Sei sulla pagina 1di 29

Comparison between Modal Analysis and Finite Element Modeling of

a Marimba Bar

Ingolf Bork
PTB Braunschweig, Germany

Antoine Chaigne & Louis-Cyrille Trebuchet


ENST, Departement Signal, CNRS URA 820, France

Markus Kosfelder
Volkswagen AG, Wolfsburg, Germany

David Pillot
ENSAE, Toulouse, France

December 4, 1997

Send proofs to:


Antoine Chaigne
ENST, Departement Signal, 46 Rue Barrault, 75634 Paris Cedex 13, France
phone: (+33) 1-45-81-75-98, fax: (+33) 1-45-88-79-35, e-mail: chaigne@sig.enst.fr

Running title: modal analysis and modeling of a marimba bar


1
Abstract

The sound spectrum of low tuned marimba bars contains many components that cannot be explained
by a simple one-dimensional model of exural vibrations. A modal analysis has been then performed
on a bass marimba bar C3 with a fundamental frequency of 130 Hz (outer dimensions: 46 cm x 6 cm
x 2.35 cm (x,y,z), material: rosewood), taking into account two spatial components of vibration. The
20 most prominent modal frequencies and mode shapes were extracted from this analysis.
A 3-D nite element (FE) analysis of this bar, based on an orthotropic model for the material, has
been conducted in parallel. Modal analysis and nite element modeling yield very similar mode shapes
and modal frequencies for the rst 12 modes, between 130 Hz and 4000 Hz, which con rms the experi-
mental results and validates the theoretical approach.
The discrepancies between measured and calculated frequencies are less than 4 % in this frequency
range. For higher frequencies, between 4000 Hz and 8000 Hz, the nite element analysis show a num-
ber of modes which were not detected in the modal analysis.

PACS: 43.75.Kk
1 Introduction

The sound of percussion instruments is governed by the radiated vibrations of their resonances. The
impulsive strike of the mallet excites the eigenmodes which usually do not form a harmonic series of
overtones since the resonance frequencies and their ratio depend on the material and shape of the
instrument. In the case of the xylophone instruments, the marimba in particular with its extended
bass range, manufacturers have developed methods to tune some of the partials of the bars; these
methods have been re ned by scientists using modern computer facilities [1]-[2]. Two approaches to
investigating the vibrational behaviour of marimba bars will be presented in this paper: On the one
side the theoretical calculation of eigenfrequencies and mode shapes using nite element methods, on
the other side the experimental way using modal analysis hardware and software tools to measure the
corresponding frequencies and mode shapes. The investigated marimba bar C3 ( f1=130.81 Hz) which
had been modally analysed in the Laboratory of Musical Acoustics of the Physikalisch-Technische Bun-
desanstalt, Braunschweig (Germany), was simulated in the Departement Signal of the Ecole Nationale
Superieure des Telecommunications in Paris (France).

2 The marimba bar

In contrast to most musical instruments, the overtones of a struck uniform rectangular bar are non-
harmonic, i.e. the frequencies of the partials, which correspond to free bending vibrations of the
bar, have no integer ratios. This results in a sound spectrum of spread partials, compared with the
harmonic spectrum of a strictly periodic sound like that of a trombone for example. Fig. 1 shows in
musical staves the partials of a harmonic tone and the sound of a rectangular bar, with the frequency
deviation from the tempered scale indicated in the unit \cent" (1 cent = 0.01 semitone). The higher
density of the harmonic sound is clearly visible, and the small deviations from the tempered scale are
apparent, too. The harmonic spectrum with integer frequency ratios 1:2:3:.. forms the basis of the
theory of harmonies and chords in western music; the de nitions of consonances and dissonances are
derived from these integer ratios which, in tempered tuning, lead to beats according to the deviations
indicated.

