Applied Acoustics 129 (2018) 322–334

Contents lists available at ScienceDirect

Applied Acoustics
journal homepage: www.elsevier.com/locate/apacoust

An improved impedance eduction technique based on impedance

models and the mode matching method
André M.N. Spillere ⇑, Augusto A. Medeiros, Julio A. Cordioli
Acoustics and Vibration Laboratory, Federal University of Santa Catarina, Florianópolis 88040-900, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: The problem of determining the acoustic impedance of liners used in turbofan engines of commercial air-
Received 21 March 2017 craft has been a point of interest for the scientific community for decades, especially in the presence of
Received in revised form 9 August 2017 grazing flows, similar to operational conditions. Different techniques have been developed to determine
Accepted 16 August 2017
liner acoustic impedance under grazing flow. More recent research have been focused on inverse meth-
ods, which consist of two steps: (i) measurement of the acoustic field in a rectangular duct with flow and
a liner sample located at one or two walls of the duct and (ii) modelling of the acoustic field and appli-
Keywords:
cation of an optimization procedure to find the impedance that minimize the difference between exper-
Acoustic liner
Impedance eduction
imental and analytic results. The process is usually carried on using optimization techniques and
Impedance model performed at each frequency step, which can lead to discontinuous impedance curves and large compu-
Aeroacoustics tational costs. In this work, the mode matching method is discussed in detail, and a new technique for
impedance determination is proposed, which incorporates a mathematical impedance model to the opti-
mization process as a mean to improve the impedance curve, therefore suppressing measurements error
and convergence issues at single frequencies. The impedance models considered here are given in the fre-
quency domain and satisfy passivity, reality and causality conditions, which allows the use of the educed
impedance in time domain simulations. Three models are considered and compared regarding different
liner samples, flow velocities and wave propagation direction. The results show that the impedance mod-
els can successfully suppress convergence errors close to the liner resonance frequency.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction noise reduction [2] and can be seen as an array of Helmholtz res-
onators. Therefore, good sound attenuation is achieved over a nar-
There are several noise sources on modern aircraft, including row frequency band, which can be chosen based on the liner
airframe noise (fuselage, wings, flaps, landing gear, etc.) and engine geometry, in order to comprise the critical BPF on a specific flight
noise (fan, turbine, jet, etc.). In many cases, specially for the phases condition. More complex liner geometries have also been
considered in the certification tests, the engine can be the domi- employed when a more broadband noise reduction is desired [3].
nating source [1]. High bypass ratio turbofan engines became lar- Liner performance is generally accounted for by means of its
gely employed in the civil aviation, mostly because of fuel acoustic impedance. However, typical aero-engine liner impedance
savings and significant reduction of jet noise. This brought atten- is known to be dependent on the liner geometry [4], temperature
tion to other engine noise sources, like the main fan and the engine [5], grazing flow [6–9] and sound pressure level [10–12]. Because
core. Since fan noise is characterized by dominating discrete tones it is not trivial to measure liner impedance under these conditions,
associated with blade passage frequencies (BPF), acoustic treat- several methods were proposed in the last decades.
ment of the fan ducts is achieved by covering the interior nacelle A group of methods, usually called impedance eduction meth-
walls with lining materials tuned to these frequencies. ods or inverse methods, have similar procedures. First, the acoustic
Most current liners are composed of three different parts: a per- field in a duct with a liner sample at the wall subject to grazing
forated plate subject to grazing flow, a honeycomb core and a rigid flow is measured. Then, a numerical or analytical model simulates
backplate. This design is the best trade-off between weight and the acoustic field in the duct, and an optimization routine is
applied to find the impedance that minimizes a certain criteria,
⇑ Corresponding author. such as the difference between experimental and simulated
E-mail addresses: [email protected] (A.M.N. Spillere), Augusto.Medeir- results.
[email protected] (A.A. Medeiros), [email protected] (J.A. Cordioli).

http://dx.doi.org/10.1016/j.apacoust.2017.08.014
0003-682X/Ó 2017 Elsevier Ltd. All rights reserved.
 A.M.N. Spillere et al. / Applied Acoustics 129 (2018) 322–334 323

