Ma Lu 2013 PDF

Comparison of Finite Element Method and
Modal Analysis of Violin Top Plate
Ye Lu
Music Technology Area, Department of Music Research

Schulich School of Music
McGill University
Montreal, Canada
June 2013
A thesis submitted to McGill University in partial fulfillment of the requirements for the
degree of Master of Arts in Music Technology.

c 2013 Ye Lu
2013/06/04
i
Abstract
This thesis presents a study of the vibrational behaviour of the violin top plate and explores
the possibility of using composite materials as a substitute for traditional wood in making
top plates. Numerical simulations and experimental tests are compared to validate the
results. The two most popular methods for numerical and experimental vibrational analysis,
the Finite Element Method (FEM) and Experimental Modal Analysis are used, respectively.
The vibrational behaviour of a spruce top plate is first studied by the two methods.
The results show high coherence. Then the same modeling and testing techniques are
applied on two composite plates. Results show that the vibrational behaviour of composite
plates differs significantly from traditional wooden plates. Thus, suggestions for further
improvement of the composite top plates are given.
ii
Sommaire
Cette these propose une etude sur le comportement vibratoire de la table dharmonie du
violon et explore la possibilite dutiliser des materiaux composites comme un substitut pour
le bois traditionnel dans la fabrication des tables dharmonies. Des simulations numeriques
et des tests exprimentaux sont compares pour valider les resultats. Les deux methodes
les plus populaires pour lanalyse vibratoire numerique et experimentale, la methode des
elements finis (FEM) et lanalyse modale exprimentale sont utilises, respectivement.
Le comportement vibratoire dune table dharmonie en epicea est dabord etudiee par les
deux methodes. Les resultats montrent une coherence forte. Puis la modelisation et meme
des techniques de test sont appliques sur deux plaques composites. Les resultats montrent
que le comportement vibratoire de table dharmonie composites diff`ere sensiblement de
plaques traditionnels en bois. Ainsi, des suggestions damelioration des tables dharmonie
composites du dessus sont donnes.
iii
Acknowledgments
I would like to gratefully thank Prof. Gary Scavone and Prof. Larry Lessard for their
supervision, guidance, and expertise during this work. I would like especially, to thank
Gary for providing the various equipment to ensure my experimental test worked smoothly
and for spending considerable amount of time on all aspects of my thesis. Also I would like
to show my great appreciation to Larry for providing me the very helpful violin top plate
prototypes and their very detailed manufacturing data. Many thanks to the luthier Peter
Purich for offering the spruce plate and the Chladni pattern photos
I would like to extend my appreciation to the students, faculty and staff of the Music
Technology Area at McGill University for their advice, insight and friendship. In particular,
I would like to thank the members of the Computational Acoustic Modeling Laboratory
(CAML) for providing me a great environment to pursue my research work. I would also
take this opportunity to acknowledge Hossein Mansour at CAML for all the meaningful
discussions and for his valued assistance. Many thanks to Darryl Cameron and Harold
Kilianski for providing necessary equipment and the technical support during my research
work. A special thanks to my girlfriend Jenny Liu for kindly providing the proof reading.
Finally, the greatest thanks go to my parents in China, for their financial and moral support
in my most difficult times.
iv
Contents
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Project Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Thesis Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 Background 5
2.1 Violin Acoustics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.1 The Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.2 The Bridge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.3 The Bass Bar and Soundpost . . . . . . . . . . . . . . . . . . . . . 8
2.1.4 The Soundbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 The Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 Procedures of the Finite Element Method . . . . . . . . . . . . . . 10
2.2.2 Node, Element and Shape Function . . . . . . . . . . . . . . . . . . 11
2.2.3 Equation of Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Experimental Modal Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.1 Mobility Measurement . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3.2 Excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.3 Coherence Function . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.4 Windows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4 Composite Music Instruments . . . . . . . . . . . . . . . . . . . . . . . . . 22
3 FEM and Modal Analysis Set-up 25

3.1 Numerical Model of a Spruce Top Plate . . . . . . . . . . . . . . . . . . . . 25
Contents v
3.1.1 CAD Model of a Top Plate . . . . . . . . . . . . . . . . . . . . . . . 26

3.1.2 Pre-Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.1.3 Normal Mode Extraction . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Modal Analysis Set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.1 Experiment Set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2.2 The Choice of the Hammer . . . . . . . . . . . . . . . . . . . . . . 37
3.2.3 Accelerometer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2.4 Selection of the Points . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2.5 Data Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2.6 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2.7 Modal Information Detection . . . . . . . . . . . . . . . . . . . . . 44
4 Comparison of FEM and Modal Analysis Results 49

4.1 FEM Results of the Spruce Plate . . . . . . . . . . . . . . . . . . . . . . . 49
4.2 Modal Analysis Results of the Spruce Plate . . . . . . . . . . . . . . . . . 51
4.3 Carbon Fibre Top Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.3.1 Choice of the Material . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4 Numerical Simulation of the Composite Top Plates . . . . . . . . . . . . . 56
4.5 Modal Analysis of the Composite Top Plates . . . . . . . . . . . . . . . . . 64
5 Discussion 70
5.1 FEM Results of the Spruce Plate . . . . . . . . . . . . . . . . . . . . . . . 70
5.2 Comparison of the FEM Results to the Modal Analysis Results of the Spruce
Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.3 Vibration and Damping of the Composite Plate #1 . . . . . . . . . . . . . 74
5.4 Vibration and Damping of the Composite Plate #2 . . . . . . . . . . . . . 76
5.5 Comparison of the FRFs by the FEM and the Modal Analysis . . . . . . . 78
5.6 FEM Experiments on Different Layer Combinations . . . . . . . . . . . . . 79
6 Conclusions and Future Work 85

6.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
6.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
References 87
vi
List of Figures
2.1 Structure of a violin [1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.2 The bridge of a violin [2] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 A single harmonic oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 A violin top plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.5 A discretized violin top plate . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.6 A coherence function in the test . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1 The spruce top plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.2 A set of selected points for defining the arch height . . . . . . . . . . . . . 28
3.3 CAD drawing of the spruce top plate . . . . . . . . . . . . . . . . . . . . . 29
3.4 Meshing of the spruce top plate . . . . . . . . . . . . . . . . . . . . . . . . 30
3.5 Collectors of the top plate . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.6 Typical setup for a roving impact hammer test on a cantilever beam [3] . . 35
3.7 The experiment set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.8 The pendulum system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.9 Impact Hammer Model 086E80 . . . . . . . . . . . . . . . . . . . . . . . . 38
3.10 Accelerometer Model 352C23/NC . . . . . . . . . . . . . . . . . . . . . . . 40
3.11 Selected hitting points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.12 Marked points on the spruce plate . . . . . . . . . . . . . . . . . . . . . . . 43
3.13 Experimental Setup #2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.14 The FRF and the coherence function of point 2 . . . . . . . . . . . . . . . 46
3.15 Overlapped FRFs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.16 Plate mode shape for mode 2 [3] . . . . . . . . . . . . . . . . . . . . . . . . 47
4.1 Mode 1 of the spruce plate by the FEM and the Chladni Method . . . . . 50
List of Figures vii
4.4 Comparison of Mode 1 of the spruce plate by the FEM and the Modal Analysis 51
4.10 Components of the second composite plate . . . . . . . . . . . . . . . . . . 55
4.11 Stage 1 of making a composite plate (layers cut and put in mould) . . . . . 56
4.12 Stage 2 of making a composite plate (air extracted) . . . . . . . . . . . . . 56
4.13 Stage 3 of making a composite plate (vacuum attached) . . . . . . . . . . . 57
4.14 Stage 4 of making a composite plate (after curing) . . . . . . . . . . . . . . 57
4.15 Stage 1 of making the urethane ring . . . . . . . . . . . . . . . . . . . . . . 57
4.16 Stage 2 of making the urethane ring . . . . . . . . . . . . . . . . . . . . . . 57
4.17 Components of the composite plate #1 . . . . . . . . . . . . . . . . . . . . 58
4.18 Comparison of Mode 1 of the composite plate #1 by the FEM and the Modal
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.24 Components of the composite plate#2 . . . . . . . . . . . . . . . . . . . . 63
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
List of Figures viii
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.1 The first six modes of the spruce plate by the FEM (Bretos) [4] . . . . . . 71
5.2 The first six modes of the spruce plate by the FEM (Molin) [5] . . . . . . . 72
5.3 The first six modes of the spruce plate by the FEM . . . . . . . . . . . . . 72
5.4 FRF of the spruce plate by the FEM . . . . . . . . . . . . . . . . . . . . . 80
5.5 FRF of the spruce plate by the Modal Analysis . . . . . . . . . . . . . . . 80
5.6 FRF of the composite plate #1 by the FEM . . . . . . . . . . . . . . . . . 81
5.7 FRF of the composite plate #1 by the Modal Analysis . . . . . . . . . . . 81
5.8 FRF of the composite plate #2 by the FEM . . . . . . . . . . . . . . . . . 82
5.9 FRF of the composite plate #2 by the Modal Analysis . . . . . . . . . . . 82
5.10 Layup 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.11 Layup 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.12 Layup 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.13 Layup 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
ix
List of Tables
3.1 Width and length of the plate . . . . . . . . . . . . . . . . . . . . . . . . . 27

3.2 A set of selected points for defining the arch height . . . . . . . . . . . . . 27
3.3 Arch height at a set of selected points . . . . . . . . . . . . . . . . . . . . . 32
3.4 Elastic parameters and density of the spruce . . . . . . . . . . . . . . . . . 32
3.5 Specifications of the hammer . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.6 Specifications of the accelerometer . . . . . . . . . . . . . . . . . . . . . . . 41
4.1 Natural frequencies of the spruce top plate by FEM . . . . . . . . . . . . . 49

4.2 Natural frequencies of the spruce top plate by modal analysis . . . . . . . . 54
4.3 Material properties of spruce, balsa and MTM45-1 . . . . . . . . . . . . . . 59
4.4 Natural frequencies of the composite plate #1 by FEM . . . . . . . . . . . 60
4.5 Material properties of Newport301 . . . . . . . . . . . . . . . . . . . . . . 64
4.6 Natural frequencies of the composite plate #2 by the FEM . . . . . . . . . 64
4.7 Natural frequencies of the composite plate #1 by the modal analysis . . . 68
4.8 Natural frequencies of the composite plate #2 by the modal analysis . . . 69
5.1 Comparison of the natural frequencies by the FEM of the spruce plate with
previous research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.2 Comparison of the natural frequencies by the FEM and the modal analysis
of the spruce plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
of the spruce plate and the composite plate #1 . . . . . . . . . . . . . . . 75
5.4 Comparison of the damping ratios of the spruce and the composite plates . 76
of the spruce plate and the composite plate #2 . . . . . . . . . . . . . . . 77
List of Tables x
5.6 Comparison of the natural frequencies by the FEM of different composite

plate layups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
xi
List of Acronyms
FEM Finite Element Method

