Acoustics 06 00004

Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

acoustics

Article
Experimental Prediction Method of Free-Field Sound Emissions
Using the Boundary Element Method and Laser
Scanning Vibrometry
Andreas Wurzinger 1,∗ , Florian Kraxberger 1 , Paul Maurerlehner 1 , Bernhard Mayr-Mittermüller 2 ,
Peter Rucz 3 , Harald Sima 2 , Manfred Kaltenbacher 1 and Stefan Schoder 1

1 Aeroacoustics and Vibroacoustics Group, Institute of Fundamentals and Theory in Electrical Engineering
(IGTE), Graz University of Technology, A-8010 Graz, Austria; kraxberger@tugraz.at (F.K.);
stefan.schoder@tugraz.at (S.S.)
2 Otto Bock Healthcare Products GmbH, A-1110 Vienna, Austria; harald.sima@ottobock.com (H.S.)
3 Department of Networked Systems and Services, Budapest University of Technology and Economics,
Magyar Tudósok Körútja 2, H-1117 Budapest, Hungary; rucz@hit.bme.hu
* Correspondence: andreas.wurzinger@tugraz.at

Abstract: Acoustic emissions play a major role in the usability of many product categories. Therefore,
mitigating the emitted sound directly at the source is paramount to improve usability and customer
satisfaction. To reliably predict acoustic emissions, numerical methods such as the boundary element
method (BEM) are employed, which allow for predicting, e.g., the acoustic emission into the free field.
BEM algorithms need appropriate boundary conditions to couple the sound field with the structural
motion of the vibrating body. In this contribution, firstly, an interpolation scheme is presented, which
allows for appropriate interpolation of arbitrary velocity data to the computational grid of the BEM.
Secondly, the free-field Helmholtz problem is solved with the open-source BEM software framework
NiHu. The forward coupling between the device of interest and BEM is based on the surface normal
velocities (i.e., a Neumann boundary condition). The BEM simulation results are validated using a
Citation: Wurzinger, A.; Kraxberger, F.; previously established aeroacoustic benchmark problem. Furthermore, an application to a medical
Maurerlehner, P.; Mayr-Mittermüller, device (knee prosthesis frame) is presented. Furthermore, the radiated sound power is evaluated and
B.; Rucz, P.; Sima, H.; Kaltenbacher, contextualized with other low-cost approximations. Regarding the validation example, very good
M.; Schoder, S. Experimental agreements are achieved between the measurements and BEM results, with a mean effective pressure
Prediction Method of Free-Field
level error of 0.63 dB averaged across three microphone positions. Applying the workflow to a knee
Sound Emissions Using the
prosthesis frame, the simulation is capable of predicting the acoustic radiation to four microphone
Boundary Element Method and Laser
positions with a mean effective pressure level error of 1.52 dB.
Scanning Vibrometry. Acoustics 2024,
6, 65–82. https://doi.org/10.3390/
Keywords: boundary element method; vibroacoustic coupling; acoustic emissions; experimental
acoustics6010004
sound prediction
Academic Editor: Yat Sze Choy

Received: 23 November 2023


Revised: 14 December 2023
Accepted: 15 December 2023 1. Introduction
Published: 3 January 2024 Sound emission quantification of complex solid mechanical systems in engineering
applications, ranging from medical applications [1], loudspeakers [2], automotive [3–6],
railway [7], and aerospace industries [8–10] to decision-making in architectural acoustics
(e.g., placement of a heat pump), is essential for their improvement towards customer
Copyright: © 2024 by the authors.
satisfaction. Specifically, the identification and evaluation of sound emissions from a knee
Licensee MDPI, Basel, Switzerland.
prosthesis constitute a major part of the development of next-generation prostheses. The
This article is an open access article
noise emissions have multiple negative effects on both the user and the user’s environ-
distributed under the terms and
ment [11]. In particular, the boundary element method (BEM) can be efficiently used to
conditions of the Creative Commons
Attribution (CC BY) license (https://
simulate sound propagation based on the surface vibration velocity [12]. On the one hand,
creativecommons.org/licenses/by/
the mechanical surface velocity may be computed by the structural finite element method
4.0/).
(FEM) applied to the solid mechanical system [13]. On the other hand, the normal surface

Acoustics 2024, 6, 65–82. https://doi.org/10.3390/acoustics6010004 https://www.mdpi.com/journal/acoustics


Acoustics 2024, 6 66

vibration velocity data can be measured using laser-doppler scanning vibrometry (LSV).
Over the years, BEM has become a standard procedure for computing the acoustic free-field
propagation efficiently [14]. Various improvements have lead to fast simulation prediction
methods in the time and time-harmonic case for the linear acoustic field (e.g., fast multipole
methods [15]) or utilizing model order reduction [16]. Based on these improvements, BEM
has become a standard method for radiating free-field sound into far distances in the
mid-frequency range [17].
In the context of the article, the surface velocity is either computed using an FEM model
in combination with openCFS (https://opencfs.org/(accessed on 23 November 2023)) [18]
or determined experimentally. While FEM simulations capture the mechanical response
of the structure under various loads regarding applied modeling simplifications of joints,
bearings, connections, material, and geometry, the measured velocity captures the real-
world surface velocity response, typically with a comparably low spatial resolution of the
surface velocity signal. Based on the data and the Helmholtz equation, BEM is used to
predict the acoustic emissions at certain microphone locations [19]. The aim of this paper is
to establish a workflow for transforming the surface velocity information into information
about the acoustic sound emissions through BEM simulation, and providing validation for
multiple test cases. The open source BEM software NiHu (https://last.hit.bme.hu/nihu
(accessed on 23 November 2023)) [20] was verified previously by several test cases. In
the first step focusing on the automotive industry, the FEM-BEM method was verified
and compared with an existing acoustic simulation of a confined flow-sound prediction
using the FEM method [21]. This example shows the accuracy of the workflow for a simple,
realistic application. Additionally, the influence of low surface resolution was investigated
systematically by virtual coarsening of the measurement grid to imitate points of missing
surface information.
Regarding this verification example, the method was finally applied to validate the
acoustic emissions of a medical knee prosthesis frame under specific test conditions. Acous-
tic microphone measurements in an anechoic chamber validated the predicted sound
emissions. The numerically obtained sound power was compared to a low-cost sound
power estimate presented in [22]. In conclusion, the results show very promising sound
prediction capabilities, with a mean error of the effective pressure level, as defined in
Section 3.2.3 (averaged across four microphone positions) of 1.52 dB.
The rest of the paper is organized as follows. In Section 2, the data acquisition using
an LSV and the successive interpolation algorithm are described. Furthermore, BEM and
its implementation in the software framework used are presented. Section 3 describes the
verification and validation of the interpolation scheme and the BEM solver. In Section 4,
an application of the workflow to an industrial use case, i.e., sound radiation of a clinically
applied knee prosthesis, is presented. Finally, Section 5 discusses the main findings.