The main difference between the many published methods is This work focus on the mode matching technique, which results
how the acoustic field is simulated in the presence of the liner in a systems of equations to be solved at each frequency and, there-
sample and grazing flow. There are methods based on Finite Ele- fore, an optimum impedance at each frequency is to be found
ment Method (FEM) simulations [13], two-port systems [14,15], (hereby called the single-frequency mode matching). A modifica-
mode-matching [16], solutions to the convected Helmholtz tion to the mode matching method is proposed to incorporate an
equation [8], to the linearized Euler equations [17], to the impedance model to the eduction process. Instead of finding the
Pridmore-Brown equation [8], among many others. Jing et al. complex impedance at each frequency, the proposed method finds
[18] also proposed a method that does not require an optimization the model parameters, that defines it for all frequencies under
in order to find the impedance, but instead calculates it directly analysis (hereby called the multiple-frequency mode matching).
from the test data. To simplify their derivation and application, This always results in smooth impedance results since the
most of these methods assume that only plane-waves propagate employed function is smooth, and facilitates interpolation and
in the hard-wall sections of the test ducts, i.e, the highest fre- extrapolation to obtain the impedance at other frequencies. Due
quency under analysis is lower then the cut-on frequency of the to the frequency domain nature of this technique, only a few min-
first transverse mode of the duct. utes are necessary to obtain an impedance curve.
Researchers at NASA have developed several indirect tech- The impedance models analysed in this work were developed
niques. One of them, the Pressure Gradient Method (PGM) [13], for time domain simulations. In order to guarantee a physical
uses a 2D FEM to model the acoustic propagation along the lined behaviour, these models have to satisfy fundamental conditions
section of a duct for a given impedance value. The impedance guess in the frequency domain, i.e. causality, reality and passivity condi-
is varied in an optimization until the calculated pressure gradients tions. Different mathematical approaches have been investigated
in the axial direction at the beginning and ending of the liner sam- to guarantee these conditions, including rational and multipole
ple match the measured ones. This is a very robust method because functions [24,25], and also an extension of the basic Helmholtz res-
of the FEM approach. It has since been extended to handle higher onator model [23]. Each model has its own advantages when trans-
order modes cut-on in the hard-wall sections [19]. This configura- lated into the time domain, but this work is focused on finding the
tion, however, requires dozens of microphones in order to decom- model that best represents the educed impedance of acoustics lin-
pose the acoustic field, which is rather cumbersome and not ers, with and without grazing flow. Semi-empirical models based
available at all laboratories. on viscous and radiation effects at small holes and non-liner cor-
The two-port method, first proposed by Roeck and Desmet [20], rections due to grazing flow [26–28] are not included since it has
uses an analytic transfer matrix to describe a lined section of a duct not been proven that they satisfy the fundamental conditions [23].
with flow. This matrix is a function of the wavenumbers in the This work is divided as following: Section 2 introduces the rel-
lined section, which are found in an optimization procedure and evant equations in order to derive the acoustic field in a duct with
then used to calculate the liner impedance. It showed promising flow. In Section 2.1 the single-frequency mode matching technique
but unstable results, and was later extended by Santana et al. is described in detail [29,16]. Section 2.2 shows how the impe-
[14] to better handle the effects of the hard-soft wall transitions, dance model can be used in the mode matching method, thus
solving most of the instabilities seen in Roeck and Desmet prelim- resulting in the multiple-frequency mode matching. Section 3
inary results. introduces the impedance models as proposed by Ozyörük and
In the mode matching technique, Elnady et al. [16] used the Long [24], Li et al. [25] and Rienstra [23]. Section 4 presents the
mode-matching technique to couple the acoustic fields in hard- experimental apparatus available at the Federal University of Santa
wall and lined sections of a duct, from which the amplitudes of Catarina, and then Section 5 shows the results obtained using
the propagating modes are found for a given impedance guess. the single-frequency mode matching method and the multiple-
The acoustic field is then calculated and compared to the measured frequency mode matching method. The main conclusions are pre-
one until the error is minimized by finding a matching impedance. sented in Section 6.
In the straightforward method proposed by Jing et al. [18], the
acoustic field along the section with the liner sample is measured 2. Impedance eduction technique
at equally-spaced positions. The well-known Pronys method is
used to approximate the acoustic field as a series of complex expo- This section will first introduce the basic equations for the
nentials, whose exponents give the axial wavenumbers, and thus acoustic field inside a duct in the presence of a lined section. Then,
the liner impedance. This method differs from the two-port the specific set of equations for the mode matching method is pre-
method and the mode matching method because it is not iterative; sented. On the following, the impedance model is coupled in the
the wavenumbers, and then the impedance, are calculated directly optimization routine, resulting in the multiple-frequency mode
from the acoustic pressure measurements. This approach has the matching method.
advantages of not requiring an optimization, thus avoiding its The basic geometry is shown in Fig. 1, which shows a rectangu-
potential drawbacks, such as dependence on initial guesses and lar duct with width b and height h and whose walls are rigid except
difficult convergence in certain situations. for a section of length L where an impedance Z w is applied at the
Some of the aforementioned methods require a high number of wall x ¼ 0. The duct can be seen as having three different sections:
microphones and lead to large computational costs, which is not a hard-wall inlet Section (1), then a lined Section (2), which repre-
always available nor desirable. Even so measurement error or sents the region with the liner in the experimental configuration,
bad convergence of the optimization at some frequencies might followed by a hard-wall outlet Section (3). Uniform flow in the pos-
result in a non-smooth impedance curve, which is not a physical itive z-direction is assumed in all sections.
result. It is necessary to include in the eduction technique an impe- In a duct with uniform mean flow in the axial ðzÞ direction, the
dance model that represents the behaviour of an acoustic liner in convected wave equation [30] for linear acoustics is given by
order to avoid such discontinuities. Such similar approach has been
used by Richter et al. [21] and Busse-Gerstengarbe et al. [22] by 1 D2 p
including the Extended Helmholtz resonator (EHR) model [23] into
r2 p  ¼ 0; ð1Þ
c20 Dt 2
an impedance eduction technique based on a time domain compu-
tational aeroacoustics method. As a drawback, the overall time for where p is acoustic pressure, c0 is the speed of sound in the fluid,
a single impedance eduction is around one day [22]. D=Dt is the material derivative and r2 is the Laplace operator.
324 A.M.N. Spillere et al. / Applied Acoustics 129 (2018) 322–334

ð1Þ ð1Þ
b frequency in the y direction, ky2i ¼ ky2r ¼ 0, and from the dispersion
relation (Eq. (4)), the wave numbers in the x direction are
(1) rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2  2
ðqÞ ðqÞ ðqÞ
h kx2i ¼ k0  Mkz2i  kz2i ð7Þ
(2)
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
(3)  2  2
ðqÞ ðqÞ ðqÞ
Zw kx2r ¼ k0  Mkz2r  kz2r ð8Þ