The sound of uniform bars, which in the low bass register contains a lot of audible inharmonic partials,
causes an ambiguous pitch sensation of the played tone. Manufacturers of xylophone and marimba
instruments therefore tried to tune the second partial to harmonic intervals by cutting arches into the
underside of the bars: the fth above the octave (ratio 3:1) or the double octave (4:1), and tuning
of the third partial to the major third above the triple octave (10:1) is often applied though experi-
ments with musically trained subjects have shown that the interval between major and minor third is
preferred if the musical context contains major and minor chords [3]. The main advantage of tuning
higher partials is the constant frequency ratio for all tones of an instrument because played scales and
melodies show parallel steps for all tuned partials and at all instruments playing in an ensemble.

The decay process of a struck marimba bar is governed by the internal structural damping, meaning
that the decay decreases inversely proportional to the frequency of the partials. Fig. 2 shows the spec-
tral development of the bass marimba bar which was chosen for this analysis: C3 with a fundamental
frequency f1 = 130.8 Hz, a rosewood bar 46 cm in length (x-direction in the coordinate system), 6 cm
in width (y-direction) and 2.35 cm in thickness (z-direction). The three tuned partials at the lower
end of the frequency scale i.e. fundamental, second partial at 524 Hz and third partial at 1391 Hz,
are clearly visible and the decreasing decay time, too, is obvious for the higher partials.
3 Modal analysis

3.1 Presentation

Modal analysis is an experimental procedure for nding the modal parameters, i.e. frequencies, dam-
ping and mode shapes of vibrating structures at their resonances. It is usually performed on computer
workstations where these parameters are evaluated using the measured mechanical transfer functions
[4] [5]. These transfer functions are obtained for example by exciting at all surface points of inte-
rest with a force F (t) and simultaneously measuring the acceleration a(t) at one xed point on the
structure under test. The complex spectral transfer functions H (f ) are calculated by Fast Fourier
Transform (FFT) analysis algorithms implemented in stand-alone FFT analysers or in the computer
used for processing these data.

For the analysis of the marimba bar an Ono Sokki Cf 350 FFT analyser was used for data acquisition,
which was connected to an HP Series 300 Workstation with the SMS Modal 6.0 software controlling
data transfer and the further processing. Proper data acquisition is the main prerequisite for the
success of the modal analysis: the excitation impulse must include all frequencies up to the highest
mode of interest and must apply enough energy to achieve a sucient signal-to-noise ratio. A small
PCB 086B80 impulse hammer mounted in a pendulum was used for the excitation to achieve good
reproducibility at the 4 averaged impacts. To obtain the maximum frequency range, the commonly
used rubber cap on the metal tip had been removed and an additional small mass of 2.7 g had been
glued on the backside of the tip to increase the impulse energy. This hammer tuning process had
been controlled by observing the time signal of the impulse to avoid double impulses, and also in the
spectral domain the energy at up to 10 kHz could be measured in the optimum case.

To measure the response of the bar, a small accelerometer (PCB 309A) of 1.1 g mass had been glued
on the surface of the marimba bar. As a time window for the FFT signal the exponential window was
used which is the rst choice for all decay processes which are longer than the time window. Since the
frequency range of the analyser had been set to 10 kHz, the time window length for the 400 spectral
lines mode used was 40 ms which includes only the rst instant of the sound attack (cf. Fig. 2). To
test the reproducibility the coherence function had to be observed after each averaging. When the
values near the resonances dropped to less than 0.99, the measurement had to be repeated.

The measurement points on the bar had been chosen taking two aspects into account: Since the
modal analysis software uses connecting lines between these points to display the vibrations of the
measured system, the points must lie on the contour lines of the marimba bar. It must, however,
be possible to hit all these points perpendicularly with the impulse hammer for a proper excitation,
therefore the point of impact had to be moved slightly away from the edges. On the other side, the
plane upper surface of the bar should be provided with a rectangular grid of points in order to show
the deformation most clearly in the animated shape deformation on the computer screen. According
to the limited facilities of the software used, 5 rows of 23 points (x-direction) each on the surface
were chosen and 52 points on the bottom close to the partially curved contour line. All points were
excited perpendicular to the top and bottom surface i.e. in the z-direction of the coordinate system
used. Furthermore one side was struck at the 46 points close to the contour lines in the y-direction.
All other points were forced by the software to show the same de ection in the y-direction as the
corresponding measured points on the opposite side of the bar ; this was achieved by programming
adequate constraint conditions.