2. Methods
In most technical applications, sound radiation can be modeled by applying forward
coupling of the structural vibration domain to the acoustic propagation domain, as depicted
in Figure 1. Consequently, any effect of the incident pressure field on the structure is
neglected [2,3]. The forward coupling is achieved by prescribing the normal component
of the acoustic particle velocity vn to the corresponding surface velocity of the structure,
i.e., applying an inhomogeneous Neumann boundary condition for the wave equation (see
Section 2.2). For simple structures, the surface velocity of the structure can be obtained,
e.g., by FE simulation of the structural dynamics. This workflow is described in Section 3.2
using the FE solver openCFS. However, in industrial applications, the high number of
contact points and joints in complex assemblies or the use of compound materials in
light-weight applications can make simulations of the mechanical structure infeasible with
reasonable effort. Therefore, when applying the proposed method to an industrial use case,
measurement data from LSV measurements are used instead as an input for the sound
prediction model, as demonstrated in Section 4.
Acoustics 2024, 6 67

Surface Sound propagation Sound


Patches of
Data acquisition Data transfer normal velocity pressure field
simulation
LSV measurement Projection &
Boundary element
FE simulation Interpolation
method

CAD
Closed surface mesh
Geometry

Figure 1. Workflow of the sound radiation simulation based on the surface velocity measurements.

2.1. Data Acquisition


The surface normal velocity, required as a boundary condition for the acoustic prop-
agation model (see Section 2.2), can be obtained efficiently by experiments using an LSV,
as depicted schematically in Figure 2. Laser vibrometry is based on the interference of
a reference laser beam with a laser beam reflected by the surface, which is frequency
modulated based on the Doppler effect [23]. Thereby, the surface velocity in the laser
direction vLSV
j = v · nLSV
j can be measured, where nLSV j denotes the unit vector in the
laser direction and the index j the index of the measured surface location. Consequently,
the three-dimensional velocity vector v can be reconstructed from three independent
excitation-synchronized measurements using unique laser directions to guarantee observ-
ability. When dealing with nearly flat surfaces, it can be sufficient to position the laser
scanning head in the dominant direction of overall normal motion and use this velocity
component to approximate the normal velocity vn .

LSV

Figure 2. Sketch of an experimental setup using a laser scanning vibrometer (LSV).

Laser scanning vibrometry allows for the subsequent measurement of an array of


points on the surface, such that the vibrometer has to be repositioned a limited number
of times to obtain data on all relevant areas. This results in N patches of measurement
data. Data on these measured patches must be transferred to a closed (and most often
refined) surface mesh used for the acoustic propagation simulation (see Section 2.3). This
data transfer comes with the following challenges:
• Imperfect representation of the geometry by the measured surface patch;
• Spatial up-sampling of coarsely distributed measurement data;
• Overlapping data patches.
To overcome these challenges, a two-staged data transfer process is proposed, consist-
ing of a projection and an interpolation step, which is combined into a linear transformation
Ti for each measurement patch i. In the following, the mesh of the closed geometry used
for the simulation is called target Γt , and patch i containing the measurement data is called
source Γsi . The algorithm to obtain the transformation matrix Ti is shown in Algorithm 1. It
is described in the following: For any given point xtk on the target mesh, the location is pro-
jected in the measurement direction onto the source mesh. For simplicity, the measurement
direction used for the projection is averaged over all points in a patch, resulting in a mean
direction nLSVi . Suppose the surface has a complex shape but is sufficiently smooth (or the
Acoustics 2024, 6 68

edges are treated separately). In that case, the projection direction can also be obtained by
averaging the normal vectors of the neighboring elements for each point on Γt . Next, linear
p
shape functions Nj ( x) are evaluated for the projected point xk in the according element e.
In this context, the (linear) shape function evaluation can also be used to check whether
the projected point is inside any element. Finally, the shape function values are used as
transformation coefficients corresponding to the source grid point indices and the index of
the target point.
h i h i
p
Ti = Tj,k = Nj ( xk ) ∀ j ∈ e ∈ Γs ∧ ∀ k ∈ Γt . (1)

Algorithm 1 Data transformation matrix


nLSV
i ← mean(nLSV j ) . Compute projection direction
for xk ∈ Γ do
t t . Loop over nodes in target grid
for xsj , xe ∈ Γs do . Loop over nodes / elements in source grid
if || xtk − xe || < maxDist then . Skip points that are too far apart
p
xk ← projectPoint(nLSV i ) . Project point onto source grid
p
if xk ∈ e then . Check if projected point is inside the element
p
Nj ← evalShapeFunction( xk ) ∀ j ∈ e . Compute interpolation weights
Tj,k ← Nj
end if
end if
end for
end for

A sketch of Algorithm 1 for a two-dimensional problem is provided in Figure 3. It


depicts an arbitrary source and target surface mesh in two dimensions, both comprised of
discrete points. The process of data transformation is illustrated, where the location from
the target mesh geometry xtk is projected onto the source mesh geometry Γsi using the mean
measurement direction nLSV i as projection direction. The black line from the target mesh
Γts in red to the source mesh Γss in blue represents the projection step. It should be noted
that the transfer direction of data values is opposite to the projection direction. After the
p
projection, the data are evaluated in xk by finite element shape functions accordingly. These
finite element shape functions define how data is distributed and interpolated across the
elements in the source mesh.

Figure 3. Sketch of an arbitrary source (blue) and target mesh (red) with indicated point projection
from the target to the source mesh.

2.2. Boundary Element Method


A brief outline of the boundary element method for solving acoustic radiation prob-
lems is provided in this section. An in-depth introduction can be found, e.g., in [19]. The
stationary sound radiation problem in a domain Ω is governed by the Helmholtz equation

∆p( x) + k2 p( x) = 0 x ∈ Ω ⊂ R3 , (2)

assuming a time harmonic dependency ejωt . Therein, p( x) denotes the scalar, complex-
valued acoustic pressure, k = ω/c is the wave number, ω is the angular frequency, and c is
Acoustics 2024, 6 69

the isentropic speed of sound. The prescribed normal velocity vn on the surface Γ can be
modeled as a Neumann boundary condition

1
vn ( x ) = − ∇ p( x) · n( x) x ∈ Γ ⊂ R2 , (3)
jωρ
with density ρ and inward facing normal vector n. Introducing the Green’s function
G ( x|y) = e−jkr /(4πr ) as the fundamental solution of the Helmholtz equation, representing
the acoustic pressure field at position x due to a unit point source at position y. Therein,
the distance r is defined as r ( x, y) = | x − y|. Transforming the Helmholtz Equation (2)
into its weak form and combining it with the Green’s function, the conventional Kirchhoff–
Helmholtz (boundary) integral equation can be derived

∂G ( x|y) ∂p(y)
Z Z
C ( x) p( x) + p(y)dΓ(y) = G ( x|y) dΓ(y) . (4)
Γ ∂n(y) Γ ∂n(y)

The coefficient C ( x) is determined by the location of x. On smooth surfaces, e.g., inside


a boundary element, we obtain C ( x) = 0.5, while at any field point in Ω \ Γ, the coefficient
is C ( x) = 1 [19]. Furthermore, the normal derivative with respect to x is applied to
Equation (4), which yields the hypersingular boundary integral equation (HBIE)

∂p( x) ∂2 G ( x | y ) ∂G ( x|y) ∂p(y)


Z Z
C ( x) + p(y)dΓ(y) = dΓ(y) . (5)
∂n( x) Γ ∂n( x)∂n(y) Γ ∂n( x) ∂n(y)
Jointly solving Equations (4) and (5) avoids the effect of ficticious eigenfrequencies.
This formulation is commonly called the Burton and Miller method, e.g., [24,25]. A descrip-
tion of the solution method involving a linear combination of the conventional Kirchhoff–
Helmholtz Equation (4) and the HBIE (5) is provided in the following.