By taking the expression for the acoustic field in the duct, Eq. (3),
y and applying the Ingard-Myers boundary condition [31] for the
L
impedance at x ¼ 0, the following equation is derived:
x
!2
z k0
ðqÞ
kz2i ðqÞ
Z w ¼ iZ 0 ðqÞ
1M cotðkx2i bÞ; ð9Þ
Fig. 1. Rectangular duct with height h and width b, whose wall at x ¼ 0 has an kx2i k0
impedance Z w along a section of length L.
where Z 0 is the characteristic impedance of the fluid, given by the
product of its density and speed of sound.
Eq. (9) is important because it relates the impedance on the wall
Eq. (1) possesses infinite solutions, each one representing a mode,
to the wave numbers in the duct, allowing to calculate the wall
that propagates only above its cut-on frequency f c , given by
impedance once the wavenumbers of the acoustic field are known.
c0 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi2ffi Conversely, it also allows one to calculate the wave numbers in the
fc ¼ 1M ; ð2Þ
2h duct for a given impedance value. Although the stability of the
Ingard-Myers boundary condition has been investigated over the
where M is Mach number of the uniform mean flow in the duct, and
past years and alternative boundary conditions have been pro-
h could be replaced by b for the modes in the other direction. The
posed [6,32,33], it is not in the scope of this work to evaluate the
acoustic pressure field in each of the three sections,
accuracy of Ingard-Myers boundary condition against others. Also,
pn ðx; y; zÞ; ðn ¼ 1; 2; 3Þ, is calculated by the summation of the Q
the classical Ingard-Myers boundary condition was present in the
modes in the duct section, whose mode shapes are given by
original work from Elnady et al. [29], and therefore this condition
Uðx; yÞ, so that
will remain unchanged.
X
Q
ðqÞ ðqÞ ðqÞ X
Q
ðqÞ
ðqÞ ikznr
pn ðx; y; zÞ ¼ ani Uni eikzni z þ aðqÞ
nr Unr e
z
; ð3Þ 2.1. Single-frequency mode matching
q¼1 q¼1

where the indexes i and r represent the incident and reflected In this section, the mode matching technique will be described
waves which propagate respectively in the zþ and z directions, q in more detail. The mode matching technique takes into account as
is the index of the mode, ordered by its cut-off frequency, aðqÞ is many modes as necessary in each section. It still assumes that the
ðqÞ only mode propagating towards both sides of the lined section is a
the amplitude of each mode, and kzn is the wavenumber in the z
plane wave mode, such that the acoustic fields in Sections 1–3 can
direction for the q-th mode, that satisfies the dispersion relation
be written as:
2 2 2 2
kx þ ky þ kz ¼ ðk0  Mkz Þ ; ð4Þ X
Q
ð1Þ ðqÞ
ð1Þ ð1Þ ðqÞ ðqÞ
p1 ðx; y; zÞ ¼ a1i U1i eikz1i z þ a1r U1r eikz1r z ; ð10Þ
where k0 ¼ x=c0 is the acoustic wavenumber. The wavenumber in q¼1
the cross-sectional directions, kx and ky , are implicit in the mode
shapes Uðx; yÞ. The eixt harmonic time dependence is omitted for X
Q
ðqÞ X
Q
ðqÞ
ðqÞ ðqÞ ðqÞ ðqÞ
brevity in Eq. (3) and all equations derived from it. p2 ðx; y; zÞ ¼ a2i U2i eikz2i z þ a2r U2r eikz2r ðzLÞ ; ð11Þ
q¼1 q¼1
An hypothesis made in the following sections is that only plane
waves propagate in the section with hard walls, i.e., the maximum
frequency under analysis should be below the lowest cut-on fre- X
Q
ðqÞ ðqÞ ðqÞ
ð1Þ ð1Þ ð1Þ
p3 ðx; y; zÞ ¼ a3i U3i eikz3i ðzLÞ þ a3r U3r eikz3r ðzLÞ : ð12Þ
quency in the duct, given by Eq. (2). That means that if only plane q¼1
waves are incident in Section 1, higher-order modes generated at
the interface with Section 2 will decay exponentially with axial In Eq. (10) the summation only occurs for the reflected modes, since
distance. The same happens with modes generated at the interface the only incident mode is the plane-wave mode (q ¼ 1). Analo-
of Section 3 to Section 2. gously, in Eq. (12) the summation only occurs for the incident
Using the hard-wall boundary condition at all walls in Section 1 modes, since the only reflected mode (which propagates in z  ,
and 3, the wave numbers and the mode shapes can be calculated. i.e., towards Section 2) is the plane-wave mode.
The wave numbers in the x and y directions for the q-th mode are Mode matching is the name of the technique used to determine
how acoustic energy is transferred from one duct section to the
kxi ¼ kxr ¼ pðq  1Þ=b;
ðqÞ ðqÞ
ð5Þ other through the propagating modes at the discontinuities, as
seen in Fig. 2. First, it is assumed continuity of pressure and axial
kyi ¼ kyr ¼ pðq  1Þ=h;
ðqÞ ðqÞ
ð6Þ particle velocity at the interface of Section 1 with the lined section
(z ¼ 0), and then at the interface of the lined section with Section 3
with the mode shapes having a cosine form for symmetric modes (z ¼ L), so that:
and a sine form for anti-symmetric modes, in each direction [16].
In the lined section, the boundary conditions are symmetric in the
p1 ðx; y; 0Þ ¼ p2 ðx; y; 0Þ; ð13Þ
y direction, and, similarly to what occurs in the hard-wall sections,
p2 ðx; y; LÞ ¼ p3 ðx; y; LÞ; ð14Þ
the wave numbers can be found from Eq. (6). Below the first cut-on
 A.M.N. Spillere et al. / Applied Acoustics 129 (2018) 322–334 325

Upstream Downstream
Loudspeaker x Loudspeaker

(1) (2) (3)