3.2 Measurements and mode shapes

A typical transfer function obtained after striking vertically at one edge of the upper surface is shown
in Fig. 3. The maxima of the curve correspond to the resonances of the bar, and it can be seen that
even up to 10 kHz clear resonances appear in the spectrum. Though the frequency resolution with
400 spectral lines, with 25 Hz per line is very low, the curve tting procedure allowed the fundamental
at 130.2 Hz to be calculated, but in general an error of 5 Hz must be assumed for all measured
frequencies, which is 65 cent for the fundamental but only 1 cent at 8000 Hz. Although the exact
tuning of the partials can be measured in other ways, too, only the results of the curve tting will be
given here.

The sound spectrum recorded by a microphone in the near eld at one end is shown in Fig. 4. The
bar was struck by a wooden mallet at one edge in order to excite also the upper modes. It can be seen
that a lot of partials also radiate sound, and it could also be shown by careful ltering that nearly all
modes up to 6000 Hz were audible separately.

A global curve tting algorithm was used to calculate the modal shapes. Frequency and damping were
estimated by using 20 transfer function where high amplitudes of the mode under investigation were
expected. Using these values, a rational fraction polynomial algorithm (SMS Modal 6.0) was applied
to all measured transfer functions revealing the complex mode shapes [4].

The calculated mode shapes which normally appear in animated form on a computer screen will be
visualized here in two di erent ways: First, as a quasi three-dimensional drawing of the bar in the
two extreme de ection positions with exaggerated amplitudes, and secondly as a scheme of nodal
lines with double-numbered indices indicating the number of nodal lines in the x and y direction. For
example the mode with only two nodal lines along the length of the bar, i.e. the fundamental bending
mode, appears as a (2,0)-mode, where these lines point to the y-direction [6]. If the nodal lines along
the x-axes point to the vertical z-direction, their number is written in square brackets [ ], indicating
that the plane of vibration is the x-z-plane; e.g the [2,0] is the rst lateral mode. Since such modes
sometimes appear in combination with vertical modes, both indices are given. The modes are shown
in the order of their resonant frequencies.

Fig. 5 shows the mode of the tuned fundamental resonance which is a clear (2,0) bending mode with
two nodal lines in the x-y-plane. These nodal lines determine the location for the support of the bar.
In order to keep mechanical damping by the bearing at a minimum, manufacturers have to apply
very carefully supporting devices such as holes for supporting horizontal strings or soft rubbers or
felt to underlay these nodal lines. The (1,1) mode is also shown in Fig. 5 which may be called the
rst torsional mode since the vibration looks like rotation about the x-axis, with two nodal lines in
the middle of the bar in each direction. This mode is radiated weakly, since the adjacent regions of
opposite phase inhibit the sound radiation at long wavelengths, here 1.06 m. The next mode (3,0) is
the second tuned partial, a bending mode with another node in the middle of the bar (Fig. 6). The
excited amplitude of vibration depends, of course, also on the point of impact since the individual
modes can be excited only at a point away from the nodal lines. This is why this partial sometimes
will not appear in the spectrum if the bar is hit exactly in the central nodal line. The frequency seems
to be slightly higher than four times the fundamental but this error is due to the uncertainty of the
frequency estimation in the frequency band chosen.

The next mode shown also in Fig. 6 is a clear lateral mode with two nodal lines along the x-axis but in
the vertical direction, i.e. the bar vibrates sidewise in the y-direction in its rst mode [2,0] (x-z-plane).
The vertical nodal lines are indicated by small circles in the x-y-plane. For the excitation of this mode
the bar has to be struck sidewise. Due to the asymmetrical undercut also a weak vertical component
can be observed.