2.3. Discrete Formulation in the Software NiHu


To solve the previously introduced problems (2) and (3), the open-source C++ BEM
library NiHu is used. The library is capable of computing the coefficients of discretized
boundary integral operators in a generic way with arbitrarily defined kernels and function
spaces [20]. To achieve maximum flexibility, the library was used to compile a MATLAB
toolbox through the MEX interface. Discretization of Equation (4) was performed em-
ploying the collocational BEM on a computational grid consisting of linear triangular and
quadrilateral elements with the collocation points being located in the element centroids,
which results in the linear system of equations

1
Ms ps − Ls qs = ps (6)
2
Mf ps − Lf qs = pf (7)

for the primal BEM problem on the surface Γ is indicated by subscript s and the evaluation
of an arbitrary number of field points in Ω \ Γ is indicated by the subscript f. Therein,
Ls/f and Ms/f denote dense matricesR representing the discretized
R single and double layer
potential integral operators Lu = G udΓ and Mu = ∂G/∂n udΓ, respectively [20].
Furthermore, ps/f denotes the acoustic pressure and qs/f = ∂p/∂n is the normal derivative
of the acoustic pressure. The normal derivative of the acoustic pressure qs can be related to
the acoustic particle velocity v by

qs = −jωρv · n . (8)
Acoustics 2024, 6 70

The arrays ps/f and qs accumulate the coefficients ps/f and qs on the discretized domain
for acoustic pressure and its normal derivative, respectively. Similarly, the discretization of
Equation (5) yields

1
Ns ps − MsT qs = qs , (9)
2

with the dense matrix MsT representing T


R
R M u = ∂/∂n G udΓ, and Ns representing the
hypersingular operator N u = ∂/∂n ∂G/∂n udΓ. Nearly singular and singular integrals
of the weakly, strongly, and hypersingular kernels are handled by a static part extraction
technique, where the singular static parts are integrated analytically [26]. According to the
Burton–Miller method, spurious eigenfrequencies can be mitigated by solving
    
1 T 1
Ms − I + αNs ps = Ls + α Ms + I qs (10)
2 2
j
with the coupling parameter α that is chosen α = − k , following the findings in [25].

3. Verification and Validation


3.1. Verification of the BEM Solver
The BEM solver was validated against various problems, including a spherical ra-
diator [20], the cat’s eye, Radiatterer, and the PAC-Man [26]. The latter three were EAA
benchmark problems (https://zenodo.org/communities/eaa-computationalacoustics (ac-
cessed on 23 November 2023)). The validation cases of the NiHu framework comprised the
verification of the conventional BEM implementation using both collocational and Galerkin
formalisms, as well as the software code for the fast multipole method. A challenging
task in implementing the BEM for the Helmholtz equation is the proper treatment of near,
weak, strong, and hypersingular integrals. The aforementioned test cases are well-suited
for verifying that the singular integration techniques are implemented properly. Finally,
the implementations of the CHIEF [27] and Burton–Miller [24] approaches for treating
fictitious eigenfrequencies arising in the case of exterior problems were also validated.

3.2. Validation Example: Pipe with Orifice


In a recent investigation, Maurerlehner et al. [21] validated a methodology employing
FEM, based on the hybrid computational aeroacoustic approach, for the simulation of the
acoustic field of confined flows with a modular test rig. In [28], Maurerlehner investigated
the aero- and vibroacoustic sound emissions of several geometries that induce turbulence
and flow instabilities, e.g., a half-moon-shaped orifice in a circular pipe. Using these
results, the acoustic emissions to the free field caused by the surface vibrations of the
duct will be the validation example for the proposed method. The bottom view of the
geometry of this validation case is sketched in Figure 4. Therein, the duct; its mounts (i.e.,
the structure); and microphone positions 1, 2, and 3 are depicted. The surface Γ used for the
BEM computations is colored in red and is the outer surface of the duct wall. The interior
diameter of the duct is Di = 50 mm and the outer diameter is Da = 56 mm.
In [28], the FE-based methodology was validated, achieving considerable agreement
between the simulations and measurements of the external pressure. Building on these
results, the BEM approach introduced in the present paper will be validated against the FE
results and measurements. While FEM requires the volumetric computational domain that
contains all points of interest to be discretized, the BEM problem is solved on the surface
mesh, and an evaluation of the unknown quantity can be performed, e.g., for a limited
number of arbitrary field points of interest during post-processing, hence reducing the
computational complexity.
Acoustics 2024, 6 71

Figure 4. Bottom view of the experimental setup of the pipe with the orifice (all measures in mm),
based on Figure 4.12 of [28]. The boundary Γ used for the BEM simulation is indicated in red.

3.2.1. Measurements and FEM Model


The measurement setup for the external pressure is described in detail in Section 4.2
of [28] and will therefore only be summarized briefly in the following. The measurement
data acquisition and processing are also described in detail in [21], in a case applied to
internal pressure data. At the three microphone positions, mic. 1 to 3, as indicated in
Figure 4, the external acoustic pressure was measured using free-field microphones (Bruel
and Kjaer 4190 with preamplifier 2669 L). From the time series data, block-averaged sound
pressure level (SPL) spectra were computed Equation (4.10) of [28]. This resulted in a set of
meas
measured amplitude spectra | p|meas
m ( f ) with a mean of | p| ( f ), where m = 1, . . . , N meas ,
with N meas being the number of spectral blocks of the measurement.
The FEM setup described in [21] was used to compute the internal acoustic pressure
and couple it to the structural mechanics domain (i.e., the duct wall), resulting in a surface
velocity field at the outer duct surface. as described in [28]. Maurerlehner [28] also used a
time-domain FEM approach for the external acoustic radiation. This resulted in a set of FEM-
FEM
simulated amplitude spectra | p|FEM
n ( f ) with their mean | p| ( f ), where n = 1, . . . , N FEM ,
with N FEM being the number of spectral blocks in the FEM simulation.