DOF Degree of Freedom
MDOF Multi-Degree-Of-Freedom
CAD Computer-Aided-Design
FRF Frequency Response Function
1
Chapter 1
Introduction
1.1 Motivation
The study of a violins vibrational and acoustical properties has been popular over the last
several decades. A typical violin is made up of several parts: a top plate, back plate, ribs,
bass bar, sound post, neck, scroll, etc. The design of the top plate has a strong influence
on the timbral quality of a violin [6]. It is not easy to make a good top plate because the
luthier must not only choose the most suitable wood (light, stiff, low friction, a specific
relation between the transversal and the longitudinal stiffness, etc.), but must also carefully
define the arching shape and thickness distribution [7].
In my thesis, both numerical and experimental methods were used to test and compare
the vibrational behavior of violin top plates. The research considered top plates made of
both spruce and composite materials. The analysis of the properties of new materials (such
as carbon fibre) were conducted under the partial supervision of Professor Larry Lessard in
the Mechanical Engineering Department. By analyzing the vibrational patterns of violin
top plates using these methods, an insight was gained into the acoustical behaviour of
the plate. The comparison of the numerical simulation and experimental tests was used
to validate the results. This research might potentially allow for the optimized design of
violin top plates and be useful as a theoretical basis for manufacturing consistently good
violins.
Research on violins is not a new topic. Indeed, one might argue that luthiers have
been conducting empirical research on violins for well over 400 years, with results from 300
years ago still considered by many to epitomize the best available specimens [8]. Hutchins
2013/06/04
1 Introduction 2
was the first one to point out the importance of the plates on the tuning of the violin
[6, 9]. Over the past few decades, several numerical and experimental techniques have
been applied to the analysis of violin top plates. Two of them, the Finite Element Method
(FEM) [10, 5, 11, 4, 12] and the Modal Analysis [13, 14, 15, 16, 17], are very popular
because they provide good results that can be compared to one another. Several researchers
have looked at different aspects of the optimization of the top plate by using these two
methods. Loen has studied the graduation patterns and thickness distributions of many
classic Cremonese violins [18]. Tinnsten et al. have used the simulated annealing method
to compensate material parameters with thickness variation [19]. Yu et al. have studied
nodal line optimization to maximize the vibration of the top plate [20]. The quest for
new materials to replace wood in the top plates has also been a hot topic during these
years [21, 22]. Due to the development of advanced manufacturing technologies, composite
materials of superior performance can be achieved, by choosing and adjusting their material
properties carefully.
1.2 Project Overview
The goal of my research project was to study the vibrational behaviour of the violin top
plate. Both numerical and experimental methods were used and compared with each other
to validate the numerical model. The normal modes (resonances) of vibration and their
frequencies were determined by both methods. Different geometry and material properties
were also tested to see how they influence the mode sensitivities.
The first step of my project was to make a geometrical model of the top plate. There are
various commercial Computer Aided Design (CAD) software packages available including
AutoCAD, SolidWorks, UniGraphics, CATIA, and Pro/Engineer. SolidWorks was selected
because it has powerful surface generating functions and it is easier to change parameters
directly in this software. The dimensions of the plate and the arch height were varied
around known values from violins considered to be of high quality.
After the mesh (a discretized geometry which is used for calculation) is made, it will be
used for the numerical simulation. The Finite Element Method (FEM), which is a numer-
ical technique particularly good at solving complex structural problems, were used. The
idea was to discretize the top plate structure into regions, each of which has many smaller
elements called nodes (intersections of mesh elements). Properties of the material (wood or
1 Introduction 3
carbon fibre) were defined in terms of elasticity and density and then the theoretical vibra-
tional behaviour of the structure was solved through matrix manipulations. This allowed
the primary eigenfrequencies of a structure to be determined, as well as the visualization
of the mode shapes. The commercial software package ANSYS was used for the calcu-
lations, together with the pre-processing software HyperMesh to mesh the structure and
define the loads and boundary conditions. Different structural parameters (thickness, arch
height, bass bar) and materials (spruce and carbon fiber) were simulated. Subsequently,
a Modal Analysis of both spruce and composite top plates was performed. The process
involved hanging each plate (to achieve a free-free boundary condition) and then record-
ing the response of the plates (using a miniature accelerometer) to a sequence of taps
at various locations (using an instrumented miniature force hammer). Both the input and
output response data was input into a Modal Analysis software, MEScope, that can extract
eigenfrequencies and visualize the modal shapes. The next step was to compare the Finite
Element results with Modal Analysis results. The parameters of the FEM were carefully
adjusted in order to get satisfactory results.
1.3 Thesis Overview
Chapter 2 will provide an introduction to violin acoustics, the idea of how to implement
FEM and Modal Analysis, and the usage of composite materials on the musical instrument.
This information will give the reader a clear idea about how things work in the later
chapters.
Chapter 3 will deal with the numerical simulation and the experimental test set-up of
the spruce plate. Comparisons with previous research will be provided, though a detailed
analysis of the results will be left for subsequent chapters.
Chapter 4 will discuss the numerical simulation and the experimental test results on
both the spruce and the composite plates. The manufacturing process of the composite
plates will also be discussed.
Chapter 5 will include an analysis of the data from the previous two chapters. Discus-
sions will focus on how the geometry and the material property variations influence the
vibrational behaviour of the plate.
Chapter 6 will summarize the results and conclusions obtained in general. A few po-
tential research goals for the future will also be suggested.
1 Introduction 4
1.4 Contributions
After carefully adjusting the parameters of the FEM, it can be used as a tool for the
design optimization of violin top plates. Instead of manually adjusting the arch height
and thickness of the plate to experiment with variations of geometric features, a numerical
simulation can be performed to estimate the resulting effect. The comparison of numerical
and experimental results can also provide some information about how a plate made of new
materials (such as carbon fiber) behaves in comparison to the traditional wooden plate.
Compared to wood, composite materials have more flexible properties and are easier to
manufacture. The results of this thesis might push forward the application of composite
materials in the music instrument industry.
5
Chapter 2
Background
2.1 Violin Acoustics
The violin is one of the most representative instruments in the string family. Over the past
three hundred years, both musicians and scientists have made great efforts to try to under-
stand the complicated vibration mechanisms behind its mellifluous sound. Their research
on violin acoustics has focused on different aspects, including the material properties of the
wood, the body vibration modes, the air cavity modes coupled with the body, the radiation
from the f-holes and so on.
A typical violin can be divided into several parts: a bow which provides excitation to
the strings; several strings which are attached at the scroll and the tailpiece with high
tensions; a bridge which supports the strings and transmits the vibration energy to the
body; a bass bar and a soundpost which provide support to the top plate and help to
better transmit the vibrations from the top plate to the soundbox; and a soundbox which
serves as a resonator. There are also several acoustically less important components, such
as the neck and the chinrest which the musician holds to make the instrument stable when
playing. A disassembled violin is shown in Fig. 2.1.
2.1.1 The Strings
The strings receive energy from the bow and are forced to vibrate. They are stretched
and fixed at both the scroll and the tailpiece, so that standing waves are generated when
they vibrate. The pitch of a fixed string is determined by its length, diameter, density and
2013/06/04
2 Background 6
Fig. 2.1 Structure of a violin [1]

2 Background 7
torsion. Several types of materials are most often used for making violin strings. Gut core
strings have the longest history. They can produce warm and rich sounds. Back in the
sixteenth century, the gut strings were made purely of sheep gut, while todays gut strings
often have only the gut core covered by copper or silver. Steel core strings are another
popular type of strings nowadays. They usually have more overtones compared to gut
strings because they have less internal damping. Due to the stable behavior of steel, these
strings are less sensitive to the environment, and can last longer. Synthetic core strings
use new materials and became available in the 1970s. They are basically made of nylon
and composite materials, which have an even warmer timbre than gut strings, as well as a
longer lifetime than steel strings.
When a string vibrates, there will be a dominant fundamental frequency and several
overtones which are harmonic to that frequency. For instance, if a violinist plays an A4
note, the fundamental frequency will be 440 Hz, and there will also be harmonic overtones
which have frequencies of 880 Hz, 1320 Hz, 1760 Hz, 2200 Hz and so on. The overtones
will drop off in amplitude with frequency but some of the lower frequency components can
be as strong or nearly as strong as the fundamental. These overtones play an important
role in constructing the timbre of the note.
Strings are changed quite often during the whole life of a violin. Strings from a certain
brand may sound fantastic on one violin but not as good on another. A violinist might
need to spend great effort finding a type of string which can produce the best sounds on
his or her violin.
2.1.2 The Bridge
The bridge, as shown in Fig. 2.2, transfers the vibration energy of the strings to the violin
body via two feet which rest on the top plate, parallel to the notches of the f-holes. One
foot is right above the bass bar and another is very close to the soundpost. The bridge
works as a filter for different wavelengths of sound. Bridges are usually made of maple,
which is also very often used to make the back plate. This wood is stiffer than spruce and
balsa and is good for withstanding the pressure applied by the strings. The two feet of the
bridge are carved to fit the curvature of the top plate. At the intersection, the bridge is
purely clamped by the strings and the top plate without glue. The four notches on the top
of the bridge and the two feet make it stand stable.
2 Background 8
Fig. 2.2 The bridge of a violin [2]
The bridge is not symmetric in either direction. From the front view, the notch of the G
string is a little bit lower than the notch of the E string, which makes it easier for the player
to control the strings. From the side view, one side is flat, another is slightly tapered. The
flat side faces the scroll and the tapered side faces the tailpiece, which makes the bridge
stand more stable on the top plate.
Since the bridge stands at the most sensitive position of a violin, even very tiny changes
in its shape or position can make perceptible differences to the timbre. Previous research
showed that modifications of its shape, density and the stiffness, as well as the position of
the two feet, will cause shifts of the natural frequencies of the violin body [23, 24].
2.1.3 The Bass Bar and Soundpost
The bass bar is glued rigidly to the top plate and it is positioned right below the G-string
foot of the bridge. The bass bar serves as a support for the top plate against the high
bending forces applied by the strings. Another function of the bass bar is to help spread
the vibration energy from the string to the upper bout and lower bout of the top plate.
The presence of the bass bar shifts the natural frequencies of the top plate. It also
makes the mode shapes asymmetric and changes the nodal line positions of certain modes.
It is found that the bass bar can be carefully tuned to match the stiffness of a top plate.
Thus when it is glued to that plate, the amplitude of the tap tone and the damping can be
adjusted to an optimum value [25].
2 Background 9
The sound post also plays an important role in the violin acoustics. It is proven that
one role of the soundpost is to introduce some asymmetry to the body vibration. On some
instruments, one foot of the bridge is simply elongated to replicate the sound post, thus
providing asymmetry [26].
Another function of the soundpost is to help better transfer the vibration energy from
the top plate to the back plate. Researchers also showed that by moving the sound post
inside the body, a phase difference can be made for the top plate and the back plate when
they are vibrating [13].
2.1.4 The Soundbox
The soundbox consists of the top plate, back plate (also called the belly), and the ribs.
The plates are carefully tuned before being assembled. Before assemblage, luthiers use
various means to get the best sound quality of the plates. The tap tone method is one
of the easiest to try, in which luthiers tap on the plate in order to determine whether it
has a similar sound compared to plates previously known to be good. In order to modify
the sound quality, tiny adjustments of the thickness and the geometry are made to the
plates. Nowadays, many luthiers also use the Chaldni method to more accurately detect
the modes of the plate [6]. The idea is to put some sand on the plate and use a shaker to
excite the structure over a range of frequencies. When the driving frequency fits the natural
frequency of the plate, the mode at that frequency is excited, and the sand will collect near
the node lines of that mode. By applying this method, luthiers can easily determine the
frequencies of the normal modes as well as the mode shapes. This method gives a clearer
idea about the quality of the top plate than the traditional tap tone method because the
latter usually requires excellent absolute pitch and aural memory.
Regarding the tonal quality, C.M. Hutchins suggested certain relationships for mode
frequencies and amplitudes. She pointed out that the ring mode (whose mode shape is
like a ring), which is usually the fifth normal mode of the top plate, should be excited as
much as possible. Also, she suggested that an acoustically good violin plate should have
octave relationships for its first, second, and fifth modes [6] (this point is doubted by some
other researchers). However, it is still unclear whether these suggestions are actually agreed
upon by luthiers or researchers.
The top and back plates are connected with ribs by glue. The ribs are very carefully
2 Background 10
shaped by heat. The ribs are made of maple and they are usually very thin (about 1mm).
The body of the violin serves as a resonator. When the string energy is transferred
to the top plate, the body helps convert the mechanical vibrations to acoustic vibrations,
radiating them into the air. The air modes inside the body couple with the body modes,
which makes the vibration behavior and the mode shapes more complicated.
Much previous research has focused on modal analysis of the whole body [13, 27, 28].
Most of these chose the excitation position at the bridge since this is the position where
energy gets transferred from the strings. Numerical methods (such as the Finite Element
Method) are also used to predict the vibrational behaviour of the violin body [29]. However,
the condition is more complicated here than analyzing the top plate since the soundpost
and the ribs are not rigidly glued to the top and back plate.
2.2 The Finite Element Method
The Finite Element Method (FEM) is a powerful numerical method for solving complex
differential equations. The idea of FEM is to discretize a complex structure into relatively
simple substructures called elements. Then the elements are connected in a certain
manner, appropriate boundary and load conditions applied, and approximate solutions to
the problem determined. Since the violin top plate has a relatively complex geometry, it
is hard to describe it with a set of differential equations and analytically solve them. In
such a case, the Finite Element Method is particularly useful for calculating the natural
frequencies and the mode shapes of the violin top plate.
2.2.1 Procedures of the Finite Element Method
The execution of a typical commercial Finite Element Method software can usually be
divided into three stages: Pre-Processing, Processing and Post-Processing.
The Pre-Processing stage includes defining the element types, meshing the system,
assigning material properties to each element and setting the boundary conditions and the
loads. The understanding of the physical principles behind the problem is always more
important than the modeling technique itself. If the boundary conditions are incorrect, or
if the element type is not properly chosen, the results could be erroneous.
Many commercial Pre-Processing software systems (such as HyperMesh and Gambit)
offer user-friendly interfaces. Usually the user does not have to directly deal with the
2 Background 11
sophisticated matrix transformations; he or she would only need to click on the screen to
choose the nodes and input all the coefficients.
The Processing stage is more straightforward. The pre-processed model is imported into
a solver, and then the solver will assemble the stiffness matrix and calculate all the Degrees
of Freedom (DOFs). This process might take several seconds to several days, depending
on the number of elements and the computational ability of the computer.
The Post-Processing stage allows visualization of the calculated data. Based on the cal-
culated DOFs, different graphs can be drawn based on the users preference. For instance,
one can choose either the stress or the strain distribution, although the calculated DOFs
are in the form of displacements. The visualization of the results provides a very direct
idea for structure modifications.
2.2.2 Node, Element and Shape Function
It is mathematically difficult to define a complex structure with a set of prescribed functions

and proper boundary conditions. And it is even more difficult to analytically solve a set of
functions like this. The FEM is particularly useful in solving complicated differential equa-
tions with numerical solutions. The FEM works by approximating the shape of a structure
with a finite number of smaller geometrical segments for which analytical functions can
be defined and solved. The overall behaviour of the complete structure is then found by
solving the functions. The continuous structure can be discretized into a set of Elements,
each containing several Nodes. Each node will have several DOFs. A DOF can be any
physical unit which must be solved in the differential equation, e.g. displacement, velocity,
pressure. A node or element can have different types of DOFs. For instance, a 4-node-8-
DOF element can have either two translational DOFs or one translational DOF and one
rotational DOF on each node.
The shape function is used to interpolate computed values between the nodes. Usually
a shape function has the form of a polynomial function. Given a differential equation of
the form:
(u(x), u(x),
u(
x), x) = 0, (2.1)
the approximate solution u(x) of the problem can be expressed by truncating n terms of
2 Background 12
its Taylor expansion:

n
u(x)
X
= a0 + a1 x + a2 x2 + a3 x3 + ...an xn = ai x i . (2.2)
i=0
This approximate solution can also be represented with a set of DOFs at the nodes:
n
u(x)
X
= N1 (x)u1 (x) + N2 (x)u2 (x) + N3 (x)u3 (x) + ...Nn (x)un (x) = Ni (x)ui (x), (2.3)
i=1
where Ni is the shape function at node i and ui (x) is the DOF at node i.
A feature of the shape function is Ni (x) = 1 at node i, and Ni (x) = 0 at other nodes.
In this manner, each shape function can represent the weight of the corresponding node.
Consider a simple 2-node-2-DOF bar element:
u1 u2
For this element, the solution can be expressed as:
1
u(x)
X
= a0 + a1 x = ai x i , (2.4)
i=1
where u(x) is the global solution, x is the global variable, and a0 , a1 are coefficients which
can be expressed by x1 and x2 (the value of x at the two nodes). The element can be
represented in matrix form as:
" # " #" #
u1 1 x1 a0
= . (2.5)
u2 1 x2 a1
where u1 = u(x1 ) and u2 = u(x2 ).