The resonant frequencies of the two modes in Fig. 7 are quite close together (61 cent) and, therefore,
sometimes cause problems in the tuning process of the third partial. The (4,0)-mode is the bending
mode which corresponds to the tuned third partial of the marimba bar. Its sound radiation is very
strong, especially when hard mallets are used and the bar is hit at its centre. The (2,1)-mode shown in
the same gure is again a torsional mode which is radiated better than the lower (1,1)-mode because
its sound wavelength is shorter. When the bar is hit vertically at the long sides, this mode disturbs
the pitch sensation of the third partial since both resonances show almost equal amplitudes in the
spectrum. Knowledge of the mode shapes helps the tuning person to isolate the bending mode (4,0)
by simply striking on the middle line. The (2,1)-mode has its longer nodal line here and is therefore
not excitable. A closer look at the de ection amplitudes of the (4,0)-mode at both sides of the bar
reveals that they are not equal. This is explained by the di erent thickness of the tuning cut at both
sides.

In Fig. 8 we again nd a lateral [3,0]-mode which is combined with a (3,1), while the purer (3,1)-mode
with less lateral de ection appears at a higher frequency (Fig. 9). For comparison, the same view is
shown for both modes. The ordinary (5,0) and (6,0)-modes are also shown in Fig. 8 and Fig. 9. In the
Figs. 10 to 14, the higher modes can be seen though identi cation becomes more and more dicult
because the spatial resolution of the grid is too rough. Furthermore, some modes are not very clear
because a lateral component often occurs, caused by the unsymmetrical undercut. All these gures
show very clearly that at the higher modes the de ection area is concentrated on the thinner middle
region of the bar which also means that, during playing of the instruments, these modes are excited
more easily by hits in this region than in the end regions. On the other side, if all modes have to
be excited as in Fig. 4 it is better to hit the edge at one of the ends since here all modes show nal
de ection and no node.

4 Finite element modeling

4.1 Basic principles

In parallel to the modal analysis, a 3-D Finite Element modeling of the bar has been conducted with
the help of the CASTEM software package [7] on HP workstations at Ecole Polytechnique and ENST.
The problem to solve is an eigenvalue problem which can be summarized by the equation [8], [9]:

([K] ?  [M]) fX g = 0 (1)


where K is the rigidity matrix and M the mass matrix. The solutions of Eq. (1) are the values of the
p
eigenfrequencies !i = i and the modal shapes (eigenvectors) Xi associated to each eigenfrequency.

The resolution of this eigenvalue problem requires rst two fundamental procedures:

 The selection of the type of element and of the discrete mesh used according to the geometry of
the structure.

 The selection of appropriate elastic parameters for de ning the material.

The selected element here is a cubic element with 8 nodes (see Fig. 15). This element has been
found to be particularly ecient in previous analysis of structures subjected to bending [10]. In order
to limit the computing time, without a ecting the accuracy of the results, an irregular mesh has been
selected. In practice, the density of elements is larger in those parts of the bar where the magnitude
of the displacement is relatively large. This is the case here in the central part of the bar where the
thickness is signi cantly lower than the thickness at both ends (see Fig. 16).

The selection of the number of elements along the x-axis has been rst validated by the modeling
of a free-free isotropic bar of constant section, allowing a comparison between simulated and known
theoretical eigenfrequencies. The density of elements along y- and z-axis and the successive re ne-
ments of the mesh have been validated by systematic tests of convergence on isotropic and orthotropic
bars of variable section (see Fig. 17 for the de nition of the axes).
In general, reducing the number of elements tends to increase arti cially the sti ness of the modeled
structure, which in turn leads to an overestimation of the modal frequencies. In our experiments, the
total number of elements has been nally set equal to 2880 leading to a relative accuracy of a few
percents on the estimation of eigenfrequencies. With this number of elements, the averaged computing
time requested for obtaining the modes in the range 100 to 8000 Hz, which corresponds to the range
of interest for musical applications, is equal to 10 hours.
One diculty of the method is to reproduce adequately the geometry of the bar. On the original
marimba bar used for the experiments the undercut was not symmetrical, with an order of magnitude
of 10 to 20 % for the deviations of thickness, as a function of the coordinate x, measured on both
sides. In the model, the thickness has been taken equal to the mean value of the measured thicknesses
on both sides.