3.2.2. BEM Simulation Model


The acoustic radiation simulation was set up following the workflow described in
Section 2. However, instead of taking measurements of the surface velocity, the availability
of a validated FE-model of the mechanical structure allowed for extraction of the surface
velocities from the (coupled) time-domain FE simulation of the pipe structure and the
fluid inside [28]. Pseudo-measurement data were generated by nearest neighbor interpo-
lation using inverse distance weighting (Shepard’s method) of the structural velocity on
a grid of realistically spaced measurement points from, e.g., an LSV measurement. This
approach allowed for variation of the measurement point density, which will be discussed
in Section 3.2.4. For an analog comparison of the simulation results to [28], the pseudo-
measurement data were split into N BEM Hann-windowed segments of N = 2000 samples
with 50% overlap, and point-wise Fourier transform was applied to each block. This
resulted in a set of velocity spectra. The computed velocity spectra were transferred to
the computational grid for the BEM simulation in the next step. The grid had equally
spaced elements with an approximate element size of 5 mm, which corresponded to 6.9 el-
ements per wavelength at 10 kHz, resulting in 10,122 degrees of freedom for the BEM
formulation. The interpolation to the BEM grid was performed using the workflow pre-
sented in Section 2.1. Because of the ideal fit of the FEM surface grid to the BEM grid,
however, the projection step could be skipped. The resulting spatially up-sampled data
were projected in the surface normal direction
 according to Equation (3) and applied as
a set of Neuman boundary conditions qs,1 . . . qs,N BEM in Equation (6) to solve the
discretized boundary element problem as described in Section 2.3. The acoustic pressure
was evaluated at the microphone positions defined in [28] (see Figure 4). This resulted
Acoustics 2024, 6 72

BEM
in a set of BEM-simulated amplitude spectra | p|BEM k ( f ) with their mean | p| ( f ), where
k = 1, . . . , N BEM , with N BEM being the number of spectral blocks of the BEM simulations.

3.2.3. Comparison of Experimental and Numerical Results


A similar evaluation procedure compared the BEM method to the established FEM
model in [28]. Maurerlehner [28] split the time domain pressure signals into segments of
N FEM = 3000 samples with 50% overlap, performed Fourier transform thereof, and com-
puted the amplitude averaged spectra. In contrast, in this work, the time domain velocity
data were split into segments of N BEM = 2000 with 50% overlap. The Fourier transform of
each segment resulted in multiple boundary sets for the BEM simulations. The resulting
pressure spectra were amplitude averaged. Figure 5 depicts the block-averaged pressure
amplitude spectrum. The legend provided an effective pressure level Leff , such that
!
k p( f )k2
Leff = 20 log √ ∀ f ∈ f interest (11)
2pref

with reference pressure pref = 20 µPa in the frequency range of interest finterest =[400, 10, 000] Hz.
Table 1 shows the effective pressure levels for the measurement, FEM, and BEM simu-
lations at the three microphone positions defined in Figure 4. The maximum deviation
of the BEM solution to the FEM solution of 0.47 dB indicated a good agreement of the
presented method to the FEM simulation approach used in [28]. The simulated data agreed
reasonably with the measurement data for all of the microphone positions. According
to [28], deviations were caused by inaccuracies of the flow simulation data inside the
pipe, the modeled mounting of the pipe, and the material damping model used, which
all altered the pseudo-measurement and therefore had a similar impact on both FEM and
BEM simulation approaches. Furthermore, using the acoustic free-field assumption rather
than modeling the acoustic chamber introduced discrepancies between the simulation and
the measurements.

Table 1. Comparison of the measured and simulated effective pressure levels in the frequency range
f interest = [400, 10, 000] Hz at the microphone positions according to Figure 4.

Microphone
Lmeas
eff LFEM
eff LBEM
eff Lmeas BEM
eff − Leff
Position
1 36.27 dB 35.95 dB 36.21 dB 0.06 dB
2 35.81 dB 35.82 dB 36.18 dB 0.37 dB
3 35.94 dB 37.87 dB 37.40 dB 1.46 dB

Figure 5 presents a frequency-resolved evaluation of the agreement among three


distinct results of the block-averaged acoustic pressure computed via the BEM and FEM
results, and the empirically obtained measurement data. The FEM results were validated
in this previous study [28]. Therefore, BEM is in excellent agreement with the FEM results,
and only minor deviations occurred. It shows the robustness of the BEM acoustic pressure
predictions. As concluded by the previous study [28], the predictions of the BEM method
were validated by the measurement data at the microphone positions. The variations from
numerical predictions to the multi-block averaged experimental results were well within
the measurement uncertainty, indicated by the envelope of spectra of measured blocks
(gray-shaded area).
Acoustics 2024, 6 73

Figure 5. Comparison of the block-averaged amplitude spectra of the acoustic pressure | p| at


microphone position 1 of the BEM results to the measurement data and FEM data. The time-domain
measurement FEM data were extracted from [28]. The gray-shaded area defined the minimum and
maximum amplitude of the block spectra | p|meas
m ( f ) used for block-averaging.

3.2.4. Variation of Measurement Point Density


The BEM solution depicted in Figure 5 was based on pseudo-measurement data that
were spatially sampled with a spacing of href = 5 mm, which matches the element size of
the computational grid used for solving the sound radiation problem. In a subsequent
study, the measurement point density was coarsened by multiplication of the grid distances
by a factor h for all of the coordinated axes equally, or h x , hy , hz for each coordinate axis
separately, resulting in NLSV pseudo measurement points. The pseudo measurement grids
considered in the study are depicted in Figure 6.

(a) (b)

(c) (d)

Figure 6. Side view of the pseudo-measurement domain with indicated measurement


points. (a) ref: h = 1 ( NLSV = 8100), (b) h = 2 ( NLSV = 2034), (c) h = 4 ( NLSV = 946),
(d) h x,y = 4, hz = 16 ( NLSV = 442).

Figure 7 depicts the acoustic pressure level L p = 20 log( p/pref ), and its deviation from
the reference BEM simulation results. The comparison was two-fold. Firstly, the acoustic
pressure levels were compared as a function of frequency. From this perspective, deviations
to the reference signals were small, and good agreement was also achieved for very poorly
resolved measured surface velocities. Secondly, to evaluate the differences in more detail,
the relative difference to the reference was plotted as a function of frequency. This shows a
Acoustics 2024, 6 74

very clear trend: the deviations between the h-dependent simulated acoustic pressure levels
and the reference simulation results (h = 1) systematically increased with the frequency.
At certain frequencies, 2100 Hz and 4750 Hz, the deviations were large and increased until
a range of 5 dB. At these frequencies, structural anti-resonances mitigated the large-scale
motion, and the measured data did not resolve the remaining motion. The increasing
deviations with frequency were expected and could be attributed to the reduced resolution
per wavelength of the interpolated surface velocity data. The second effect, which was
visible, was that with decreasing resolution (increasing h) of the measured surface velocities,
the deviations became larger, and in general, the radiated sound pressure levels were
underestimated. Again, this can be explained by the reduced resolution of the structural
motion and due to the linear interpolation scheme, which underestimated the movement
of convex curved surfaces. In summary, Figure 7 demonstrates how the acoustic pressure
level deviations in a simulation changed with both the frequency and spatial resolution
of the structural motion. As the structural wavelengths were large compared with the
acoustic wavelength, the surface velocity required less spatial resolution in the presented
benchmark problem.

