 1.02
x 0.132 m). Evaluating a thick orthotropic plate requires consideration of the third dimension
noted as thickness or depth, the definitions and differences between thick and thin plates are
detailed in section 2.4. The refined NDE system aimed at accurately predicting elastic and shear

12
moduli in both major and minor directions using a similar inverse iterative approach as outlined
through the previously noted literature. Zhou et al. (2020) investigated the accuracy between
three sets of resonance frequency combinations for (i) lower 5 frequencies (f1 to f5), (ii) higher
5 frequencies (f6 to f10), and (iii) frequencies f1 to f7. The frequency combinations showed that
while all combinations were able to predict both Ex and Ey, combinations (i) and (iii) produced
the lowest error on average. The elastic moduli in the major direction were predicted within
7% of the static values although the minor direction determination requires further verification.
It was concluded that combination (iii) provided a higher accuracy than (i) and reported that
increasing the number of frequencies can enhance the accuracy of the NDE system prediction.
Implementing the equation of motion defined by Kirchoff-Love (defined in section 2.4.2.1) the
Ex and Ey as well as Gxy is determined. From this approach the mechanical properties predicted
are directly related to the transient response, density and Poisson’s ratio. The results presented
through the study found that Ex produced the stiffer result compared to Ey. Furthermore, Ex
produced results 2 magnitudes higher than the Gxy values. Opazo-Vega et al. (2021)
investigated the application of the transverse vibration NDE method to predict the elastic and
stiffness characteristics of CLT. Opazo-Vega et al. (2021) selected four 3-layer CLT panels at
2.596 x 1.196 x 0.096 m for a Lx/h ratio of 27. Unlike several referenced works above, Opazo-
Vega et al. (2021) investigated the variability between the global and local performance of low
grade CLT panels. Adapting the Mindlin-Reissner theory for rectangular orthotropic plates a
strong correlation with the shear analogy method (SAM) for CLT elastic properties was
obtained.

Zhang et al. (2021) investigated the efficacy of an NDE system for predicting mechanical
properties of CLT through various degrees of deterioration and decay. The evaluated materials
have been separated into two sections of 16 CLT beam strips (2.2 x 0.4 x 0.1 m) and 16 CLT
panels (4.2 x 2.3 x 0.1 m). Through conducting experimental vibration analysis in combination
with non-contact transverse vibrations, measurements were conducted experimentally and Ex
was determined through the Rayleigh-Ritz method. Results indicated a 30% difference in Ex
values between the two analysis techniques as a result of the direction of measurements being
taken. Furthermore, comparisons of vibrational analysis with static testing resulted in a 5 to
20% difference for Ex as the specimens’ decayed (Zhang et al., 2021).

As presented through the cited literature through this section, the transient vibrations have been
measured through surface mounted accelerometers with the aim of measuring the materials

13
natural frequency. Considering the focus being to observing the materials natural frequency,
other applicable methods for measuring a materials response to transient impulse have been
reviewed (Robin et al., 2021). Robin et al. (2021) introduced another method of measuring a
materials response to transient vibration, non-contact imaging. Employing the use of a high
speed (HS) camera system, an aluminium bar was evaluated using a cantilever beam scenario
with the bar fixed in place at one end. The experiments compared the performance of the non-
contact approach to a surface mounted, single axis accelerometer. Using the two approaches,
the first 5 natural frequencies (f1 to f5) were measured, and Ex calculated using the formula Eq.
2-1. Comparing the results measured from the HS cameras and the accelerometer, an average
error of 11.2% was measured. While the accuracy of either approach was not confirmed with
the use of a FEA model, limitations noted in the HS camera system were presented in the
frequency response function (FRF). The FRF reported by Robin et al. (2021) presented a high
noise spectrum as the frequencies increase. While this does not increase the prediction error
between the two sets of measured frequencies, it can effect the selection accuracy whether it
be automated or manual.

The literature reviewed above presents a cross section of the available literature in regards to
NDE techniques and technology for the evaluation of wood and engineered wood products
(EWPs). The findings of this section in particular identify a series of study with results deemed
acceptable and in agreement with a variation between the NDE and static methods of 10 to
20%. As noted by many, the results of NDE experiments consists of the evaluation of transient
impulse signals passing through a material, from the point of impact to the sensor (Bucur,
2006). Due to the natural material properties inherent to timber it is expected that these NDE
techniques will provide a higher value for predicted properties compared to the static validation
methods. Furthermore, the literature presented through this section gives weight to the
complexity of these NDE systems and the various design stages and evaluation methods to be
conducted.

2.3 Boundary Conditions

The boundary conditions (BCs) of an NDE system refer to the supports for which the material
being measured is held or restricted from translation and/or rotation. Ensuring the experimental
supports achieve a BC identical to the modelled conditions is critical for ensuring the accuracy
of the acquired signals. External influences on the conditions such as rigid supports rather than
free supports, or inconsistent contact with the measured material can attribute to erroneous

14
signal information being collected. A study conducted by Maheri (2010) interpreted the
accuracy of modal analysis from a freely supported and fixed (CCCC or SSSS) BC where a
free support condition translates to supports such that vertical movement is not restricted at any
point along the panel. A fixed support is, at its core, the opposite to a free support where the
outer edges of the panel are restricted from movement. Maheri (2010) Measurements on fibre
reinforced plastic (FRP) panels was conducted using thin plate theorem so focus could be
concentrated towards the performance differences between the two BCs. FRP plates were
manufactured with a range of grain angle differences between the outer-most layers and found
the results to be irrespective of the BCs indicating the two conditions could be confidently
applied for the experimental modal analysis of thin plates (Maheri, 2010).

Concentrating on wood-based panels, Neiderwestberg et al. (2014) investigated a selection of

BCs for the accuracy and effort required to develop experimentally when comparing the
resultant NDE values against the static results. To control the specimen grade and quality,
Neiderwestberg et al. (2014) manufactured nine 3-layer CLT panels for the modal experiments
at 0.57 x 0.57 x 0.05 m. As an added control method, the boards were edge bonded to minimise
the amount of board movement and distorting possible. The CLT was exposed to modal testing
under FFFF, FFFS, SFSF, FSFS, and SSSS (where F stands for free and S represents simply
supported edge with translational restraint only) BCs to predict the elastic moduli in major and
minor directions as well as the transverse shear moduli of the panels. Results found the FFFF
BC to be the most repeatable and accurate setup with an average error of 0.3%, 11.9% and
10.6% for Ex, Ey, and Gxy respectively when compared against static results. In addition to the
evaluated BCs based on the modal and static comparison, Neiderwestberg et al. (2014) assessed
the practicality of implementing some of the BCs and what restrictions they may encounter in
industry. It was hypothesised that while the FFFF BC was identified as the most suitable
configuration, the tested arrangement hanging vertically could introduce erroneous results
regarding stabilisation and safety, alternative variations to the FFFF BC were to be considered.

Expanding on the findings from Neiderwestberg et al. (2014), Zhou et al. (2017) conducted
modal testing on a FFFF, SFFF, and SFSF BCs for 20 OSB panels (1.21 x 0.61 x 0.01 m) and
20 MDF panels (1.22 x 0.62 x 0.016 m). In conducting modal analysis, the first 15 harmonic
of the fundamental frequency were collected and the NDE system developed through the
Rayleigh-Ritz approach for thin plates where three elastic constants were to be measured. The
results indicated that the FFFF BC produced the higher accuracy with a < 10% error for the

15
three elastic constants determined. A finding identified was the appearance order of modes was
not consistent for all BCs and therefore should be considered for automation of such an NDE
system if adopting said BC approach. From a detailed review of research similar to this project,
it was found that changes to the BCs used for experimental testing and modelling of the
specimen have a great impact on the accuracy of the results. The BCs refer to the way in which
the specimen is restricted (or not) from movement during analysis, three BCs were chosen to
investigate on a laboratory scale. A completely free (FFFF) system, a system that is free on
opposite ends and simply supported on the others (SFSF) and a simply supported system on
one side and free on all others (SFFF).

2.4 Theoretical Background Information

2.4.1 Introduction to Beam Theory

WBMPs can be utilised as load-carrying beam or plate like structures therefore, it is necessary
to explore the governing equations of both beam and plate theory. Beam theory is most
commonly used to calculate the materials response to vibration; it can be classified into two
categories: thin or thick beam theory, which is based on their length to thickness ratio. A beam
is typically considered thin when the L/h ratio is >10 based on the slenderness ratio derived by
Han et al. (1999).

2.4.1.1 Euler-Bernouli Beam Theory

The “Euler-Bernoulli beam theory” (1750) neglects the effects of shear forces distributed
through the cross section and thus does not account for the effects of transverse shear
deformation and corresponding strains. Consequently, it under-predicts deflections and over-
predicts natural frequencies. Thin beam theory is the simplest method, which does not predict
the bending modes with sufficient accuracy as it neglects the transversal shear deformation and
the rotary inertia of the specimens’ cross section. For thin beams (beam L/h ratio >20) these
effects are of minor importance although, for thick beams, these effects can be significant and
are therefore important to recognise. More advanced beam theories such as the Timoshenko
beam theory (Timoshenko, 1922) have been developed to account for these effects. For a
dynamic Euler–Bernoulli beam, the Euler–Lagrange equation is:

𝜕 𝜕 𝜔 𝜕 𝜔
𝐸𝐼 = −𝜇 + 𝑞(𝑥) Eq. 2-3
𝜕𝑥 𝜕𝑥 𝜕𝑡

16
Here the left side of the equation represents the potential energy due to internal forces, and the
first term of right side represents the kinetic energy, where 𝜇 is the mass per unit length (𝜌𝐴),
and 𝑞(𝑥) is the external load. When the beam is homogeneous and the load is uniform 𝐸, 𝐼,
and q are independent of 𝑥, and the beam equation is simpler:

𝜕 𝜔 𝜕 𝜔 Eq. 2-4
𝐸𝐼 = −𝜇 +𝑞
𝜕𝑥 𝜕𝑡

2.4.1.2 Timoshenko Beam Theory

As introduced earlier, Timoshenko (1922) beam theory takes into account shear deformation
and the effects of rotary inertia of the beams cross section. The influence of the beam cross
section in rotary deformation is considered through beams moment of inertia (Majkut, 2009).
This makes the approach developed by Timoshenko suitable for describing the behaviour of
thick beams. The equation of motion for the vertical deflection 𝑤 for a beam with a constant
cross section is given by (Van Damme et al., 2017):

𝜕 𝑤 𝜕 𝑤 𝐸𝐼𝜌 𝜕 𝑤 𝜌 𝐼𝜕 𝑤 Eq. 2-5

𝐸𝐼 + 𝜌𝐴 − 𝜌𝐼 + + = 𝑞(𝑥)
𝜕𝑥 𝜕𝑡 𝐾𝐴𝐺 𝜕𝑥 𝜕𝑡 𝐾𝐺 𝜕𝑡

where 𝐸 is young’s modulus, 𝐼 is the second moment of area, 𝜌 is the material mass
density, 𝐺 is the shear modulus, 𝐴 is the cross sectional area of the beam, and 𝐾 is a constant
that introduces the shear force variation through thickness; the last two terms take the influence
of shear into account.

This can be verified by removing the last two terms and obtaining the Euler-Bernoulli beam
equation as seen above in Eq. 2-4. In contrast, Timoshenko’s thick beam theory, for three-layer
sandwich beams can be modelled depending on the strip grain direction (Van Damme et al.,
2017). Van Damme et al. (2017) showed that consideration of beam geometry is important for
estimating the orthotropic material constant and model the bending modes. Timoshenko
developed the “Timoshenko beam theory” during the early 20th century. The Euler-Bernoulli
beam theory states that the planar section of the beam after deformation is perpendicular to the
beam's neutral axis. This is true for cases where the applied load results in a pure bending based
response. As displayed in Figure 2-3, this geometrical condition is outlined by disregarding the
shear deformation force in the cross section of said beam.

17
However, as the L/h ratio decreases the effect of shear on the beam becomes more influential.
For CLT panels, the Euler-Bernoulli beam method accounts for deformation indirectly by
calculating the effective bending stiffness based on the efficiency of the connection between
longitudinal layers. This approach makes use of the transversal layers within the panel
providing the connection through shear stiffness (Christovasilis et al., 2016). Consider a book
bending parallel to its major axis, as the bending increases the pages shift along one another;
this is an example of the Timoshenko beam theory. As the book bends rotation occurs between
the cross section and the point of bending; indicated by the blue line in Figure 2-3. The Euler-
Bernoulli theory considers the cross section to be perpendicular to the bending line and
therefore stiffer than Timoshenko; shown by the red line in Figure 2-3.

Figure 2-3: Shows difference between planar section being both perpendicular and not to the
deflected beam (FEA, 2019).

2.4.2 Plate Theory

Mechanics defines plate theory as the mathematical description for the mechanics of flat plates.
This can be closely related to the Euler-Bernoulli, or Timoshenko beam theories. Plates are
defined as a planar structural element with a small thickness (h) in comparison to the
dimensions of the plate. Plate theory can also be categorised as either thin or thick, similar to
beam categorisation. Euler (1750) first formulated the mathematical representation of a very
thin plate in 1766 (Love, 1906). Kirchhoff then developed the first complete theory of plate
bending based on assumptions made for forces applied to the middle surface as remaining
normal to the deflected mid-plane (Love, 1906). Since this theory neglects the deformation
caused by transverse shear, it would lead to considerable errors if applied to moderately thick
plates. Reissner and Mindlin arrived at somewhat different theories for moderately thick plates
to eliminate the above mentioned deficiency of the classical plate theory. The theory developed

18
by Reissner (1976) includes the effects of shear deformation by assuming uniform shear-stress
distribution through the thickness of the plate. Mindlin (1951) also improved the classical plate
theory for vibrations by considering the effect of the rotational inertia in addition to the shear
deformation forces (Szilard, 2004). Leissa (1993) provided fundamental equations and theories
for the vibration of plates including circular and rectangular plates.

Thin plate theory was initially developed from the concept of an isotropic material (identical
elastic properties in all axial directions). Thin plate theory originates from the concept of an
isotropic material where it is assumed that the transverse shear stresses in the material are zero.
This is due to an isotropic material having identical elastic properties in all directions. The
origin of anisotropy can be regarded as the variation in the materials response to excitation with
respect to direction of the applied force (USDA, 2010). Anisotropy refers to the change in
material properties depending on the orientation from which they are measured. The definition
of an orthotropic material is one that simulates the best approximation for a real wood structures
due to the variation in the material properties along three orthogonal axes (Bodig et al., 1993).
Somewhat of a continuation from anisotropy, orthotropic material properties vary depending
on the measuring direction. The concept of both isotropic and orthotropic materials will be
investigated through this section on relevant plate theorem. A rectangular plate as shown in
Figure 2-4 is defined as a plane structural element with a small thickness (h) compared to the
planar dimensions (along the axis x and y). This difference in length scale allows reducing the
full three-dimensional solid mechanics problem to a two-dimensional problem. In Figure 2-4,
the bending moments are noted as Mx and My (associated with the rotation angles x and y);
the torsional moment as Mxy, and the transverse shear forces as Txz and Tyz.

Figure 2-4: Geometric definition of a rectangular plate with moments and forces (Lekhnitskii,
1968).

19
The expression of Hooke’s law for an orthotropic material in a two-dimensional problem is
written as (Lekhnitskii, 1968):

σ C C 0 ε
σ = C C 0 ε Eq. 2-6
σ 0 0 C 2ε
𝑬𝟏
⎧𝑪𝟏𝟏 =
⎪ 𝟏 − 𝝂𝟏𝟐 𝝂𝟐𝟏
⎪ 𝑬𝟐
𝑪𝟐𝟐 =
𝟏 − 𝝂𝟏𝟐 𝝂𝟐𝟏 Eq. 2-7
⎨ 𝑬𝟏 𝝂𝟐𝟏 𝑬𝟐 𝝂𝟏𝟐
⎪𝑪𝟏𝟐 = =
⎪ 𝟏 − 𝝂𝟏𝟐 𝝂𝟐𝟏 𝟏 − 𝝂𝟏𝟐 𝝂𝟐𝟏
⎩𝑪𝟑𝟑 = 𝑮𝟏𝟐
where Ei is the modulus of elasticity along the axis i, ij is the Poisson’s coefficient, and
Gij is the shear modulus (i, j = 1, 2). The equilibrium equations of an element of plate with no
external forces leads to the expression of the governing equation of motion (Brancheriau,
2021):

𝜕 𝜃 𝜕 𝜃 𝜕 𝜃 𝜕 𝜃
𝐼 𝐶 + (2𝐺 +𝐶 ) + +𝐶
𝜕𝑥 𝜕𝑥𝜕𝑦 𝜕𝑥 𝜕𝑦 𝜕𝑦
Eq. 2-8
𝜕 𝜃 𝜕 𝜃 𝜕 𝑤
− 𝜌𝐼 + + 𝜌ℎ =0
𝜕𝑥𝜕𝑡 𝜕𝑦𝜕𝑡 𝜕𝑡

With : density, 𝐼 = . Eq. 2-8 is valid for thin and thick plates.