4.2 Modeling of the orthotropic bar

The density of the bar has been simply obtained by dividing its total mass by its volume. It is assumed
that the bar is homogeneous. The density is found equal to 1.080 Kg/m3 , which corresponds to the
order of magnitude for this family of Brazilian Rosewood.

A preliminary modeling has been conducted under the assumption that the bar was isotropic with
an unique Young's modulus of elasticity set equal to the longitudinal modulus EL . This modulus was
taken equal to 23 GPa, which corresponds to a mean value for this wood [11]. The equivalent Poisson's
ratio eq was taken equal to 0.1. This \equivalent" ratio is not known with great certainty in the case
of wood. However systematic variations of eq between 0.1 and 0.3 didn't make appreciable changes
in the estimation of eigenfrequencies.
With this simpli ed isotropic model, it has been found that the FE modeling was ecient in repro-
ducing the frequencies and modal shapes of the rst three vertical bending modes of the bar. On
the contrary, the isotropic model failed in reproducing accurately the other modes of vibrations. The
predicted lowest frequency corresponding to a torsional motion, for example, was found equal to nearly
twice the measured frequency. The discrepancies between predicted and measured torsional modes of
higher orders were even larger.
In conclusion, this preliminary investigation con rmed that the assumption of an isotropic material
is valid for predicting the lowest vertical bending modes of the bar, only. However, if the purpose is
to clearly identify other modes of vibration, such as torsion and lateral exion, then another more
elaborated model has to be developed.

For that reason, the bar is now assumed to be made of orthotropic material. Therefore, a set of
nine independent elastic constants is necessary for the modeling of the bar. This set is made of three
elastic moduli (EL , ER and ET ), three shear moduli (GLR, GRT , GLT ) and three Poisson's ratios
(LR , RT and LT ) [12]. The de nitions of these elastic constants are brie y reviewed in Appendix A,
whereas the de nition of longitudinal, radial and tangential directions of the wood can be seen in
Fig. 17. The values of the orthotropic elastic constants, which are listed in Table I, are taken equal to
averaged values found in the literature for Brazilian Rosewood and similar species (see, for example,
[13]).

The in uence of variations of some parameters (shear moduli and Poisson's ratios, in particular)
on the values of the eigenfrequencies has been tested within a reasonable range (50 %). It has been
found that these variations don't signi cantly alter the eigenfrequencies.

4.3 Results of simulations

The results presented in Table II show that the eigenfrequencies are correctly estimated with the
orthotropic model. The predicted frequency obtained for the lowest torsional mode (mode (1,1),
325.5 Hz), in particular, is in very good agreement with the measured frequency derived from the
modal analysis (323.4 Hz), the relative error being equal to 0.6 % only. This result is a consequence
of the fact that both radial and tangential moduli are now signi cantly lower than the longitudinal
modulus, whereas with an isotropic model, no distinctions are introduced between the sti nesses along
and across the bers. The other FE-simulations are in very good agreement with the modal analysis,
particularly for modes 1 to 12, which corresponds to frequencies lower than 4 kHz. In this frequency
range, the relative error between measured and calculated eigenfrequencies remains less than 4 % and
the modal shapes are clearly identical (compare the set of gures from Fig. 5 to Fig. 10 to the Figs. 18
to 20).
The comparison between measured and simulated modes becomes more dicult for the modes 14 to
23, i.e. for frequencies lying in the range 4 to 8 kHz. In this range, the ordering of the calculated modes
doesn't correspond exactly to the ordering of the measured modes, mainly because of the increasing
error due to the limited number of mesh elements in the directions 0y (vertical) and 0z (width). Other
discrepancies may be due do some inhomogeneities in the structure of the wood which are not taken
into account in the model.
In addition, it turns out that a number of the modes where not measured (modes 14, 19, 20, 21 and
24), probably because of their relatively small magnitude. However, even in this high frequency range,
the eigenfrequencies of the main exural modes along 0x, such as the (7,0) and the (8,0) modes, are
correctly predicted by the FE-modeling. Here again, the comparison between measured and predicted
modal shapes becomes harder with increasing order of modes (compare the set of gures from Fig. 11
to Fig. 14 to Figs. 21 to 23). Also the designation of the modes is more dicult in this range since
the three components of the bar motion become signi cant. As a consequence, the description of the
modes in the form of two indices, inspired by rectangular plates, may be not fully justi ed in this
range.