Figure 7. Comparison of the block-averaged acoustic pressure level at microphone position 1 for
varying grid size of the LSV measurement.

4. Application: Prosthesis Frame


The previously verified and validated simulation setup consisting of the interpolation
algorithm and the acoustic propagation simulation using BEM was applied to an industrial
problem. Thereby, the acoustic radiation of a clinically applied knee prosthesis frame was
investigated by means of measurements and simulations. The prosthesis frame (device
unter test—DUT) had dimensions of (270 × 80 × 100 mm) and was made of carbon-fiber
reinforced plastic with titanium inserts. The complex material behavior made it diffi-
cult to accurately predict the structural dynamics using FE simulation and experimental
acquisition of the surface vibration was preferred.
Acoustics 2024, 6 75

4.1. Measurement Setup


The measurement environment for both the microphone and LSV measurements was
an acoustically treated chamber of type Studiobox Premium with a size of 3.0 × 2.02 × 2.3 m.
The prosthesis frame (device under test—DUT) was positioned centrally in the room on a
table together with the shaker (DynaLabs DYN-PM-100), which was connected to the DUT
by a stinger. The shaker was mounted on rubber elements to decouple it mechanically
from the table. Furthermore, an acoustically treated box was placed over the shaker to
minimize the influence of the sound emissions of the shaker on the measurement setup.
To measure the input force, an impedance head (DJB AF/100/10) was applied at the stinger–
DUT connection, allowing for measurements of the input force F (t) on the DUT. The
excitation signal was a pseudo-random noise that was generated by the control unit of the
LSV for a frequency range from 2 Hz to 6.4 kHz with a desired frequency resolution of 2 Hz
(corresponding to 3200 spectral lines). Together with a sampling frequency of 16 kHz, this
resulted in a measurement time of 0.5 s. Using this setup, the microphone measurements
obtaining the acoustic pressure and the surface velocity measurements with the LSV were
conducted as described in the following.

4.1.1. Acoustic Pressure Measurements


The setup of the acoustic measurements using four microphone positions is depicted
schematically in Figure 8 in a side and top view. Four microphones (Bruel and Kjaer 4189
free-field microphones ) were placed at a distance of lmic = 0.7 m around the DUT, obtaining
the acoustic pressure. Microphones 1, 2, and 3 were located in the x/z-plane at y = 0.
Microphone 4 was at a 45° elevation. The microphone signals in time domain were denoted
as pmic1 (t), pmic2 (t), pmic3 (t), and pmic4 (t) for microphones 1 to 4, respectively.

Mic. 1

Impedance head Mic. 4 Impedance head


Mic. 4 Mic. 2
Stinger (elevated) Stinger
(elevated)
Acoustically Acoustically
treated box treated box
Mic. 2
Mic. 3
Shaker Shaker Mic. 3
DUT

DUT
Table
Rubber mount

Mic.Figure
1 8. Sketch of the experimental setup for the microphone measurements. Mic. 1, 2, and 3 are
located in the x/z plane. Mic. 4 is at 45° elevation above mic. 2. The coordinate system’s origin is
located in the center of the device under test (DUT), such that the excitation force vector lies inside
the x/z plane.

The microphones were connected to a data acquisition unit (HEAD acoustics SQuadriga
III) handling the analog/digital conversion. The time series measurements of the acoustic
pressure and the input force with a sampling frequency of f S = 48 kHz were buffered
into blocks with a length of Nsamp = 2400 samples and 50% overlap, corresponding to a
target frequency resolution of ∆ f = 20 Hz. A Hann window function was applied on each
block. For each block, the Nsamp -point Discrete Fourier Transform (DFT) was performed
as implemented in MATLAB’s stft()-command [29], which resulted in pressure spectra
p̃mic1
i ( f ), p̃mic2
i ( f ), p̃mic3
i ( f ), and p̃mic4
i ( f ) for microphones 1 to 4, respectively, as well as
F̃i ( f ) for the input force spectrum. Thereby, i = 1, . . . , Nblocks , where Nblocks is the number
of blocks, which is equal to the number of spectra after DFT. The transfer function H1u ( f )
from the input force F to a quantity of interest u was computed using

GuF
H1u ( f ) = , (12)
GFF
Acoustics 2024, 6 76

where GuF is the cross power spectral density between the quantity of interest u and the
input force F, and GFF is the auto power density spectrum of the input force F. GuF and
GFF are computed from

Nblocks Nblocks
1 1
GuF =
Nblocks ∑ ui ( f ) F̃i∗ ( f ) , GFF =
Nblocks ∑ F̃i ( f ) F̃i∗ ( f ) , (13)
i =1 i =1

where superscript ∗ denotes the complex conjugate. The quantity of interest ui ( f ) is one of
p
the pressure spectra p̃mic1
i ( f ), p̃mic2
i ( f ), p̃mic3
i ( f ), or p̃mic4
i ( f ), such that H1 ( f ) is the desired
vibroacoustic transfer function. This resulted in one transfer function for each microphone
position and, thus, four transfer functions in total.

4.1.2. Surface Velocity Measurements Using LSV


The surface velocity of the DUT was measured using an LSV (Polytec PSV-500 Xtra),
as described in Section 2.1. The LSV was positioned in four positions, as depicted in
Figure 9, such that the whole DUT surface could be measured. This resulted in an overall
number of measurement points NLSV = 688. Figure 10 shows the distribution of scanning
points for each measurement patch. The four patches of the measurement data were
combined during the data transfer process described in Section 4.2.

LSV Pos. 1
LS
V
Po
s. 2 Impedance head
Stinger LSV Pos. 4 DUT LSV Pos. 3

Impedance head
LSV Pos. 1 Stinger

Shaker
Shaker
DUT

Shaker

Table
Rubber mount

Figure 9. Sketch of the experimental setup with the LSV. The coordinate system’s origin is located in
the center of the device under test (DUT). The LSV is positioned approximately 0.7 m from the DUT.

Figure 10. Discretized surface of the DUT with the indicated positions of scanning points during
LSV measurement. The colors red, purple, green, and orange refer to LSV positions 1, 2, 3, and 4,
respectively.

From the measured surface velocity and the input forces, a transfer function from input
force F to surface velocity v was computed with Equations (12) and (13). Thereby, the quan-
tity of interest u is the velocity spectrum. This form of the H1v ( f ) transfer function is the
Acoustics 2024, 6 77

so-called mobility (p. 52, [30]). The measurement data post-processing was performed using
the software Polytec PSV Software, which directly computed the mobility transfer functions.