(q)
p(1)
1i
(q)
p1r p2i
(q)
p2r p(q)
3i p(1)
3r

Flow
1 2 3 4 Liner sample 5 6 7 8
z
Microphones Microphones

Fig. 2. Representation of the acoustic field in the mode matching method and schematic view of the test rig. The microphones are numbered from 1 to 8, from the left to the
right. When the downstream loudspeaker is used the indexes i and r are switched.

uz1 ðx; y; 0Þ ¼ uz2 ðx; y; 0Þ; ð15Þ 2.2. Multiple-frequency mode matching

The problem of convergence to local minima can be aggravated

uz2 ðx; y; LÞ ¼ uz3 ðx; y; LÞ: ð16Þ when using experimental data, because experimental error can
mask the minima or create other minima to which the optimiza-
It must be noted that imposing continuity of pressure and axial par- tion might converge. This issue is commonly observed in impe-
ticle velocity in the presence of mean flow may lead to singularities dance curves obtained experimentally using the mode-matching
at the liner discontinuity in the mode matching technique. A better technique or other impedance eduction methods, where the impe-
approach is to assume conservation of mass and momentum [34] dance curve display abrupt spikes or valleys that are not physical.
or, alternatively, continuity of potential displacement and its In general, these problems are related to a low signal-to-noise ratio
derivative [15]. However, this analysis falls outside the scope of this in the experimental data. This is specially critical at frequencies
work. close to the liner maximum attenuation, since the acoustic field
By applying the boundary conditions (Eqs. (13)–(16)), and con- is considerably attenuated by the liner but the flow noise is still
sidering the orthogonality between modes it is possible to end up present. It is important to note that the impedance at the liner
with a system of 4Q equations and 4Q unknowns: the modal maximum attenuation frequency range is usually of most interest
ðqÞ ðqÞ ðqÞ ðqÞ since it is the range for which the liner was design for.
amplitudes a1r , a21 , a2r and a3i , with q ¼ 1; . . . ; Q .
In this section, a modification to the mode matching method is
The required inputs to the system of equations are the incident
ð1Þ
proposed. In the original version, described in Section 2.1, the
plane-wave amplitude in Section 1, a1i , and the exit reflection objective function is given by the sum of the squared percentage
ðqÞ ðqÞ
coefficient ReðqÞ
¼ a3r =a3i ,
which is zero for all q > 1 since the only error between measured and calculated acoustic pressure at
reflected mode is the plane-wave mode, as per Eq. (12). Both inputs the j-th microphone position at each frequency and for a given
can be easily computed using the procedure outlined in Appendix impedance guess. However, the optimization problem can be
A. It is also required to know the wave numbers for each mode, in rewritten considering a frequency-dependent impedance function
each direction, in each duct section. In the hard sections, they are Zðx; f Þ, that gives the impedance for all frequencies for a given set
easily computed from equations Eq. (5), (6) and (A.2). In the lined of parameters x. The acoustic field, then, is calculated using the
section, they have to be computed from an expected impedance impedance for each frequency, and the objective function,
value by solving together Eqs. (7)–(9). Eq. (17), becomes a function of x, and is summed over all frequen-
From solving the aforementioned system of equations for an cies to give the global error, according to:
expected impedance value, the modal amplitudes are found, and 0 !2 1
the acoustic field can be computed at any position in the duct with X fN Xk
pj;exp ðf Þ  pj;MMM ðZðx; f ÞÞ
eðxÞ ¼ @ A: ð18Þ
Eqs. (10)–(12). It makes sense then to use the microphones posi- pj;exp ðf Þ
ð1Þ f ¼f j¼1
ReðqÞ
0
tions already used to compute and a1i
to compare the calcu-
lated acoustic field to the measured one. From that, a cost The parameters in vector x are the new optimization variables,
function is built, which are varied to minimize eðxÞ. By finding x, the impedance is
defined for all tested frequencies, solving a single, but larger, opti-
!2 mization problem. This new method requires the definition of a
Xk
pj;exp ðf Þ  pj;MMM ðZ; f Þ
eðZ; f Þ ¼ ; ð17Þ model for the impedance curve Zðx; f Þ.
j¼1
pj;exp ðf Þ If the Mach number dependency of a parameter is known and
present in the model, it could be included in the optimization pro-
cess, such that the scaling factor is sought. Therefore, the acoustic
and by minimizing it the unknown impedance can be found. In the
field could be compared to the measurements at different flow
current work, the MATLAB fsolve minimizer [35] with the
velocities, and the objective function given by Eq. (18) would be
Levenberg-Marquardt [36] algorithm was used, with mostly default
rewritten as a sum of the cost function over all the flow velocities.
options. The procedure is repeated for each frequency, so that the
results are independent, and a non-smooth impedance curve may
be obtained. The fact that the optimization is repeated at each fre- 3. Impedance models
quency step can also lead to a considerable computational cost, spe-
cially for noisy input data. This reinforces the idea of using an The impedance models presented here were developed for
impedance model in the optimization process to guarantee a translating impedance values in the frequency domain into an
smooth and physical behaviour and reduced processing time. equivalent relation in the time domain, mostly for use in numerical
A detailed derivation of the full system of equations in the mode methods. This problem is not trivial because of the necessity to
matching method can be seen in the original papers [29,16]. guarantee that the resulting relation is physical, i.e., obeys the
326 A.M.N. Spillere et al. / Applied Acoustics 129 (2018) 322–334