It has been noticed in the FE-modeling that the values of the eigenfrequencies are very sensitive
to small changes (typically 0.1 mm) in the thickness of the bar in its central part. This observation is
in agreement with the current practice of instrument makers who achieve a ne tuning of the bar by
polishing it slightly in its central region.
5 Conclusion

In this paper, the complex vibration pattern of a low-tuned marimba bar has been presented from
both an experimental and numerical point of view. The comparison between the results obtained
through modal analysis and nite element modeling, respectively, show a high degree of similarity in
general for the modal frequencies and shapes, and particularly for frequencies below 4 kHz, i.e. in the
prime range of interest for musical applications. This excellent agreement is clearly seen in Fig. 24
which summarizes the comparison between measured and calculated resonance frequencies.
Both techniques are limited by the spatial resolution. As a consequence the comparison becomes
harder for higher frequencies. In addition, since the damping factor of the wood usually increases
with frequency, some of the higher modes cannot be easily measured through modal analysis. On the
contrary, the three-dimensional nite element analysis, which is achieved with the assumption of no
damping, yields all possible \candidates" for the modes in a speci c frequency range.
It must be emphasized that the nite element modeling has to be conducted under the assumption
of an orthotropic model in order to obtain a good agreement with experimental data, which implies
to deal with nine elastic parameters. However, it turns out that the additional shear moduli and
Poisson's ratio need not to be de ned with great accuracy and that variations of 50 % of these
parameters don't cause appreciable variations on the eigenfrequencies.
Finally, it is hoped that the results of this investigation will give a better understanding on the
acoustics of marimba and that they will be of interest for instrument makers.

Acknowledgement

The authors wish to thank Voichita Bucur for helpful comments and references on elastic properties
of wood for percussive instruments.
APPENDIX A: Elastic constants of orthotropic material

In its reference system of coordinates (see Fig. 17), an orthotropic material has three orthogonal planes
of symmetry. Writing the linearized elastic equations for this class of materials between the strain
tensor " and the stress tensor  in this system yields:

0 1 0 10 1
B
B "LL CC BB E1L ? ERLR ? ETLT 0 0 0 CC BB LL CC
B
B CC BB CC BB CC
B
B CC BB CC BB CC
B
B CC BB  CC BB CC
B
B "RR CC BB ? ELRL E1R ? ETRT 0 0 0 CC BB RR CC
B
B CC BB CC BB CC
B
B CC BB CC BB CC
B
B CC BB CC BB CC
B
B "T T CC BB ? ELTL ? ERTR E1T 0 0 0 CC BB T T CC
B
B CC BB CC BB CC
B
B CC = BB CC BB CC (A.2)
B
B CC BB CC BB CC
B
B "RT CC BB 0 0 0 1 0 0 CC BB RT CC
B
B CC BB 2GRT
CC BB CC
B
B CC BB CC BB CC
B
B CC BB CC BB CC
B
B "LT CC BB 0 0 0 0 1
0 CC BB LT CC
B
B CC BB 2GLT
CC BB CC
B
B CC BB CC BB CC
B
@ CA B@ CA B@ CA
"LR 0 0 0 0 0 1
2GLR LR

The equations (A.2) involve 9 independent elastic constants:

 3 Young's moduli: EL , ER and ET

 3 Poisson's ratios: LR , RT and LT


 3 Shear moduli: GLR, GRT and GLT

The three remaining Poisson's ratios de ned in (A.2) are obtained by taking advantage of the symmetry
properties of the elasticity tensor:
RL LR T L LT T R RT
ER
= EL
; ET
= EL
; ET
= ER
(A.3)
References
[1] I. Bork. Zur Abstimmung und Kopplung von schwingenden Staben und Hohlraumresonatoren.
PhD thesis, Dissertation, Technischen Universitat Carolo-Wilhelmina, Braunschweig, 1983.