Solving for a0 and a1 , we have:
u2 u1
a0 = u1 x1
x2 x1
u2 u1 (2.6)
a1 =
x2 x1
2 Background 13
The solution u(x) can be represented as:

u2 u1 u2 u1
u(x) = u1 x1 + x
x2 x1 x2 x1

x2 x x x1 (2.7)
= u1 + u2
x2 x1 x2 x1
= N1 u1 + N2 u2
and the shape function N can be represented as:

" #
x2 x
x2 x1
N= xx1
, (2.8)
x2 x1
where x1 and x2 are the coordinates at the two ends of the bar and x could be any point
on this element. In this case we have the simplest 2-Node-2-DOF 1D Bar Element. In
practical engineering problems, 2D and 3D elements are used very often. In a commercial
software, a 2D element can usually have up to 8 nodes and 48 DOF, and a 3D element can
usually have up to 27 nodes and 60 DOFs (rotational DOFs of 3D element are often omitted
in commercial software). No matter how large the element is, it is always necessary to find
an expression for the shape functions and these shape functions are used to calculate the
mass and stiffness matrices of the element, as shown in the next chapter.
2.2.3 Equation of Motion
A vibration problem can be described by a general equation of motion, which follows

Newtons laws of motion. Lets first consider the equation of motion of a single harmonic
oscillator:
d2 x dx
m 2 + c + kx = F, (2.9)
dt dt
where m is the mass of the oscillator, c is the damping coefficient of the oscillator, k is
the stiffness of the oscillator and F is the external force. This equation can describe the
vibrational behavior of a damped oscillator, as shown in Fig. 2.3:
Most of the vibration problems in real life are far more complicated than simply solving
the vibration of an oscillator. Usually a complex structure (such as violin top plate) has
more than one natural frequency. If the problem is assumed linear, the structure can
be considered as the sum of a bunch of harmonic oscillators, and the structure will have
2 Background 14
Fig. 2.3 A single harmonic oscillator
multiple DOFs. Then the equation of motion can be written in matrix form as:
+ [C][X]
[M ][X] + [K][X] = [F ], (2.10)
where M is the mass matrix, C is the damping matrix, K is the stiffness matrix and F is
the external force. These three matrices are very important in determining the vibrational
properties, including the natural frequencies and damping ratios of a structure.
If the vibration problem has a complex structure and intricate boundary conditions, it
is hard to accurately define the mass, damping, stiffness matrix, as well as the loads and the
boundary conditions in Eq. 2.10. It will be more straightforward to represent the complex
geometry with a set of elements. For instance, a violin top plate can be discretized to
small elements as shown in Fig. 2.4 and Fig. 2.5:
Fig. 2.4 A violin top plate
Each element has its own mass, stiffness, and damping properties. By assembling the
small matrices in each element into the big matrix in Eq. 2.10, all the natural frequencies
2 Background 15
Fig. 2.5 A discretized violin top plate
and displacements at each node can be calculated.

In Eq. 2.10, all the mass, damping and stiffness matrices are square symmetric. In
general, the complete equation is a quadratic eigenvalue problem. For the modal analysis,
the damping terms are usually ignored. Thus we have:
+ [K] [X] = [F ] .
[M ] [X] (2.11)
Considering the free vibration, in which the load on the system equals zero, the equation
becomes:
+ [K] [X] = 0.
[M ] [X] (2.12)
It is assumed the solution has the form:
[x] = [a]ej[]t , (2.13)
where [a] is the vector of displacement amplitudes, [] are the vector of natural frequencies
that need to be determined, and t is the time. After substitution into the equation, this
becomes:
([K] [M ])[X] = 0, (2.14)
where = 2 . In order to get non-trivial solutions, the left multiplier has to be singular,
so we have:
det([K] [M ]) = 0 (2.15)
The eigenvalues are defined as the values of which satisfy the equation and can be
2 Background 16
calculated by solving the equation:
([K] i [M ])[a]i = 0, (2.16)
where [a]i are called the eigenvectors of the equation.

If we want to solve Equation 2.16, [K] and [M ] must be explicitly expressed. In order to
get the mass and stiffness matrices for the whole structure, the mass and stiffness matrices
need to be first calculated for each element, which can be represented as:
Z Z Z
[Mi ] = [Ni ][Ni ]T V (2.17)
V
Z Z Z
[Ki ] = [Bi ]T E[Bi ]V, (2.18)
V

where Bi = x [Ni ] is the strain-displacement matrix, Ni is the shape function and E is the
elasticity matrix of element i. The value of E is related to a set of material properties,
including Youngs Modulus, Poisson Ratio and Shear Modulus. The elasticity matrix for
an orthotropical material can be written as:
Eyxy Ezxz
1

Ex
0 0 0
Exyx 1
Ezyz 0 0 0

Ey

Exzx Eyzy 1
Ez
0 0 0
[E] = , (2.19)

1

0 0 0 Gyz
0 0

1
0 0 0 0 0

Gxz
1
0 0 0 0 0 Gxy
where Ex , Ey , Ez are the three dimensional Youngs Modulus values, xy , yx , yz are the
Poisson Ratios and Gxy , Gxz , Gyz are Shear Modulus values.
Since all the parameters can be explicitly expressed, the equation can be directly solved.
A set of natural frequencies for each mode can then be calculated. The eigenvectors of each
natural frequency, which are displacements of each node under a certain mode can also be
calculated.
Consider again the bar element in Section 2.2.2, as shown below.
2 Background 17
u1 u2 u3
The shape function (Eq. 2.8) can be used to calculate the mass and stiffness matrices of
the element. By substituting Eq. 2.8 into Eq. 2.17 and Eq. 2.18, the mass and stiffness
matrices of each element can be represented as:
" # " #
k1 k1 k2 k2
[K1 ] = [K2 ] = (2.20)
k1 k1 k2 k2
" # " #
m1 m1 m2 m2
[M1 ] = [M2 ] = , (2.21)
m1 m1 m2 m2
and then assembled into the big matrix for the whole structure:

k1 k1 0
[K] = k1 k1 + k2 k2 , (2.22)

0 k2 k2

m1 m1 0
[M ] = m1 m1 + m2 m2 . (2.23)

0 m2 m2
After substituting these two matrices into Eq. 2.16, the natural frequencies and the eigen-
vectors can be extracted. For a violin top plate, the natural frequencies and the mode
shapes are important indicators that may help inform the design.
2.3 Experimental Modal Analysis
FEM can be used to calculate the natural frequencies of a structure. This technique is
also called Finite Element Modal Analysis. However, Modal Analysis more often refers
to Experimental Modal Analysis. Unlike the Finite Element Modal Analysis, which uses
numerical algorithms to calculate the mode information, an Experimental Modal Analysis
uses physical tests to calibrate the real structure and the mode extraction is based on the
experimental data. In my thesis, modal analysis refers to Experimental Modal Analysis
only.
2 Background 18
Modal analysis is a popular technique for detecting and predicting dynamic behaviours
of a structure. The modal parameters, including eigenfrequencies, damping, and mode
shapes, are of the most interest and can be calculated by applying modal analysis.
The central idea of modal analysis is to get the Frequency Response Function (FRF)
of the structure. FRF is the quantitative measure of the output spectrum of a structure
in response to a stimulus and is often used to characterize the dynamic behaviour of the
system. To compute the FRF, a measurable excitation needs to be applied to the struc-
ture at a certain position and the responses will be measured at several other positions.
Subsequently, both signals are fed into an analyzer for calculating the FRFs.
2.3.1 Mobility Measurement
This important aspect of modal analysis involves measuring a set of FRFs at different
chosen positions. Each FRF is defined by the ratio of the output response to the input
signal. The input and the response are measured simultaneously. By applying a Fourier
Transform to both the input and output signals, they can be transformed from the time
domain to the frequency domain. The function will be complex. Both the magnitude and
the phase can be calculated from the complex function. There are three types of FRFs with
varying vibration responses: compliance FRF, mobility FRF and accelerance FRF, which
correspond to displacement, velocity and acceleration, respectively in response to a force
input. In my experiment, the vibration response was measured with acceleration since the
accelerometer is one of the most convenient motion transducers. The equation of motion
(Eq. 2.9) can then be rewritten in the form of:
d2 x dx 2 02 F
+ 20 + 0 x = , (2.24)
dt2 dt k
where is the damping ratio, 0 is the natural frequency, k is the stiffness of the oscillator
and F is the external force. The accelerance function can be represented in the form of:
2 02

[
x()] 1
H() = = . (2.25)
[F ()] k 02 2 + j(20 )
2 Background 19
The magnitude of this function can be calculated as:

" 2 2
#
x() 1 0
F () = k , (2.26)
p
2
(0 ) + (20 )2
2 2
or: " #
2

x() 1
F () = m p( 2 2 )2 + (2 )2 , (2.27)

0 0
and the phase can be represented as:

20
= arctan 2 . (2.28)
0 2
2.3.2 Excitation
In general, the excitation applied on the structure should have sufficient magnitude and a
proper frequency bandwidth of interest. There are many choices for excitation. Two most
commonly used excitation techniques impact testing and shaker testing usually fulfill
the need for different types of excitation forces. A new technique called operational modal
analysis [30] has also been popular these years.
For impact testing, an impact hammer is used to generate a short-term excitation signal.
A series of selected points are hit by the hammer. The impact hammer can excite a wide
frequency range quickly and is the most widely used modal analysis technique due to its
operational flexibility and low cost. However, the crest factor (the peak-to-average ratio of a
waveform) in this technique is usually higher than for shaker testing, which will decrease the
signal-to-noise ratio and perhaps cause some non-linear behaviours in the structure. Also,
the repeatability of the measurements and potential double hits need to be considered,
although they are not problems in shaker testing.
Shaker testing can produce more types of excitation forces. The structure is vibrated
by the shaker at a series of selected points. This technique is often used in more complex
structures. Since several shakers can be used simultaneously as the sources of input, the
shaker excitation can be used for Multi-Input-Multi-Output(MIMO) analysis. However,
the test setup is more complex than for hammer testing. Usually more equipment and
channels are needed.
The operational modal analysis method uses the natural excitation of the structure
2 Background 20
itself. It can be done while the structure is in operation, so that no specific boundary
conditions are needed. However, only unscaled modal analysis data can be obtained, and
the calculation process is usually time-consuming.
Impact excitations can produce a transient impulse signal. The maximum amplitude
is at zero seconds. The bandwidth of the frequency and the duration of the impulse is
determined by the mass and stiffness of both the hammer and the structure. The range
of the frequency response is related to the tip material. A metal tip will have a higher
cut-off frequency than a rubber tip. To do the measurement, a force transducer needs
to be attached to the tip of the hammer. The measured force of an impact equals the
measured acceleration multiplied by the mass of the impactor behind the piezoelectric disc
of the transducer. The true force will be the measured force multiplied by the ratio of the
total mass over the mass behind the force transducer piezoelectric disc. Unlike the shaker
excitation, the hammer excitation does not add variable mass loadings on the structure.
Thus, it is particularly good for light structures. The change of the mass loadings from one
point to another will lead to modal frequency shifts at different measurement positions.
Given the lightweight nature of violins and their plates, the impact excitation is more
suitable than the shaker excitation.
2.3.3 Coherence Function
In order to get more accurate results for impact testing, it is better to hit each point
several times and take an average. Thus, the coherence function (as shown in Fig. 2.6) is
an important criteria for judging whether a set of tests is accurate enough. It is defined as:
|Dxy |2
Cxy = (2.29)
Dxx Dyy
where Dxy is the cross-spectral density (the Fourier transform of the cross-correlation func-
tion) between x and y, and Dxx and Dyy the auto-spectral density (which describes how
the power of a signal or time series is distributed over the different frequencies) of x and y
respectively.
Ideally, the coherence function should show a value of 1, which means the recorded
impact signal of each hit is exactly the same. However, in the real case, incoherence cannot
be avoided. There are usually several notches in the coherence function, usually at anti-
2 Background 21
Fig. 2.6 A coherence function in the test
resonances of the FRF. Although the notches in the coherence function are due to the
poor signal-to-noise ratio, no particular attention needs to be paid to this problem. By
simply taking the average of several hits, the curve should be smooth enough. Usually the
coherence is considered good if the value is higher than 0.95 at the resonances.
When choosing the test points, node points should be avoided. Hitting on a node point
can cause very poor coherence.
Doing the impact test manually will nevertheless produce some incoherence since it
is impossible to manually hit the plate exactly at the same point every time. In the
experiment, a pendulum system was used for doing the hit instead of using hands. The
pendulum can, to a great extent, avoid incoherence between two hits by providing consistent
impacts.
2.3.4 Windows
Compared to the response signal length, the impulse signal generated by the impact is
relatively short. Thus, a time window (a function that imposes zero values outside the
chosen interval) can be useful to minimize the electrical noise and the vibrational noise of
the hammer after the impact. In my experiments, the rectangular window was used for
the input signal and the exponential window was used for the output signal, which will be
discussed in detail in the later chapters. The window can effectively smooth the signal. A
double hit is another problem that should be avoided during the impact test. It is caused by
2 Background 22
either the lack of skill of the experimentalist, or the rebound of the structure. A double hit
can not be compensated by a window function. Although a double hit can be compensated
when computing the FRF, it can introduce some ripples as an artifact.
2.4 Composite Music Instruments
When choosing raw materials for making musical instruments, a variety of candidates are
available to accommodate different types of instruments. For string instruments, wood is
definitely the most commonly used material, from the very early masterpiece violins to the
recent factory-made instruments.
Composite materials are not conventional materials for making musical instruments.
However, with the development of material science, the application of composite materi-
als have been greatly expanded due to their extraordinary behaviour and their reasonable
prices. The appearance of composite musical instruments was the result of collaborations
between luthiers and material engineers. During the past several decades of research and ex-
ploration, they have managed to design and manufacture various kinds of high-performance
musical instruments with composite materials. Without sacrificing tonal qualities, these
instruments not only have a fancy appearance but also excellent material properties in
stiffness, durability, thermal expansion, and dimensional stability.
Charles Kaman was one of the pioneers in using composite materials for musical instru-
ment design. In the 1960s, he founded a company named Ovation Instruments which was
one of the first to design, make, and sell glass-fibre reinforced composite guitars. Later in
the 1970s, they also built guitars with carbon fibre top plates. After years of experiments
and modifications, they managed to reduce the thickness of the guitar top plate to 0.110
inches. The company believed that a thinner top plate would contribute to a better sound
quality (which is also believed by many violin makers). The company also made wood
necks reinforced with inserted carbon fibre layers.
Dr. John Decker was another carbon fibre guitar designer. He founded a company
called Rainsong Graphite Guitars. Since the mid 1990s, the company has made a series
of graphite guitars with a diverse array of body shapes and neck styles. They consid-
ered the sensitivities of woods to the temperature and the humidity, and tried to handle
these problems (which exist even in the best guitars) by applying different structures of
carbon/epoxy layers. Unlike Ovation Instruments, who makes roundback guitars, the
2 Background 23
Rainsong company uses traditional guitar body geometries. They believe that by applying
new materials to the traditional guitars, they can expand the sound volume range over a
traditional acoustic guitars limit.
In addition to their use in guitars, composite materials were also applied to other
instruments in the string family. In the 1990s, Luis Leguia founded a company called Luis
and Clark String Instruments, which manufactured a set of different string instruments
including violin, viola, cello, and bass. According to their report, carbon fibre cellos were
the hottest on the market. The choices of materials in different layers gave their cellos rich
reverberations. The dimensional stability was also one of the spotlights in their design.
As they said, the high tension at the bridge position usually leads to deformations of the
wood plate after many years of playing. Since the carbon fibre had a much higher stiffness,
the plate was more resistant to the tension, thus keeping the rich sound quality of the new
instrument.
Composite materials also have an application in wind instruments. One example would
be the very old traditional instrument: bagpipes. The traditional bagpipe tubes are made of
blackwood. Although this is one of the most ideal materials for making wood instruments,
it is quite sensitive to a change of temperature and humidity. Rob Gandara, a material
engineer, founded a company called Pipe Makers Union LLC which manufactured carbon
fibre bagpipes, flutes, and whistles. His products had good tonal qualities and stronger
radiation than the traditional wood ones. Also, by applying a tapering diameter, his flutes
had more powerful low registers than the commonly-seen silver alloy flutes.
Another company called Acrobaatti Oy/MATIT Flutes also has several flute products
with different composite materials. The founder Matti Kahonen discovered that the carbon
fibre flutes frequency spectrum range was richer than the traditional ones. The high
stiffness of carbon fibre might lead to less energy absorption compared to other materials
[31].
People at McGill University are also designing musical instruments with composite ma-
terials. Antoine Lefebvre from the Computational Acoustic Modeling Lab (CAML) made
some saxophone necks with carbon fibre. People from the Structures and Composites Lab-
oratory have spent several years exploring the possible applications of different composite
materials to musical instruments. A natural fibre ukulele was first made by Steven Phillips
in Professor Lessards group [22]. This innovative ukelele was moulded in one piece directly
and had good tonal quality. Several carbon fibre violin top plates were also made by dif-
2 Background 24
ferent layer combinations. Many of them were considered good by a luthier after being
assembled with a traditional body. Current projects in Lessards lab include designing a
carbon fibre guitar board and an Indian drum.
25
Chapter 3
FEM and Modal Analysis Set-up
This chapter discusses the set-up of the Finite Element Method (FEM) and the Modal
Analysis, which is used to obtain the FEM and Modal Analysis results for both the spruce
top plate and the composite plates in the next chapter.
3.1 Numerical Model of a Spruce Top Plate
A numerical model of a spruce top plate was developed. If the numerical model is accu-
rate enough, simple changes considering either the material properties or the geometrical
parameters can be made to evaluate design modifications. It is far more convenient and
economical to design using a numerical model than to change the geometries and material
properties of the prototype and manufacture a lot of pieces.
Several Computer-Aided-Design (CAD) models were built according to the geometry
and material parameters of the tested plates (Fig. 3.1). The first model was based on the
spruce plate originally crafted by Peter Purich, which had neither varnish nor a bass bar.
Other models, to be discussed in Chpt. 4, were based on composite plates, manufactured in
the Structures and Composite Materials Laboratory in McGill University. In this chapter,
the numerical simulation of the spruce top plate will be discussed. The prototype is shown
in Fig. 3.1.
2013/06/04
Fig. 3.1 The spruce top plate
3.1.1 CAD Model of a Top Plate
In order to build the FEM model, all the important geometrical parameters of the top
plate needed to be determined first. A commercial CAD software system, SolidWorks
2012, which was available in Professor Larry Lessards lab, was used for making the drafts
on computer.
Unlike the guitar top plate, which is flat and has a uniform thickness distribution, the
varying curvature of the violin top plate is hard to define. Small changes in the surface
geometries can lead to large differences in the sound. Since the spruce plate from the
luthier was cut and polished manually, no defined geometrical parameters were available.
As a result, an electronic digital caliper (Fowler, Model EDP 13501 54-100-024-1 RS232,
24) was used to measure several feature points on the top plate. The dimensions of this
top plate are defined in Table 3.1.
The arch height is also a very important characteristic feature of the top plate because
it has an important influence on the bending modes and the node line positions. The arch
height is measured by picking several feature points on the plate. The positions of the
selected points are shown in Fig. 3.2 and the arch heights of those points are shown in
Table 3.2.
The f-holes play an important role in the sound radiation of the body. Thus, the
Table 3.1 Width and length of the plate
Position of the Dimension (mm)