causality, reality and passivity conditions. Violation of these condi- 3.2. Rational functions
tions would manifest as non-physical results, like generation of
energy at the wall or other non-physical instabilities. Özyörük and Long [24] proposed the use of rational functions to
Of particular interest are the Extended Helmholtz resonator represent the impedance in frequency domain. This can be easily
[23] and the rational function models [24,25], as they have shown translated into time domain by means of the z-transform. Such
good agreement with measured data [37,21]. Other impedance an example of impedance function is given by
models are available in the literature, such as the mass-springer-
r2  r1 i xr 4
damper model [38] (also called the effective impedance model ZðxÞ ¼ r 1 þ þ þ ixr 7 ; ð22Þ
1 þ ixr 3 ð1  x2 =r 26 Þ þ ixr 5
[21]), given by
which was used to curve fit the parameters r 1;...;7 to experimental data
X
ZðxÞ ¼ R0 þ iX þ x  i ; ð19Þ from an ceramic tubular liner [37]. There is no physical interpretation
x of the parameters, although some of them resemble terms from the
where R0 ; X þ and X  are positive coefficients. This model in partic- EHR, e.g. r1 and face sheet resistance R, and r7 and mass reactance
ular is not representative of the educed impedance because of the m. According to Özyörük et al. [37], although resistance and reac-
constant resistance ReðZÞ ¼ R0 (for instance see [37,17,21,16] and tance can be accurately predicted when compared individually to
Fig. 5b, where the resistance is frequency-dependent), and therefore experimental data, the model does not satisfy the stability criterion,
is not considered here. Semi-empirical models [26–28] were also and therefore both parts have to be simultaneously considered in the
analysed, but their expressions are, in general, too complicated or optimization routine. The vector of parameters is given by
simplified to guarantee that the fundamental conditions are not x ¼ fr 1 ; r 2 ; r 3 ; r4 ; r 5 ; r6 ; r 7 g: ð23Þ
violated [23]. Therefore, these models are not included in the
and the upper and lower bounds are found in Table 2.
multiple-frequency mode matching.
Notice that Eq. (22) is not the only valid function. Özyörük et al.
Any flow effect is corrected by the Ingard-Myers boundary con-
[37] combined filter type functions in such a way that it could fit
dition [31] and therefore the general shape of the impedance
the experimental results from a ceramic tubular liner. There is
model remains the same. The parameters are allowed to vary with
not guarantee that this function is still valid for the single degree
Mach number, but not with frequency. This will be further dis-
of freedom liners investigated in this work (see Section 4). Never-
cussed in Section 5.
theless, this function is used in the multiple-frequency mode
matching method.
3.1. Extended Helmholtz resonator
3.3. Multipole model
The EHR, as the name suggests, is based on a simple Helmholtz
resonator, whose impedance satisfies the fundamental conditions. The multipole model was suggest by Reymen et al. [39] in order
More details on this matter and the derivation can be seen on the to use the recursive convolution for time domain simulations.
original paper [23]. The resulting frequency-dependent impedance However, the parameters had to be chosen carefully, otherwise
expression is: passivity condition would not be satisfied [25]. Li et al. [25]
  improved the multipole model by rewriting it as a sum of residues
1 1
ZðxÞ ¼ R þ ixm  ib cot xT l  i e ; ð20Þ and poles and adding the effective impedance model (given by Eq.
2 2 (19)), thus resulting in
!
where R is the face-sheet resistance, m is the mass reactance of the X X
S es
C e
C
ZðxÞ ¼ R0 þ iX þ x  i þ  s
; ð24Þ
air in the holes, b is a parameter to account for different cavity reac- x s¼1 x  ~ns x þ ~ns
tances, T l is the response time, and e is the damping in the cavity’s
fluid. In the original paper, advice is given on how to choose the where Ce s and ~
ns are complex parameters. Troian et al. [40] took a
parameters to generate curves that resemble liner impedance, i.e., similar approach by including the resistance R0 from the effective
without abrupt variations in the frequency range of interest and impedance model to the multipole model,
within coherent impedance values. !
X
P
Ap X
S es
C e
C
From Eq. (20) it is straightforward to see that the vector x, ZðxÞ ¼ R0 þ þ þ s
; ð25Þ
which define the impedance for a given frequency in the modified k  ix s¼1 ~ns  ix ~ns  ix
p¼1 p
mode matching, is given by 5 parameters:
where Ap is a real coefficient, and kp is a positive coefficient, in order
x ¼ fR; m; b; T l ; eg: ð21Þ to satisfy the fundamental conditions. Still, the passivity condition
can be violated and the coefficients must be checked. P and S are
As previously discussed, all the parameters have physical interpre- selected according to the type of liner, for instance P ¼ 5 and
tations. It is important therefore to limit the upper and lower S ¼ 0 for single degree of freedom (SDOF) liners [40]. The multipole
boundaries of the variables to feasible values. For instance, negative model investigated here is given by P ¼ 1 and S ¼ 1 plus the coeffi-
or too high resistances, oscillating curves, and other problems can cients R0 and X þ from the EFI model, such that
be avoided. Some of these bounds were defined empirically while
A e
C e
C
care was taken not to overly limit the possible curves to be gener- ZðxÞ ¼ R0 þ iX þ x þ þ þ ; ð26Þ
ated by Eq. (20). The defined upper and lower bounds for all param- k  ix n  ix n  ix
~ ~ 

eters are given in Table 1.

Table 1
Upper and lower bounds for the optimization variables in the modified mode matching method.