[2] F. Ordu~na Bustamante. Nonuniform beams with harmonically related overtones for use in per-
cussion instruments. J. Acoust. Soc. Am., 90(6):2935{2941, 1991. Erratum J. Acoust. Soc. Am.
91(6), June 1992.

[3] I. Bork and J. Meyer. Zur klanglichen Bewertung von Xylophonen. Das Musikinstrument,
31:1076{1081, 1982. Traduit en anglais par Th. D. Rossing et U. J. Hansen dans Percussive
Notes : 23(6) 48-57 (1985).

[4] R.J. Allemang & D. L. Brown. Experimental Modal Analysis. Shock and Vibration Handbook.
McGraw-Hill, New York, 1988. Editor C. M. Harris.

[5] D. J. Ewins. Modal testing: Theory and practice. J. Wiley & Sons, New York, 1986.

[6] N. H. Fletcher and Th. D. Rossing. Physics of musical instruments. Springer-Verlag, New York,
1991.

[7] L. Bohar and A. Millard. Guide d'utilisation CASTEM 2000. CEA/DMT/LAMS, 1992.

[8] E. Dhatt and G. Touzot. Une presentation de la methode des elements nis. Maloine, 1981.

[9] O. Zienkiewicz. The Finite Element Method. McGraw-Hill, 1977.

[10] S. J. Fenves. Numerical and Computer Methods in Structural Mechanics. Academic Press, 1973.

[11] V. Bucur. Acoustics of Wood. CRC Press, 1995.

[12] R. F. S. Hearmon. An Introduction to Applied Anisotropic Elasticity. Oxford University Press,


1961.

[13] D. Guitard. Mecanique du materiau bois et composites. Cepadues Editions, 1987.


TABLE I: Values of the parameters used in the FE modeling of the marimba bar C3

Marimba bar C3 : 130 Hz - Rosewood


Length L = 46.0 cm
Width b = 6.0 cm
Maximum thickness (edge) h = 2.35 cm
Minimum thickness (center) h = 0.68 cm
Total mass MB = 457.8 g
Young's modulus (Longitudinal) EL = 2:3  1010 N m?2
Young's modulus (Radial) ER = 2:3  109 N m?2
Young's modulus (Tangential) ET = 1:15  109 N m?2
Density  = 1080 Kg m?3
Poisson's ratio LR = 0.3
Poisson's ratio RT = 0.6
Poisson's ratio LT = 0.45
Shear modulus GLR = 3:0  109 N m?2
Shear modulus GRT = 1:0  109 N m?2
Shear modulus GLT = 3:0  109 N m?2

Number of elements along 0x (length): 160


Number of elements along 0y (width): 6
Number of elements along 0z (thickness): 3
Total number of elements: 2880
TABLE II: Comparison between measured and calculated eigenfrequencies

range FE modeling Modal Analysis mode x-y mode x-z


(Hz) (Hz) (lateral)
1 134.9 130.2 (2,0)
2 325.5 323.4 (1,1)
3 521.6 524.2 (3,0)
4 942.3 911.6 [2,0]
5 1367.7 1391.2 (4,0)
6 1464.4 1441.3 (2,1)
7 2278.9 2197.2 (3,1) [3,0]
8 2529.7 2519.6 (5,0)
9 2826.6 2740.3 (3,1)
10 3174.8 3196.1 (6,0)
11 3566.5 3432.8 (4,1)
12 3983.1 3999.9 (6,0) [4,0]
13 4076.8 4429.3 (2,2)
14 4309.7 ? [4,0]
15 4358.3 4221.2 (5,1)
16 5019.9 5037.0 (7,0)
17 5367.1 5814.9 (3,2)
18 5441.0 5271.2 (6,1)
19 6088.4 ? (4,2)
20 6433.0 ? [5,0]
21 6466.4 ? (5,2)
22 6507.7 6264.1 (8,0)
23 6700.0 6410.4 (7,1)
24 7271.5 ? (6,2)
25 ? 7859.0 (8,1)
Captions for gures