Measured Points
Total Length 358

Lower Bout Width 204.5
Upper Bout Width 165.5
Upper Corner Width 159
Lower Corner Width 182
Middle Line Width 108.7
f-hole Inner
Notches Distance 67.3
Table 3.2 A set of selected points for defining the arch height
Arch Heights of the Points Dimension (mm)

from Bottom to Top
Point 1 0
Point 2 4.0
Point 3 9.7
Point 4 12.1
Point 5 16.1
Point 6 18.7
Point 7 17.5
Point 8 14.2
Point 9 13.0
Point 10 6.8
Point 11 0
Fig. 3.2 A set of selected points for defining the arch height
positions and the shapes of them need to be defined as accurately as possible. However,
the complex shapes of the f-holes make it hard to define their geometrical parameters by
simple measurements. The geometries of the f-holes were borrowed from a previous model
made by Hossein Mansour in CAML. The geometries of the f-holes in that model were
copied into my model and projected on the top plate. Fine adjustments for the bridge
position and the distance between the two f-holes were then subsequently performed.
In my model, the top plate was divided into several surfaces, with each surface governed
by feature points and lines. In this manner, surfaces can be generated separately and the
curvature of each surface can be controlled more precisely. In this model, reference surfaces
were first drawn as shown in Fig. 3.3 (i.e. those surfaces with lighter colored lines). The
feature lines in each reference surface were drawn by using the feature points in Table 3.1
and Table 3.2. Then the surfaces were generated by using these feature lines. Only half of
the planes were first generated; the other half were simply drawn by reflecting the created
half plane. In this manner, the model is symmetric. Note that this would not work if the
plate has a bass bar.
Usually if a CAD draft for manufacturing is needed, we can make a 3-D model by
extracting the 2-D model with an assigned thickness. In this manner, the model will not
only look more realistic, but also contain the necessary geometry parameters for manufac-
turing. Different kinds of manufacturing methods can be defined based on the 3D model
in the CAD software. For instance, if we want to use a lathe to cut the prototype, the
manufacturing information, including the geometrical features and the cutting path, can
be saved in a file and exported to a Computer Numerical Control (CNC) machine. Then
the machine will automatically cut the material according to the defined path. However,
Fig. 3.3 CAD drawing of the spruce top plate
for the FEM, it is not necessary to define the geometry details of a model exhaustively. In
most cases a too-detailed CAD model can even produce negative effects for Finite Element
modeling because many minute features (for example, small thickness) either do not have
great influence on the vibrational behaviour or are hard to model. The geometry in these
areas is often too complicated if relatively coarse elements are needed to do the simulation.
Of course it would be possible to apply a lot of tiny elements to the complex-geometrical
areas, but the computing time will be increased significantly. One rule of thumb for doing
FEM modeling is to simplify the model as much as reasonably possible.
For very thin shell structures like the violin top plates, shell elements (which are 2-D
elements) are more often used than 3-D elements. In this case, it is not necessary to assign
a thickness to the model. A 2-D model is accurate enough for finding the important mode
parameters of the top plate.
3.1.2 Pre-Processing
The Finite Element package used for the simulation was ANSYS 10.0. It is very powerful
and accurate in solving dynamic structural problems, although its pre-processing module is
not very user-friendly. Thus, the pre-processing was done in another commercial software
package, HyperMesh, which shares interfaces with most of the mainstream CAD software
and Finite Element software. The CAD model was first imported from ProE to HyperMesh
in igs format. Then the model was preprocessed and was exported to ANSYS in cdb
format.
In FEM, the first thing to consider is choosing the right element type. Applying im-
proper elements on a model would give bad or even wrong results. In my model, shell
elements were used for the spruce top plate. Solid elements were also used for the bass bar,
which will be discussed in later chapters.
A shell element is the combination of the bending element and the membrane element.
At each node, the shell element has both translational DOFs and rotational DOFs in
each direction. When the structures thickness is far less than the other dimensions, shell
elements can usually give more accurate results than solid elements. The reason is that a
shell element has six DOFs at each node while a solid element only has three. One layer
of solid elements would not be fine enough to represent the strains in the wall direction.
Although accurate results can be obtained by applying enough layers of solid elements, the
computation time would probably be much longer than the time needed for shell elements.
ANSYS provides plenty of elements for the users convenience. For quadrilateral shell
elements, both 4-node and 8-node elements can be chosen. In our model, the 4-node element
is fine enough to calculate the modes. The one used in the simulation for the spruce plate
was Shell63, which has six DOFs at each node: three translational and three rotational
DOFs in the x, y, and z directions. Since the thickness can vary continuously for this
element, they were defined separately at each node.
The meshed model is shown in Fig. 3.4.
Fig. 3.4 Meshing of the spruce top plate
The meshing around the f-holes was made smaller than those at the upper bout and
the lower bout not only due to the complex geometry of the f-holes, but also due to the
important modal information in the central region of the violin top plate. The meshes at
the other parts on the top plate are relatively coarse in order to save computation time.
Excessively fine meshing in these regions is unnecessary since the modes are less sensitive
in the upper and lower bouts than at the centre area around the f-holes.
The meshes were intentionally made symmetric to ensure that the nodal displacements
and the nodal lines be symmetric. To do this, half of the plate was first meshed. Then the
meshing on this half plate were reflected along the y-axis. In this manner, the plot of the
calculated results would be purely symmetric (although in the real case, it is hard to make
symmetric nodal lines because the plate is not homogeneous).
The varying thickness of the plate was also taken into consideration. A Micrometer
(Fowler EDP 10641 522-229-201-0 0-1 MICROMETER .0001) was used to do the cali-
bration of the plate thickness. Several papers have discussed the influence of the thickness
on the natural frequencies of the violin top plate [18, 32]; it was proven that by varying
the thickness, the mode shapes can undergo changes. The meshes were divided into several
regions (called Components in HyperMesh) as shown in Fig. 3.5.
Fig. 3.5 Collectors of the top plate
The thickness corresponding to different colors are shown in Table 3.3. The thickest
areas are the center and the edges. At the upper bout and the lower bout area, the
thicknesses are relatively small. There are also transient areas between the center area
and the upper and lower bouts. Varying the thickness is important for modeling the plate
accurately.
The material properties also need to be defined. For the violin top plate, spruce is a
commonly used material, and maple is more often used in making the back plate. The
Table 3.3 Arch height at a set of selected points
Arch Heights of the Points Dimension (mm)

form Bottom to Top
Brown 4.0
Yellow 3.3
Blue 3.4
Red 3.8
Pink 4.0
material properties of spruce from a previous study of the top plate was used here [4]. The
results are shown in Table 3.4.
Table 3.4 Elastic parameters and density of the spruce
Parameters Value
(density) 420 Kgm3

Ex (Youngs Modulus in x direction) 15.13 GPa
Ey (Youngs Modulus in y direction) 1.2 GPa
Gxy (Shear Modulus in xy direction) 0.70 GPa
Gyz (Shear Modulus in yz direction) 0.059 GPa
Gxz (Shear Modulus in xz direction) 0.70 GPa
vxy (Poisson Ratio in xy direction) 0.3
vxz (Poisson Ratio in xz direction) 0.3
vyz (Poisson Ratio in yz direction) 0.3
Generally, spruce is an orthotropic material which has different stiffnesses in the longitu-
dinal and horizontal directions. The existence of grains is the cause of such difference. The
relationship between the stress and the strain can be expressed by using the 3-D Hookes
Law: 1 yx zx

xx Ex
Ey
Ez
0 0 0 xx
xy 1 zy
yy Ex Ez 0 0 0 yy

Ey
xz yz 1
0 0 0
zz Ex Ey Ez zz
= 1
. (3.1)
2yz 0 0 0 0 0 yz

Gyz
0 G1xz 0 zx
2
zx 0 0 0

2xy 0 0 0 0 0 G1 xy
xy
In orthotropic materials, the stiffness modulus in the z-direction can be assumed to be

the same as in the y-direction.
In terms of the boundary conditions, many previous studies on the guitar top plate
assumed that the board was clamped at the edge [12, 33, 34]. Since the guitar has a
relatively more regular geometry than the violin, the mode shapes are symmetric after
assembling the soundboard on the body. Also, the top and back board of the guitar
vibrates in phase. Fixed boundary conditions are assumed to predict the whole bodys
vibration patterns. However, the violin is a more complicated system. Neither the ribs
nor the soundpost is rigidly adhered to the top and back plates (due to the attach angles).
And due to the existence of the bass bar, the mode shapes are asymmetric. The sound
post can also introduce a phase difference between the top and back plate. Thus, fixing the
boundaries of the top plate cannot simply tell us the vibration patterns of the whole body.
In my experiment, since the aim is to compare the finite element simulation results with
modal analysis data of the plate, and since the boundary condition in the modal analysis
is free-free (which will be discussed in the next chapter), the boundary conditions in the
FEM model were also made free-free. Thus, the edges do not have any constraints on
their displacements.
There is no need to apply any load on the free plate since we are interested in the natural
frequencies of the top plate, which are its built-in properties. The idea of extracting the
normal modes involves solving the eigenvalue problem:
+ [K][X] = 0,
M [X] (3.2)
which can be transferred to:

([K] i2 [M ])[X] = 0, (3.3)
where the i are the natural frequencies we are interested in.
3.1.3 Normal Mode Extraction
The preprocessed model was transmitted to the commercial FEM package ANSYS. Several
matrix transformation algorithms are available in ANSYS for modal analysis, including the
Block Lanczos Method, the Subspace Method, the PowerDynamics Method, the Reduced
Method, the Unsymmetric Method, and the Damped Method. The Subspace Method
seems to be the most suitable algorithm for our problem. If the number of modes to
solve is relatively small (less than 40), this method can give faster calculations than other
methods. Also, this method is particularly suitable when the model consists of well-shaped
elements. The Subspace Method uses iterations to solve full stiffness and mass matrices.
It gives more accurate results than other methods.
It is not necessary to calculate all the modes for the top plate. Usually the natural
frequencies below 4000 Hz are of the most interest. High frequency normal modes are very
dense and difficult to differentiate. And many of them are also heavily coupled, making
the mode shapes difficult to distinguish from one to another. Another technique called
Statistical Energy Analysis (SEA) is more often applied for high frequency analysis.
3.2 Modal Analysis Set-up
In order to validate the results of the FEM simulation, an experimental test was conducted
on the spruce plate using the roving hammer method, as a comparison. Considering the
light weight of the top plate, the hammer is more suitable than the shaker as the excitation
source in our experiment.
A hammer test is also called an impact test. Lets take a look at an example. If there is
a structure with three points marked on it (as shown in Fig. 3.6), nine Frequency Response
Functions (FRFs) in total can be calculated. It is obvious that each FRF has a different
shape and that there are some peaks in one FRF which do not show up in another. If
hitting at one point, the response will not be the same at different points on the structure,
and vice versa. The reason is that if the excitation or response point happens to lie on the
nodal line of a certain mode, the input energy might not be well transferred to excite the
mode. Thus, by synchronizing several FRFs, we can avoid missing any important features
in the frequency response. Ideally, if a structure has n points, at most n2 FRFs can be
extracted by roving both the hammer and the accelerometer to every point. However, such
exhaustive measurements are usually unnecessary. Taking a look at Fig. 3.6, it is obvious
that a synchronization of any row or column (by overlapping the FRFs, e.g. h31 + h32 +
h33 or h11 + h12 + h13) can give similar results. This means that if only one row or one
column of the FRFs is chosen, they will probably be enough to accurately represent the
vibration behaviour of the whole structure.
In practice, people choose either to rove the hammer or the accelerometer. There is
Fig. 3.6 Typical setup for a roving impact hammer test on a cantilever beam
[3]
not much difference between these two choices. The idea is to choose either the row or
the column in the FRF matrix. As shown in Fig. 3.6, a roving hammer test with the
accelerometer fixed at point-3 is done by choosing the third row of the matrix. In my
experiment, repositioning an accelerometer could be a time-consuming process because it
was attached with wax. So only the hammer was roved, which was faster and easier to
perform.
3.2.1 Experiment Set-up
The experiment was done in the Spatial Audio Lab in the Centre for Interdisciplinary
Research in Music Media and Technology (CIRMMT) at McGill University. There are sev-
eral commonly used boundary conditions for modal analysis, including free-free, fixed and
mass-added boundary conditions. Among these choices, the free-free boundary condition
is most commonly used, not only because it is the most easy to handle in practice, but also
because it best describes the dynamic behaviour of a single component without assembly.
In my test, the free-free boundary condition was applied on the top plate. To achieve
this, a nylon string was strung over a round beam near the ceiling and then attached to
the top of the plate by glue. The string was intentionally made very long for two reasons.
First, using very long strings allows the top plate to be close to the ground, allowing hits
to be made more easily. Second, since the movement of the swinging plate on the string
is an artifact of the setup, a long length of the string can decrease these motions to very
1
low frequencies. Generally, if the swinging frequencies are less than 10 of the lowest body
mode, they will not affect the modal information extraction. Since the frequency of the
first mode of the top plate is usually about 80-100 Hz, these movements at low frequencies
can be simply filtered out as noise. The whole setup of the experiment is shown in Fig.
3.7.
Fig. 3.7 The experiment set-up
At each selected test point, an average was taken by hitting the point several times.
In a typical modal test, people often use a hand to hold the hammer and hit the selected
points manually. However, this will nevertheless introduce some differences between each
hit.
In order to make each hit each time as similar as possible, a pendulum system was
introduced in our experiment. The top of the hammer handle was clamped in the pendulum.
The rotational axis of the pendulum was perpendicular to the hit direction of the tip. The
pendulum was clamped on a chemistry stand, so that it could be easily moved to different
heights. The pendulum system is shown in Fig. 3.8.
Fig. 3.8 The pendulum system
3.2.2 The Choice of the Hammer
The hammer was also carefully chosen. Several hammers with different sizes are available
in our lab. A light hammer should be used to minimize mass loading and to achieve a
larger bandwidth of the result. The one used in the test was a PCB Piezotronics Miniature
Impact Hammer Model 086E80. The hammer is shown in Fig. 3.9.
This hammer is the smallest in the lab, having a tip diameter of 2.5 mm and a length
of 107 mm. The size is proper for providing enough impact energy on the plate, so that
Fig. 3.9 Impact Hammer Model 086E80
the modes can be well excited. A force sensor is pre-installed behind the tip, in order to
record the impulse signal. The specifications of this hammer are shown in Table 3.5.
There are both rubber and steel tips available for this hammer. A rubber tip is obviously
softer than a steel tip, which could generate a more gentle impact on the structure.
However, using a soft tip made of rubber would lead to a lower cut-off frequency than
using a steel tip. In some previous modal analysis research on the whole violin, people used
plastic tips. George Bissinger explained this as a means to protect the violin, since many
of their specimens were quite expensive (a few were even Cremona violins). The steel tip
could easily leave some indentations on the spruce-made plate. Unfortunately, using soft
tips sacrifices the range of the frequency response. Although we are mostly interested in
a relatively low frequency range (i.e. probably below 4000 Hz) for the violin top plate, it
is still necessary to widen the excitation frequency range as much as possible. In my test,
the price of the plate was not a concern, thus the steel tip was chosen. In this case, a high
cut-off frequency was ensured. It should be noted that the plate was not damaged during
the measurements.
The impact signal was recorded in voltages and was transformed to a signal conditioner
before being processed by the software. The gain of the signal conditioner which connected
to the hammer was set to 1.
Table 3.5 Specifications of the hammer
PERFORMANCE Unit
Sensitivity(20%) 22.5 mV/N

Measurement Range 222 N
Resonant Frequency 100 kHz
Non-Linearity 1%
ELECTRICAL
Excitation Voltage 20 to 30 VDC

Constant Current Excitation 2 to 20 mA
Output Impedance 100 Ohm
Output Bias Voltage 8 to 14 VDC
Discharge Time Constant 100 sec
PHYSICAL
Sensing Element Quartz

Sealing Epoxy
Hammer Mass 4.8 g
Head Diameter 6.3 mm
Tip Diameter 2.5 mm
Hammer Length 107 mm
Electrical Connection Position Side
Extender Mass Weight 1.25 gm
Electrical Connector 5 44 Coaxial
3.2.3 Accelerometer
The accelerometer measures the acceleration of the excited structure. Similar to the impact
hammer, an accelerometer generates an output signal in the form of voltage. A signal
conditioner was also used to connect the accelerometer and the computer in order to transfer
the analog output signal to digital form. The gain of the signal conditioner was set to 100,
so that the frequency response plot would have the same scale as in previous papers.
The accelerometer used in the testing was a PCB Piezotronics Shear Accelerometer
Model 352C23/NC Serial LW147466, as shown in Fig. 3.10. The specifications of this
accelerometer are shown in Table 3.6.
Fig. 3.10 Accelerometer Model 352C23/NC
Note that the accelerometer should be able to faithfully record the acceleration of the
measured structure without distorting the amplitude and phase shift. However, if the
structure has some nonlinear properties, they will also be recorded by the accelerometer.
3.2.4 Selection of the Points
For the roving hammer test, the idea is to fix the position of the accelerometer at one point
and use the hammer to hit at a set of selected points. In my experiment, 43 points in total
were picked from the nodes on the Finite Element model, as shown in Fig. 3.11. Since
Table 3.6 Specifications of the accelerometer
PERFORMANCE Unit
Sensitivity (20%) 0.5 mV/(m/s2 )

Measurement Range 9810 m/s2
Frequency Range (5%) 2.0 to 10, 000 Hz
Frequency Range (10%) 1.5 to 15, 000 Hz
Frequency Range (3dB) 0.7 to 25, 000 Hz
Resonant Frequency 70 kHz
Broadband Resolution (1 to 10, 000 Hz) 0.03 m/s2 rms
Non-Linearity 1%
Transverse Sensitivity 5%
ENVIRONMENTAL
Overload Limit (Shock) 98,000 m/s2

Temperature Range (Operating) 54 to +121C
ELECTRICAL
Excitation Voltage 18 to 30 VDC

Constant Current Excitation 2 to 20 mA
Output Impedance 200 Ohm
Output Bias Voltage 7 to 11 VDC
Discharge Time Constant 0.24 to 1.0 s
Settling Time (within 10% of bias) <3s

Spectral Noise (1 Hz) 14, 715 ( m/s2 )/ Hz



Spectral Noise (1 kHz) 294 ( m/s2 )/ Hz

Spectral Noise (10 kHz) 196 ( m/s2 )/ Hz
Electrical Isolation (Base) > 108 Ohm
PHYSICAL
Size (Height x Length x Width) 2.8 mm x 8.6 mm x 4.1 mm

Weight 0.2 g
Sensing Element Ceramic
Sensing Geometry Shear
Housing Material Anodized Aluminum
Sealing Epoxy
Electrical Connector 3 56 Coaxial Jack
Electrical Connection Position Side
Mounting Adhesive
the centre area has a complex geometry and would probably contain more mode features
than the edges (which was verified by the Finite Element results), more points were chosen
there. The points were also made symmetric to the y-axis, so that the mode shapes and
nodal lines would also be symmetric when synchronizing the mode shapes and making the
animations. A full-scale plot of the plate was printed out, and the points were mapped to
the real plate, as shown in Fig. 3.11 (the red points) and Fig. 3.12. The marked points
were used as references for the impact hammer.
Fig. 3.11 Selected hitting points
3.2.5 Data Acquisition
The data acquisition was performed by using a National Instruments PCI-4472 8-channel
Dynamic Signal Acquisition Board which was pre-installed in a PC. Since the roving ham-
mer method was used, only two channels were needed. The hammer was connected to the
input channel and the accelerator was connected to the first output channel. The board
can be recognized by the modal analysis software MEScope 5.
At the design stage, several data acquisition methods were taken into consideration.
Other choices included using a National Instruments USB-4432 4-channel Dynamic Signal
Acquisition Board with Matlab code developed by Professor Gary Scavone, or using a
Fig. 3.12 Marked points on the spruce plate
sound card with a software package by George Stoppani. The advantage of using a USB-
4432 card is that it can be easily connected to any PC or laptop. However, from a software
perspective, only the Matlab code supports that card. Also, the data from Matlab would
need further treatments. In order to animate the mode shapes, it is necessary to manually
edit all the frequency response function data to a .uff file and then transmit the edited
file to another modal analysis software system. Stoppanis software is able to acquire the
data and do the synthesis of the mode shapes at the same time. However, the animation
functions in that software are relatively weaker than MEScope.
Considering all these factors, the PCI-4472 Board was the final choice. The hammer
and the accelerometer were connected to the first two channels of the board. The excitation
and the response signals were transferred as voltage changes by the PCI board and then
fed into the software for calculating the frequency response.
3.2.6 Data Analysis
The commercial software MEScope was used to calculate the frequency response functions
and to synthesize the mode data, including the mode shapes, the natural frequencies, and
the damping.
The frequency response function of each hit was calculated simultaneously upon receipt
of the impact signal. In order to minimize differences between hits, an average of 4 hits was
made. The average was automatically updated after each hit, and the coherence function
was also computed.
An ideal coherence function would have a value of 1 throughout the whole range of the
frequency spectrum (if each hit was exactly the same). However, this is not achievable in a
real experiment. There are often some notches on the coherence function, at points where
the frequency response function also has notches. These notches are usually acceptable.
However, if there are too many notches at other places, the data are considered bad. In
my experiment, the plate was suspended by a string and each hit made the plate move a
bit in different directions. In order to decrease the difference in the coherence function, it
was necessary to wait a few seconds for the plate movement to dampen to an acceptable
degree. Another option was to take some measures to increase the damping of the plate.
In my experiment, another string was attached to the long string in order to constrain the
movements of the long string. By doing this, the plate would thus move in a smaller range
and return to a stable status faster. The constraining string is shown in Fig. 3.13.
One of the averaged FRFs and the coherence function in the test is shown in Fig. 3.14.
As shown in the graph, the coherence function was quite smooth. Despite a few notches in
the coherence, the majority of the values were close to 1. Thus, the hits at this point were
considered good.
3.2.7 Modal Information Detection
After measurements at all the 43 selected points, 43 FRFs were generated. The next
step was to synchronize them and extract the modal information. If all the FRFs were
overlapped, several peaks would be seen on the graph, as shown in Fig. 3.15:
Important mode features can be extracted from these FRFs. The peaks on the over-
lapped graph are the natural frequencies of the structure. At each natural frequency, a
certain vibration pattern of the plate, called the mode shape at that frequency, would
be anticipated. The mode shape of a certain natural frequency can be calculated by using
the amplitude of each FRF at that frequency. For instance, if we want to extract the mode
shape of the second mode of a plate, as shown in Fig. 3.16, we can simply pick the ampli-
tudes of the second natural frequency of each FRF and assign them to the corresponding
node. Since these amplitudes are different in each FRF, the plate will be defined into a
Fig. 3.13 Experimental Setup #2