R m b Tl e
Upper bound 1 2  104 2 2  104 1
Lower bound 0 0 0 0 0
 A.M.N. Spillere et al. / Applied Acoustics 129 (2018) 322–334 327

Table 2
Upper and lower bounds for the optimization variables using the rational function model.

r1 r2 r3 r4 r5 r6 r7
2 2 1
Upper bound 1 1  10 3
1  10 1  10 1  10 1  10 7
1  103
Lower bound 0 0 0 0 0 0 0

Therefore, the vector of parameters is given by 4.2. Test samples

n o
e ; ~n ;
x ¼ R0 ; X þ ; A; k; C ð27Þ Two liner samples were considered in this study. Both are typ-
ical single degree of freedom (SDOF) locally reacting liners used in
and the upper and lower bounds, defined empirically, are found in aeronautical applications, as sketched in Fig. 3. The open area is
Table 3. defined as the ratio between the area of the holes and total face-
sheet area and it accounts for partial honeycomb blockage of the
holes in the assembly process.
4. Experimental procedure The main geometrical parameters of each sample, according to
Fig. 3 are summarized in Table 4. Both test samples have the same
The method described in the previous section have very specific cavity depth, which is related to liner reactance by  cotðxL=c0 Þ
requirements regarding test apparatus in order to be used. Most [26], and so similar reactances are expected. However, the different
geometrical specifications of the test duct have already been geometries of perforated plates may lead to different educed resis-
detailed, i.e., a rectangular duct, with a liner sample on one wall, tances, and consequently different levels of acoustic energy dissi-
and hard-wall sections before and after the lined section. Micro- pation in the test rig. As it is explained in the next section, this
phones have to be positioned in the hard-wall sections, flush to may result in bad convergence at some frequencies in the single-
the duct walls. A system able to measure and control flow velocity frequency mode matching method, specially in the presence of
in the duct is required, as well an acoustic excitation system. flow.

4.3. Experimental issues

4.1. Test rig

In order to illustrate the problems related to the single-

The test rig built at the Federal University of Santa Catarina is
frequency mode matching technique the following test case is
made of several modular, interchangeable sections, as illustrated
considered: test sample B, downstream acoustic source and Mach
in Fig. 2. The test section consists of 5 ducts, whose position can
0.20. The signal-to-noise ratio (SNR) is defined here as the differ-
be switched to accommodate different test configurations. For
ence between tonal excitation level and flow-induced noise. The
instance, the source section, where the speakers are connected,
latter is an effect of the turbulent boundary layer since the
can be positioned upstream or downstream to the liner sample
microphones are flush mounted to the wall. This results typically
in order to evaluate flow direction effects [9]. There are 8 compres-
in a broadband noise level around 85 dB at Mach 0.20. An alter-
sion drivers for a single duct section in the test rig, which are able
native solution is to recess the microphones and use a wire-mesh
to generate sound pressure levels exceeding 140 dB at the test sec-
flush mounted to the wall, which results in a flow-induced noise
tion. All results shown in this work were made using a single
reduction up to 5 dB [41]. Nonetheless this gain in SNR may be
speaker with pure discrete tones limited to not exceed 130 dB at
insufficient at frequencies close to the liner maximum attenua-
the microphone located immediately before the liner in order to
tion frequency.
avoid non-linear effects.
As seen in Fig. 4, microphone 5 (downstream do the liner)
The sections before and after the sample section are the two
shows a good SNR over all the frequency range. It is a section
microphone sections. The liner sample is 0.20 m long, covering
where the acoustic field is close to 130 dB and has not been
the entire duct height. The cross-section of the duct is 0.04 m by
affected by the liner sample. Microphones 4 (upstream to the
0.10 m, which results in a no-flow cut-on frequency of 1700 Hz
liner) shows a large reduction in SNR due to liner attenuation
for the first transverse mode, as per Eq. (2). Since the microphones
specially in the frequency range of 2100 Hz and 2600 Hz. Micro-
are positioned on half of the duct height, which is exactly the nodal
phone 5 has a slightly higher flow-induced noise level than
line of the first transverse mode, the effect of high order modes is
microphone 4 because the boundary layer has been affected by
not captured until the excitation frequency reaches the second
the presence of the liner, resulting in a more turbulent boundary
transverse mode at 3400 Hz. As a consequence, the frequency
layer and a small reduction in SNR, which can be seen at low
range under analysis is limited from 500 to 3000 Hz.
frequencies in Fig. 4.
Excitation signal generation, microphone signal acquisition, and
The problem of low SNR can be clearly seen in terms of insertion
controlling of the flow velocity, are executed by a custom software
loss (IL) in the duct, defined as (in the case of a downstream acous-
implemented in LabView. Liners were measured from no flow to
tic source):
Mach 0.25. The mean flow velocity value used in the impedance
!
eduction algorithms is calculated from the acoustic measurements ð1Þ
a3i
at the microphones using a custom routine and assuming uniform IL ¼ 20log10 ð1Þ
: ð28Þ
flow [41]. a1i

Table 3
Upper and lower bounds for the optimization variables using the multipole model.

R0 Xþ A k e
Reð CÞ eÞ
Imð C Reð~
nÞ Imð~
nÞ
3 6 6 2
Upper bound 1 1  10 1  10 1  10 0 0 1  10 1  103
Lower bound 0 0 0 0 1  106 1  106 0 0
328 A.M.N. Spillere et al. / Applied Acoustics 129 (2018) 322–334

70

d 60
t

Insertion Loss [dB]

50

40

30

L 20

10

0
500 1000 1500 2000 2500 3000
Frequency [Hz]
Fig. 3. Sketch of a single degree of freedom acoustic liner used in aeronautical
applications. The main geometrical parameters are the hole diameter d, the plate
thickness t, the cavity depth L and the percentage of open area. The perforated plate
(a) Insertion loss of the test sample
is subject to a grazing flow and incident acoustic waves. 4