2.4.2.1 Kirchoff-Love Thin Plate Theory

Classical plate theory for thin plates originated from beam theory. Bernoulli was the first to
devised this in the 1750s although it was Love and Kirchhoff that first applied it to plates and
shells. During this process, four major assumptions are made to reduce the problem down from
a three-dimensional representation of the systems elasticity to two-dimensions (Lekhnitskii,
1968) . These are (1) the mid-plane does not undergo in-plane deformations, (2) displacements
on the mid-surface are small compared to the thickness, (3) stresses in the thickness direction
are neglected to reduce the 3D problem down to a 2D one, and (4) transverse shear forces are
assumed as zero. The shear deformation is important to consider when the thickness ratio of
the panel ranges from 10 to 20 for the span of the plate bending curvature. For a two-
dimensional plate structure, thin plate theory assumes both that the shear effects in the thickness
and the rotational inertia of sections are neglected. With these assumptions, Eq. 2-8 is
simplified as (Brancheriau, 2021):

20
 𝜕 𝜃 𝜕 𝜃 𝜕 𝜃 𝜕 𝜃 𝜕 𝑤
𝐼 𝐶 + (2𝐺 +𝐶 ) + +𝐶 + 𝜌ℎ =0 Eq. 2-9
𝜕𝑥 𝜕𝑥𝜕𝑦 𝜕𝑥 𝜕𝑦 𝜕𝑦 𝜕𝑡
The displacement w along the axis z is linked with the angles according to the relationship as
shown below (valid for thin plates):

𝜕𝑤
⎧𝜃 =
𝜕𝑥
𝜕𝑤 Eq. 2-10
⎨𝜃 =
⎩ 𝜕𝑦
The equation of motion for an orthotropic material can be expressed as a function of the
displacement w by substituting Eq. 2-10 in Eq. 2-9 which is valid for thin plates (Brancheriau,
2021).

𝜕 𝑤 𝜕 𝑤 𝜕 𝑤 𝜕 𝑤
𝐼 𝐶 + 2(2𝐺 +𝐶 ) +𝐶 + 𝜌ℎ =0 Eq. 2-11
𝜕𝑥 𝜕𝑥 𝜕𝑦 𝜕𝑦 𝜕𝑡
2.4.2.2 Mindlin-Reissner Thick Plate Theory

Although Kirchhoff’s hypothesis provides a comparative solution for most cases, it is also
bound by a number of limitations related to the transverse shear deformation. One of these
issues arises when problems with the rotary inertia about the cross section come into question.
Independent rotation of the mid-surface is not possible which affects the boundaries of plates
considered as thick, where the undefined transverse shear stresses are important to consider. It
is known that as a plate-like structure deforms or deflects its transverse stiffness changes,
therefore only small changes to the stiffness characteristics can be assumed as constant.
Exceeding these small changes in deflection influence from cross sectional rotation contributes
to stiffness predictions.

The shear effects in the thickness are not neglected. Following the rationale and the proposed
correction by Timoshenko, the rotation angles are written as follows (K the Timoshenko’s
shear coefficient):

𝜕𝑤 𝑇
⎧𝜃 = −
𝜕𝑥 𝐾ℎ𝐺
Eq. 2-12
⎨𝜃 = 𝜕𝑤 − 𝑇
⎩ 𝜕𝑦 𝐾ℎ𝐺
This last equation must be used with Eq. 2-8 to describe the motion of an orthotropic thick
plate. Thick plate theory is generally recommended as it tends to be more accurate though
slightly stiffer, as it is susceptible to large aspect ratios. In general, the shear forces become

21
significant when the difference between the span and plate bending curve become large (Abell
et al., 2016). The Mindlin-Reissner theory is often called the first-order shear deformation
theory (FSDT) of plates. The Mindlin theory is applied for moderately thick plates with an
approximate L/h ratio of 10 (Szilard, 2004). The Mindlin theory is the most widely used
displacement-based theory (Szilard, 2004). It is based on FSDT using the kinematic
assumptions for the in-plane displacements. It considers all shear deformation forces as well as
decoupling rotary and lateral deformations. The detailed expression for strain and stress
resultant of the Mindlin approach can be found in Hanna (1991).

2.4.3 Evaluation of CLT

Of the parameters previously investigated when evaluating CLT panels, the modelling
techniques rarely vary (F. Bos et al., 2003; Steiger et al., 2010; Zhou et al., 2017). The reviewed
research considers the panel material as a solid orthotropic plate, consistent with the theorem
outlined in the previous section. Due to the construction strategies used for CLT, laminates are
bonded face-to-face only, therefore having possible discontinuations between boards edgewise.
Furthermore, although these theorems can account for the orthogonal properties of timber, it is
not known whether accuracy could be increased by considering a modelled panel comprised of
individual boards. Santoni et al. (2017) noted some reservations in applying Kirchhoff’s plate
theory to the evaluation of CLT as the panel is not a ‘perfect plate’. Santoni et al. (2017)
investigated the sensor positioning across a single thin 3-layer CLT panel (4.2 x 2.9 x 0.08 m)
with the intent in measuring variation as a result of phase angle; due to sensor position. While
the position was observed to have an effect of prediction of the elastic material properties, it
was noted that the variation could be due to the inherent discontinuities in the CLT.

Shahnewaz et al. (2015) evaluated CLT panels for their performance under in-plane loading.
By considering the individual boards, layering technique, and bonded interfaces of the panel,
a finite element analysis (FEA) of a three layer CLT panel was generated. Shahnewaz et al.
(2015) reported the results of the FEA and the experimental testing produced results ranging
from 15 to 25% in variation. Albostami et al. (2020) evaluated a series of CLT panel models
developed using ANSYS when subjected to a 4-point bending test. With known design
properties of the 5-layer CLT panels obtained from previously conducted experimental testing,
the percentage difference between the ANSYS models and CLT static experiments were
conducted (Albostami et al., 2020). Comparisons indicate an average difference between the
experimental and FEA results of 5 to 15% of Ex. This percentage difference in results is

22
consistent with the literature presented through section 2.2. As introduced in section 2.4 a NDE
system is comprised of both an experimental and theoretical component. The accuracy of the
theoretical is based on the realistic simulation compared to the real test specimen. Manufacture
questions were raised as to the versatility and representativeness of current published research
in regard to the modelled material geometry.

Chang et al. (2011) conducted a series of static tests on shear walls to determine the frictional
characteristics between the edge laminated boards. The frictional tests consisted of
approximately 40 biaxial experiments on shear walls with Teak (Tectona grandis)
reinforcement. Calculated values for frictional coefficients indicated a range of 0.2 and 0.5.
Based on the information presented throughout this section there is a defined gap between the
inherent effects of modelling CLT as a composite structure rather than a solid plate in regard
to vibrational performance.

2.5 Modal Theory

Modal analysis, also known as mode superposition, can model any structural vibration as the
summation of the individual contributions of each natural mode. The following section will
review the fundamental theorem and process that is undertaken to accurately determine the
modal shapes and natural frequency. The basics of modal analysis can be defined from the
modal expansion theorem, which makes the assertion that you can represent any type of motion
{𝑥(𝑡)} as the superposition of each contributing mode for a stationary wave. Therefore, the
motion can be defined as:

mẅ + ςẇ + kw = f(x) Eq. 2-13

where m is the mass matrix, 𝜍 is the damping matrix, 𝑘 is the stiffness matrix, 𝑤̈ refers
to the acceleration vector, 𝑤̇ represents the velocity vector, and 𝑤 is the displacement vector.
The equation of motion expressed above can then be further simplified into the following:

{w(t)} = {u } × q (t) Eq. 2-14

where {𝑢} represents the mode shapes ({𝑢 } is the first mode shape, {𝑢 } is the second
and so on), and 𝑞(𝑡) refers to the time and amplitude dependant behaviour (natural coordinates)
which define how much mode 1 contributes to the system as a wholes’ vibratory motion and
time dependence. Therefore, the proposition is that the total response of the system can be

23
represented as the superposition of each of the natural modes within the system. The number
of degrees of freedom (DoF) are equal to that same number of natural modes. For n number of
modes in a system the equation of motion can be defined as

{w(t)} = {u } × q (t) + {u } × q (t) + ⋯ + {u } × q (t) Eq. 2-15

This can then be compartmentalised into a matrix of modal values, and a vector of natural
coordinates:

q (t)
{w(t)} = [{u } … {u }] … Eq. 2-16
q (t)

Modal expansion theorem states that a vector of n generalised coordinates (the same set of
coordinates used to derive the equation of motion) can be summarised as:

x = u × q(t) Eq. 2-17

where 𝑥 is a vector for the equation of motion, u is a matrix of the mode shapes, and
𝑞(𝑡) is the modal and natural coordinates in the system. The best and easiest to comprehend
example of mode shapes, natural frequency and vibration interacting together is the principle
of a plucked guitar string. Guitar strings are, at their most basic level, long varying in thickness
shapes that are in a constant state of tension. When they are plucked, depending on the string
cross-section they will vibrate until they revert to their static position. This vibration continues
through the string and into the body of the guitar where the vibration is then converted into
sound in a similar process known as the reverberant method. This same principle can be applied
to almost all long, thin shapes in a constant state of tension. In a 2D problem, consider a string
tied tight at one end and a pulse applied at the other; once a certain amplitude or frequency is
reached, the following figure will be produced.

Figure 2-5: First mode example.

24
Figure 2-5 represents the first mode, a parabolic curve closely related to a skipping rope. To
get the second mode shape as shown in Figure 2-6 the frequency would need to increase by
double the first mode shapes’ frequency.

Figure 2-6: Second mode example.

Looking at both Figure 2-5 and Figure 2-6 it can be said that by doubling the frequency the
wavelength of the plot is halved. As the modes are functions of time, one complete cycle is
referred to as one period (𝑇) which is equal to one whole wavelength. Furthermore, Eq. 2-18
defines the period as:

1 Eq. 2-18
T=
f

Therefore, if 𝑓 is doubled then the period will be halved and so on. This would continue for n
amount of mode shapes to be evaluated, the equation for which can be summarised as:

nπx Eq. 2-19

{u} = sin
L

For a string fixed at both sides, here 𝑢 is the modal response, 𝑛 is the modal increment, and 𝐿
is the length of the shape. Mode shapes are made up of nodes and anti-nodes, the nodes are
where the plot crosses zero (i.e. the first mode has two nodes, third has three, etc.), and the anti-
nodes are points where the signal reaches maximum or minimum amplitude. The distance
between the first and last node can be expressed as a number of wavelengths of a sine function
(i.e. the first mode is half a sine functions wavelength with two nodes, second mode is a full
sine wave with three nodes, etc.).

2.5.1 Experimental Modal Analysis

Having this understanding of how the mode shapes are produced and behave under specific
conditions gives the ability to approximate them and their resonance by simply observing how

25
the material responds to vibration. Gu (2017) designed a CLT-Glulam composite floor system
and conducted a non-destructive vibration test to evaluate its flexural and shear properties. The
heel-drop excitation was used to simulate a human footstep, similar to the analysis system used
in ISO10140-3 (2010). Based on their findings, they concluded that CLT-Glulam composite
floor systems are governed by mechanical performance serviceability and not ultimate strength
(Gu, 2017). Therefore, it is recommended that in addition to designing for the ultimate
strengths, the application be taken into consideration with equal importance. Zhou et al. (2017)
completed a comparative study on the measurement of elastic properties of wood-based panels
using modal testing with the best performing BC. Zhou et al. (2017) analysed OSB and MDF
using both NDE and static tests. The differences between the dynamic and static results were
approximately 10, 20, and 15% variation for 𝐸 , 𝐸 and Gxy respectively.

Steiger et al. (2012) performed modal analysis using the sweep response spectrum of samples
with varying length and cut in the two fibre directions of a CLT plate. The experimental results
are then compared against the eigen-frequencies obtained from an analytical model of free
vibrations for thick, sandwich beams. For all samples that were identified through the process
Van Damme et al. (2017) found that the Ex, Ey and Gxy were optimized to match the theoretical
value measurements. Steiger et al. (2012) compared bending stiffness of CLT by modal
analysis of full-size panels. Panels 2.5 × 2.5 m at a range of thicknesses were evaluated against
static bending tests on 100 and 300 mm wide strip specimen. They found that with 100 mm
wide strip-based specimen cannot provide correct assessment due to a large variation in the
results, whereas single 300 mm wide strips lead to acceptable results. This is due to the lamella
used in manufacturing the tested CLT having core boards of 100 to 150 mm in width, causing
test specimens to consist of either a whole or two parts of a core board. These findings will be
taken into account for the static validation section as discussed in section 3.2.

Frédéric Bos et al. (2003) presented an NDE apparatus, called VibraPann especially developed
for wood-based panels. The system as seen in Figure 2-7 makes use of a single accelerometer
placed at the receiving point, a stimulation device (impact hammer), a number of nodal
supports, data collection hardware and processing software. A number of OSB, plywood and
MDF panels (2.4 × 1.2 m approximately) were analysed with all samples varying in thickness
from 10 to 22 mm. Each of the grid points shown in Figure 2-7 were recorded with the impact
location fixed to point (10,1) as indicated in the schematic. The impulse was measured using
an accelerometer which was placed at each of the points along the panel. This technique is

26
known as roving sensor as the excitation source is fixed and the sensor moves around the panel.
The time signals collected are then processed using a FRF; using the resulting magnitudes of
the imaginary part of each measured natural frequency, the mode shapes can be reconstructed.

Figure 2-7: ‘VibraPann System’ diagram of modal testing and modal analysis for full sized
wood-based panels.

VibraPann determines the modal parameters and conducts the mode shape simulation or
reconstruction simultaneously. The software returns the mode points, shapes, and natural
frequencies of the first nine vibratory modes for the panel tested. The results obtained by
Frédéric Bos et al. (2003) show the limitation of the VibraPann system being the size of the
panel. All experiments were conducted with 1.2 × 2.4 m panel sizes, where the common
dimensions for CLT are 2.8 × 3.2 m in surface area.

Figure 2-8: mode shapes of thin isotropic plate (F. Bos et al., 2003).

27
Figure 2-8 shows a variety of different mode shape responses from a thin plate simulation. (i,
j) represent the number of half-periods in each principal direction through the panel (i.e. 0,2
forms a cosine wave and 3,0 forms a sine wave both in direction of the major axis). The
numbering present in Figure 2-8 relates to the number of deflections that occur in the axial
directions of the plate. From an industrial point of view, the MoE and MoR along the bending
axis contains important properties that are commonly used as quality control measures. Based
on the findings of Frédéric Bos et al. (2003) as well as several other authors (Steiger et al.,
2010; Vacher et al., 2010; Zhou et al., 2020), it can be said that using modal analysis to
determine the mechanical properties of wood-based panels and reconstruct the experimental
signals returns a good correlation between dynamic and static measurements. Frédéric Bos et
al. (2003) found their results to have an R2 = 0.85 and R2 = 0.64 in both Ex and Ey respectively.
Different approaches have been attempted by altering specific testing parameters and BCs with
most conclusions coming to the same result of +10% when comparing static and dynamic
results. Although all indicators stipulate that these systems are effective and efficient, there is
no suitable system today for in-line evaluation of CLT or other mass timber panels. This is a
result of two main problems with attempting to implement vibration analysis being (i) the BCs,
and (ii) ensuring identification is consistent and systematic. The BCs are often not repeatable
or very difficult to do, when transitioning from a laboratory environment to an in-line
manufacturing application. An additional hurdle this method faces is not every panel is
identical due to the natural defects each piece of timber in the panel may contain. Therefore, is
it difficult to implement a systematic or automatic identification process for determining the
resonance frequencies.

To account for the identified assumptions in the above-mentioned literature a common

technique employed for experimental modal test verification is the modal assurance criterion
(MAC) analysis. MAC analysis investigates the agreement between the experimentally
generated mode shapes against theoretical mode shapes generated through FEA. Eq. 2-20 is
the formula used to obtain this agreement between the experimental and theoretical vectors for
each mode.

|{𝜑 } {𝜑 } |
MAC( , ) = Eq. 2-20
({𝜑 } {𝜑 } )({𝜑 } {𝜑 } )

where {𝜑 } and {𝜑 } are the experimental mode vectors for the 𝑥th and 𝑦th mode,
{𝜑 } and {𝜑 } refer to the theoretical modal vectors for the 𝑥th and 𝑦th mode. Vacher et

28
al. (2010) defines the MAC criterion as being the angle between two vectors which can be
applied to either real or complex value vectors. Figure 2-9 presents an example from Vacher et
al. (2010) of a MAC plot comparison between two modes. The MAC value obtained through
these experiments stipulates that a ‘1’ corresponds to a 100% match of compared modes;
alternatively, a poor match will be close to 0 or 0%. Based on the results of Vacher et al. (2010)
two modes are considered to be in agreement when the MAC value is greater than 0.75 (75%).
Furthermore, they are considered to be not in agreement or uncorrelated when the MAC value
is lower than 0.6 (60%).

Figure 2-9: Comparison of two models through the MAC criterion (Vacher et al., 2010).

2.6 Standardised Static Validation Methods

The mechanical properties of wood and wood composite materials are traditionally detected
through static deformation tests by using mechanical testing machines working on small-scale
specimens known as small-clears. The mechanical behaviour of a CLT panel is complex,
mainly due to the orthogonality in the grain direction of the successive layers and the inherent
variations in timber itself. The structural response of CLT panels has been thoroughly studied
for the case of loads applied both perpendicular to the plane and in-plane (Sikora et al., 2016).
Most of these testing specifications are based on four-point bending tests according to
EN16351 (2015). The mechanical properties that are of interest when conducting the bending
tests are the bending strength (MoR) and global stiffness properties (MoE). The board layering
method used introduces a high tensile strength along the x and y-axis of the panel; parallel and
perpendicular to the major axis of the face grain respectively. Various theories have been

29
proposed for the analysis of CLT panels under loads perpendicular to plane and the majority
focus on the calculation of the effective bending and stiffness based on the panel’s layer
characteristics (Sikora et al., 2016). Mechanically jointed beam theory (Gamma method) is a
common analytical approach used across Europe and is outlined in Appendix B of Eurocode 5
EN1995-1 (2004). It is most effectively implemented with the Euler-Bernoulli beam element.
Although no shear deformations are considered in this approach, it accounts for them indirectly
by calculating the effective bending stiffness based on the efficiency of the connection between
the longitudinal layers (the layers which grain direction is parallel to the length of the beam).
The testing method will be derived from EN16351 (2015) to ensure the static testing follows
closely to the practices used in the industry. The Gamma method makes allowances for the
shear deformation through a connection efficiency factor (Gagnon et al., 2011).