Fig. 3.14 The FRF and the coherence function of point 2
Fig. 3.15 Overlapped FRFs

physically meaningful shape. In Fig. 3.16, the plate is twisted. Thus the second mode of
this plate is a twist mode.
Fig. 3.16 Plate mode shape for mode 2 [3]
More intuitively, since the idea of the impact test was to hit each point and generate a
corresponding FRF, the amplitudes can be considered as the response of each hit. Recall the
Chladni method which stated that if the plate is governed by the second natural frequency,
some areas would have large displacements (or to say, higher sensitivities) while some
would be stationary. For the impact test, if a point in the sensitive area is hit, more
vibrations of the second mode would be excited opposed to hitting a point in the stationary
area. From the frequency spectrum, we should also be able to see large amplitudes of
the second natural frequency at the points in the sensitive area and relatively small
amplitudes at the points in the stationary area. Thus, the amplitudes of the second natural
frequency in each FRF can be considered as the displacements of each corresponding node
at that frequency. In this manner, the mode shapes of each mode can be simply plotted.
In Modal Analysis, the damping values of each mode were also estimated, and it is
actually a task which would have been difficult to perform in FEM. The damping values
of the spruce top plate were also calculated simultaneously once the peaks were detected.
The damping ratio at each natural frequency can be expressed as:
3dB
= (3.4)
20
where 3dB is the bandwidth of the 3dB decay of the peak, and 0 is the frequency value
at the peak. For a good violin top plate, small damping ratios should be anticipated, so
that the sound would not be too dry. Comparison of the spruce plate and the carbon
fibre plate will be shown and discussed in Chapter 4 and Chapter 5.
The inverse of the damping ratio can also be used, which is called the quality factor (or
Q factor) to describe the damping properties of a structure.
0
Q= (3.5)
3dB
It should be noted that, under usual circumstances, there should not be a large disparity
in the modal damping values unless there are some special damping mechanisms that affect
one mode more than others. A large difference between two close modal dampings may be
due to the curve-fitting error.
49
Chapter 4
Comparison of FEM and Modal

Analysis Results
4.1 FEM Results of the Spruce Plate
The calculated natural frequencies are shown in Table 4.1. The mode shapes of the three
most prominent modes, Mode 1, Mode 2, Mode 5, are shown in Figs. 4.1 4.3, with a
comparison to the Chladni Method results by the luthier Peter Purich. As can be seen, the
two set of results are in general agreement.
Table 4.1 Natural frequencies of the spruce top plate by FEM
Mode 1 2 3 4 5 6
Frequency 92.401 144.92 193.99 225.25 283.70 338.39
Mode 7 8 9 10 11 12
Frequency 407.48 430.80 468.22 552.66 566.03 672.37
Mode 13 14 15 16 17 18
Frequency 686.34 747.53 810.25 835.18 859.34 969.06
The FEM results also correspond well to those from previous papers [4, 6]. Thus, it
can be proven that the FEM model of the spruce plate is reasonably accurate. The results
calculated by FEM will also be compared to the Modal Analysis results in Section 4.2.
2013/06/04
Fig. 4.1 Mode 1 of the spruce plate by the FEM and the Chladni Method
4.2 Modal Analysis Results of the Spruce Plate
In MEScope, the software can detect the peaks of each FRF and generate the mode shapes
automatically. The first six modes of the spruce plate as determined by the Modal Analysis
are shown in Figs. 4.4 4.9, in a comparison to the FEM results calculated previously.
The mode shapes on the left side are calculated by the FEM, and the mode shapes on the
right side are determined by the Modal Analysis.
Fig. 4.4 Comparison of Mode 1 of the spruce plate by the FEM and the
Modal Analysis
Modal Analysis
Modal Analysis
A list of all the natural frequencies and damping values of the modes below 1000 Hz by
Modal Analysis is shown in Table 4.2:
4.3 Carbon Fibre Top Plate
Possible geometry variations of the violin have been explored for centuries. Although the
current shape of the violin might already be optimal, there is still a lot of space for modifying
the material properties of the violin. Composite materials are a possible alternative choice
for making the violin. Considering the complex geometry of the top plate, the wood crafting
Modal Analysis
Modal Analysis
Modal Analysis
Table 4.2 Natural frequencies of the spruce top plate by modal analysis
Mode 1 2 3 4 5 6
Frequency 87.2 140 236 242 303 370
Damping 0.579 0.97 1.65 1.57 1.94 2.58
Mode 7 8 9 10 1 12
Frequency 381 404 501 528 558 631
Damping 2.81 2.96 3.41 3.39 3.91 4.14
Mode 13 14 15 16 17 18
Frequency 648 695 819 851 935 964
Damping 4.43 4.51 6.28 5.3 2.75 7.46

process often requires high expertise and great patience. Therefore, composite materials
could be an alternative to wood because they are easier to handle during fabrication. They
are also less sensitive to humidity and temperature change compared to wood, which makes
the instrument more stable in different environments.
4.3.1 Choice of the Material
Many carbon fibre violin top plates were previously made in the Structures and Composites
Laboratory at McGill University. As mentioned in a report from Professor Lessards group,
they paid attention to both the material behaviour and the manufacturing process when
choosing the proper materials [35]. Many designs have shown good material properties but
were hard to manufacture. The pieces made in this project all have sandwich structures
with different materials for each layer. Some of them developed cracks, which are obviously
undesirable for commercial products. The two plates from Professor Lessards group were
free of fabrication defects. The first one (we will call it plate #1 hereafter) was made up
of three layers. The top and bottom layers were made of MTM45-1 epoxy matrix woven
pre-preg, and the core layer was made of balsa. The second one (we will call it plate #2
hereafter) was composed of seven layers of Newport 301 unidirectional carbon fibre pre-preg
and a urethane ring, as shown in Fig. 4.10. Plate #1 had a bassbar assembled on it while
the spruce plate and composite Plate #2 did not.
Fig. 4.10 Components of the second composite plate
Pre-preg is a terminology referring to pre-impregnated composite material. For this

type of material, the epoxy matrix is impregnated in advance with either woven or uni-
directional composite fibres. The epoxy is partially cured so that the material still has a
certain degree of durability and flexibility so that it can be turned into the desired shape
during manufacturing. The pre-preg can be preserved in cold temperatures for a long time.
When using pre-preg material, we first cut it into the desired shape, and then put it in the
oven for several hours to fully cure it. The cure time depends on the temperature. For
instance, for MTM45-1, which was used to make plate #1, the cure time is about 20 hours
at 80o C, 4 hours at 120o C and 2 hours at 130o C.
The whole process of making a composite plate is shown in Figs. 4.11-4.14. First, all
the layers were cut and put in an aluminum mould which was CNC-machined according to
the CAD drawing. Then a vacuum bag was put over the mould to extract the air inside
and provide pressure (Fig. 4.11 and Fig. 4.12). Following the extraction, the mould and
the materials were sent into the oven together for full curing (Fig. 4.13). After waiting for
the desired time, the final prototype was ready. The violin plate was then removed from
the mold and the edges trimmed slightly to its final size (Fig. 4.14).
Fig. 4.11 Stage 1 of making a com- Fig. 4.12 Stage 2 of making a com-
posite plate (layers cut and put in posite plate (air extracted)
mould)
One problem of the early prototypes was that they had sharp edges which often cut the
players neck. In an attempt to mitigate this problem, a urethane ring was also made and
assembled to the top plate in the later prototypes, as shown in Fig. 4.15 and Fig. 4.16.
4.4 Numerical Simulation of the Composite Top Plates
The Finite Element Method (FEM) was applied to determine the natural frequencies and
mode shapes of the two composite plates. The modeling process was quite similar to the
Fig. 4.13 Stage 3 of making a com- Fig. 4.14 Stage 4 of making a com-
posite plate (vacuum attached) posite plate (after curing)
Fig. 4.15 Stage 1 of making the ure- Fig. 4.16 Stage 2 of making the ure-
thane ring thane ring
spruce plate. The structure was first meshed, then the model was transmitted to the solver
for calculation and the mode shapes were plotted finally.
The CAD model of the two plates, which was given by Professor Larry Lessard, was
used to make the CNC mould. Since this model did not have f-holes in it, the f-holes
were added from the spruce plate model which I had built previously and the new model
was transmitted into HyperMesh for FEM Pre-Processing. Modeling Plate #1 was rather
difficult because the plate consisted of several different materials. In general, the plate had
a sandwich structure, with MTM45-1 weave carbon fibre pre-preg material on the top and
the bottom, and a layer of wood as the core. Balsa was used around the edges and spruce
was used in the central area. A bass bar, which was made of spruce, was also attached to
it.
Fig. 4.17 Components of the composite plate #1
As shown in Figure 4.17, the model was first divided into several regions and named
as components in HyperMesh. Each component was assigned with different properties.
On the top plate, Shell99, a 2D 8-Node-48-DOF element, was used to model composite
materials. This element allows the input material properties to be either in matrix form
or in layer form. In my modeling, the layer form was used. We can define the material
properties and the thickness of each layer separately. The yellow component contains three
layers (MTM45-1: 0.4 mm; Balsa: 4.5 mm; MTM45-1: 0.4 mm), the green component
contains two layers (MTM45-1: 0.4 mm; MTM45-1: 0.4 mm), and the blue component
also contains three layers (MTM45-1: 0.4 mm; spruce: 4.5 mm: MTM45-1: 0.4 mm). The
properties of these materials are shown in Table 4.3.
The bass bar was modeled with Solid45 Elements, which are 3D 8-Node-24-DOF ele-
ments. Spruce was the material of the bass bar. The joints of the bass bar and the plate
were forced to have the same DOFs so that the bass bar could be considered rigidly glued
Table 4.3 Material properties of spruce, balsa and MTM45-1
Spruce Balsa MTM45-1
Density (kg/m3 ) 420 170 1600
Ex (MPa) 15130 3900 58530
Ey (MPa) 1200 600 54500
vxy 0.3 0.4 0.3
vxz 0.03 023 0.4
vyz 0.3 0.23 0.3
Gxy (MPa) 700 500 2600
Gyz (MPa) 590 300 2600
Gxz (MPa) 590 300 2600

on the plate. After meshing the structure and defining all the properties, it was transmitted
to ANSYS for mode extraction. The calculated natural frequencies of this plate are shown
in Table 4.4. The first six calculated mode shapes by the FEM are shown in Fig. 4.18
4.23, with a comparison to those by the Modal Analysis. The calculated mode shapes had
smaller amplitudes than the spruce plate, which may have been due to the presence of the
bass bar as well as a large stiffness difference between layers. Also, the natural frequencies
were in general higher than those of the spruce plate. These patterns can also be seen in the
experimental results, which will be discussed in the next section. More detailed discussions
about these results will be provided in the next chapter.
Table 4.4 Natural frequencies of the composite plate #1 by FEM
Mode 1 2 3 4 5 6
Frequency 77.5 149 198 269 293 358
Mode 7 8 9 10 1 12
Frequency 375 436 498 520 547 620
Mode 13 14 15 16 17 18
Frequency 647 661 687 740 759 816
Fig. 4.18 Comparison of Mode 1 of the composite plate #1 by the FEM

and the Modal Analysis





The simulation of Plate #2 was similar to Plate #1. As shown in Fig. 4.24, the blue
component was made of seven layers of Newport301 unidirectional carbon fibre pre-preg
material at the central part. Each layer was positioned at a different angle, as shown in Fig.
4.10. The red component contained only two layers of Newport301 pre-preg material, posi-
tioned in zero angle to each other. The yellow component marked where the urethane ring
was positioned. Since there was no bass bar on this plate, Shell99 Elements was used only.
By defining different layers, angles, and material properties for each component respec-
tively, the natural frequencies can be calculated. The material properties of Newport301
and the urethane provided by the manufacturer are shown in Table 4.5. Only one Shear
Modulus value (G) is provided by the manufacturer.
Fig. 4.24 Components of the composite plate#2
The calculated natural frequencies are shown in Table 4.6. The first six calculated mode
shapes by the FEM are shown in Fig. 4.25 4.30, with a comparison to those by the Modal
Table 4.5 Material properties of Newport301
Newport301 Urethane
Density (kg/m3 ) 1227 750
Ex (MPa) 131000 7600
Ey (MPa) 8963 7600
vxy 0.3 0.25
vxz 0.3 0.25
vyz 0.3 0.25
G (MPa) 3448 3200
Analysis.
Table 4.6 Natural frequencies of the composite plate #2 by the FEM
Mode 1 2 3 4 5 6
Frequency 49.8 108 145 183 202 249
Mode 7 8 9 10 1 12
Frequency 334 357 446 469 542 558
Mode 13 14 15 16 17 18
Frequency 648 711 800 872 907 944
4.5 Modal Analysis of the Composite Top Plates
The Modal Analysis process of the composite plates was similar to that used for the spruce
plate. The test of Plate #1 was done in the Spatial Audio Lab in the Center for Interdis-






ciplinary Research in Music Media and Technology (CIRMMT) in McGill University. The
equipment setup was the same as for the spruce plate, for which a roving hammer was used
to hit 43 points on the plate. The synchronized mode shapes are shown in Fig. 4.18 4.23
and a list of all the natural frequencies and damping values below 1000 Hz calculated by
Modal Analysis is shown in Table 4.7.
Table 4.7 Natural frequencies of the composite plate #1 by the modal anal-
ysis
Mode 1 2 3 4 5 6
Frequency 123 196 291 333 421 432
Damping 0.839 2.81 2.91 5.45 3.92 6.28
Mode 7 8 9 10 1 12
Frequency 452 532 625 692 704 710
Damping 4.07 8.54 3.37 2.06 0.663 0.284
Mode 13 14 15 16 17 18
Frequency 716 722 727 734 828 842
Damping 0.21 1.39 0.701 0.0475 0.592 1.28
The Modal Analysis of Plate #2 was not done at the same time as the analysis for
Plate #1. Since the results of Plate #1 were quite different from the spruce plate (it was
considered to have a bad tonal quality by the luthier), we decided to borrow Plate #2,
the good tonal quality plate recommended by the luthier. Since we wanted to test the
structural vibrations of the plate itself, the room condition was not really important. As
Professor Gary Scavone suggested, the test was done in CAML this time, where most of
the equipment was housed. The synchronized mode shapes are shown in Fig. 4.25 4.30
and a list of all the natural frequencies and damping ratios below 1000 Hz calculated by
Modal Analysis is shown in Table 4.6.
Table 4.8 Natural frequencies of the composite plate #2 by the modal anal-
ysis
Mode 1 2 3 4 5 6
Frequency Missing 86.2 170 236 273 377
Damping Missing 0.362 0.844 2.27 2.29 1.48
Mode 7 8 9 10 1 12
Frequency 401 450 493 532 594 678
Damping 3.20 2.38 3.76 2.87 2.36 2.54
Mode 13 14 15 16 17 18
Frequency 732 774 804 898 937 966
Damping 3.82 3.93 4.51 5.60 7.46 5.84