Resistance [−]
3
Table 4
Geometrical parameters of the tested liner samples 2
Test sample Liner A Liner B
1
Type SDOF SDOF
Hole diameter d [mm] 1.0 2.0
0
Face-sheet thickness t [mm] 0.65 0.8 500 1000 1500 2000 2500 3000
Cavity depth L [mm] 19.0 19.0
Open area [%] 5.18 8.63 2
Hole placement Staggered Staggered
Reactance [−]

−2
60
−4
Signal-to-Noise Ratio [dB]

50
−6
40 500 1000 1500 2000 2500 3000
Frequency [Hz]
30

20 0 5 10 15 20
SNR at microphone 4 [dB]
10 Microphone 4
Microphone 5 (b) Impedance eduction
0
500 1000 1500 2000 2500 3000 Fig. 5. (a) Insertion loss and (b) impedance eduction result based on the classical
Frequency [Hz] mode matching technique for the following test case: test sample B, downstream
acoustic source and Mach 0.20. The colored dots represent the data quality based on
Fig. 4. Signal-to-noise ratio at microphones 4 (red dot) and 5 (blue dot) for the the signal-to-noise ratio from microphone 4. (For interpretation of the references to
following test case: test sample B, downstream acoustic source and Mach 0.20. color in this figure legend, the reader is referred to the web version of this article.)
Microphone 5 is located downstream to the test sample, whereas microphone 4 is
located upstream to the liner sample. (For interpretation of the references to color
in this figure legend, the reader is referred to the web version of this article.) (Eq. (17)) at the frequencies of 1500 Hz and 2100 Hz, respectively.
The former have a well defined region of minimum, leading to sim-
ilar results with the single- and multiple-frequency mode match-
Typical SDOF liners exhibit a peak in insertion loss, which is not ing techniques. The latter is a frequency with poor SNR, and the
clearly defined in Fig. 5a. Therefore, it is not possible to quantify region of minimum is not well defined. Therefore, the educed
the maximum attenuation nor the frequency where it occurs. This impedance from the single-frequency mode matching may not
analysis can be extended to the impedance eduction, as shown in represent the correct liner impedance, as the results with the
Fig. 5b. The discontinuities and oscillations in the resistance and multiple-frequency mode matching are also in a region of small
reactance between 2100 Hz and 2600 Hz are a non-physical beha- cost function.
viour of the liner and consequence of the poor SNR in this frequency
range, justifying the use of an impedance model in the eduction
method. Measurements with different test samples and acoustic 5. Results and discussion
source position exhibit the same trend, even at lower flow veloci-
ties, and therefore the use of an impedance model have to be a stan- In this section, the results using both the single-frequency and
dard procedure. multiple-frequency mode matching methods are shown. Data for
In order to understand the discontinuities and oscillations in upstream and downstream acoustic sources are available, thus
the impedance curve, a closer look at some frequencies is given both results are shown independently. A full list of the parameters
by Figs. 6a and b. which show the contour plot of the cost function found for each impedance model is given in Appendix B.
 A.M.N. Spillere et al. / Applied Acoustics 129 (2018) 322–334 329

−1 14 In general, the differences between the impedance models and

140
the single-frequency mode matching occur at low frequencies,
120 which are usually out of the range of interest. There is a good

180
0 agreement overall above 1000 Hz, including the frequency of max-
−1,5 14
100

0
16
12
0 imum attenuation in the test rig. Any impedance model can be
Reactance [−]

80
80 used in the multiple-frequency mode matching without any signif-
0
10 60 icant difference, specially above 1500 Hz, even at Mach 0.25.
−2 60 Fig. 8 shows the results for downstream acoustic source. Over-
all, the level of agreement is very similar to seen for upstream con-

40
dition. However, the instabilities in the results obtained with the
120

80

−2,5 60 single-frequency method may occur at different frequencies. At

10
0 no flow condition, the single-frequency mode matching method
14

80
0

16 100
overpredicts the resistance at low frequencies when compared to
0 120
the multiple-frequency mode matching method, independent of
−3
0,6 0,8 1 1,2 1,4 the impedance model chosen. This is expected to be due to poor
Resistance [−] signal-to-noise ratio in the experimental data at this frequency
(a) 1500 Hz range. At Mach 0.10, the multiple-frequency mode matching
method improves the results around 1900 Hz and 2300 Hz as some
80 40 scattering can be seen in the single-frequency mode matching. This
60
40
frequency range is close to the liner resonance frequency i.e. the
−0,5 frequency when the liner has the highest attenuation, as previously
mentioned. In this frequency range the sound wave amplitude
after the liner is considerable lower and may be close to the flow
Reactance [−]

noise, resulting in a low signal-to-noise ratio. This situation is

40

40
−1 aggravated for upstream propagating wave (downstream acoustic
source) as the propagation velocity relative to the liner is lower
60 40
than in the downstream propagating wave case, thus the acoustic
80
dissipation is higher.
12800220240

−1,5 10 60
16

14 0
0

0 At Mach 0.20 there is a very good agreement between the

0

12 80
0 single-frequency and multiple-frequency mode matching methods
100 over most of the frequency range. At Mach 0.25 the reactance at
260