Various studies have investigated the efficacy of validating NDE approaches on timber and
timber composite products using static tests as described above. Ross et al. (1994) examined
transient vibrations through clear and composite timber products and found wave theory can
accurately observe the dynamic response of samples. They compiled results from a number of
researchers that compared values for MoE they obtained from both impulse vibration testing
and static tests (Ross et al., 1994). Measuring the mechanical properties of timber products
through vibration analysis has been proved a successful method (Moslemi, 1967; Ross et al.,
1994). The results have shown that the MoE tested through vibration has a good, consistent
relationship in regard to that tested through traditional static bending.

2.7 Summary of Reviewed Literature

The literature reviewed throughout this section of the thesis outline the areas of represented
research that have been investigated through this study. From the reviewed information it is
clear that attempts in this field have been made to establish an in-line NDE system for mass
panel QA. The results generated through these studies appears to achieve the intended purpose
of measuring mechanical properties of CLT with a perceived good agreement with validated
testing in the form of static analysis. Although there has yet to be any in-line established system
used by industry to date and to the knowledge of the research team. From the reviewed
literature several key variables of focus have been identified that were used as the focus or
research questions for this study. They are as follows:

 Boundary Conditions: From all literature reviewed, a conclusive statement from all
researchers has been the effect of the experimental setup is that of great importance.

30
 The way in which the panel responds to vibrations/ impulse will affect the observed
response measured from the attached sensors. This in turn effects the accuracy of
comparing the theoretical model developed. Zhou et al. (2017) evaluated a variety of
BCs for their resultant effects on accuracy when determining mechanical properties for
a variety of WBMPs. While it was concluded that all BCs evaluated could be
implemented within an NDE system, the clarity of the signals for each BC type
appeared to vary. As presented in Figure 1-3 (b) the magnitude and bandwidth of the
peak frequencies (resolution) can affect the detection of said peaks. This could be a
contributing factor to false peak detection specifically for an automated NDE system
approach. It was also observed through the reviewed studies that a freely supported
system has been implemented in various ways experimentally.
 Modelling Inaccuracies: Majority of reviewed studies implemented MAC analysis as a
validation factor for the experimental accuracy. This allows for additional similarities
to be drawn between the experimental and theoretical results. Although as noted
through these studies, theoretical evaluation is commonly based on plate theory where
the geometry of the modelled panel is a solid plate (Steiger et al., 2010; Zhou et al.,
2017). While considering composite wood panels (OSB, MDF, etc.) as an orthotropic
plate element is acceptable the third research question focuses on the validity of
considering a multi-layer material such as CLT as a solid plate element. Due to the
perpendicular layering and variations in the presence of edge bonding, it is hypothesised
that through considering additional variables higher accuracy correlation can be
achieved when comparing experimental and theoretical results.
 System Procedural Differences: From the reviewed studies presented throughout this
section a number of differences in regard to the procedural techniques used through the
experimental process have been highlighted. The number of frequencies detected and
its effect on the results has been noted in a number of studies such as (Zhou et al., 2017)
which appears dependant on the governing theory. Based on the required data for
predicting mechanical properties of thick plates the number of initial conditions
(frequency values) is hypothesised as required to be high to minimise the prediction
error. Although reviewed studies collect frequency values ranging from 5 to 15
frequencies. The more frequency values to be collected presents more areas for
introduced errors, notably for an automated system.

31
Based on the findings presented throughout this section, it is clear some gaps exist in the field
of NDE development for CLT evaluation. Based on the developed research questions and thesis
structure the acquisition system development, BC selection and modal evaluation for
alternative modelling types discussed above will be evaluated. From the outputs of these
developmental studies the system validation and results of which will be discussed attesting to
the accuracy for the tested materials. The following section discusses the purpose developed
acquisition system to be used throughout the experimental component of this study.

2.8 Research Questions

2.8.1 Development of a Non-Destructive Evaluation System for CLT

Can an NDE system be developed for mass timber panels (CLT) that can accurately
approximate their mechanical properties and with what degree of accuracy for an in-line
application can this be achieved?

2.8.2 Assessment of Appropriate Boundary Conditions

What boundary conditions (BC) result in the most accurate and achievable system for
determining the stiffness properties of mass timber panels?

2.8.3 Modelling Approaches to Identify Potential Inaccuracies

Can mode shapes be accurately generated and compared against FEA for a series of alternative
modelling techniques?

2.9 Research Objectives

From the reviewed literature presented and explored through the following section, a noticeable
gap has been identified for the focus of developed NDE systems on their intended application
and implementation with industry, and more specifically where in the manufacturing supply
chain would such a system be situated. Based on the research questions the following objectives
have been established.

2.9.1 NDE system Development

Previous studies have investigated the use of transient vibration measurements in NDE
approaches with some success although no such system has been established for the in-line use

32
on CLT. The specific aim of this NDE system would be through the QA process of CLT
manufacture, providing manufacturers with a robust and rapid NDE system.

2.9.2 BC Evaluation
One of the primary influencers on vibration analysis is the way in which the test specimen is
supported. The selection of an appropriate boundary condition support for testing large scale
CLT (weight ranging from 300 to 500 kg) must take into account the practical aspects of the
experiments. Through this methodology establishment, a series of boundary conditions (BC)
noted as successful for laboratory scale panel testing by other researchers were evaluated for
their accuracy as well as application. The selected BC needs to be suitable for use by industry.

2.9.3 Modelling of CLT

As noted above, previous research focuses on the assumption that CLT reacts to vibrations as
a solid plate element, as is how they are modelled using numerous FEA approaches and plate
theory equation of motions. Although CLT manufacture can vary between the inclusion of or
lack thereof edge bonding between laminates, as the materials secured for evaluation through
this study were not edge bonded, modelling hurdles specific to non-edge bonded CLT was
considered. This section of the work will look at comparing a classical plate model of a CLT
panel versus a FEA model of a CLT panel consisting of multiple layers and board
arrangements. These theoretical results will be compared against the experimental modal
analysis, MAC evaluation will be used to compare the frequency as well as the mode shape
similarities.

2.10 Thesis Outline

This thesis is comprised of 7 sections. Section 1 contains the introduction and outlines the aims
of the study and focus of the research. Section 2 presents the literature review of the study
which encapsulates the evidence leading to the identified gap in the available research as well
as the methodologies adopted through the study for experimental modal testing. Section 3
details the development of the acquisition system used to collect and post-process specific
vibratory information across the experimental components of the research. The 4th section
presents the developed testing methods and a review of suitable boundary conditions for
incorporating within the experimental method. The 5th section investigates the theoretical
development component of the method and the modelling scenarios to be evaluated for
potential inaccuracies in common CLT evaluation techniques. Section 6 presents the

33
comparison between the obtained NDE properties for a series of CLT panels and thicknesses
against the static test data. Finally section 7 of this thesis details the conclusions of the
presented research and discusses opportunities for further investigation. The flow chart shown
in Figure 2-10 presents the stages extracted from the literature reviewed for the development
of the NDE system and parameters that will be developed either alongside, prior to or post
experimental work. The highlighted areas of the flow chart were conducted in collaboration
with both DAF and CIRAD.

Figure 2-10: Flow chart of systematic approach to NDE system development.

34
3 Acquisition System and Testing Method
3.1 Software Development

A large component of the project is the development of an NDE system for the assessment of
CLT panels for an in-line application. As an innovative approach in the field of WBMPs
assessment and to allow for easy operation and control the digital signal processing (DSP)
techniques used on analysing the acquired signals the research team adapted a software
platform LabVIEW (ver. 2017). The classical dynamic NDE system discussed in section 2.2 is
comprised of hardware and software. The hardware consists of sensors for measuring vibration,
impacting device, and acquisition loggers. This section will discuss the software component of
the NDE system to be used throughout this study. The signal collection is dependent on a
number of factors:

 Sensor attachment: To ensure signals are measured accurately it is important the

sensors are attached appropriately and accurately. To do so the sensor attachment has
been evaluated between a magnetic fixture and a tack substance to attach the sensors.
From these two sensors types it was anticipated that the tack substance approach would
provide a more absorptive connection between the sensor and the material surface. This
sensor type also supports a more rapid attachment approach.
 Signal settings: Based on the impulse signal type the nature of the signal is short in
duration from point of impulse to the decay of the majority of the signal. Figure 3-1
presents an example time domain signal showing the majority of the signal consists of
a 1 to 2 second window of activity. While this activity window is short, the settings
presented in Figure 3-1 will ensure a sufficient resolution in frequency to observe the
resonance accurately.

35
 Figure 3-1: Signal duration example.

 Post processing techniques: Through the use of the LabVIEW system the signal can be
acquired with no external processing required. Once digitised the post processing
techniques and their internal settings available allow for the rapid modification of the
signal.

Once the signal is acquired through the data acquisition (DAQ), filtering and conditioning is
conducted to remove unwanted/ unrequired portions of the signal. The filter type is set to a low
pass filter (LPF), which as shown in Figure 3-2 fits a tapered curve over the frequency spectrum
to remove high-frequency noise and avoid signal aliasing which can influence the frequency
spectrum.

36
 Figure 3-2: LPF example.

For acquiring the signal it is important to ensure the sampling rates and filter settings are
appropriately tuned to ensure all observable modes are able to be obtained. Through initial
theoretical modelling of a two-dimensional plate structure (3.0 × 2.8 m), the first modal
frequency was observed between 8 to 34 Hz; depending on thickness and material density.
Settings were chosen based on achieving a sampling interval and frequency resolution suitable
for obtaining these low frequencies. Table 3-1 shows the software settings used for the large-
scale testing trials.

Table 3-1: Signal acquisition and DSP settings required for large scale trials.

Signal Processing Parameter Setting

No. Samples 51,200
Sampling Frequency (Hz) 8,000
Frequency Resolution (Hz) 0.156
Signal Durations (s) 6.4
Sampling interval 125 µs
Cut-off Frequency (LPF setting) 500 Hz

As the frequency resolution is the sampling frequency divided by the number of samples, a
small frequency interval is obtained through having a low sampling frequency and a large
number of samples. The settings in Table 3-1 were selected to provide a small frequency
interval as desired to ensure low modal frequencies were obtained. The cut-off frequency was
set high for the preliminary assessments to accurately observe the vibrational response of the

37
panel and any effect from the BCs. The cut-off frequency, sampling frequency, and number of
samples measured are all variable within the software. Figure 3-3 displays the flow chart for
which the data passes through once collected. The flow chart in Figure 3-3 visually depicts the
stages of processing as well as the individual code block settings used to achieve them.

Figure 3-3: Flow chart of signal processing and test diagram.

3.2 Sensor Calibration

Prior to the implementation of the accelerometer as a vibration measurement device, angular
calibration testing was performed using a standard pendulum device, as shown in Figure 3-4.
Using the pendulum with the accelerometer positioned along each axial direction x, y, and z;
also referred to as radial, tangential and zero, oscillations were recorded. Testing was
conducted using the settings displayed in Table 3-1. Tests were conducted by positioning the
pendulum at 90° and allowing the arm to swing at gravity with a pendulum mass of 40g. Test
durations were long enough to acquire full amplitude oscillation and signal damping. Based on
the intended use orientation (surface mounted), the results of the radial direction are considered
most representative. The results of the radial direction tests are presented in Figure 3-5.

38
 Figure 3-4: Experimental setup for pendulum test of accelerometer.

Figure 3-5: Pendulum test results for (top) the phase angle of the pendulum in radial axis, (b)
velocity of pendulum, and (c) the measured acceleration (orange) and predicted response
(blue).

The results of Figure 3-5 (a) shows the correct measured phase angle of the accelerometer for
the angular displacement for a pendulum starting angle of 90°. The measured acceleration
indicates the maximum amplitude measured was approximately 9.8 m/s2 with a decaying
amplitude over the measuring period.

39
3.3 Analysis Method

Prior to conducting large scale tests on the target product (CLT panels), it was necessary for
the acquisition system and experimental methodology to be trialled through using a scaled test
sample. Through this stage a Cast3M (ver. 19.1, 2019) model of a thin orthotropic plate was
generated and the mode shapes observed for the first four mode shapes as shown in Figure 3-6.

Figure 3-6: Four mode shapes showing the bending and torsional modes.

Based on the mode shapes in Figure 3-6 points of null deflection were observed at
approximately ¼ lengths from the edges of the simulated panel. The support points were
positioned so that no mode shapes were dampened or restricted from deflection. Based on these
results, the setup shown in Figure 3-7 contains the support points. A plywood panel was
selected and suspended at four points using elastic supports and impacted at diagonally
opposite ends as depicted in Figure 3-7.

Figure 3-7: Free-Free boundary condition supports; recommended impact and sensor point.

40
As defined by F. Bos et al. (2003) and using the data extracted from the tests described in
Figure 3-3, the Ex and Ey values are calculated as shown in Eq. 3-1 and Eq. 3-2.

12(1 − 𝜈 ) 𝜌 𝐿
𝐸 = 2𝜋𝑓 Eq. 3-1
501 ℎ
12(1 − 𝜈 ) 𝜌 𝐿
𝐸 = 2𝜋𝑓 Eq. 3-2
501 ℎ
where 𝑓 is the bending modes for the modal number i, 𝜈 is the Poisson’s ratio, 𝜌 is
the material density, ℎ is the thickness measurement, and 𝐿 and 𝐿 are the length dimensions
for the major and minor axial directions respectively. The support locations were identified
through preliminary modelling of a plate structure responding to vibration by identifying these
nodal points for the first mode shapes in bending. Nodes are points on the panel, which
experience zero deflection during vibratory response. The plywood panel described earlier is
of 1.2 × 0.8 × 0.01 m and tested as shown in Figure 3-6. The plywood panel was exposed to
the same testing methodology developed for large scale panels with the panel being first tested
as a whole, cut to sections for evaluating Ex and Ey, tested using NDE beam analysis, and finally
mechanically tested in accordance with EN789 (2004). Through experimental identification of
the first four modes and comparisons with the theoretical resonance frequencies the results
displayed in Table 3-2 were calculated.

Figure 3-8: Experimental test setup for preliminary samples

The overall error between the static and dynamic tests is 11.3% and 0.67% in Ex and Ey
respectively. Although the variation for the parallel test specimens is high this is considered to
be attributed to the low sampling numbers as these tests were based on a single panel test. Static
testing was conducted on sections of the plywood panel depicted in Figure 3-6 at a length of
0.48 × 0.10 × 0.01 m in both the parallel and perpendicular directions to the face grain of the
plywood. Two test pieces for each direction were taken from the tested plywood panel. The

41
test setup as shown in Figure 3-9 presents the 4-point testing setup where deflection was
measured at the mid-point of the test specimen. The bending modulus of elasticity (MoE) was
determined using Eq. 3-3:

(𝐹 − 𝐹 )𝑙 𝑙
𝐸= Eq. 3-3
16(𝑢 − 𝑢 )𝐼
( )
where ( )
is the load versus displacement curve obtained from the test along the

linear portion of the slope, 𝐼 is the second moment of inertia ( ), and 𝑙 and 𝑙 refer to the

upper bending span length (mm) and the distance from an upper span point to the outer span
(mm) respectively.

Figure 3-9: Static bending test used for evaluating plywood panel EN789 (2004).

Table 3-2: Results of plywood NDE predicted mechanical properties (on full sheet), and the
static values.

Sample Static MoE NDE MoE (%)

Direction (MPa) (MPa) difference
Ex 14,817 (5.1) 13,137 11.3
Ey 4,463 (7.2) 4,493 0.7

The results presented in Table 3-2 present the percentage difference between the initial NDE
testing of a single plywood sheet against static values obtained through testing as shown in
Figure 3-9. The results reported for the static values are the average of 2 tests per axial
direction. The difference between the two values averaged is presented in parenthesis for each
static property. While this data set is not sufficient to draw a defined conclusion the results of
Section 6 provide a larger population series for validation of the NDE method.

42
 4 Experimental Modal Analysis of Appropriate Boundary
Conditions for the Evaluation of Cross-Laminated Timber
Panels for an In-Line Approach

STATEMENT OF CONTRIBUTION TO CO-AUTHORED PUBLISHED PAPER

This section includes a co-authored paper. The bibliography details of the co-authored paper,
including all authors, are:

A. Faircloth, L. Brancheriau, H. Karampour, S. So, H. Bailleres, and C. Kumar, (2021).

‘Experimental modal analysis of appropriate boundary conditions for the evaluation of cross-
laminated timber panels for an in-line approach’. Forest Product Journal 71(2), pp. 161 – 170.
doi:10.13073/FPJ-D-20-00062.

My contribution to this paper involved: literature review, experimental method development,

experimental planning, setup and testing, analysis and discussion of results as well as writing
and editing the manuscript.

43
 Experimental Modal Analysis of Appropriate Boundary
Conditions for the Evaluation of Cross-Laminated Timber
Panels for an In-Line Approach
4.1 Abstract

Transverse modal analysis of timber panels is a proven effective alternative method for
approximating a materials’ elastic constants. Specific testing configurations such as the
boundary conditions (BC) and location of sensor and impact play a critical role in the accuracy
of the results obtained from the experimental assessment. This section investigated the signal
specific details such as the signal quality factor, which directly relate to the damping properties
and internal friction, as well as frequency shifting obtained from six different BCs. A freely
supported (FFFF), opposing minor sides (shorter length) simply supported and major sides
(longest length) free (SFSF) as well as the reverse of the SFSF configuration with minor sides
free and major lengths simply supported (FSFS), and all sides simply supported (SSSS) setup
were investigated. Variations into the proposed methods used to achieve a FFFF supported
system were also to be considered. A combination of experimental testing in parallel to finite
element analysis was conducted to recreate the setup that would be used within a manufacturing
facility for non-destructive assessment of large-scale cross-laminated timber (CLT) panels. The
differences between all BC configurations for their resonance frequency quality and location
indicate that a freely supported system provides higher resolution results, good comparison of
less than 10% error with the FEA and experimental results as well as advantages in a simple
experimental setup for the intended application.