70
Chapter 5
Discussion
5.1 FEM Results of the Spruce Plate
As discussed in Chapter 1, many researchers have studied the vibrational behavior of the
violin top plate by using either FEM or Modal Analysis [9, 10, 5, 11, 4, 13, 14, 15, 16, 17].
All these studies have considered the vibrational behaviour of the traditional wooden plate.
The results can be used as a comparison to the results of the spruce plate. In general, the
vibration of the violin top plate contains both twisting and bending modes. The mode
shapes all have similar patterns in the reference papers, although the natural frequencies
differ. The results of the spruce plates FEM simulation corresponds well with previous
research.
As shown in Fig. 5.1 and 5.2, the first mode of the violin plate is a twisting mode,
with two perpendicular nodal lines across the centre. The second mode is a bending mode,
with two parallel nodal lines aligned in the longitudinal direction. The mode shapes of the
third mode and the fourth mode are interchangeable for different plates. As can be seen
here, the third mode shape in Fig. 5.1 is similar to the fourth mode shape in Fig. 5.2, and
the fourth mode shape in Fig. 5.2 is similar to the third mode shape in Fig. 5.1. Similar
patterns can also be found in my simulation results, in which the order of these two modes
often change when changing the material properties and plate geometries. The fifth mode
is usually called the ring mode because its nodal line has the shape of a ring. Many
researchers and luthiers believe that this mode is one of the most influential on the tonal
quality and that it should have an octave relation with the second mode for a good plate.
There is a more intuitive nomenclature for the mode shapes. The first mode, as shown
2013/06/04
5 Discussion 71
in Figs. 5.1 5.3, can be called the (1,1) plate mode, which means there is one node in both
the transverse and longitudinal direction. Similarly, the second mode can be called the (2,2)
plate mode. The third and the fourth modes correspond to the (3,2) and (4,1) plate modes,
respectively. The numbers of the modes represent the half-wavelengths contained in each
direction. For instance, the (1,1) mode contains half a wavelength in both longitudinal and
transverse directions, while the (2,2) mode contains one wavelength in both longitudinal
and transverse directions. This can explain why the natural frequency of the (2,1) mode is
higher than the (1,1) mode, because the wavelength is one time shorter in the transverse
direction in the (2,1) mode than in the (1,1) mode.
The fifth mode can be called (1) ring mode, the vibrational behaviour of which is
different from the other plate modes. Higher ring modes should also be anticipated in the
high frequency range, where more rings are present in other mode shapes.
Figure 5.3 shows the FEM simulation results of the spruce plate. The mode shapes
correspond very well with previous studies [4, 5]. The calculated natural frequencies are
shown in Table 5.1. The value differ a lot from paper to paper. Generally, the natural
frequency of the first mode lies between 60-100 Hz. If the calculated natural frequencies
are extraordinarily high or close to each other, then there may be errors in the simulation.
If a model gives reasonable results, we can then simply adjust the material properties and
the geometries to modify the design, which would be much faster and easier to manage
than manually crafting the prototype.
Fig. 5.1 The first six modes of the spruce plate by the FEM (Bretos) [4]
5 Discussion 72
Fig. 5.2 The first six modes of the spruce plate by the FEM (Molin) [5]
Fig. 5.3 The first six modes of the spruce plate by the FEM
5 Discussion 73
Table 5.1 Comparison of the natural frequencies by the FEM of the spruce
plate with previous research
Spruce Bretos[4] Molin[5]
Mode 1 92.4 86.1 106
Mode 2 144.9 168.2 196
Mode 3 194.0 252.8 339
Mode 4 225.3 281.3 343
Mode 5 283.7 346.2 471
Mode 6 338.4 403.5
5.2 Comparison of the FEM Results to the Modal Analysis

Results of the Spruce Plate
The FEM is powerful for predicting a structures behaviour. However, people might doubt
whether the results are convincing enough, since the modeling process involves assumptions
and any approximation or simplification of the physical condition might cause errors. Al-
though the FEM results for the spruce plate already fit the previous research studies quite
well, the modal analysis of the plate could help further validate the simulation results. Figs.
3.20 - 3.25 show the comparison of the first six modes of the spruce plate by FEM and
Modal Analysis. The positive and negative signs are used to point out the moving direction
of the plate. The natural frequencies of the first six modes are shown in Table 5.2. Since
most of the important conclusions of violin acoustics are based on the low frequency range
(and no previous papers have shown the plate mode shapes up to the sixth mode), the
scope here is limited to the first six modes for comparison. As evident, both the mode
shapes and the natural frequencies fit quite well comparing the two.
Now the numerical model can be considered accurate enough. We can use it as a basis
for design experiments.
5 Discussion 74
Table 5.2 Comparison of the natural frequencies by the FEM and the modal
analysis of the spruce plate
FEM Modal Analysis
Mode 1 92.4 87.2
Mode 2 144.9 140
Mode 3 194.0 236
Mode 4 225.3 242
Mode 5 283.7 303
Mode 6 338.4 370
5.3 Vibration and Damping of the Composite Plate #1
Composite plate #1, which was made by Professor Larry Lessards group at an early stage
in their experiments with violin plates, had spruce and balsa at the core. During that stage,
they mainly explored material combinations of the violin top plate which would have higher
stiffness and less weight than the traditional plate. His student, Gyu-hyeong Lim, found
that those plates with carbon fibre outer layers and a wooden core (including the one we
used in the test) can achieve the same stiffness as the traditional wooden plate with a much
thinner thickness and less weight [36].
Although the plate had extraordinary mechanical behaviour, the acoustical properties
were not as desired. As judged by the luthier, Peter Purich, this plate did not produce
satisfactory sounds when assembled on the violin body. Thus, one of our goals was to find
the reason behind it.
The first six mode shapes of this plate, calculated by both FEM and Modal Analysis,
was first compared, as shown in Fig. 4.9 4.16. Due to the presence of the bass bar,
the mode shapes are asymmetric . This might not be too obvious in the Modal Analysis
results. However, when we view the mode shapes by FEM and then compare them with
those by Modal Analysis, we are able to see that they fit quite well and are all biased to
one side.
5 Discussion 75
The presence of the bass bar not only makes the plate asymmetric, but it also increases
the stiffness of the plate. This could probably explain why the natural frequencies of this
plate were generally higher than those of the spruce plate, as shown in Table 5.3. My
previous experimental FEM test showed that the presence of the bass bar could increase
the natural frequency of each mode by at least 10%. However, each mode showed different
sensitivities to the presence of the bass bar. Some modes (e.g. the second and the third
mode) had an increase in natural frequency of over 30%. From my point of view, the
sensitivity of a mode is due to the mode shape itself. If the bass bar happens to stand
near the nodal line (where all the points are almost stationary), the mode shape would not
change much. In this case, the natural frequency would also experience a relatively small
increase. However, if the bass bar happens to be near the antinodes of the mode (where
maximum displacements occur), the mode shape may change a lot, hence producing even
more influence on the corresponding natural frequency.
analysis of the spruce plate and the composite plate #1
FEM of Modal Analysis FEM of the Modal Analysis

the Spruce of the Spruce First Composite of the First
Plate Plate Plate Composite Plate
Mode 1 92.4 87.2 101 123
Mode 2 144.9 140 265 196
Mode 3 194.0 236 340 291
Mode 4 225.3 242 360 333
Mode 5 283.7 303 375 421
Mode 6 338.4 370 509 452
Another notable difference for this plate is the damping ratio, as shown in Table 5.4. In
general, the damping ratios of each mode on this plate are much higher than those of the
spruce plate (some of them are even more than twice that of the spruce plate). We usually
desire small damping ratios so that the violin can be more easily excited to produce more
resonances in the plate vibration. Large damping ratios would make the vibration decay
fast. The sound would be perceived as dry, a quality which is not desired for a musical
5 Discussion 76
instrument. In fact, tapping on this plate could clearly reveal that its sound was more dull
than the spruce plates. One possible cause of the high damping ratio could be the large
difference in material properties between the layers, since carbon fibre is much stiffer than
spruce and balsa. The high damping ratios could explain why this plate was considered to
have bad tonal quality by the luthier.
Table 5.4 Comparison of the damping ratios of the spruce and the composite
plates
Damping ratio Damping ratio Damping ratio

of the of the First of the Second
Spruce Plate Composite Plate Composite Plate
Mode1 0.579 0.856 0.362
Mode2 0.97 2.7 0.844
Mode3 1.65 2.82 2.27
Mode4 1.57 5.28 2.29
Mode5 1.94 3.98 1.48
Mode6 2.58 5.88 3.2
5.4 Vibration and Damping of the Composite Plate #2
Composite plate #2, which consisted of seven carbon fibre layers, was also made by Pro-
fessor Larry Lessards group, but at a later stage. This plate, which did not have a bass
bar on it, was considered by the luthier to be a good plate.
A comparison of the first six mode shapes by FEM and Modal Analysis can be seen in
Fig. 4.16 4.21. The mode shapes agree quite well with each other even though the first
mode was not excited in the experimental test. In modal analysis, it is possible to miss
some modes due to the structural and material property deviations of the real prototype
to the theoretical model. In general, the mode shapes of this plate were also somehow
different from the previous plate. For this plate, the ring mode appeared as the fourth
mode even though it usually appears as the fifth mode of a violin plate.
Since there was no bass bar on this plate, the mode shapes were all symmetric. If we
5 Discussion 77
look at Table 5.5, the natural frequencies of this plate were, in general, much lower than
Plate #1s; some of the modes were even lower than those of the spruce plate. The absence
of the bass bar, which reduces the stiffness of the plate, could explain this. Another reason
could be due to reduced thickness, which also serves a function of decreasing the natural
frequencies of the plate. Plate #1 has a thickness of about 4 mm at the central part
whereas the thickness of the central part of Plate #2 is only about 2.2 mm. By applying
a set of FEM experiments on each plate, it was found that the natural frequencies were
quite sensitive to thickness change of the plate. A thickness reduction of 0.5 mm can make
all the natural frequencies decrease by about 10%, provided that all the other parameters
remain constant. Thus, the low natural frequencies of Plate #2 could also be attributed to
its reduced thickness.
analysis of the spruce plate and the composite plate #2
FEM of Modal Analysis FEM of the Modal Analysis

the Spruce of the Spruce Second Composite of the Second
Plate Plate Plate Composite Plate
Mode 1 92.4 87.2 65.0
Mode 2 144.9 140 116.7 86.2
Mode 3 194.0 236 270.9 170
Mode 4 225.3 242 205.1 236
Mode 5 283.7 303 213.0 273
Mode 6 338.4 370 361.8 377
The damping ratios of this plate are shown in Table 5.4. The damping ratios are
around the same level as that of the spruce plate, which means Plate #2 would have
stronger resonances than Plate #1. Simply tapping on this plate reveals the difference.
The low damping ratios could have been caused by the homogeneous material properties
of this plate since all the layers were made of the same material. Also, the damping ratios
here validated the luthiers judgment that Plate #2 possessed better tonal quality than
Plate #1.
5 Discussion 78
5.5 Comparison of the FRFs by the FEM and the Modal

Analysis
The eigenvalue extraction process in the FEM can only give us the natural frequencies
of the top plate. Since the mode information in the Modal Analysis was represented in
the form of FRFs, it is necessary to also calculate the FRFs by FEM. Thus, we can more
clearly see the transition of each mode and the peak values at natural frequencies. To do
this in ANSYS, the Harmonic Response Analysis was applied. The idea was to add a
harmonic sweep excitation on the structure and calculate the response at each node. In the
numerical experiment, both the excitation and the response was defined at the middle point
of the bridge. The sweep frequency range was 0-1000 Hz, which covered enough modes for
analysis.
The computed FRFs are shown in Figs. 5.4, 5.6 and 5.8. Figures 5.5, 5.7 and 5.9 show a
comparison of the FRFs calculated by the Modal Analysis. As can be seen, the amplitude
peaks have reasonable coherence at the first several modes (below 500 Hz). At the higher
frequency range (500 - 1000 Hz), the FRFs calculated by the FEM show more amplitude
peaks than those by the Modal Analysis. This phenomenon is particularly obvious for
the composite plate #1. As discussed previously, it is possible to miss some modes in the
Modal Analysis because of the distribution of the hitting points and the manner of hitting.
A main difference of the composite plate #1 to the other two plates is the presence of the
bass bar, which is anticipated to have a set of natural frequencies itself distinct from the
top plate. This assumption was validated by the FEM simulation of the composite plate
#1. It is obvious in the animation of the FEM results that although the bass bar was
adhered to the top plate, it had some featured vibration behaviour itself. In the Modal
Analysis of composite plate #1, however, no point was selected on the bass bar. Thus, the
Modal Analysis of this plate would not only tend to miss some of the bass bar modes, but
also show less plate modes than the real case (because the bass bar can kill certain plate
modes). This could potentially explain why there are less modes in Fig. 5.7 than in Fig.
5.6.
Since the modal damping ratios differed a lot for each plate, I also thought of comparing
them by both the FEM and the Modal Analysis. However, the damping ratio in the FEM
is an input but not output (by solving the damping matrix [C] in Eq. 2.10). Thus, only
the damping ratios by the Modal Analysis were presented.
5 Discussion 79
5.6 FEM Experiments on Different Layer Combinations
Several parallel FEM experiments were also made to study how different layer combinations
would possibly affect the natural frequencies. Professor Lessards group had a lot of designs.
Some of them were considered to have good tonal quality by the luthier, as shown in Fig.
5.10-5.13. Since these designs all had a similar structure to Plate #2 but with different
layer combinations, the simulation was done based on Plate #2s FEM model. The results
are shown in Table 5.6.
From the experimental results, it became evident that thickness played an important
role in adjusting the plates natural frequencies. Even adding a 0.3 mm layer to the plate
can result in an apparent increase in natural frequencies. This further validated that the
relatively low natural frequencies of Plate #2 were reasonable. From Table 5.3, it seems
Plate #1 has a very different vibration behavior to the other plates, because most of its
natural frequencies are much higher than the other plates. Thus, in the future design, the
thickness of each layer might be restricted under a certain value, to give the desired natural
frequencies. Another conclusion on Table 5.3 is that the increment of the layer number
does not necessarily increase the natural frequencies. Layup 1 - 4 all have more layers
than Plate #1, but with smaller natural frequencies than that plate. One should also note,
that it was difficult to tell the influence of angle change on the natural frequencies from
examining the extrapolated data.
5 Discussion 80
Fig. 5.4 FRF of the spruce plate by the FEM
Fig. 5.5 FRF of the spruce plate by the Modal Analysis