4.2. BEM Simulation Model


The simulation model was set up according to Section 2.3. Therefore, the surface of a
CAD representation of DUT, Γs , was discretized with linear quadrilateral and triangular
elements with an approximate edge length of 6 mm, resolving the maximum frequency
of interest at 5 kHz with approximately 11.4 elements per wavelength. This resulted in
7158 degrees of freedom. The computational grid is depicted in Figure 10. The mobility
transfer function H1v was transferred to Γs using the presented workflow in Section 2.1
for each measurement patch individually, and subsequently summed up. It should be
noted that the summation of the velocity components was feasible, as the measurement
directions of the measurement patches that shared overlapping regions were approximately
orthogonal, and thus the resulting velocity vector did not overestimate the true velocity in
any direction. Furthermore, inside non-overlapping regions, the measurement direction
was chosen to match the normal direction of each surface patch as well as possible. One-
dimensional LSV measurement of oblique-angled motion led to a geometrically induced
systematic error [31]. Concerning these systematic errors, deviations in measurement
direction from the surface normal direction led to an underestimation of the radiated sound.
For high wave numbers, a phase error could also be significant. In general, a comparison of
the obtained surface velocity data to single-point measurements or a 3D-LSV measurement
might be required to approve the usage of 1D LSV data. Areas of the DUT that were
inaccessible for measurement were set to zero. Thin structures were assumed as thin plates,
based on the Kirchhoff–Love plate theory, and, therefore, only flexural plate waves were
considered. Consequently, the inner wall surfaces were projected onto the measurement
patches of the outer surfaces. All of the aforementioned simplifying assumptions should
result in underestimating the radiated sound power, as parts of the DUT’s oscillation
were neglected. Potentially, neglecting destructive local acoustic effects could also lead to
overestimating the radiated sound; however, this would be unlikely at low frequencies.
The mobility data on Γs were then interpolated onto the element centers and projected in
the normal direction of the respective element. Finally, the Neumann boundary condition
was computed according to Equation (8), modeling the acoustic radiation due to a unit
p
force excitation yielding the vibroacoustic transfer function H1 .

4.3. Comparison of Measurement and Simulation


p
The vibroacoustic H1 transfer functions computed from the force measurement and
microphone measurements or LSV measurement in combination with BEM simulation, re-
spectively, are depicted in Figure 11. For all four of the evaluated positions, good agreement
between the two workflows was achieved. Resonance peaks in the transfer function due to
resonance of the structural-mechanical system, the acoustic system, or coupled resonance
were captured equally well, especially at medium frequencies. In the low-frequency range
below 500 Hz, the microphone measurement overestimated the transfer function compared
with the simulation approach. While the used anechoic room was found to guarantee
sufficient insulation from the environment above 100 Hz, acoustic free-field conditions
were violated with a significant increase in reverberation time below 250 Hz, based on the
manufacturer’s information. Furthermore, reference measurements of the sound field with
a removed stinger connection from the shaker to the DUT revealed significant inefficiency
in the decoupling of the shaker with SPL being approximately 10 dB higher below 500 Hz
to the higher frequency range. Especially at 300 Hz and 480 Hz, resonance peaks could be
observed. This behavior was most definitely caused by the inefficient decoupling of the
shaker from the measurement environment both mechanically via induced vibrations to
the table and acoustically due to insufficient acoustic insulation of the shaker.
A quantitative measure of similarity between two frequency response functions was
provided by the Cross Signature Scale Factor (CSSF) [32], which is a frequency-dependent
Acoustics 2024, 6 78

comparison yielding a value in the interval [0, 1]. From the measured transfer function
p p
H1,meas and the simulated transfer function H1,sim , CSSF was computed with Equation (2)
of [32], such that
 ∗
p p
2 H1,meas ( f ) · H1,sim ( f )
CSSF( f ) =  ∗  ∗ , (14)
p p p p
H1,meas ( f ) · H1,meas ( f ) + H1,sim ( f ) · H1,sim ( f )

where superscript ∗ denotes the complex conjugate. A CSSF value close to 100% indicates
excellent agreement of the transfer functions. In Figure 11, CSSF is plotted together with
the measured and simulated transfer functions. Therefrom, it is clear that the simulation
was able to predict the transfer function with a high degree of conformity over a large
frequency range.

100 100

10!1 10!1
jH1p j (Pa/N)

jH1p j (Pa/N)
CSSF (%)

CSSF (%)
10!2 100 10!2 100

50 50

10!3 0 10!3 0
simulation (Le, =97.58 dB) simulation (Le, =96.97 dB)
measurement (Le, =97.70 dB) measurement (Le, =99.91 dB)
10!4 10!4
1 2 3 4 5 1 2 3 4 5
f (kHz) f (kHz)
(a) Mic. 1 (b) Mic. 2
100 100

10!1 10!1
jH1p j (Pa/N)
jH1p j (Pa/N)

CSSF (%)
CSSF (%)

10!2 100 10!2 100

50 50

10!3 0 10!3 0
simulation (Le, =98.18 dB) simulation (Le, =96.69 dB)
measurement (Le, =98.09 dB) measurement (Le, =99.61 dB)
10!4 10!4
1 2 3 4 5 1 2 3 4 5
f (kHz) f (kHz)
(c) Mic. 3 (d) Mic. 4
p
Figure 11. H1 transfer functions based on microphone measurements and LSV measurements in
combination with the BEM simulation. The effective pressure level (according to Equation (11), in this
case for a unit force excitation) was evaluated for f interes = [500, 5000] Hz.

Table 2 lists the effective pressure level for a unit force excitation, as defined in (11) in
the frequency range f interes = [500, 5000] Hz. The maximum deviation of the simulation
approach to the acoustic measurement of 2.94 dB indicates a good prediction of the average
vibroacoustic transfer path in the medium- to high-frequency range above 500 Hz. In
general, a slight underestimation of the measured amplitude spectra could be observed in
this frequency range. This behavior could be explained by the simplifications discussed
Acoustics 2024, 6 79

in Section 4.2. Furthermore, the oblique laser direction, which does not coincide with the
normal velocity at every measurement point, could contribute to the deviations. Larger
deviations at certain microphone positions and frequencies, e.g., at the position of Mic. 2 in
the interval [2, 3] kHz, could be due to the imperfect free-field in the experimental setup.

Table 2. Effective pressure level for a unit force excitation in the frequency range f interes =
[500, 5000] Hz at the microphone positions according to Figure 8.