4.2 Introduction

As mass timber construction becomes a more widely adopted and accepted building method
for large scale residential and commercial constructions (McGavin et al., 2020), the
requirements for consistent and graded cross-laminated timber (CLT) panels from
manufacturers is increasing, and thus so shall a method by which these panels can be rapidly
evaluated (Steiger et al., 2010). Non-destructive evaluation (NDE) techniques are used as quick
and effective methods for approximating the elastic properties of timber beams and boards with
work towards a system for timber panel elements encroaching on a solution for in-line
applications (Guan et al., 2017). An ideal solution to evaluate these panels would be a non-

44
destructive grading system for CLT panels as they are produced from the manufacturing
facility, therefore a system designed around the application (Zhou et al., 2020). With the
increased size and weight of these mass timber panels, the boundary conditions (BCs) used for
NDE testing of these panels plays a crucial role in their evaluation and need to be considered
in large scale applications.

The BCs directly affect the ability to acquire and observe the harmonic responses required to
calculate mechanical properties (Damme, Schoenwald, et al., 2017). The dynamic response of
these timber panels can contain reactions other than bending modes such as torsion and rigid
body motion; based on the measurement method employed these bending modes will occur at
smaller or greater amplitudes affecting the identification process (Zhou et al., 2017). Although
completely free (FFFF) BCs are simple physically to setup, there exists no analytical solution
to the BCs (Zhou et al., 2017). In contrast, a completely simply supported (SSSS) setup can be
difficult to implement due to the constant contact requirements of the support conditions
although it has an exact analytical frequency equation.

Zhou et al. (2017) undertook a comparative study on four BCs proposed for the non-destructive
assessment of wood-based panels within an in-line application and to compare the elastic
constant variations across the proposed BCs. Modal testing was conducted on panels 1.2 x 0.6
m with varying thickness for the four BCs, all sides free (FFFF), one side simply supported
and the other three free (SFFF), one side clamped and the other three free (CFFF), and opposite
parallel sides along minor strength direction (panel widths) simply supported and the remaining
sides free (SFSF). Although during experimental testing it was quickly realised that the support
conditions required for a SFFF and CFFF system are not applicable to in-line use due to the
physical difficulty in setting up the supports. It was concluded that all BCs are applicable for
laboratory scale experiments with the FFFF and SFSF configurations returning a 10 %
difference between static and NDE techniques used. Zhou et al. (2017) also noted that the non-
destructive assessment of these panels was sensitive to the supporting conditions and setup of
the experimental tests. Following on from this, (Zhou et al., 2020) conducted a body of research
aiming to quantify the bending and shear properties of CLT through modal analysis under a
SFSF BC. Comparing experimental and theoretical results for the stiffness constants gave a 1.6
% and 0.2 % error in major and minor axial directions respectively, although a discrepancy was
found between the experimental and theoretical shear moduli for the minor axial direction. This
discrepancy is not unexpected due to the widths of the panel (minor dimension) being the
simply supported dimension.

45
In a branch of the previous discussed study, Neiderwestberg et al. (2014), conducted modal
analysis on single layer CLT panels for BCs with the span parallel to the face grain direction,
and with the span perpendicular to the grain direction (SFSF and FSFS), as well as a FFFF and
SFFF systems. It was found that conducting these experiments was difficult to setup as the
simply supported conditions require constant contact being made between the supports and
specimen surface. Neiderwestberg et al. (2014) compared the results from NDE testing of
single layer glue-laminated timber panel with results from static tests for determining the elastic
properties of the axial directions parallel (Ex) and perpendicular (Ey) to the face grain. The
FFFF BC retuned errors of 1.0 % and 18.6 % difference for Ex and Ey respectively. The SFFF
system returned an acceptable error for Ex of -8.0 % although the error for Ey was 40 %. The
SFSF condition produced the smallest average error of -3.5 % for Ex and -3.7 % for Ey. These
results reinforce the argument that the BCs have a direct effect of the accuracy in measuring
key mechanical calculation descriptors such as fundamental frequencies.

A comparative study conducted by Damme and Zemp (2017) found from the evaluation of
bending wave dispersion curves in CLT that the BCs dictated the acquisition of the resonant
frequencies of tested beams. The result of the studies discussed above conclude that the
conditions applied to restrict, suspend or support the test specimen have a direct impact on the
measuring accuracy of the experimental approach and therefore displaying the importance of
an investigation into such methods. Neiderwestberg et al. (2014) and Zhou et al. (2017) did
investigate the effect BCs have on the comparison between static testing methods and the
calculated elastic parameters although validation through this method requires a large number
of repeated experiments to be considered accurate. Concerning plate-like structures the
vibrational modes need to be considered in two-dimensions and are a product of the samples
density (kg/m3) and the fundamental frequencies, therefore concerning changes in density and/
or geometry are expected to have an effect on the materials behaviour to vibrational excitation
as well as fundamental frequency locations (Zhang et al., 2021).

The quality factor (Q factor) is a representation of the resonant frequency peak resolution; non-
destructive assessment of timber elements relates to specific material properties and natural
frequencies of the measured samples, therefore it is important to ensure the frequencies used
for this assessment are clearly defined in the frequency spectrum (Shirmohammadi et al.,
2020). These descriptors are commonly used for the evaluation of acoustics and sound
properties of materials where the BCs remain consistent and material properties are evaluated
(Labonnote et al., 2013). This study aimed to fix the material (CLT) and evaluate a number of

46
BCs; rather than characterising the material through the NDE method applied with the BCs,
the supporting conditions will be evaluated to determine their suitability. From the above, it is
the conclusion of the research team that there is a noticeable gap in relating the BCs to the
intended application of an in-line evaluation system for CLT panels as well as assessing the
effects each BC has on accurately measuring the vibratory response of these panels. Therefore,
the objective of this study was to investigate a number of BCs as selected based on previously
conducted and relevant research and determine a best suited BC for application to an in-line
NDE system for assessing the elastic constants of CLT panels.

4.3 Materials and Methods

4.3.1 Test Specimens

The samples used for assessment were CLT panels comprised of three layers, with each
proceeding layer perpendicular to the last. The grade quality of the panels could be considered
as non-visual with a target grade of 8GPa (10GPa – face/ 6GPa – core/ 10 GPa – face) as
detailed in the XLam Australia technical design guide. The panels were 3.2 x 2.8 x 0.075 m in
size made up of Radiata Pine (pinus radiata) with an average density of 475 kg/m3 measured
from the test specimens. As introduced with (Zhang et al., 2021) the density and corresponding
geometrical properties will affect the frequency locations, therefore to limit the number of
variables within the study, a single species type, and grade of CLT was selected; the
geometrical properties, density, and estimated elastic constants were kept consistent for the
FEA. Woods major stiffness properties are related to the axial direction parallel to the grain
(Steiger et al., 2010), therefore the major axial direction is recorded as the direction parallel to
the face board layer orientation (Lx); the minor axial direction is considered as perpendicular
to the face board orientation (Ly). CLT is comprised of laminated boards bonded to each
surface, although the edges between adjacent boards are not bonded for the panel evaluated in
this section. One assumption made in this study is that the discontinuities between adjacent
boards are negligible to the stress wave propagation and the CLT panels can be modelled as a
single plate with elastic properties Ex = 9.1 GPa, Ey = 1.8 GPa, and Gxy = 0.8 GPa (Oregon
State University OSLU freeware).

47
4.3.2 Experimental Protocol

The BCs selected for the study were achieved by using a range of readily available products
and materials that would be applicable to an in-line manufacturing scenario as is the intended
purpose of the system. The SSSS system was achieved by using a simple timber border of a 50
mm width around the perimeter of the panel (Figure 4-1 (a)); this was the most practical and
achievable arrangement to ensure a sturdy setup and minimal contact with the panel. Due to
the simplicity in the setup requirements, this BC may be attractive to industry; it is important
to ensure that all edges are making contact with the boards or the BCs will not be accurately
simulated. The SFSF and FSFS BCs were constructed using raised supports (Figure 4-1 (b))
placed as close to the edges as practical where support was required. This configuration was
also simple in its setup with workers able to lift the panels into place with ease; it was found
equally important with this setup to ensure that the SS edges were in constant contact with the
supports so not to introduce errors. These configurations are presented in Figure 4-1. To
investigate the most effective and practical setup, three configurations of a freely supported
system were selected and are also depicted in Figure 4-1. The three configurations consist of
airbag supports (FFFF-1) underneath the panel at four points (Figure 4-1 (c)), nylon recovery
straps to suspend the panel flatwise (FFFF-2) at four points (Figure 4-1 (d)), and the nylon
straps attached to the edge of the panel to suspend it edgewise (FFFF-3) at two points (Figure
4-1 (e)). Each of the six BCs were investigated using the setups shown in Figure 4-3 and Figure
4-4 with findings discussed through this section. The adopted configurations of a SFSF BC
consist of two parallel sides of the panel supported. The placement of the supports and width
of the contact area have been arranged so that Figure 4-1a the simply supported (SS) sides are
supported as close to the edge as practical, and Figure 4-1b the contacting area is minimal; two
scenarios are investigated where the major axial lengths are SS (3.2 m lengths) and where the
minor lengths are SS (2.8 m lengths).

48
Figure 4-1: (top) Experimental setup of a) SSSS, b) SFSF (shown) and FSFS BCs, c) Airbags
(FFFF), d) Suspended on flat (FFFF), e) Suspended on edge (FFFF), (bottom) diagram of the
BCs shown in a) through e).

49
4.3.3 Vibration Assessment

A typical modal testing setup was employed for assessment of the BC configurations to
determine the vibrational characteristics of the testing setup. As shown in Figure 4-2, the panel
is excited by an impacting hammer (IEPE Brüel & Kjær type 8206 impact hammer) on the
surface of the panel; this impulse is then measured through a surface mounted accelerometer
(BCP Piezoelectronics model 352C33 single axis accelerometer) affixed to the panel. The
impulse propagates through the panel as indicated by the curved lines in the figure, where the
response is collected by the accelerometer.

Figure 4-2: Impulse location and stress wave propagation through a CLT panel.

The raw signal obtained from the impact hammer and accelerometer is then processed through
the LabVIEW based programming environment with a sound and vibration card (National
Instruments Sound and Vibration Module NI 9234 24-bit ADC) where the signal is sampled at
a rate of 8 kHz, for a total 51200 samples acquired, giving a sampling resolution of 0.156 Hz,
for an acquisition duration of 6.4 seconds (National Instruments, 2018). From the initial finite
element analysis (FEA) presented in Section 4.3.4, the first natural frequency (f1) ranges from
4 to 10 Hz. Therefore, to ensure the natural frequencies are accurately observed, a sampling
interval of 125µs was targeted. The deconvolution equation is defined below by determining
the result of 𝑦(𝑛) of the following convolution formula (Eq. 4-1):

𝑦(𝑛) = ℎ(𝑘) ∗ 𝑥(𝑛 − 𝑘) Eq. 4-1

50
Where 𝑥(𝑘) is the input (impact hammer) signal, ℎ(𝑘) is the output (accelerometer) signal, and
𝑦(𝑛) is the convolved response. Assessment of the signal quality is conducted on the FRF
output; the response was generated with no additional post processing techniques such as
windowing to assess raw data response. The formulas present in Eq. 4-2 and Eq. 4-3 is the
process of assessment used for determining the resonant peak resolution.

𝑓
𝑄= , Eq. 4-2
𝑓 −𝑓

𝐴
location of 𝑓 , = ± Eq. 4-3
√2

Where 𝑄 is the quality factor, determined by the natural frequency 𝑓 , divided by the difference
between the upper and lower peak frequency boundaries 𝑓 and 𝑓 respectively. 𝐴 refers to
the amplitude of the natural frequency 𝑓 (SproBmann et al., 2017). Prior to experimental
testing of the selected BCs, FEA modelling was conducted to model modal responses of a three
dimensional plate to determine locations on the panel that experience minimum and maximum
deflections to ensure the support locations for FFFF conditions will not obstruct the modal
response. Bos et al. (2003) began with a similar approach by modelling the response of a panel
under free supporting conditions to determine appropriate locations for the nodal supports. The
areas that experience minimal deflection, referred to as the nodal points are to be the support
locations and areas of maximum deflection, anti-nodes, as the impact and sensor locations; by
observing up to four mode shapes as seen in Figure 4-3, the best suited locations for supports
were determined. This modelling was undertaken for all BCs to determine appropriate
impacting and sensing locations as well as support points for the FFFF configuration. Figure
4-3 shows the first several modes obtained through theoretical modelling of a thin, orthotropic
plate with dimensions, density and BCs the same as those of the CLT panels assessed through
this study. It can be seen in the four BC setups areas of maximum deflection and minimal
movement.

4.3.4 Finite Element Analysis

Modelling conducted to obtain the following modal responses was done through the software
platform Cast3M (CEA, 2020). A three dimensional rectangular plate simulation was used to
simulate the CLT panels where the length (Lx) is the first dimension, width (Ly) is the second
dimension, and the thickness is the third dimension. Poisson’s ratio material properties has

51
been set to a value of 0.3, as regularly used for the theoretical analysis of timber (Santoni et al.,
2017).

Figure 4-3: First four mode shapes for the 4 boundary conditions (a) a freely supported
system (FFFF), (b) simply supported widths and free lengths (SFSF), (c) free widths and
simply supported lengths (FSFS), and (d) all sides simply supported (SSSS).

The mode shapes above show that for (a) ¼ distances from the corners of the panel appear to
experience minimal movement, more prominent in modes 2 and 4. Similarly, the corners of the
panel experience maximum deflection; expected with the FFFF BCs. The modes in Figure 4-3
(b), (c) and (d) show the panels response with either two opposing, or all sides SS. Looking at
the responses from these three configurations it can be seen that at ¼ distances from the corner
of the panel deflections can be measured for the majority of modes; contradictory to the
observed response for Figure 4-3 (a). The results of the FEA modelling leads to the
experimental test setup for each BC is as depicted in Figure 4-4, containing support, impact
and sensor locations.

52
 Figure 4-4: BC supports and recommended impact locations (a) – FFFF, (b) – SFSF, (c) –
FSFS, (d) - SSSS.

4.4 Results and Discussion

The tests consisted of six BC variations and the accuracy of the signal in obtaining the
maximum number of modes. Theoretical simulations of thin orthotropic plates were conducted
as displayed in Figure 4-3 using Young’s moduli estimations of 10 GPa for the major direction
(Ex), 1.5 GPa for the minor direction (Ey) determined from the design properties of the CLT
panels and the assumption that Young’s modulus along the grain direction is > 10 times that
measured perpendicular to the grain direction (Damme & Zemp, 2017). Shear moduli
estimations of 0.8 GPa were used based on the design properties of the CLT. From these results
for the selected BCs, it was found that the first 12 mode shapes fit within the range of 10 Hz to
200 Hz frequency range with frequencies lower than 10 Hz producing the rigid body motion
(RBM) of the system. Zhou et al. (2017) found that to accurately approximate the elastic
constants of timber composite panels it is ideal to acquire 10 to 15 modes, therefore based on
these findings and the results of the theoretical component, a minimum of 12 modes were
targeted for all BC investigations. Figure 4-5 shows the overlayed responses from the six BCs.
Figure 4-5 (top plot) shows the variation within the FFFF BCs, although the frequencies appear
within a consistent range of each other, a shift in the frequencies for all modes is observed;
these results are from tests conducted with a single panel to minimise introduced variations. It
can also be found that the BCs affect the amplitude of the FRF, which can impact the resonance

53
detection and the signal quality. The bottom plot contains the SFSF, FSFS, and SSSS BC
responses; both SFSF and FSFS responses appear to contain resonant frequencies within the
range of each other with some minor shifting present. Differences between SFSF and FSFS are
expected due to the change in configuration as the panel dimensions of length and width do not
differ greatly between one another. The FRF in Figure 4-5 (bottom plot) for SSSS is clearer
and more resolvable, allowing for confident peak detection.

Figure 4-5: (top) – FFFF variation comparison, (bottom) – SFSF, FSFS, and SSSS
comparison.

Figure 4-5 contains torsional modes and bending modes along both major and minor directions;
for the modal determination process proposed in this study, bending modes in both major and
minor directions will be more pronounced although all are viewed in the spectrum shown in
Figure 4-5. The location of resonant frequencies can be obstructed by the overlap of some
modes, noted as double peaks in (Zhou et al., 2017); this is due to the proximity of the
resonance modes, coupled with a high damping ratio. This occurrence is validated through
successful comparison of the experimental data and detected resonant peaks with the
theoretical simulated modal frequencies presented in Figure 4-6 showing the comparative error.
Comparing the identified resonant frequencies with the corresponding mode shape frequencies
acquired from the theoretical analysis will determine the effect the frequency shift has on
identifying the correct modes as well as determining the most appropriate BC that matches
with the theoretical response. Figure 4-6 represents the error percentage comparison for each
of the six BCs against their respective simulated response for a panel comprised of the same
physical properties (dimensions, density and BCs). The results presented in Figure 4-6 show

54
that as the frequency increases, so does the error for the three FFFF BCs with the FFFF-1
configuration being able to measure up to the 10th mode with a maximum variation in results
of +5%.

Figure 4-6: Comparative results between simulation and experimental resonance frequencies
for the 6 BCs (top) modes 1 to 6, (bottom) modes 7 to 12.