Fig. 1 Harmonic partials and resonances of a uniform bar

Fig. 2 Spectral development of the partials of a marimba bar C3 (130.8 Hz)

Fig. 3 Transfer function a(f )=F (f ) at an edge struck perpendicular to the surface

Fig. 4 Sound spectrum of the marimba bar C3 (130.8 Hz) in the near eld struck at one edge

Fig. 5 Deformation shapes and nodal lines of the fundamental (2,0) and (1,1) -modes

Fig. 6 Shapes of double octave (3,0) and rst lateral modes [2,0]

Fig. 7 Shapes of the third partial (4,0) and (2,1)-modes

Fig. 8 Shape of the (3,1)-mode in combination with [3,0] and the (5,0) -modes

Fig. 9 Shape of the (3,1) and (6,0)-modes

Fig. 10 Shape of (4,1) and the combined (6,0)[4,0]-modes

Fig. 11 Shape of the (5,1) and (2,2)-modes

Fig. 12 Shape of the (7,0) and (6,1)-modes

Fig. 13 Shape of the (3,2) and (8,0)-modes

Fig. 14 Shape of the (7,1) and (8,1)-modes

Figure 15: 3-D cubic element with 8 nodes


Figure 16: Irregular mesh used for the FE modeling of the marimba bar
Figure 17: Reference system of coordinates for an orthotropic material. (L): longitudinal ; (R): radial
; (T): tangential.
Figure 18: Finite Element modeling (mode shapes). (a) 134.9 Hz (2,0) ; (b) 325.5 Hz (1,1) ; (c)
521.6 Hz (3,0) ; (d) 942.3 Hz [2,0]
Figure 19: Finite Element modeling (mode shapes). (a) 1367.7 Hz (4,0) ; (b) 1464.4 Hz (2,1) ; (c)
2278.9 Hz (3,1)+[3,0] ; (d) 2529.7 Hz (5,0)
Figure 20: Finite Element modeling (mode shapes). (a) 2826.6 Hz (3,1) ; (b) 3174.8 Hz (6,0) ; (c)
3566.5 Hz (4,1) ; (d) 3983.1 Hz [4,0]+(6,0)
Figure 21: Finite Element modeling (mode shapes). (a) 4076.8 Hz (2,2) ; (b) 4309.7 Hz [4,0] ; (c)
4358.3 Hz (5,1) ; (d) 5019.9 Hz (7,0)
Figure 22: Finite Element modeling (mode shapes). (a) 5367.1 Hz (3,2) ; (b) 5441.0 Hz (6,1) ; (c)
6088.4 Hz (4,2) ; (d) 6433.0 Hz [5,0]
Figure 23: Finite Element modeling (mode shapes). (a) 6466.4 Hz (5,2) ; (b) 6507.7 Hz (8,0) ; (c)
6700.0 Hz (7,1) ; (d) 7271.5 Hz (6,2)
Figure 24: Comparison between measured and calculated resonance frequencies of the marimba bar C3
5 8

6 7

1 4

2 3

Figure 15:
Figure 16:
EL

ER ET z
x

y 0

Figure 17:
(a) (b)

(c) (d)

Figure 18:
(a) (b)

(c) (d)

Figure 19:
(a) (b)

(c) (d)

Figure 20:
(a) (b)

(c) (d)

Figure 21:
(a) (b)

(c) (d)

Figure 22:
(a) (b)

(c) (d)

Figure 23:

Potrebbero piacerti anche