Microphone Position Lmeas


eff LBEM
eff Lmeas BEM
eff − Leff
1 97.70 dB 97.58 dB 0.12 dB
2 99.91 dB 96.97 dB 2.94 dB
3 98.09 dB 98.18 dB 0.09 dB
4 99.61 dB 96.69 dB 2.92 dB

4.4. Radiated Sound Power


One of the most important quantities to assess
H the sound emissions of a machine
or product is the radiated sound power P = Γ I · ndΓ , with the acoustic intensity
I = 12 <{ pv∗ }, and the conjugate complex indicated by superscript ∗ . Having solved the
BEM problem as defined in Section 4.2, and achieved acceptable agreement with validation
measurements, as discussed in Section 4.3, the sound power can be calculated by

ps qs∗
 
1
Z
PBEM = < dΓ , (15)
2 Γs −jωρ

using Equation (8). Figure 12 shows the predicted sound power in comparison with the
common low-cost approximations based on the equivalent radiated power (ERP) PERP or
based on the volume velocity PVV , as defined in Equations (10) and (19) of [22], respectively.
The two low-cost approximations can be understood as qualitative upper and lower bounds
for vibroacoustically emitted sound power. ERP uses the far-field approximation, resulting
in a constant radiation efficiency σ = 1 for all surface areas and neglecting local acoustic
effects. Therefore, dipole effects (antiphase vibration of the sources) were not captured,
and the radiated sound power was overestimated in the low-frequency range. Towards
higher frequencies, these effects were less significant, and thus the deviation decreased.
In contrast, the sound power approximation based on the volume velocity assumed an
acoustically compact source, which was only viable in the very low-frequency range. Both
estimates could be evaluated explicitly as simple summations without the need to assemble
a system matrix and thus could be computed in the order of seconds on any state-of-
the-art notebook. In contrast, BEM, as presented in Section 2.2, required an assembly
of system matrices, and resulted in an algebraic system of equations that needed to be
solved. The presented simulation of the prosthesis’ frame required approx. 14 h on a
single CPU-core of the used system, but it scaled well on a multi-core CPU, reducing the
actual computation time to approx. 14 min. The computations were performed on a single
compute node of the Vienna Scientific Cluster (VSC) equipped with an AMD EPYC 7713
(64 core) CPU, and 512 GB RAM. Consequently, an estimate of where to use the presented
BEM approach was viewed as useful to the authors. The frequency range in which the
radiated sound power based on BEM simulation deviated from both approximations is
given by the Helmholtz number He = λL = kL ≈ 1 with length scale L. Consequently, if the
frequency range of interest intersects with
c
f ≈ , (16)
2πL
the BEM approach has to be used to produce accurate predictions. Choosing L as the main
dimensions of DUT results in He ≈ 1 in the frequency range f ≈ [200, 700] Hz, which
agrees with the results shown in Figure 12.
Acoustics 2024, 6 80

Figure 12. Radiated sound power of DUT computed from the BEM simulation in comparison with
common low-cost approximations equivalent radiated power PERP , and sound power based on
volume velocity PVV .

5. Discussion
This paper presents a simulation approach based on BEM, which utilizes interpolated
surface velocities as the boundary conditions. The interpolation algorithm for the surface
velocities is based on evaluating the finite element basis functions at the target mesh.
A verification of the BEM solver was achieved in a previous study. For validating the
simulation approach, surface velocities simulated by FEM were taken as inputs with
varying mesh sizes. The acoustic propagation results of the validation simulation computed
with BEM were in very good agreement with measurements and the comparison simulation
with FEM, as shown in Table 1. After applying the verified and validated simulation
approach to compute the acoustic radiations of a clinically applied knee prosthesis frame,
very good agreements were found between the simulated and measured effective acoustic
pressure levels, as shown in Table 2.
Acoustic propagation simulations using BEM implies ideal free-field radiation condi-
tions for the whole frequency range. For low frequencies, this can not be easily achieved
in real-world measurement environments, such as anechoic chambers. Therefore, the pre-
sented method allows for the prediction of free-field acoustic emissions based solely on LSV
measurements, which can also be performed under non-ideal acoustic conditions, i.e., a
general room that is not anechoic, provided that no acoustic feedback occurs. In addition,
based on the sound emission simulation results, further acoustic quantities, such as the
free-field radiated acoustic power, can be evaluated. This can be especially useful, if the
frequency of interest is close to He ≈ 1, where other low-cost approximation methods fail
to provide accurate predictions for the radiated acoustic power.
Finally, the fact that the validation example and its application to a real-world prob-
lem exhibit very good agreements between (i) the BEM simulation results, (ii) the FEM
results, and (iii) acoustic pressure measurements suggests that the presented approach may
generalize well to other problems with a similar geometric complexity.
Acoustics 2024, 6 81

Author Contributions: Conceptualization, A.W. and S.S.; methodology, A.W.; software, A.W. and
P.R.; validation, A.W. and F.K.; formal analysis, A.W.; investigation, A.W., F.K., B.M.-M. and P.M.;
resources, P.M. and H.S.; data curation, A.W. and F.K.; writing—original draft preparation, A.W.,
F.K., P.R. and S.S.; writing—review and editing, A.W., F.K., P.M., P.R., B.M.-M., H.S., M.K. and S.S.;
visualization, A.W. and F.K.; supervision, S.S., M.K. and H.S.; project administration, S.S.; fund-
ing acquisition, S.S., A.W. and P.R. All authors have read and agreed to the published version of
the manuscript.
Funding: The support from the Austrian Research Promotion Agency (FFG) under no. 874784 and
from Otto Bock Healthcare Products GmbH are gratefully acknowledged. P.R. acknowledges the
support by the Hungarian National Research, Development and Innovation Office under contract
No.K̃–143436. Open Access Funding by the Graz University of Technology.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The simulation data presented in this study are available on request
from the corresponding author. The LSV measurement data and the geometry data presented in
Section 4 are not publicly available due to a non-disclosure agreement with Otto Bock Healthcare
Products GmbH, Vienna, Austria.
Acknowledgments: The computational results presented have been achieved in part using the
software NiHu [20] and in part using the software openCFS [Nearest-Neighbor Interpolation] [18].
The authors acknowledge computational resources provided by the Vienna Scientific Cluster (VSC).
Conflicts of Interest: The authors declare no conflicts of interest.