While all FFFF conditions produced similar comparative results, the error varies with each
configuration with FFFF-1 producing the lowest error percentage, followed by FFFF-3, and
FFFF-2 with the highest error of the three. The SFSF BC has a considerable error for lower
modes most-likely due to possible discrepancies in the BC setup. SFSF’s third mode was
recorded although as it was difficult to detect due to low amplitude; this in combination with
the error between frequencies could indicate an error in the physical setup. Observing Figure
4-5 (bottom plot) of the experimental response from the FRF below 50 Hz showing the low
amplitude mode. From the modal responses displayed in Figure 4-3, mode 3 is a torsional mode
which could be another contributor to the low amplitude measured. The FSFS BC returns a
similar error percentage to SFSF from modes 1 to 4 and as shown in Figure 4-5 (bottom plot)
has a low amplitude making it difficult to detect. It can also be noted from FSFS that resonant
frequencies above 100Hz start to decay in resolution and amplitude. The SSSS BC shows high

55
error from modes 1 to 3 although it is able to accurately identify modal frequencies with errors
below +10% for modes 4 to 12.

Figure 4-7: Quality factor vs mode number for BCs.

Comparing these results with signal Q factors provides a better understanding of the effects
each of the nominated BCs are having on the experimental evaluation. Using Eq. 4-2 and Eq.
4-3, the Q factors were determined and are presented in Figure 4-7, displaying the six BCs and
their mean and range to the calculated Q factors. These values show that a high quality factor
will correspond to a clearly defined peak due to the calculation method, therefore BCs that
obtain consistent and high Q factors are ideal for peak detection. This assumption is proven
correct in Figure 4-7 with the FFFF displaying higher Q factors than SFSF, FSFS and SSSS as
expected. The Q factor plot indicates that for the three FFFF configurations, the FFFF-2
produces the highest average Q factor although the average difference between FFFF-2 and
FFFF-1 is small with FFFF-1 having a smaller variation. This can be further reinforced by
comparing the quality of the signals with the comparative data displayed in Figure 4-6. SFSF,
FSFS, and SSSS have Q factors in the similar range as each other with FSFS having the lowest
Q factor average. SSSS has a similar average although a larger range of values; SFSF returned
a higher average although a large variation in results. From these results, it can be concluded
that the three FFFF BCs provide clear results and the setup developed through this study is
repeatable; established from the consistent Q factors. Of the 3 FFFF configurations, FFFF-1
produced a consistent Q factor in comparison to the other 5 configurations tested. Additionally

56
supports used for FFFF-1 are simple to operate, allowing for rapid setup and testing, as well as
being able to support a wide range of weights from 300 kg to several tonnes allowing for
various sample sizes and weight to be assessed. Testing with FFFF-1 was conducted using an
inflation pressure of 180 kPa based on manufacturers recommendations, as part of this study a
sensitivity assessment of the pressure used for FFFF-1 corresponding to the dynamic
assessment of the CLT panels was conducted. For FFFF-1 tests, pressure variations from 70
kPa ranging to 700 kPa in increments of 70 kPa were conducted; Figure 4-8 presents the
comparative data from these sensitivity tests.

Figure 4-8 shows that with the increasing pressure values the error between simulated and
experimental increases slowly over the incremental tests. As the pressure increases, the
resonant frequency positions shift higher with some frequencies decreasing in amplitude
(beginning to dampen). The RBM varies from 3.506 Hz to 6.177Hz from 70 kPa to 700 kPa
respectively; the amplitude also increases as shown in Figure 4-10. This change in RBM
properties especially amplitude can cause errors during peak identification and should be noted
to ensure frequencies of this range are neglected. Comparing 70, 350, and 700 kPa, it can be
seen in Figure 4-10 that modes 6 and 10 decrease in amplitude and quality as the pressure
increases. From these findings of frequency shifting and amplitude decrease it is concluded
that the supports reach a point where the pressure supplied causes the supports to become
stiffer, allowing for less vibratory movement and damping some modal responses; thus causing
the supports to act less as free support conditions and more as simply supported configurations
at those points. Figure 4-9 shows the change in Q factor with the increasing pressure. The Q
factor results show that as the pressure increases, the range of the measurements decreases up
until 350 kPa, and then begins to increase again. The average Q factor value is consistent
although due to the variations in range, the error is lowest for pressure measurements from 70
to 210 kPa between modes 1 to 5 as shown in Figure 4-8.

57
Figure 4-8: Comparative results for airbag pressure investigation (top) modes 1 to 6, (bottom)
modes 7 to 12.

Figure 4-9: Pressure sensitivity investigation, changes in signal quality with rising pressure.

58
Figure 4-10: Rapid decrease in peak quality, leading to unwanted error in peak identification
(presented in 70, 350 and 700 kPa).

4.5 Conclusions

Six BCs identified in this article are suitable applications for a NDE system for CLT panels, a
freely supported system was identified as being the most repeatable configuration providing
consistent quality factors, a good comparison of <+8 % error between theoretical and simulated
modal frequency, and a simple setup for manufacturers to implement in-line. Of the three FFFF
variations assessed, FFFF-1 returned clear and easy to interpret results as well as being a robust
and repeatable configuration. Average error between modes 1 to 10 was <+5 %. Based on the
results of the sensitivity analysis conducted on the pressure used to inflate the airbags used for
FFFF-1, the optimal pressure range was > 70 kPa and < 210 kPa; a pressure of 180 kPa was
selected for the comparative BC tests. By considering not only the accuracy and performance
of the BC but the application to industry as well, this study outlines FFFF-1 as a more suitable
configuration for NDE testing of CLT panels in an in-line environment. The BC allows for
rapid and effective assessment of the material to a large scale while being a suitable difficulty
to prepare/ setup with little equipment needed (air bags, accelerometer, DAQ, and signal
acquisition software); the FFFF-1 BC configuration with a pressure of 70 kPa to 210 kPa was
found to accurately represent a FEA model of a freely supported BC. Assumptions made during
this study have been the effects modelling the CLT panel as a solid plate element and not the
accumulation of the individual boards (edge bonding effects not taken into account).

59
5 Mode Shape Regeneration
5.1 Background

The use of non-destructive evaluation (NDE) techniques for the assessment of timber products
has been well established and published for solid timber boards, with some information on the
assessment of engineered wood products (EWPs) such as cross-laminated timber (CLT) panels.
A consistent assumption made in the published literature for the NDE of CLT is to model the
CLT as layers of shell elements with thickness equal to the thickness of the laminates of which
the CLT is comprised and predict resonance frequencies. CLT consists of various cross layered
boards face bonded together to form a panel and by considering such a non-homogeneous
construction as a single orthotropic plate element could affect the accuracy of the frequency
analysis. A series of modelling techniques have been explored throughout this section to
investigate the effects of modelling CLT as (i) a solid plate, (ii) several bonded layered plates,
and (iii) as a panel formed through a series of boards bonded together. Within the final
parameter of modelling the CLT for the individual boards within it, the effect of friction
between edge contacting boards and inversely the effect of a small edge spacing between
boards has also been investigated. All results have been compared against experimental modal
testing using surface mounted accelerometers and vibration analysis.

5.2 Materials and Methods

5.2.1 Test Specimens

Similarly to Faircloth et al. (2021) presented in section 4 of this thesis, the approach taken for
the experimental testing has been to fix the specimens material properties by evaluating a single
CLT panel under a FFFF BC to investigate the variation between the experimental results and
the various FEA techniques. The CLT panel used for the experiments and modelling was
provided by the primary CLT panel manufacturer in Australia, the material was comprised of
3 layers of radiata pine (Pinus radiata) boards and an average global density of 475 kg/m3. The
panel was 3.2 x 2.8 x 0.075 m and of the same properties of the tested panel in Section 4. As
noted by the manufacturers design guide and described above the grade variation between face
and core layers of the CLT target a GL10 grade (10 GPa) for the face layers and GL6 (6 GPa)
for the core layer with a panel layup of 20 mm / 35 mm / 20 mm; the panel layup is detailed in
Figure 5-1.

60
 Figure 5-1: CLT panel layering and design.
5.2.2 Experimental Protocol

As introduced by Steiger et al. (2010) the method of vibratory assessment was implemented
for assessing the CLT specimen described in the above section for a grid array of points across
the surface of the panel under a flatwise FFFF BC. The point locations, size and support points
are shown in Figure 5-2. A roving sensor approach was used for evaluating the CLT panel by
impacting the panel at a fixed position (highlighted in Figure 5-2) and re-position the
accelerometer after each test. The testing equipment used was an electronic impacting hammer
(IEPE Brüel & Kjær type 8206 impact hammer), a surface mounted accelerometer (PCB
Piezoelectronics model 352C33 single-axis), sound and vibration acquisition data logger
(National Instruments Sound and Vibration Module NI 9234 24-bit ADC), and LabVIEW
acquisition software. Signal acquisition was conducted using a sampling frequency of 8 kHz
for 51,200 samples resulting in a sampling resolution of 0.156 Hz. While impacting force was
intended to be kept consistent by the operator, the acquisition of this data allows for
deconvolution of the two signals (impact and accelerometer) to remove any basis that may
influence the results. Once the data was collected from the 24 points, the data was then
evaluated using MATLAB (ver. 2020a MathWorks, CA, USA) to observe the frequency
spectrum for the resonance frequencies. Using numerical integration, the acceleration
information collected through the accelerometer positions was converted to displacement. To
reconstruct the mode shapes, the displacement related with each peak frequency point is plotted
to give a topographical representation of the panel for each attenuated frequency. From the
reconstructed mode shapes obtained, the mode type (bending or torsional) and occurring
sequence for the corresponding resonance frequencies was overlayed against the FEA results
obtained from the various scenarios as outlined in the following section.

61
 Figure 5-2: (top view) Highlighted modal testing points on panel.

5.2.3 Theoretical Techniques

In conjunction with the experimental testing, two theoretical analysis methods were
implemented for validating the experimental results and reconstructed mode shapes. These
analysis methods make use of ANSYS and MATLAB. The material properties of Loblolly Pine
(pinus taeda) were considered comparable to Radiata Pine (USDA, 2010) and therefore have
been used through this study for Poisson’s ratio values of µLR of 0.33, µLT of 0.29, and µRT of
0.38. The variables of interest for the FEA study are outlined in the following section.

5.2.4 ANSYS FEA

The research team has identified a number of variables that can influence the materials response
to vibration that are currently overlooked through previous research outputs for CLT analysis.
These changes in the vibrational response can have a direct effect on the mode shapes
generated. Properties of interest and how they were intended to be investigated were (a) panel
construction type, (b) stiffness grade, and (c) influence of edge positioned boards. The stiffness
variation investigation was devised due to the inherent variation in mechanical properties

62
through a single board and more so a single grade that has been assigned through AS1720.1
(2010). For example, it is not uncommon for a 6 m length board to have a 20 % variation in
MoR for the results (Wood Solutions, 2021). Therefore, the following scenarios have been
targeted whereby the core and surface grade layers vary higher and lower than the commercial
target of G10/G6/G10. The panel grade compositions will follow the traditional CLT make up
with higher grades at the outer layers and a lower grade in the core as presented in Table 5-1.

Table 5-1: Assumed Stiffness compositions.

Composition Type Layer thickness

Layer
A (MPa) B (MPa) C (MPa) (mm)
Face 8,000 6,000 10,000 20
Core 6,000 4,000 6,000 35
Face 8,000 6,000 10,000 20

These will be compared against the current assumption of CLT reacting as a solid plate. These
parameters consider each layer as a solid plate, face bonded (each layer mated with the
opposing to ensure a full contact in ANSYS) with each other to form a 3-layer panel. The
results of these theoretical parameters are to be used to identify the modelling approach most
representable of a CLT panel and its reaction to vibration based transient impulse. The MoE
and G values used for the stiffness component of the study are presented in Table 5-2.

Table 5-2: Material properties for 3-layer shell CLT model in FEA (USDA, 2010).

Grade MoElongi MoEtang MoEradial GLR GLT GRT

classification (MPa) (MPa) (MPa) (MPa) (MPa) (MPa)
G4 4,000 312 452 328 324 52
G6 6,000 468 678 492 286 78
G8 8,000 624 904 656 648 104
G10 10,000 780 1,130 820 810 130

Comparisons with these lay-ups will also consider a solid plate element which will be obtained
by modelling a plate element of the sample dimensions as the CLT panel although with a single
set or mechanical parameters as outlined in Table 5-3 where parallel and perpendicular refer to
the face board grain directions.

63
Table 5-3: Single layer CLT scenario, material properties obtained from classical laminate
theory (Oregon State University OSLU freeware).

Composition MoEpara MoEperp GLR

type (MPa) (MPa) (MPa)
A 7,280 1,430 639
B 5,453 1,018 475
C 9,108 1,830 803

Further to the investigative parameters of the sensitivity study discussed above, the common
construction method for CLT is face bonded laminates of graded timber boards layered
perpendicular to the proceeding layer where these boards are not bonded edgewise. Variables
linked with the natural occurrences in CLT panels have been identified and are to be compared
alongside the parameters discussed. The edge distance or gap between these boards was
investigated between 0 and 1.5 mm as well as the consideration of a frictional coefficient (µ)
between the boards for the 0 mm gap were tested for values of 0.1 and 0.3. The benchmarking
of edge gap boards in non-edge bonded CLT was observed by Zhou et al. (2020) and (2016)
of approximately 2 to 3 mm. Based on visual inspection of the CLT panel, a 1.5 mm gap was
selected to investigate the progressive difference between a gap (1.5 mm) and no gap (0 mm).
To consider the non-bonded edges between boards, as is the case with the CLT measured for
the experiments in section 6, a contact between boards was established with a frictional
coefficient of 0.1 and 0.3. Furthermore, Chang et al. (2011) conducted 40 bi-axial static tests
to measure the frictional performance between teak wood surfaces when the grain direction of
the two samples were perpendicular to one another. Calculated values indicated a frictional
coefficient range of 0.2 and 0.4; therefore, selected frictional values of 0.1 and 0.3 have been
chosen for evaluation.

5.2.5 MAC Analysis

The correlation between detected resonance frequencies through the experimental approach
against the theoretical results obtained from the serval parameters discussed above is an
important consideration for validating the accuracy of the FEA results. Ensuring the mode
types are consistent between both experimental and FEA testing, the accuracy of these two
methods is further investigated using MAC analysis. As discussed in Pastor et al. (2012) and
Lewis et al. (2016) the MAC is defined as the statistical indicator to identify differences

64
between two mode shapes. Eq. 2-20 shows the formula used to calculate the MAC value; the
result is between 1 and 0, 1 being a high correlation and 0 being low.

To account for the differences in displacement magnitudes, the experimental and theoretical
data sets were normalised prior to MAC calculations. The MAC analysis will assist in
frequency correlation between the two analysis methods to identify mode shape agreement.
Through the use of the respective displacement of each mode shape (eigenvector) at the
experimental test points, the product of the experimental and theoretical eigenvector values
correlates to the degree of agreement between the two mode shapes. Modes shapes generated
through FEA with the ANSYS modelling software will extract the eigenvector information and
eigenvector information from the experimental data has been extracted. Using MATLAB, the
percentage results (0 to 1) will be obtained for the agreement between the two as well as a
graphical representation of the agreement.

5.3 Results and Discussion

Table 5-4 presents the frequency results from the theoretical mode shapes of the three stiffness
compositions (A, B, and C) for both the 3-layer plate and solid plate considerations described
above, the values are expressed as the percentage difference (%) between the experimental
frequencies obtained from vibration assessment and the corresponding FEA frequency for each
composition. The experimental frequencies are presented in the third column.

Table 5-4: Results summary of stiffness compositions and experimental modal testing.

FEA Results
Measured
Mode Mode A B C
Frequency
ID Shape 3-layer Solid 3-layer Solid 3-layer Solid
(Hz)
(%diff) (%diff) (%diff) (%diff) (%diff) (%diff)
1 1,1 f1 (11.7) 15 18 16 30 5 8
2 0,2 f2 (16.5) 7 11 12 11 6 11
3 1,2 f3 (27.9) 10 5 13 12 1 6
4 2,0 f4 (34.2) 16 6 17 7 6 5
5 2,1 f5 (40.6) 15 9 16 13 4 5
6 0,3 f6 (44.7) 5 14 10 9 8 14
7 2,2 f7 (55.3) 8 5 11 8 3 2

65
 8 1,3 f8 (58.4) 13 1 16 5 1 2
9 3,0 f9 (85.6) 12 5 15 11 2 2
10 2,3 f10 (88.3) 14 7 16 11 3 2
11 3.1 f11 (94.0) 15 9 17 13 5 6
12 0,4 f12 (112.8) 27 13 29 18 17 12

The data presents a varying relationship across the 12 obtained modal frequencies for the three
stiffness compositions with some consistencies between the 3-layer consideration and the solid
plate simulation although composition C presents the closest relationship with the lowest
average error across all modes. C also produces a consistent range of difference between the
two simulation scenarios although based on average error and range of error, C for the 3-layer
scenario appears to be the closest match with the experimental data. The results from f1 to f11
show a consistent relationship with no frequency discrepancy greater than +8 %; the 12th mode
(f12) produced a large difference in between experimental and FEA frequency which is explored
through the mode shape reconstruction and MAC analysis presented in Figure 5-3. Individual
graded board information was not available for correlation with the assumed properties
matching with a G10/G6/G10 composition. From the experimental modal analysis, the 12
mode shapes have been reconstructed using MATLAB and are compared against the mode
shapes exported from ANSYS below in Figure 5-3.

The comparisons shown in Figure 5-3 provide a reinforcing argument for the agreement and
low error of the experimental modal tests against the modelled stiffness sensitivity of a 3-layer
CLT simulation. All modes were able to be reconstructed and the visual comparison and MAC
analysis shown for modes 1 to 7 indicate a close agreement of 82 % to 85 %. The results of
modes 8 to 12, showed increasing disparity where the MAC analysis returns a percentage
accuracy of 75 % to 65 %. This could be a result of limited data from the experimental tests as
the modes obtained through ANSYS appear to contain a larger number of distortions that would
benefit from greater than 24 points of data.