5 Discussion 81
Fig. 5.6 FRF of the composite plate #1 by the FEM
Fig. 5.7 FRF of the composite plate #1 by the Modal Analysis

5 Discussion 82
Fig. 5.8 FRF of the composite plate #2 by the FEM
Fig. 5.9 FRF of the composite plate #2 by the Modal Analysis

5 Discussion 83
Fig. 5.10 Layup 1
Fig. 5.11 Layup 2
Fig. 5.12 Layup 3

5 Discussion 84
Fig. 5.13 Layup 4
Table 5.6 Comparison of the natural frequencies by the FEM of different

composite plate layups
Plate #1 Plate #2 Layup 1 Layup 2 Layup 3 Layup 4
Mode 1 92.4 65.0 51.2 55.2 70.3 59.9
Mode 2 144.9 116.7 92.4 100 125 108
Mode 3 194.0 270.9 143 153 181 162
Mode 4 225.3 205.1 154 172 225 187
Mode 5 283.7 213.0 160 174 233 194
Mode 6 338.4 361.8 235 256 327 279

85
Chapter 6
Conclusions and Future Work
6.1 Conclusions
The aim of this thesis was to study the vibrational behaviour of the violin top plate and
to explore the possibility of using composite materials as a substitute for traditional wood
in making the top plate. Both numerical simulations and experimental tests were applied
and compared to one another, thus making the results more convincing. A violin top
model based on a real spruce top plate was first built. The Finite Element Method (FEM)
was applied to this model to calculate the natural frequencies and mode shapes. Then
an experimental Modal Analysis was also conducted on the real plate to get not only the
natural frequencies and mode shapes, but also the damping ratios of each mode. The two
sets of results were carefully compared to prove that they not only fit with each other quite
well but also strongly agree with many previously published works by other researchers.
Thus, we can say that the spruce plate model is reasonably accurate. This model could
also be helpful for luthiers who may want to modify their designs. The high coherence of
the experimental results with the previous work indicates that the experiment design is
appropriate.
The two composite plates were obtained from Professor Lessard in order to compare
their behaviours to the traditional spruce plate. The work on these plates was original since
there are no published works to date which study the vibrational behaviour of violin top
plates made of new materials. Based on the results of the spruce plate, it is reasonable to
believe that the modeling process and the experiment design can give meaningful results.
The first composite plate (Plate #1) had wood as the core, carbon fibre as the coverage,
2013/06/04
6 Conclusions and Future Work 86
and an attached bass bar. The second composite plate (Plate #2) was made by several
layers of carbon fibre and a urethane ring around the plate without the bass bar. Both FEM
and Modal Analysis were applied again on the two composite plates. The simulation and
experiment results fit well with each other for both plates. The results indicated that Plate
#1 had higher natural frequencies than the spruce plate, which may have been caused by
the presence of the bass bar and the large stiffness differences between the layers. Plate #1
also had much larger damping ratios than the other two plates, which was not desirable.
The high damping ratio could explain why it was judged by the luthier as having bad tonal
quality. As for Plate #2, its lower thickness may have generated lower natural frequencies
than that of the spruce plate. The damping ratio of Plate #2 was quite similar to the
spruce plate, which could have been a reason for it to be judged by the luthier as having
good tonal quality.
Several parallel FEM experiments were done based on Plate #2s model to explore
the possible material variations for modification. Based on the prototypes we had, results
showed that the main difference between the designs was due to the natural frequency
change that accompanied the thickness change of each plate, a finding which may be used
as a frequency adjustment measure for future manufacturing.
6.2 Future Work
The vibrational behaviour study of the violin top plate gives us some hints about possible
modifications regarding the geometry and the use of new materials to improve or reproduce
the violin tonal quality. Similar numerical simulation and experimental tests can also be
applied to the back plate, which is usually made of maple. Such work can possibly allow
us to study how a composite plate behaves on a whole assembled violin. The results of
the plates can form a baseline for the modeling of the whole body of the violin. Some
researchers have performed a modal analysis of the violin body; however, the numerical
simulation of this process would be far more complex. Since the violin plates are not rigidly
glued on the ribs, the boundary conditions would be undoubtedly more difficult to deal
with. Also, the air cavity inside the violin body vibrates in a certain pattern. The coupling
of the air cavity modes and the body modes would be another challenge.
87
References
[1] H. Johannsson, Violin making: Construction. http://aiweb.techfak.

uni-bielefeld.de/content/bworld-robot-control-software/. [Online; ac-
cessed 04-December-2012].
[2] Violin student central: The bridge. http://www.violinstudent.com/bridge.

html. [Online; accessed 19-January-2013].
[3] P. Avitabile, Experimental modal analysis, Sound and vibration, vol. 35, no. 1,
pp. 2031, 2001.
[4] J. Bretos, C. Santamaria, and J. Moral, Vibrational patterns and frequency responses
of the free plates and box of a violin obtained by finite element analysis, The Journal
of the Acoustical Society of America, vol. 105, p. 1942, 1999.
[5] N. Molin, L. Lindgren, and E. Jansson, Parameters of violin plates and their influence
on the plate modes, The Journal of the Acoustical Society of America, vol. 83, p. 281,
1988.
[6] C. Hutchins, The acoustics of violin plates, Scientific American, vol. 245, no. 4,
pp. 170186, 1981.
[7] J. Moral and C. Hutchins, From properties of free top plates, of free back plates and
of ribs to properties of assembled violins, Report STL-QPSR, pp. 129, 1984.
[8] C. Hutchins, A history of violin research, The Journal of the Acoustical Society of
America, vol. 73, p. 1421, 1983.
[9] C. Hutchins and D. Voskuil, Mode tuning for the violin maker, CAS Journal, vol. 2,
no. 4, pp. 59, 1993.
[10] N. Molin, M. Tinnsten, U. Wiklund, and E. Jansson, FEM-analysis of an orthotropic

shell to determine material parameters of wood and vibrations modes of violin plates,
Report STL-QPSR, vol. 4, 1984.
References 88
[11] R. Schumacher, Compliances of wood for violin top plates, The Journal of the Acous-
tical Society of America, vol. 84, p. 1223, 1988.
[12] M. Elejabarrieta, A. Ezcurra, and C. Santamaria, Vibrational behaviour of the gui-

tar soundboard analysed by the finite element method, Acta Acustica united with
Acustica, vol. 87, no. 1, pp. 128136, 2001.
[13] K. Marshall, Modal analysis of a violin, The Journal of the Acoustical Society of
America, vol. 77, p. 695, 1985.
[14] E. Jansson and B. Niewczyk, Experiments with violin plates and different boundary
conditions, The Journal of the Acoustical Society of America, vol. 86, p. 895, 1989.
[15] G. Bissinger and K. Ye, Automated hammer-impact modal analysis with a scanning
laser vibrometer: working example-a violin, in SPIE proceedings series, pp. 943949,
Society of Photo-Optical Instrumentation Engineers, 2000.
[16] D. Oliver, V. Palan, G. Bissinger, and D. Rowe, 3-dimensional laser Doppler vibra-
tion analysis of a stradivarius violin, in Proceedings of the 25th International Modal
Analysis Conference, Society for Experimental Mechanics, Bethel, CT, U.S.A., 2007.
[17] G. Bissinger and D. Oliver, 3-D laser vibrometry on legendary old italian violins,
Sound and Vibration, vol. 41, no. 7, pp. 1015, 2007.
[18] J. Loen, Reverse graduation in fine cremonese violins, CAS Journal, vol. 4, no. 7,
pp. 2739, 2003.
[19] M. Tinnsten and P. Carlsson, Numerical optimization of violin top plates, Acta
Acustica united with Acustica, vol. 88, no. 2, pp. 278285, 2002.
[20] Y. Yu, I. Jang, I. Kim, and B. Kwak, Nodal line optimization and its application to
violin top plate design, Journal of Sound and Vibration, vol. 329, no. 22, pp. 4785
4796, 2010.
[21] A. Okuda and T. Ono, Bracing effect in a guitar top board by vibration experiment
and modal analysis, Acoustical science and technology, vol. 29, no. 1, pp. 103105,
2008.
[22] S. Phillips and L. Lessard, Application of natural fiber composites to musical in-
strument top plates, Journal of Composite Materials, vol. 46, no. 2, pp. 145154,
2012.
[23] O. Rodgers and T. Masino, The effect of wood removal on bridge frequencies, Acous-
tical Society Journal, vol. 1, no. 6, pp. 610, 1990.
References 89
[24] E. Jansson, Violin frequency responsebridge mobility and bridge feet distance,
Applied Acoustics, vol. 65, no. 12, pp. 11971205, 2004.
[25] G. Bissinger and C. Hutchins, Tuning the bass bar in a violin plate, CAS Newsletter,
vol. 26, pp. 1012, 1976.
[26] L. Condax, The sound post of the violin, Catgut Acoust. Soc. Newsletter, vol. 2,
pp. 67, 1964.
[27] N. Molin, A. Wahlin, and E. Jansson, Transient wave response of the violin body,
The Journal of the Acoustical Society of America, vol. 88, p. 2479, 1990.
[28] N. Molin, A. W ahlin, and E. Jansson, Transient wave response of the violin body
revisited, The Journal of the Acoustical Society of America, vol. 90, p. 2192, 1991.
[29] G. Knott, A modal analysis of the violin using MSC/NASTRAN and PATRAN.,
tech. rep., DTIC Document, 1987.
[30] P. Mohanty and D. J. Rixen, Operational modal analysis in the presence of harmonic
excitation, Journal of Sound and Vibration, vol. 270, no. 1, pp. 93109, 2004.
[31] M. LeGault, Acoustical architecture: Making beautiful music,

http://www.compositesworld.com/articles/acoustical-architecture-making-beautiful-
music, 2012.12.02.
[32] O. Rodgers, Relative influence of plate arching and plate thickness patterns on violin
back free plate tuning, Catgut Acoust. Soc. J., vol. 1, p. No.6 (SERIES II), 1990.
[33] I. Curtu, M. Stanciu, C. Itu, and R. Grimberg, Numerical modeling of the acous-
tic plates as constituents of stringed instruments, in Proc. of the 6th International
Conference of DAAAM Baltic Industrial Engineering, pp. 5358, 2008.
[34] B. Richardson, Guitar makingthe acousticians tale, Proceedings of the second Vi-
enna talk on music acoustics, 2010.
[35] M.-N. Lam, Design and fabrication of a japanese tea table and production of violin
tops and tailpieces, tech. rep., McGill University, 2012.
[36] G. Lim, Optimization of carbon-fiber violin top for maximum loudness, tech. rep.,
Structures and Composite Materials Laboratory, McGill University, 2010.

Ma Lu 2013 PDF

Uploaded by

Copyright:

Available Formats

Ma Lu 2013 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ma Lu 2013 PDF

Uploaded by

Copyright:

Available Formats

Comparison of Finite Element Method and

Modal Analysis of Violin Top Plate

Music Technology Area, Department of Music Research

3 FEM and Modal Analysis Set-up 25

3.1.1 CAD Model of a Top Plate . . . . . . . . . . . . . . . . . . . . . . . 26

4 Comparison of FEM and Modal Analysis Results 49

6 Conclusions and Future Work 85

2.1 Structure of a violin [1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3.1 The spruce top plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.1 Width and length of the plate . . . . . . . . . . . . . . . . . . . . . . . . . 27

4.1 Natural frequencies of the spruce top plate by FEM . . . . . . . . . . . . . 49

5.6 Comparison of the natural frequencies by the FEM of different composite

FEM Finite Element Method

1.2 Project Overview

1.3 Thesis Overview

2.1 Violin Acoustics

2.1.1 The Strings

Fig. 2.1 Structure of a violin [1]

2.1.2 The Bridge

Fig. 2.2 The bridge of a violin [2]

2.1.3 The Bass Bar and Soundpost

2.1.4 The Soundbox

2.2 The Finite Element Method

2.2.1 Procedures of the Finite Element Method

2.2.2 Node, Element and Shape Function

It is mathematically difficult to define a complex structure with a set of prescribed functions

its Taylor expansion:

For this element, the solution can be expressed as:

where u1 = u(x1 ) and u2 = u(x2 ).

The solution u(x) can be represented as:

and the shape function N can be represented as:

2.2.3 Equation of Motion

A vibration problem can be described by a general equation of motion, which follows

Fig. 2.3 A single harmonic oscillator

Fig. 2.4 A violin top plate

Fig. 2.5 A discretized violin top plate

and displacements at each node can be calculated.

It is assumed the solution has the form:

[x] = [a]ej[]t , (2.13)

calculated by solving the equation:

([K] i [M ])[a]i = 0, (2.16)

where [a]i are called the eigenvectors of the equation.

2.3 Experimental Modal Analysis

2.3.1 Mobility Measurement

The magnitude of this function can be calculated as:

and the phase can be represented as:

2.3.3 Coherence Function

Fig. 2.6 A coherence function in the test

2.4 Composite Music Instruments

FEM and Modal Analysis Set-up

3.1 Numerical Model of a Spruce Top Plate

Fig. 3.1 The spruce top plate

3.1.1 CAD Model of a Top Plate

Table 3.1 Width and length of the plate

Position of the Dimension (mm)

Total Length 358

Arch Heights of the Points Dimension (mm)

Fig. 3.3 CAD drawing of the spruce top plate

Fig. 3.4 Meshing of the spruce top plate