Abbreviations
The following abbreviations are used in this manuscript:

BEM Boundary element method


CAD Computer-aided design
DUT Device under test
ERP Equivalent radiated power
FEM Finite element method
HBIE Hypersingular boundary integral equation
LSV Laser scanning vibrometry/vibrometer
SPL Sound pressure level

References
1. Christensen, R. Acoustic Modeling of Hearing Aid Components. Ph.D. Thesis, Syddansk Universitet, Odense, Denmark, 2010.
2. Candy, J. Accurate calculation of radiation and diffraction from loudspeaker enclosures at low frequency. J. Audio Eng. Soc. 2013,
61, 356–365.
3. Nuraini, A.; Mohd Ihsan, A.; Nor, M.; Jamaluddin, N. Vibro-acoustic analysis of free piston engine structure using finite element
and boundary element methods. J. Mech. Sci. Technol. 2012, 26, 2405–2411. [CrossRef]
4. Panda, K.C. Dealing with noise and vibration in automotive industry. Procedia Eng. 2016, 144, 1167–1174. [CrossRef]
5. Oberst, S.; Lai, J.; Marburg, S. Guidelines for numerical vibration and acoustic analysis of disc brake squeal using simple models
of brake systems. J. Sound Vib. 2013, 332, 2284–2299. [CrossRef]
6. Bates, T.P.; Bacon, I.C.; Blotter, J.D.; Sommerfeldt, S.D. Vibration-based sound power measurements of arbitrarily curved panels.
J. Acoust. Soc. Am. 2022, 151, 1171–1179. [CrossRef] [PubMed]
7. Yeh, F.; Chang, X.; Sung, Y. Numerical and Experimental Study on Vibration and Noise of Embedded Rail System. J. Appl. Math.
Phys. 2017, 5, 1629. [CrossRef]
8. Vlahopoulos, N.; Vallance, C.; Stark, R.D., III. Numerical approach for computing noise-induced vibration from launch
environments. J. Spacecr. Rocket. 1998, 35, 355–360. [CrossRef]
9. Rossignol, K.S.; Lummer, M.; Delfs, J. Validation of DLR’s sound shielding prediction tool using a novel sound source. In
Proceedings of the 15th AIAA/CEAS Aeroacoustics Conference (30th AIAA Aeroacoustics Conference), Miami, FL, USA, 11–13
May 2009 ; p. 3329. [CrossRef]
10. Lummer, M.; Akkermans, R.A.; Richter, C.; Pröber, C.; Delfs, J. Validation of a model for open rotor noise predictions and
calculation of shielding effects using a fast BEM. In Proceedings of the 19th AIAA/CEAS Aeroacoustics Conference, Berlin,
Germany, 27–29 May 2013 ; p. 2096. [CrossRef]
Acoustics 2024, 6 82

11. Kumar, S.; Forster, H.M.; Bailey, P.; Griffiths, T.D. Mapping unpleasantness of sounds to their auditory representation. J. Acoust.
Soc. Am. 2008, 124, 3810–3817. [CrossRef]
12. Preuss, S.; Gurbuz, C.; Jelich, C.; Baydoun, S.K.; Marburg, S. Recent advances in acoustic boundary element methods. J. Theor.
Comput. Acoust. 2022, 30, 2240002. [CrossRef]
13. Fritze, D.; Marburg, S.; Hardtke, H.J. FEM–BEM-coupling and structural–acoustic sensitivity analysis for shell geometries.
Comput. Struct. 2005, 83, 143–154. [CrossRef]
14. Kirkup, S. The boundary element method in acoustics: A survey. Appl. Sci. 2019, 9, 1642. [CrossRef]
15. Li, Y.; Meyer, J.; Lokki, T.; Cuenca, J.; Atak, O.; Desmet, W. Benchmarking of finite-difference time-domain method and fast
multipole boundary element method for room acoustics. Appl. Acoust. 2022, 191, 108662. [CrossRef]
16. Chen, L.; Lian, H.; Natarajan, S.; Zhao, W.; Chen, X.; Bordas, S. Multi-frequency acoustic topology optimization of sound-
absorption materials with isogeometric boundary element methods accelerated by frequency-decoupling and model order
reduction techniques. Comput. Methods Appl. Mech. Eng. 2022, 395, 114997. [CrossRef]
17. Deckers, E.; Atak, O.; Coox, L.; D’Amico, R.; Devriendt, H.; Jonckheere, S.; Koo, K.; Pluymers, B.; Vandepitte, D.; Desmet, W. The
wave based method: An overview of 15 years of research. Wave Motion 2014, 51, 550–565. [CrossRef]
18. Schoder, S.; Roppert, K. openCFS-Data: Data Pre-Post-Processing Tool for openCFS–Aeroacoustics Source Filters. arXiv 2023,
arXiv:2302.03637. [CrossRef]
19. Marburg, S. Boundary Element Method for Time-Harmonic Acoustic Problems. In Computational Acoustics; Kaltenbacher, M., Ed.;
Springer International Publishing: Cham, Switzerland, 2018; pp. 69–158. [CrossRef]
20. Fiala, P.; Rucz, P. NiHu: An open source C++ BEM library. Adv. Eng. Softw. 2014, 75, 101–112. [CrossRef]
21. Maurerlehner, P.; Schoder, S.; Tieber, J.; Freidhager, C.; Steiner, H.; Brenn, G.; Schäfer, K.H.; Ennemoser, A.; Kaltenbacher, M.
Aeroacoustic formulations for confined flows based on incompressible flow data. Acta Acust. 2022, 6, 45.[CrossRef]
22. Fritze, D.; Marburg, S.; Hardtke, H.J. Estimation of radiated sound power: A case study on common approximation methods.
Acta Acust. United Acust. 2009, 95, 833–842. [CrossRef]
23. Rothberg, S.; Allen, M.; Castellini, P.; Di Maio, D.; Dirckx, J.; Ewins, D.; Halkon, B.; Muyshondt, P.; Paone, N.; Ryan, T.; et al. An
international review of laser Doppler vibrometry: Making light work of vibration measurement. Opt. Lasers Eng. 2017, 99, 11–22.
[CrossRef]
24. Burton, A.J.; Miller, G.F. The application of integral equation methods to the numerical solution of some exterior boundary-value
problems. Proc. R. Soc. Lond. 1971, 323, 201–210. [CrossRef]
25. Marburg, S. The Burton and Miller Method: Unlocking Another Mystery of Its Coupling Parameter. J. Comput. Acoust. 2016,
24, 1550016. [CrossRef]
26. Fiala, P.; Rucz, P. NiHu: A multi-purpose open source fast multipole solver. In Proceedings of the DAGA 2019. German
Acoustical Society (DEGA), Rostock, Germany, 18–21 March 2019; pp. 1262–1265.
27. Schenck, H.A. Improved Integral Formulation for Acoustic Radiation Problems. J. Acoust. Soc. Am. 1968, 44, 41–58. [CrossRef]
28. Maurerlehner, P. Aero- and Vibroacoustics of Confined Flows. Ph.D. Thesis, TU Graz, Graz, Austria, 2023.
29. Mathworks. Matlab documentation: Stft(), 2023 . Available online: https://de.mathworks.com/help/signal/ref/stft.html
(accessed on 11 October 2023) .
30. Avitabile, P. Modal Testing: A Practitioner’s Guide; John Wiley & Sons Ltd: Chichester, UK, 2017. [CrossRef]
31. Neumann, M.N.; Hennings, B.; Lammering, R. Identification and Avoidance of Systematic Measurement Errors in Lamb Wave
Observation With One-Dimensional Scanning Laser Vibrometry. Strain 2013, 49, 95–101. [CrossRef]
32. Dascotte, E.; Strobbe, J. Updating finite element models using FRF correlation functions. In Proceedings of the 17th International
Modal Analysis Conference, Kissimmee, FL, USA, 8–11 February 1999 ; pp. 1169–1174.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.

You might also like