The high correlation suggested by the MAC analysis confirms the bending modes assigned in
Table 5-4 are correct based on the experimental data collected. The difference in results
recorded for some modes is to be investigated. The differences recorded, while within +7 %
for the majority of recorded data suggest that the setup or material was influencing the results
and thus was investigated by considering the impact of the unbonded edge gap. Alternatively,
it is suggested that the edgewise contact between neighbouring boards could influence the

66
results through a frictional perspective and thus has been investigated. The results of which are
presented and discussed in Table 5-5. This stage of the results looks at a series of additional
conditions under the stiffness grade composition C as it produced the closest relationship
between experimental and theoretical frequencies. FEA has been conducted on a CLT model
with an edge-gap distance of 1.5 mm; for a gap of 0 mm a frictional coefficient of 0.1 and 0.3
has been modelled. Percentage difference values between the experimental and FEA results is
presented in Table 5-5.

67
68
69
 Figure 5-3: Mode shapes obtained through (left) experimental modal analysis and mode
shape reconstruction, (right) FEA of a 3-layer plate CLT composite panel (G10/G8/G10) in
ANSYS, and MAC analysis for each paired mode shape comparison.

Following on from the results presented in Figure 5-3, Figure 5-4 (a and b) presents the MAC
values for the 12 x 12 matrix of modal agreements. The MAC values as presented in Figure
5-3 were determined based on the displacement matrix obtained from MATLAB on the
experimental results and the FEA obtained from ANSYS.

a)

70
 12 0% 27% 1% 3% 1% 4% 0% 1% 0% 19% 0% 69%
b) 11 48% 0% 5% 0% 2% 30% 5% 0% 0% 4% 67% 0%

Mode Number (Experimental)

10 3% 2% 1% 1% 17% 5% 2% 8% 0% 65% 1% 1%
9 1% 3% 3% 0% 4% 3% 9% 0% 75% 0% 1% 0%
8 11% 7% 0% 4% 0% 4% 2% 75% 2% 0% 1% 10%
7 3% 5% 1% 5% 9% 2% 87% 6% 16% 3% 4% 2%
6 0% 4% 0% 1% 0% 82% 1% 1% 1% 8% 13% 9%
5 5% 1% 3% 1% 82% 0% 4% 2% 1% 7% 1% 1%
4 2% 6% 34% 85% 6% 1% 25% 2% 1% 11% 10% 6%
3 4% 1% 84% 22% 3% 0% 2% 0% 1% 6% 2% 6%
2 2% 84% 0% 5% 0% 5% 4% 5% 1% 16% 1% 34%
1 82% 0% 5% 0% 3% 4% 3% 0% 3% 4% 32% 0%
1 2 3 4 5 6 7 8 9 10 11 12
Mode Number (FEA)

Figure 5-4: a) 3D representation of MAC values, and b) 2D representation of MAC values.

While the matched diagonal modes (1-1, 2-2, 3-3, etc.) produce a high agreement as shown in
Figure 5-4, some off diagonals show some match variation. Comparing these results with the
information contained in Table 5-4 and Figure 5-3 it can be noted that these variations from
the expected responses are due to close resonance frequencies. The frequencies f4 and f5, f7 and
f8, and f9 and f10 are within 3 to 6 Hz of one another and appears to cause disparity in mode
reconstruction. A noticeable difference between the results presented in Table 5-4 and Table
5-5 appears to be the result of mode shape swapping or change in modal sequence. The results
appear to have less of a relation with the experimental data than the results of the C composition
scenario as a 3-plate-layer. Looking at the mode shape occurrences presented in Table 5-5, all
conditions produce the same modes the first 3 frequencies although after this point they differ
from the experimental reconstructed mode shapes (Exp. Mode from Table 5-5). All conditions
follow the same mode type for the first four frequencies and then begin to shift. Of the
compositions presented in Table 5-5, the 0.3 friction FEA modes match closest with the
experimental frequencies.

Table 5-5: Results summary for CLT scenario considerations of gap size and board friction.

Mode Frequency Exp. 1.5 mm Edge- Gap 0.1 µ 0.1µ 0.3µ 0.3µ
ID (Hz) Mode Gap (%) Mode (%) Mode (%) Mode
1 f1 (11.7) 1,1 15 1,1 14 1,1 4 1,1
2 f2 (16.5) 0,2 16 0,2 25 0,2 25 0,2
3 f3 (27.9) 1,2 1 1,2 14 1,2 7 1,2

71
 4 f4 (34.2) 2,0 26 0,3 2 2,0 2 2,0
5 f5 (40.6) 2,1 9 2,0 25 0,3 13 2,1
6 f6 (44.7) 0,3 7 2,1 4 2,1 12 0,3
7 f7 (55.3) 2,2 19 1,3 24 1,3 20 1,3
8 f8 (58.4) 1,3 8 2,2 2 2,2 3 2,2
9 f9 (85.6) 3,0 67 0,4 61 0,4 6 3,0

The results of Table 5-5 show the frictional coefficient of 0.3 produced similar modes to the
experimental results. From MAC analysis of the FEA from the 0.3 model described above as
the apparent closest representation the following agreements were found. As below, ‘B’ refers
to bending, ‘T’ refers to torsion, and ‘swap’ refers to the swapping of mode shapes where it
appeared the sequence of modes change between the results of Table 5-4 and Table 5-5.

Mode 1: 83% (T) Mode 2: 51% (B) Mode 3: 87% (B) Mode 4: 74% (B)
Mode 5: 68% (B&T) Mode 6: 50% (B&T) Mode 7: 5% (swap) Mode 8: 1% (swap)
Mode 9: 58% (B)

The MAC outputs indicate less of an agreement than was measured for the 3-layer shell model
presented in Figure 5-3 and Table 5-4. Occurrences were modes changed sequence were found
to have low agreement percentages as would be expected. The mode type being bending,
torsional, or a combination of the two also appears to effect the accuracy. Figure 5-5 presents
the MAC values for the 9 x 9 matrix presented in Table 5-5.

a)

72
 9 2% 3% 9% 3% 5% 4% 4% 0% 58%
b) 8 2% 4% 3% 20% 8% 3% 59% 1% 14%
7 25% 1% 0% 3% 0% 0% 5% 59% 0%

(Experimental)
Mode Number
6 5% 1% 3% 1% 78% 50% 5% 3% 1%
5 0% 0% 7% 0% 68% 52% 1% 1% 0%
4 1% 3% 35% 74% 5% 0% 24% 3% 1%
3 5% 3% 87% 3% 1% 0% 3% 0% 1%
2 2% 51% 1% 8% 0% 5% 5% 4% 2%
1 83% 1% 2% 1% 3% 1% 2% 1% 6%
1 2 3 4 5 6 7 8 9
Mode Number (FEA)

Figure 5-5: a) 3D representation of MAC values, and b) 2D representation of MAC values.

5.4 Conclusion
The aim of this study was to investigate a range of assumptions currently being made through
the non-destructive assessment of CLT panels. These assumptions consider CLT as a solid,
single element plate rather than as the multi-layer, multi-grade panelised system. NDE
techniques rely on the accuracy of theoretical modes to assist in the processing and analysis of
the experimental data obtained through vibrational assessment. This study compared the
frequencies from experimental modal analysis against a range of stiffness assumptions as these
are the initial conditions to any FEA model. This was conducted for both single element plates
and 3-plate-layer shell elements face bonded to represent each layer in the CLT panel that was
tested.
The grade or stiffness compositions contained in Table 5-2 and Table 5-3 formed the A, B and
C composition types and from comparison against experimental data C produced a +7% error
on average for all modes for the two simulation conditions. The two scenarios of a solid plate
and a 3-plate-layer panel produced consistent errors between one another although the 3-plate-
layer panel scenario produced slightly lower variation. This was further explored by the
reconstruction of modes shapes from the experimental modal testing, compared against the
theoretical mode shapes for composition C. The agreement between the two was shown
through MAC analysis as presented in Figure 5-3 and found which presented an approximate
agreement of 85 % for f1 to f7 with agreement accuracy decreasing for the more complex modes.
Modes at f8 to f12 presented an agreement accuracy of 75 % to 65 %.

73
While these results suggested a modest agreement was found for C as a 3-plate-layer panel, the
discrepancies present were thought to be caused by the model not considering all variables
within the CLT panel. Considering the scenario where layered boards have a consistent spacing
of 1.5 mm as a result of board distortion, using the stiffness conditions from C, a model
considering the individual board make up of CLT was created. A large variation from the
experimental data and mode shape sequence was found. Considering a gap between all boards
negatively impacted the comparison with the experimental results as well as shifting the mode
types and corresponding sequence of them which has already been confirmed through Figure
5-3.
Frictional coefficients of 0.1 and 0.3 were tested for in the inverse scenario that boards placed
tightly with edges together could induce friction between one another and effect the vibrational
response; this found conflicting results for the two coefficients. 0.1 retuned a similar degree of
inaccuracy to the 1.5 mm gap whereby frequency analysis was found to be difficult due to the
mode shape sequence shifting. For 0.3, while the results of Table 5-5 indicate a number of
frequencies with large variation from the experimental results, there was minimal mode shape
shifting noted for the first 6 modes. this was further confirmed through MAC results indicating
an average agreement of 68 %. Overall, the 3-layer shell method appears to produce a slight
increase over the solid shell plate method. The consideration of small variations to the
construction such as board gap, or friction between laminates appears to provide no added
accuracy. More evaluation into the types of variables for consideration (such as cut outs and
other defects) is recommended for a deeper understanding.

74
6 Evaluation of NDE System
6.1 Background

This section outlines the development of a non-destructive evaluation (NDE) system based on
the methodology investigated through section 4 and 5. The NDE system has developed with
the aim being to determine the global elastic parameters such as young’s and shear moduli for
cross laminated timber (CLT) panels. This study focuses on the industrial sectors’ anticipated
adoption of such a rapid panel grading system and therefore has been designed, developed, and
investigated based on the in-line requirements for such a system to be successfully
implemented. The system and testing method have investigated the effects of ‘thin’ and ‘thick’
plate theorem for orthotropic materials at 3.0 x 3.0 m size CLT panels and the prediction
accuracy obtained through 5, 8 and 12 resonance frequencies. The perceived true mechanical
properties were determined through static testing to validate the predicted values from the NDE
system. A post-processing optimisation algorithm has been developed and tested to allow for
the rapid assessment of mechanical properties against a modelled simulation allowing the
system to forgo the requirement of initial conditions unlike most models. Results indicated that
the presence of the shear deformation force influences the accuracy of the system when
predicting elastic parameters for thicker panels. This has been supported by reviewed literature
presented in section 2.2 and 2.4. A solution to orthotropic thick plates has been developed and
incorporated within this approach as introduced in section 2.4. In addition to this, attempts were
made to correlate the number of input resonance frequencies selected with NDE method
accuracy.

6.2 Materials and Methods

6.2.1 Test Specimens

All experiments were conducted on Australian manufactured Radiata Pine (pinus radiata) CLT
panels with varying parameters such as layering, size and thickness. Table 6-1 outlines the
number of panels tested, nominal sizes, layers, and governing plate theory (Section 2.4). All
testing was conducted within a conditioned testing laboratory at a nominal room temperature
of 21.5 °C and 65 % relative humidity.

75
Table 6-1: Table of CLT specimen constructions, nominal sizes, and densities.

Panel No. No. Panel Governing

Dimensions (m)
ID Panels layers Theory
P75 3.0 x 2.4 × 0.75 6 3 Thin
P105 3.0 x 2.4 x 0.105 6 3 Thin
P115 3.0 x 2.4 x 0.115 4 3 Thin
P125 3.0 x 2.4 x 0.125 9 3 Thin
P150 2.8 x 1.8 x 0.15 3 5 Thick

The material was sourced as part of an initial production run conducted by XLam, supplied to
DAF, Griffith University, and the University of Queensland in 2016. Furthermore, the
specimens will be referred to through this section as their corresponding panel IDs as assigned
in the first column of Table 6-1.

6.2.2 Dynamic Assessment Protocol

The testing was divided into two sequential streams of dynamic vibration and static testing.
The dynamic vibration testing process for the assessment of the CLT samples listed in Table
6-1 followed an impulse dynamic assessment whereby the panel is situated under the freely
supported BC (following the FFFF BC assessment in section 4, Faircloth et al. (2021)) as
presented in Figure 6-1. Testing setup shows a surface mounted accelerometer (PCB
Piezoelectronics model 352C33 single-axis accelerometer) and electronic impacting hammer
(IEPE Brüel & Kjӕr type 8206 impact hammer) which were used to conduct the impulse
measurements. Finite element analysis (FEA) of an orthotropic plate was used for each CLT
panel size (based on dimensions listed in Table 6-1) with the Cast3M software. This was
conducted to validate the support point positions for the areas experiencing maximum and
minimum deflection (anti-nodes and nodes). This method introduces the assumption that the
CLT panel reacts to bending and torsion as a solid plate element and not comprised of
individually laminated boards, as investigated in section 5. As shown in Figure 6-1 sensor and
impactor locations at diagonally opposite corners allow for bending modes to be measured
accurately.

76
 Figure 6-1: Experimental BC setup for vibration assessment of CLT panels.

Impulse data is acquired through the National Instruments Sound and Vibration module (NI
9234 24-bit ADC) using LabVIEW at a sampling rate of 8 kHz and 51,200 samples, resulting
in a frequency resolution of 0.157 Hz for 6.4 seconds. The time signal and frequency response
function are shown in Figure 6-2. From the acquired signal, processing time was minimised to
limit the introduction of noise errors from longer signal durations. The signals were re-sampled
at a resolution of 0.5 Hz for 16,000 samples and a duration of 2 seconds. It can be seen in
Figure 6-2 that the impulse occurs over a short period of time where the signal begins to decay
after approximately 1 second.

Figure 6-2: (left) impulse signal in time domain, and (right) the result of the frequency
response function (FRF).

From the above right-hand-side of Figure 6-2 the resonance frequencies from the impulse
response of the panel can be determined. The frequency domain will produce a peak at the
frequency at which a mode occurs. From the preliminary analysis and results of previous

77
sections it is clear a frequency range of 0 to 50 Hz presents 8 to 10 resonance frequencies for
all panel types with the first modal frequency occurring between 3 to 5 Hz. The amplitude of
the observed peaks will be influenced by the type of mode and position of the accelerometers.
In combination with the experimental dynamic results, theoretical analysis in MATLAB is used
to assist in the peak frequency selection and identification to remove errors introduced from
user selection in the form of possible false modes. False modes can be caused by reactions to
vibrational impulse due to something other than the material, i.e. the BCs, environmental
conditions (ground vibrations from nearby machinery), and user contact with material.

6.2.3 Optimisation Algorithm

FEA was used for modelling the dynamic behaviour of thin and thick orthotropic plate with
the theory of Hencky-Mindlin (Onate, 2013) presented in section 2.4. A CLT panel was defined
as a rectangular plate with a small thickness h (z-axis) compared to the planar dimensions (Lx
along the x-axis, and Ly along the y-axis). Nodal displacement was approximated using a
fourth-order polynomial evaluated with MATLAB (Hutton, 2004). The Hencky-Mindlin
theory took into account the transverse shear effect; this effect was negligible for thin plates
(Kirchoff theory). Thus, the main elastic parameters were the elastic moduli (Ex along the x-
axis, Ey along the y-axis) and the in-plane shear modulus (Gxy). The transverse shears (Gxz and
Gyz) were considered as secondary parameters because their effect depended on the ratios
longitudinal dimensions. From the dimensions, the weight and the five elastic parameters, the
numerical model allows for the prediction of the resonance frequencies for the panel. The
inverse problem (inversion procedure) sought to find the elastic parameters values (Ex, Ey, Gxy,
Gxz, Gyz) using the experimental frequencies. The following criterion was thus computed in
order to minimise the difference between the experimental and computed frequencies
(Brancheriau, 2021):

𝑓 −𝑓
𝛿= Eq. 6-1
𝑓

where 𝛿 is the criterion, 𝑓 is the experimental frequency of peak frequency (i), and
𝑓 is the computed frequency of peak frequency (i).

This criterion of relative errors (Eq. 6-1) allowed taking into account the different scales
between frequencies. For example, a difference of 1 Hz for a frequency of 3 Hz had a higher
influence with this criterion than the same difference of 1 Hz for a frequency of 300 Hz. This

78
criterion was minimised by an optimisation method (Nelder et al., 1965). This optimisation
method assumed that only one solution existed or that the initial solution was chosen close to
the true solution. The uniqueness of the solution is presented in the results, along with the
robustness of the inversion procedure. The initial solution was a key point that ensured a fast
and reliable convergence toward the true solution. The number of resonance frequencies
obtained for analysis has been stipulated to have a direct influence on the accuracy of predicting
the mechanical properties of beams and panels. These are commonly referred to as ‘ranks’ as
they consist of the ranking of resonance frequencies along the frequency spectrum.

6.2.4 Static Evaluation Protocol

Validation of the dynamic NDE results obtained for MoE in both axial directions (Ex and Ey)
and the primary shear moduli (Gxy) approximations were validated through static testing
methods adapted from EN16351 (2015) and EN408 (2010). Test specimens were taken from
the panel along each axial direction. A specimen strip width of 300 mm is taken from the two
panel edges for major and minor axial directions respectively as shown in Figure 6-3. Testing
for the above previously mentioned properties requires both flat and edgewise testing. The
ensure the test specimens conform to the requirements of (EN408, 2010), the 300 mm strips
were ripped to 3 ~100 mm wide sub-strip pieces in order to test for Gxy. The Gxy value reported
in further sections is the average of these 3 values.

Figure 6-3: Cutting plan applied to CLT panels for test strips.

79
A similar study by Steiger et al. (2012) compared static CLT test specimen sizes of 300 and
100 mm wide strips for validating a modal analysis method to determine MoE of CLT panels.
It was found that the 100 mm wide specimens could not accurately determine the MoE values
without a large variation in results. By comparison a single 300 mm specimen strip can be used
in place of 5 100 mm wide specimens for acceptable comparative results. MoE and G were
measured during the one instance of a four-point bending test using a Shimadzu Universal
Testing Machine (AG-100X; Shimadzu Corporation, Kyoto, Japan) with a 100 kN load cell
attachment. Here three methods of data collection are implemented as shown in Figure 6-4.

Figure 6-4: 4-point bending test setup in accordance with EN408 (2010) cl. 10 and 11 for
measuring MoE and Shear.

The mid-point deflection is measured using a set of linear variable deflection transducers
(LVDTs). This mid-point deflection is used to determine the panel stiffness (N/mm) and
furthermore the MoE from Eq. 6-2 (EN408 (2010)):
3𝐿 𝐿 − 4𝐿
𝑀𝑜𝐸 =
∆𝐹
2𝑏ℎ 2 ∆𝑤 −
6𝑎 Eq. 6-2
5𝐺𝑏ℎ
∆
here ∆ is the load versus displacement curve (N/mm), 𝐿 is the upper span, 𝐿 is the lower

span, and 𝑏 and ℎ represent the width and thickness of the sample respectively. The
denominator term allows for the correction of shear calculated through the method

described in Eq. 6-3 where 𝐺 is the shear modulus. The lasers in Figure 6-4 (L1 and L2) at the
outer points of the panel measure any global movement during the test at these points to allow
for correction of the stiffness value. This test setup was used to determine MoE in both axial
directions (Ex and Ey). Using non-contact, image processing techniques, a digital image
correlation (DIC) system was implemented to measure the strain distribution at locations as
shown in Figure 6-4 and expanded on in Figure 6-5. The Correlated Solutions DIC system uses

80
the VIC-3D program to capture images from 2 pairs of USB-C cameras at diagonals to each
pair to observe the movement on each side of the test specimen for the duration of the test. The
placement of the camera systems has been determined due to the outer third spans being
regarded as high shear zones due to the opposing loading directions on either side in a 4-point
bending scenario. The analysis of the outer thirds of the bending sample allows for observation
of a point on the beam that experiences a high moment force. During post-processing, virtual
extensometers are added to the specimen sides diagonally to give a representation of the
deflection between top and bottom layers of the CLT. This deformation is shown in Figure 6-5
and is used to calculate G in accordance with Eq. 6-3:
ℎ (𝑉 , − 𝑉 , )
𝐺= 𝛼 Eq. 6-3
𝑏ℎ (𝑤 )

(| , | | , |)
𝑤ℎ𝑒𝑟𝑒, 𝛼 = − ,𝑤 =

Figure 6-5: DIC setup for deformation measurement to determine shear moduli.

Here 𝑤 is the mean deformation of the two digital extensometers during the test and (𝑉 , −
𝑉 , ) is the load increments (N).

81
6.3 Results and Discussion
6.3.1 Robustness of the NDE Optimisation Procedure

The resonance frequencies along with the geometrical information of the CLT panels
(dimensions and weight) are used as input variables of the optimisation procedure. The
optimisation procedure cycles through estimates for material properties until the difference
between the experimental and theoretical frequencies are minimal. To investigate the effects
of the number of resonance frequencies for NDE of full-size CLT panels, three sets of
numerical simulations were performed. To do so the dimensions and density associated to a
P150 CLT panel were used (Lx = 2.8 m, Ly = 1.8 m, Hz = 0.15 m and density = 465 kg/m3).
The elastic parameters values (true solution) were defined as the static average values (Ex =
8,554 MPa, Ey = 995 MPa, Gxy = 389 MPa, Gxz = 96 MPa, and Gyz = 62 MPa). The
theoretical frequencies were then computed from FEA. To evaluate the robustness of the
optimisation procedure, the generated frequencies were biased by an error value adjusted using
a randomiser across n frequencies; where n refers to the number of frequencies or ranks. This
error (in Hertz) mimicked the error induced by the frequency resolution of the experimental
signals. The rank frequencies have been separated into 3 groups of 5, 8 and 12 for 30 test
iterations per rank group. Figure 6-6 shows the increased error value within each of the rank
groupings (here, the error value is the average value of the absolute difference between the
biased frequency and the theoretical frequency). The optimisation procedure was finally used
on the biased frequencies to determine the elastic parameters values (to be compared with the
true solution).

Figure 6-6: Robustness error for rank groupings.

82
The values presented in Figure 6-6 and their corresponding, calculated mechanical properties
have been compared against the true solution discussed above. Of the 30 test iterations per
rank, iterations have been separated into blocks of 10 sets of values for comparison against the
true solution. The absolute percentage difference values are presented in Table 6-2.
Table 6-2: Robustness percentage difference values.
Rank Avg. f
Ex Ey Gxy Gxz Gyz
composition error (Hz)
5-1 11% 14% 1% 0% 27% 0.47
5-2 7% 3% 2% 22% 13% 0.82
5-3 7% 11% 7% 2% 44% 1.55
8-1 1% 0% 2% 1% 1% 0.46
8-2 1% 1% 0% 0% 1% 0.92
8-3 1% 3% 0% 5% 38% 1.58
12-1 1% 1% 0% 1% 5% 0.43
12-2 0% 1% 2% 0% 8% 0.85
12-3 6% 5% 1% 6% 21% 1.53

From the results shown in Table 6-2 it can be interpreted that as the number of frequencies
included in the robustness calculations increases, the agreement with the true solution becomes
closer. Furthermore, the error adjustments as separated into three groups on average indicate
the low and mid frequency error values for the 8 and 12 ranks produce the closest agreement
with the true solution. This indicates that while both 8 and 12 ranks produce estimations within
5% for all 5 mechanical properties, the agreement is affected by the variation of the
experimental frequencies from the theoretical.

6.3.2 Experimental Testing

Based on the results of dynamic and static testing on the products detailed in Table 6-1 the
following section presents the comparison between the two testing methods for Ex, Ey, and Gxy.
Figure 6-7 presents the comparison between static and dynamic testing for Ex with a linear
regression line. The results are separated by colour into 5 groups to represent the 5 thicknesses.
The data set mean for each thickness is also plotted and reported against the data sets as shown
in both figures. The plot shows a number of outliers as well as thicknesses with higher
variability than others; such as the P125 test specimens.

83
 Figure 6-7: Ex comparison with static and dynamic assessment for the linear fitted data set.
From the results of Figure 6-7 the variability appears to be concentrated to the P125 values
showing a larger variation through the dynamic NDE values than the static. From further
observation of the panel data, the density of panels 29, 44, and 49 are considerably lower than
the average of the rest of the subset. It is suspected that the P125 specimens contain a higher
number of defects and due to the maintained 3 layering in the P125 panels resulting in increased
board thicknesses. These panels have therefore been removed from the analysis. The corrected
results are presented in Figure 6-8 where (a) refers to the linear fitted data.

84
Figure 6-8: Corrected Ex for outliers with (a) a linear response, and (b) a balanced response of
the data set.
The constant term in Figure 6-8 (a) of 470 is statistically non-significant and therefore Figure
6-8 (b) shows the exact relationship between the two testing parameters with an overestimation
of 12% from NDE. The results presented in Figure 6-9 display the relationship between the
static and dynamic results for Ey. The results show a 76% relationship between the two
approaches for Ey although several thicknesses present an increased variation.

Figure 6-9: Ey comparison with static and dynamic assessment for the linear fitted data set.

85
Closer inspection of the data presented in Figure 6-9 determined the bulk of the variation is
concentrated to the P125 specimens, as is also noted for Ex. As noted for Ex, panels 44 and 47
have a considerably lower density compared to the average population. These panels have been
removed from the analysis for Ey. The results shown in Figure 6-10 (a) and (b) show the
corrected responses when panels 44 and 47 are removed from the analysis. The relationship
between the two testing conditions increases from 76% to 86%. The constant term in Figure
6-10Figure 6-8 (a) of 97 is statistically non-significant and therefore Figure 6-8 (b) shows the
exact relationship between the two testing parameters with an overestimation of 6% from NDE.

Figure 6-10: Corrected Ey for outliers with (a) a linear response, and (b) a balanced response
of the data set.

86
The results of Ex and Ey are presented below in Table 6-3 where the mean and coefficient of
variation are tabled for the two approaches of dynamic NDE and static testing. The values
presented in Table 6-3 show a range of 7.22 to 11.01 GPa for Ex and 1.32 to 0.55 GPa for Ey.
It can also be observed that the P125 specimens contain a high variation in the results with a
28% and 38% variation in the dataset across NDE and static testing respectively. This high
variation explains the relationship shown through Figure 6-7 to Figure 6-10 where the P125
samples are causing the majority of the skewed data.
Table 6-3: Summary of NDE and static results for MoE and G.

Mean MoE Mean CoV

Test Lx/h
Panel Type (GPa) Gxy (%)
Method ratio
Ex Ey (MPa) Ex Ey Gxy
NDE 10.44 1.32 550.4 2 18 7
P75 32
Static 8.08 1.20 423.6 7 7 6
NDE 8.11 0.61 291.6 24 11 11
P105 23
Static 7.22 0.64 216.5 12 15 10
NDE 7.41 0.55 189.5 13 26 12
P115 21
Static 7.21 0.60 160.9 13 19 15
NDE 9.94 0.85 507.0 15 28 66
P125 19
Static 9.37 0.80 304.5 8 38 23
NDE 11.01 0.59 423.3 16 14 5
P150 12
Static 9.90 0.62 357.1 18 9 5

The relationship between dynamic and static results for the primary shear variable Gxy is shown
below in Figure 6-11. A similar trend is apparent from the plot where the P125 samples have a
high variation in comparison to other thicknesses. It is particularly apparent that the variation
due to the P125 specimens is affecting the linear relationship across the population.

87
Figure 6-11: Gxy comparison with static and dynamic assessment for the linear fitted data set.
As the static results for Gxy are determined off of an average of 3 sub-strip pieces 100 mm in
width, the percentage difference for each sub piece against the dynamic NDE value has been
determined and is presented in Figure 6-12. This chart shows the large range at which the
results vary for P125, followed by P115 and P105. Due to the variation across the majority of
the P125 samples, these have been removed from the analysis of the whole population with the
comparison with static and dynamic presented in Figure 6-13.

Figure 6-12: Gxy percentage difference within sub-strip pieces for static testing and NDE.

88
Figure 6-13: Corrected Gxy for outliers with (a) a linear response, and (b) a balanced response
of the data set.
The constant term in Figure 6-13 (a) of 46 is statistically non-significant and therefore Figure
6-13 (b) shows the exact relationship between the two testing parameters with an underestimate
of 18% from NDE. The corrected data sets shown in Figure 6-13 show minimal variation across
the remaining thicknesses which is reinforced in Table 6-3. The table below presents the mean
and coefficient of variation within each of the thickness groups. The results presented in the
table show a consistent CoV percentage of 6 to 15% for P75, P105, P115, and P150. The P125
samples show a very high variation for both dynamic and static approaches.

89
Throughout this section, interpreting the results has been made difficult through the noted
variation inherent to these CLT panels. While the trends are skewed by large variations in the
test material, a relationship between both NDE and static testing methods can be observed,
indicating the suitability of such a system for this application with further refinement. Figure
6-14 shows an image of the surface finish of some CLT test pieces, suggesting a number of
factors that contribute to the high variation in some thicknesses.

Figure 6-14: Example of CLT panel defects noted through testing.

Considering the observations in section 6.3.1 where the results of which were based on the
correct input frequencies being selected, the vibrational frequency spectrum for the transient
analysis of CLT is composed by several modes (bending in OX, in OY and torsion) which are
difficult to identify (depending on the physical and mechanical characteristics of the pane). The
analysis and peak selection is influenced by the presence of double or overlapping peaks. This
is made obvious through the results of the algorithm convergence assessed through MATLAB.
The convergence will occur towards the resultant minima where the minima is the elastic
prediction based on the experimental frequencies. Figure 6-15 presents the convergence
response developed by the experimental frequency values for the measured CLT panels.

90
 Figure 6-15: Values of the minima for the elastic moduli values (a) 3D view and (b)
topographic view.
As seen in Figure 6-15 (a) there are two distinct minima which can bias the solution based on
the initial conditions or guess values. Therefore suggesting the results are heavily dependent
on the correct frequency values and frequency rank order being inputted into the algorithm and
further are greatly influenced by the panel variability. This is further reinforced by the
transverse shear convergence shown in Figure 6-16.

Figure 6-16: Values of the minima for the transverse shear (a) 3D view and (b) topographic
view.

91
6.4 Conclusion

Five CLT thicknesses were evaluated against a developed dynamic (NDE) approach for
determining key mechanical properties such as modulus of elasticity for major and minor axial
directions (Ex and Ey) as well as shear moduli perpendicular to the face of the panels (Gxy). The
results of the NDE assessments were combined with the use of an optimisation algorithm which
allows the system to forgo input parameters as most FEA models require. Rather the system
operates on an iterative process where conditions are cycled and compared against the obtained
frequency spectrum until the error is minimal, returning the optimal response. Testing was
conducted to i) validate the overall approach of using vibrational measurements in combination
with a CIRAD developed solution to the Hencky-Mindlin thick orthotropic plate theory. This
solution when fed the frequency results from experimental testing, determines mechanical
properties of ‘thin’ plates, and investigate the use of the optimisation algorithm to predict the
mechanical properties of thick plates. NDE results obtained from the panel assessments were
validated against standardised testing methods for determining mechanical properties of CLT
and found results to be within 2 to 24% of one another for Ex across P75, P115, P125, and
P150. This range of results has been echoed through other properties with a percentage
difference of 7 to 26% for Ey and 5 to 15% for Gxy for the same panel types. The P125 panels
produced the largest percentage differences for the three mechanical properties indicating a
cause for further investigation. Investigating panel quality and collected details, density
variation across the 9 tested panels ranged from 410 kg/m3 to 510 kg/m3 for the large scale
panels, although strip specimens taken from the parent panels did not indicate such a variation.
Thus, it is suspected that the 300 mm wide strip specimen may not be sufficient to produce a
representative result on the panels’ performance.
The robustness assessment reported in section 6.3.1 summarised the comparison between
having 5, 8 and 12 theoretical frequencies as input parameters for the optimisation algorithm.
The findings of previous studies indicated that the higher number of ranks used through the
mechanical properties’ determinations correlated to a higher accuracy. The results in Figure
6-6 and Table 6-2 present a good comparison with the true solution for both 8 and 12 ranks
while. The application of an error value to the theoretical frequencies reinforced these findings
suggesting for theoretical approaches 8 to 12 frequencies would be sufficient in predicting Ex,
Ey, Gxy, Gxz, and Gyz within 5% of the true solution.

92
7 Overall Conclusions

The aim of the study has been to evaluate a range of variables considered to be of great
influence in the application of a non-destructive evaluation (NDE) system in a quality
assurance scenario for cross-laminated timber (CLT) panel manufactures. This thesis
comprised of 7 individual sections explores the development of a robust prototype acquisition
system and testing methodology to conduct NDE tests on large scale CLT panels. The study
compared several boundary condition (BC) supports to be comprised within the nominated
testing method as well as validated the analysis method through modal reconstruction using
the modal assurance criterion (MAC) analysis. Finally, the collaborative efforts of a developed
optimisation algorithm have been initially validated using the developed testing method and
acquisition system against static testing methods.

The outcomes of these efforts have led to the development of a prototype acquisition system
using the LabVIEW testing software platform and a suitable BC ensuring a repeatable
configuration and easy setup practical for industry to adapt. The results of the BC evaluation
found while the majority of tested conditions were able to collect the required data, there were
a number of practical considerations the researchers identified with some of the more complex
setups. The focus for all aspects of this study has been to validate the system on a semi-
commercial scale and therefore the appetite for industry to invest in the setup needed to be
considered. Combining this with the low damping and consistent signal quality from the freely
supported system using the air bags (FFFF-1), this BC setup was considered the most suitable
as well as most practical for a production line scenario.

Following these tests, section 5 conducted experimental modal testing across the surface of a
CLT panel when supported using the selected BC from section 4. From the results of these tests
the mode shapes were regenerated using MATLAB and the MAC analysis was conducted to
evaluate the agreements. The first stage of the FEA found the 3-layer shell model to have the
lowest error against the experimental modal results. Expanding on these results the CLT panel
was then modelled for each individual laminate and its assumed properties to resemble a CLT
panel. The MAC comparisons indicated a loss in accuracy with this method with some
interesting observations. It appeared that by considering the CLT as a series of laminates rather
than solid plates resulted in similar frequency ranges although a shifting of mode shapes. The
mode shape types refers to the bending, torsion or combination of the two shape displacement.
The shifting of these shapes or change in their sequence is important in frequency analysis. It

93
is suggested that a detailed model coupled with an in-depth physical testing approach could
provide insight into the potential effects in modelling CLT in its entirety.

From the method development discussed through sections 3 to 5, section 6 presented the
validation testing data from the dynamic NDE experiments on large scale CLT panels for 5
thicknesses. The results were interpreted using a CIRAD developed theoretical algorithm and
compared against static test results on each of the panels. The results on the algorithm
robustness suggest a successful solution can be found when the correct experimental
frequencies are collected and inputted into the code. Although it was also noted in Figure 6-15
and Figure 6-16 that the evaluation of CLT is not without complexities. The results of FEA
shown in Figure 6-15 and Figure 6-16 present the convergence minima of the algorithm for
both E and G values. The results concluded that there exist multiple minima where the
prediction could converge and thus the prediction of mechanical properties would be heavily
dependant on the initial conditions.

The comparison of dynamic NDE with static testing presents both this effect as well as the
influence of physical panel properties such as size and density. The results contained a number
of data points that were removed from the analysis due to this variation which appeared to be
concentrated to the P125 samples. While variation was present in some thicknesses this could
be due to density variation, panel age and quality, and the effect of edge gaps. From the results
of Gxy it was also determined that strips of 100 mm wide CLT pieces varied up to 35% in a
300 mm wide section. With variation to this degree noted the discrepancies between the
predicted and static results could rest with the material. As these panels were secured as part
of the sites initial production, expansion of this work with a more scrutinous view on material
type would be suggested.

7.1 Thesis Findings

As concluded above, the research outlined in this thesis has been to investigate variables of
interest to contribute towards a in-line NDE system for CLT panel evaluation and quality
assurance. Providing manufacturers with the rapid ability to effectively and accurately check
their CLT products for the inherent mechanical performance establishes peace of mind. This
relates to the manufacture knowing their products are as designed, the developer to ensure they
are designing for the correct material properties, and the user knowing the structure and the

94
materials being used within it are fit for purpose. The research questions established at the
beginning of this report outlined the aims and objectives being to:

 Develop a prototype NDE system and validate for a range of CLT sizes.
 Evaluate the BC’s available based on their suitability for the application as well as their
performance.
 Investigate suspected inaccuracies that theoretical components of these studies may be
introducing by considering CLT as a solid plate structure.

Based on the information contained within sections 3 to 6 these questions have been evaluated,
researched, and concluded on for the material evaluated. The developed NDE system
established through section 3 provides a working platform for which further development and
refinement can be conducted to fully automate and trial within industry. Its proven
effectiveness in collecting, processing and delivery data to the MATLAB codes indicates its
proof of concept in the planned aim. The published results of section 4 present the planning
and evaluation stages that contributed to the large scale evaluation of 6 BCs. The results were
evaluated not only based on the accuracy of the data collected but also on the practical setup
requirements to determine whether industry would be able to implement the nominated setup.
The air bag configuration in FFFF-1 is an easy setup, requires minimal equipment and provides
repeatable and high-resolution results shown through the low damping and high-quality factor
values.

Prior to trialling the developed method, the research team had reservations on considering a
multi-layer segmented product such as CLT as a solid plate structure as classically done for
theoretical plate analysis. Through the investigation stages of section 5 a multi-layer and solid
plate element were compared against experimental modal analysis. While the two models
produced similar results the 3-layer shell was compared against re-constructed mode shapes
form the experimental data where MAC analysis was used to assess the agreement. The results
of the MAC analysis indicated bending modes to be relatively high in agreement
(approximately 85 %). Although higher modes and torsional modes presented less accurate
agreements of 75 to 65 %. It is hypothesised that this is due to the surface mounted
accelerometers being less likely able to measure in-plane deflections than bending based
movement. It is although expected due to the size of the panels and a MATLAB/ ANSYS
modal reconstruction application being used that some inaccuracies are being introduced
through the method.

95
The results of section 6 present the findings from implementing the testing method from section
4 along with the acquisition system from section 3 to validate the CIRAD developed theoretical
algorithm. By evaluating 5 different thicknesses of CLT the system has been validated for
width to thickness ratios of 32 to 12. The results also investigated the settings used within the
algorithm to investigate the robustness of the solution in regard to number of frequencies
required (5, 8 and 12 frequencies) as inputs to the system. These presented an increase in
accuracy with the increased frequency numbers, consistent with findings from the literature.

7.2 Recommendations for Future Work

From the findings of this work a number of areas have been identified as expandable or as
requiring more research to further understand the effects being had. The following points as
recommendation for future research have been nominated:

1. Modelling variables: The MAC analysis and FEA of CLT panel provides a introductory
view on the design advantages to a working detailed CLT model. An extension of this
work into the effects of considering the range of defects, joints and unbonded areas
within a CLT panel could provide manufacturers with a method for observing the
performance limiting characteristics involved in the design and manufacture process.
2. Validation of Experimental NDE system: The results presented throughout section 6
provide an insight into the application of NDE systems for the quality assurance testing
of CLT. The current status of the system trialled through this study is not yet suitable
for industry implementation although provides a proof of concept for the use of such a
test method in the future. An extension of this work into a larger validation trial would
be advantageous for system development.
3. Prototype Establishment: The system developed in section 3 provides a working proof
of concept although is not a standalone system ready for industry implementation.
Future work is needed on incorporating the developed algorithm and prediction
methods into the LabVIEW testing platform and improving the user interface.
4. In-situ and Vibrational Research: Based on reviewed literature regarding vibration
analysis, the consideration of a structure rather than individual elements is aligned with
the serviceability requirements of structural design. Further research into the
application of a similar method as adopted through this study, with the intent of
measuring in-situ performance (condition monitoring) is a logical next step.

96
8 Bibliography
Abell, M., & Kalny, O. (2016). Technical Knowledge Base, Thin vs. Thick shells.
Albostami, A., Wu, Z., & L., C. (2020). Elastic Response of Cross-Laminated Timber Panels using Finite
Element and Analytical Techniques. Canadian Journal of Civil Engineering, 48(10).
doi:10.1139/cjce-2020-0205
AS1720.1. (2010). Timber structures - design methods. Standards Australia.
AS/NZS4357.2. (2006). Structural laminated veneer lumber (LVL), Part 2: Determination of structural
properties. Standards Australia/ Standards New Zealand.
Avitabile, P. (2001). Experimental Modal Analysis - A simple non-mathematical presentation. Sound
and Vibration Magazine.
Bodig, J., & Jayne, B. A. (1993). Mechanics of Wood and Wood Composites. Krieger Publishing
Company, Melbourne, FL, U.S.A.
Bos, F., & Bos Casagrande, S. (2003). On-line non-destructive evaluation and control of wood-based
panels by vibration analysis (Vol. 268).
Bos, F., & Casagrande, S. B. (2003). On-line non-destructive evaluation and control of wood-based
panels by vibration analsysis. Journal of Sound and Vibration, 403-412.
Brancheriau, L. (2021). Non-destructive evaluation of elastic material properties and acoustic
characterisation of wood based massive panel. Centre for International Research and
Agricultural Development, milestone 4, 31.
Brancheriau, L., & Bailleres, H. (2002). Natrural Vibration Analysis of Clear Wooden Beams: A
theoretical Review. Wood Science and Technology, 36, 347-365. doi:10.1007/s00226-002-
0143-7
Brandner, R. (2013). Production and Technology of Cross Laminated Timber (CLT): A state-of-the-art
Report. European Conference on Cross Laminated Timber (CLT), 1.
Brandner, R., Flatscher, G., Ringhofer, A., Schickhofer, G., & Thiel, A. (2016). Cross Laminated Timber
(CLT): overview and development. European Journal of Wood and Wood Products, 74(3).
Bucar, D. G., & Bucar, B. (2011). Strength grading of structural timber using the single mode
transverse damped vibration method. Wood Research, 56(1), 67-76.
Bucur, V. (2006). Acoustics of Wood. Spinger Series in Wood Science, 2.
Cawley, P., & Adams, R. D. (1988). The mechanics of the coin-tap method of non-destructive testing.
Journal of Sound and Vibration, 122(2), 299 - 316. doi:doi.org/10.1016/S0022-
460X(88)80356-0
Chang, W. S., Hsu, M. F., & Komatsu, K. (2011). A new proposal to reinforced planked timber shear
walls. Journal of Wood Science, 57, 493-500.
Christovasilis, I. P., Brunetti, M., Follesa, M., Nocetti, M., & Vassallo, D. (2016). Evaluation of the
mechanical properties of cross laminated timber with elementary beam theories.
Damme, B. V., Schoenwald, S., & Zemp, A. (2017). Modeling the bending vibration of cross-laminated
timber beams. European Journal of Wood and Wood Products, 75. doi:10.1007/s00107-016-
1152-9
Damme, B. V., & Zemp, A. (2017). Measuring Dispersion Curves of Bending Waves in Beams: A
Comparison of Spatial Fourier Transform and Inhomogeneous Wave Correlation. Acta
Acustica United with Acustica, 103.
EN408. (2010). Timber structures - Structural timber and glued laminated timber - Determination of
some physical and mechanical properties. British Standard.
EN789. (2004). Timber Structures - Test Methods - Determination of Mechanical Properties of Wood
Based Panels. British Standard.
EN1995-1. (2004). Eurocode 5: Design of timber structures, Part 1-1: General - common rules and
rules for buildings. British Standard, 1(1), 124.
EN14374. (2004). Timber structures. Structural laminated veneer lumber. Requirements. British
adopted European standard.

97
EN16351. (2015). Timber Structures - Cross Laminated Timber - Requirements. BSI Standards
Publication.
Faircloth, A., Brancheriau, L., Karampour, H., So, S., Bailleres, H., & Kumar, C. (2021). Experimental
Modal Analysis of Appropriate Boundary Conditions for the Evaluation of Cross-Laminated
Timber Panels for an In-line Approach. Forest Products Journal, 71(2), 161-170.
doi:doi:10.13073/FPJ-D-20-00062
FEA, H. (2019). Thin Beam vs. Thick Beam. Retrieved from
http://habituatingfea.blogspot.com/2012/11/thin-beam-vs-thick-beam.html
Gagnon, S., & Popovski, M. (2011). Cross Laminated Timber Handbook - Structural design of cross-
lamianted timber elements. Forest Product Innovations - Canadian Edition, Chapter 3.
Gu, M. (2017). Strength and Serviceability Performances of Southern Yellow Pine Cross-Laminated
Timber (CLT) and CLT-Glulam Composite Beam. Clemson University,
Guan, C., Zhang, H., Wang, X., Miao, H., Zhou, L., & Liu, F. (2017). Experimental and Theoretical
Modal Analysis of Full-Sized Wood Composite Panels Supported on Four Nodes. MPDI.
Han, S., Benaroya, H., & Wei, T. (1999). Dynamics of Transversely Vibrating Beams using four
Engineering Theories. Journal of Sound and Vibration, 255(5), 935-988.
Hanna, N. F. (1991). Thick plate theories, with applications to vibration. The Ohio State University,
Hutton, D. (2004). Fundementals of Finite Element Analysis. McGraw-Hill, 494.
ISO10140-3. (2010). Acoutics - Laboratory measurements of sound insulation of building elements,
Part 3: Measurement of impact sound insulation. International Standards Organisation, 3.
Labonnote, N., Ronnquist, A., & Malo, K. A. (2013). Semi-analytical prediction and experimental
evaluation of material damping in wood panels. Holzforschung, 67(3), 333-343.
doi:10.1515/hf-2012-0095
Leissa, A. (1993). Vibration of plates (NASA, Washington, DC, 1969). Reprinted by Acoustical Society
of America, Woodbury NY.
Lekhnitskii, S. G. (1968). Anisotropic Plates. Gordon & Breach, 534.
Lewis, K., Basaglia, B., Stirling, R., & Crews, K. (2016). The use of Cross Laminated Timber for Long
Span Flooring in Commerical Buildings. WCTE 2016 - World Conference on Timber
Engineering, 4813-4821.
Love, A. E. H. (1906). A treatise on the Mathematical Theory of Elasticity. Cambridge University - at
the university press, 2, 3-29.
Maheri, M. (2010). The effects of layup and boundary conditions on the modal damping of FRP
composite panels. Journal of Composite Materials, 45(13). doi:10.1177/0021998310382314
Majkut, L. (2009). Free and Force Vibrations of Timoshenko beams Described by Single Difference
Equation. Journal of Theoretical and Applied Mechanics, 47(1), 193-210.
McGavin, R. L., Dakin, T., & Shanks, J. (2020). Mass-timber Construction in Australia: Is CLT the only
answer? bioresources.com, 15(3), 4642-4645.
Middleton, R. (2017). Timber Construction Using CLT. Retrieved from
https://www.reidmiddleton.com/reidourblog/timber-construction-using-clt/
Mindlin, R. D. (1951). Influence of rotary inertia and shear on flexural motions of isotropic, elastic
plates. J.of Appl.Mech., 18, 31-38. Retrieved from https://ci.nii.ac.jp/naid/10004540844/en/
Morandi, F., Santoni, A., Fausti, P., & Garai, M. (2021). Determination of the dispersion relation in
cross-laminated timber plates: Benchmarking of time- and frequency-domain methods.
Applied Acoustics, 185.
Moslemi, A. (1967). Dynamic viscoelasticity in hardboard. Forest Products Journal, 17(1), 25-33.
Neiderwestberg, J., Zhou, J., & Chui, Y. H. (2014). Influence of Boundary Conditions in Modal Testing
on Evaluated Elastic Properties of Mass Timber Panels. World Conference on Timber
Engineering 2014.
Nelder, A., & Mead, R. (1965). A simplex method for function minimization. Computer Journal, 7(4),
308-313.

98
Onate, E. (2013). Structural Analysis with the Finite Element Model. Volume 2. Beams, Plates and
Shells. Springer, 864.
Opazo-Vega, A., Benedetti, F., Nunez-Decap, M., Maureia-Carsalade, N., & Oyarzo-Vera, C. (2021).
Non-Destructive Assessment of the Elastic Properties of Low-Grade CLT Panels. MDPI,
12(1734).
Pastor, M., Binda, M., & Harcarik, T. (2012). Modal Assurance Criterion. Procedia Engineering, 48,
543 - 548.
Reissner, E. (1976). On the theory of transverse bending of elastic plates. International Journal of
Solids and Structures, 12(8), 545-554.
Robin, O., O'Donoughue, P., Berry, A., Farley, V., & Prithipaul, K. (2021). Full field measurements on a
cantilever beam under impact using visable and infrared deflectometry. Applied Acoustics,
183.
Ross, R. J., & Pellerin, R. F. (1994). Nondestructive testing for assessing wood members in structures:
A review. (General technical report FPL; GTR-70): 39 p.: ill.; 28 cm., 70.
Santoni, A., Schoenwald, S., Damme, B. V., & Fausti, P. (2017). Determination of the elastic and
stiffness characteristics of cross-laminated timber plates from flexural wave velocity
measurements. Journal of Sound and Vibration, 387-401.
Shahnewaz, M., Tannert, T., Alam, M., & Poppovski, M. (2015). Experimental and finite element
analysis of cross laminated timber (CLT) panels. First international conference on advances in
civil infrastructure and construction materials, MIST, Dhaka, Bangladesh, 14-15 December
2015.
Shirmohammadi, M., Faircloth, A., & Redman, A. (2020). Determining acoustic and mechanical
properties of Australian native hardwood species for guitar fretboard production. European
Journal of Wood and Wood Products, 78, 1161-1171.
Sikora, K. S., McPolin, D. O., & Harte, A. M. (2016). Effects of the thickness of cross-laminated timber
(CLT) panels made from Irish Sitka Spruce on mechanical performance in bending and shear.
Construction and Building Materails, 116, 141-150.
SproBmann, R., Zauer, M., & Wagenfuhr, A. (2017). Characterization of acosutic and mechanical
properties of common tropical woods used in classical guitars. Results in Physics, 7, 1737-
1742.
Steiger, R., Gluzow, D., & Czaderski, C. (2012). Comparison of Bending Stiffness of Cross-Laminated
Solid Timber derived by Modal Analysis of Full Panels and by Bending Tests of Strip-Shaped
Specimens. European Journal of Wood and Wood Products, 70, 141-153.
Steiger, R., Gulzow, D., & Gsell, A. (2010). Non-destructive evaluation of elastic materail properties
of cross-lamianted timber (CLT). World Conference on Timber Engineering 2010.
Suzuki, J.-i., Mizukami, T., Naruse, T., & Araki, Y. (2016). Fire Resistance of Timber Panel Structures
Under Standard Fire Exposure. Fire Technology, 52(4), 1015-1034. doi:10.1007/s10694-016-
0578-2
Szilard, R. (2004). Theories and applications of plate analysis : classical, numerical, and engineering
methods: Hoboken, NJ : John Wiley.
Timoshenko, S. P. (1922). X. On the transverse vibrations of bars of uniform cross-section. The
London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, 43(253), 125-
131.
USDA. (2010). Wood Handbook - Wood and an Engineering Material. United States Department of
Agriculture - Forest Products Laboratory, 101-102.
Vacher, P., Jacquier, B., & Bucharles, A. (2010). Extensions of the MAC criterion to complex modes.
Proceedings of ISMA2010 including USD2010, Noise and Vibration Engineering(24th), 2713-
2716.
Van Damme, B., Schoenwald, S., & Zemp, A. (2017). Modeling the bending vibration of cross-
laminated timber beams. European Journal of Wood and Wood Products, 75(6), 985-994.

99
Wang, J. Y., Stirling, R., Morris, I., Taylor, A., Lloyd, J., Kirker, G., Lebow, S., Mankowski, M., Barnes, H.
M., & Morrell, J. J. (2018). Durability of Mass Timber Structures: A review of the biological
risks. Wood and Fiber Science, 50.
Xia, B., O'nill, T., Zuo, J., Skitmore, M., & Chen, Q. (2014). Percived Obstacles to Multi-Storey Timber-
Frame Construction: An Australian Study.
Zang, Z., Jiang, Z., Hse, C., & Liu, R. (2017). Assessing the impact of wood decay fungi on the modulus
of elasticity of slash pine (Pinus elliottii) by stress wave non-destructive testing. International
Biodeterioration & Biodegradation, 117. doi:10.1016/j.ibiod.2016.12.003
Zhang, L., Tiemann, A., Zhang, T., Gauthier, T., Hsu, K., Mahamid, M., Moniruzzaman, P. K., & Ozevin,
D. (2021). Nondestructive assessment of cross-laminated timber using non-contact
transverse vibration and ultrasonic testing. European Journal of Wood and Wood Products.
Zhou, J., Chui, Y. H., Gong, M., & Hu, L. (2016). Simultaneous measurement of elastic constants of
full-size engineered wood-based panels by modal testing. Holzforschung, 70(7).
doi:10.1515/hf-2015-0117
Zhou, J., Chui, Y. H., Gong, M., & Hu, L. (2017). Comparative study on measurement of elastic
constants of wood-based panels using modal testing: choice of boundary conditions and
calculation methods. Journal of Wood Science. doi:10.1007/s10086-017-1645-0
Zhou, J., Chui, Y. H., Neiderwestberg, J., & Gong, M. (2020). Effective bending and shear stiffness of
cross-laminated timber by modal testing: method development and application. Composites
Part B.

100