−2 low frequency given by the multiple-frequency mode matching

0,2 0,4 0,6 0,8 1
shows the same behaviour as in the upstream acoustic source sit-
Resistance [−]
uation, which could be a limitation of the EHR and rational func-
(b) 2100 Hz tion models, although the resistance is well predicted.
Fig. 6. Contour plot of cost function at (a) 1500 Hz and (b) 2100 Hz given by the
mode matching technique for the following test case: test sample B, downstream 5.2. Liner B
acoustic source and Mach 0.20. At this frequency the signal-to-noise ratio is high
and the region of minimum is well defined. Markers represent the educed The single-frequency mode matching results for liner B, as seen
impedance from the single-frequency mode matching (black), and multiple- in Figs. 9 and 10, highlight the necessity to include an impedance
frequency mode matching using the following impedance models: EHR (blue),
rational functions (green) and multipole model (red). (For interpretation of the
model in the optimization routine. In several frequencies over all
references to color in this figure legend, the reader is referred to the web version of Mach numbers the impedance found is clearly not consistent with
this article.) the trend and cannot represent a physical behaviour of the liner.
For the upstream acoustic source (Fig. 9), at no flow condition a
good agreement can be seen for the reactance, whereas the
resistance is underpredicted at low frequency in the multiple-
5.1. Liner A frequency mode matching when compared to the single-
frequency mode matching method. At Mach 0.10 both curves show
Results for liner A are shown in Fig. 7 for upstream source posi- the same trend. Some substantial scattering can be seen at Mach
tion. At no flow condition and Mach 0.10 the multiple-frequency 0.20 along the whole frequency range, but the multiple-
mode matching method results in a smooth curve with same trend frequency mode matching results in a smooth curve as a clear
from the single-frequency mode matching. Therefore, it can be example of the advantage of this method. At Mach 0.25 the reac-
considered that the improved technique gives better results when tance shows the same behaviour as seen for liner A, and both
the single-frequency mode matching method fails to find the cor- methods show different results at low frequencies due to the
rect impedance due to a bad initial guess, poor optimization at mathematical approach of each model.
low frequencies or when the results start to deviate from the trend, In the downstream acoustic source condition, a very similar
as seen at frequencies around 2200 Hz for instance. analysis to liner A can be carried out. At no flow condition, a good
At higher Mach numbers the methods show good agreement agreement is seen between both methods. Some deviations from
above 1000 Hz between all impedance models and the single- the trend around 2300 Hz are seen in the single-frequency mode
frequency mode matching, but some discrepancies can be observed matching, whereas the multiple-frequency mode matching pro-
below this frequency, specially for the reactance. In the EHR model, vides smooth impedance curves. At Mach 0.10 and 0.20 a good
as x ! 0, then ImðZ w Þ ! 0, and therefore the curve is forced to agreement is seen overall, which allows to correct the impedance
deviate from the trend found in the single-frequency mode match- found in the single-frequency mode matching between 2000 Hz
ing method. This is not the case for the multipole model, which can and 2500 Hz. Likewise liner A this is the frequency range covers
follow the trend in the reactance found using the single-frequency the liner resonance frequency, therefore experimental data is not
mode matching. entirely reliable at these frequencies. This reinforces the advantage
330 A.M.N. Spillere et al. / Applied Acoustics 129 (2018) 322–334

No flow Mach 0.10 Mach 0.20 Mach 0.25

5 5 5 5
4 4 4 4
Resistance [−]

3 3 3 3
2 2 2 2
1 1 1 1
0 0 0 0
−1 −1 −1 −1
1 2 3 1 2 3 1 2 3 1 2 3
2 2 2 2

0 0 0 0
Reactance [−]

−2 −2 −2 −2

−4 −4 −4 −4

−6 −6 −6 −6

−8 −8 −8 −8
1 2 3 1 2 3 1 2 3 1 2 3
Frequency [kHz] Frequency [kHz] Frequency [kHz] Frequency [kHz]
Fig. 7. Impedance eduction result for liner A and upstream acoustic source using the single-frequency mode matching method (black) and the multiple-frequency mode
matching with the following impedance models: EHR (blue), rational function (green) and multipole (red). (For interpretation of the references to color in this figure legend,
the reader is referred to the web version of this article.)

No flow Mach 0.10 Mach 0.20 Mach 0.25

5 5 5 5
4 4 4 4
Resistance [−]

3 3 3 3
2 2 2 2
1 1 1 1
0 0 0 0
−1 −1 −1 −1
1 2 3 1 2 3 1 2 3 1 2 3
2 2 2 2

0 0 0 0
Reactance [−]

−2 −2 −2 −2

−4 −4 −4 −4

−6 −6 −6 −6

−8 −8 −8 −8
1 2 3 1 2 3 1 2 3 1 2 3
Frequency [kHz] Frequency [kHz] Frequency [kHz] Frequency [kHz]
Fig. 8. Impedance eduction result for liner A and downstream acoustic source using the single-frequency mode matching method (black) and the multiple-frequency mode
matching with the following impedance models: EHR (blue), rational function (green) and multipole (red). (For interpretation of the references to color in this figure legend,
the reader is referred to the web version of this article.)

of using the multiple-frequency mode matching to find the correct tion routine. Therefore, a smooth impedance curve with frequency
impedance. At Mach 0.25 the reactance at low frequency shows is to be found, instead of a single result at each frequency.
again different behaviours on both methods. Results for two liners show the improvement in the impedance
curve as the multiple-frequency mode matching method is able to
6. Conclusion capture the overall trend. The method finds more physical impe-
dance values at frequencies where measurement errors are present
In this work, an improved impedance eduction technique was or the optimization routine performs poorly. For instance, the fre-
evaluated. It is a combination of the classical mode matching quency range of 2000 Hz and 2500 Hz covers the liner resonance fre-
method [29] with an impedance model [24,23,25] in the optimiza- quency and experimental data is not entirely reliable, which results
 A.M.N. Spillere et al. / Applied Acoustics 129 (2018) 322–334 331

No flow Mach 0.10 Mach 0.20 Mach 0.25

5 5 5 5
4 4 4 4
Resistance [−]

3 3 3 3
2 2 2 2
1 1 1 1
0 0 0 0
−1 −1 −1 −1
1 2 3 1 2 3 1 2 3 1 2 3
2 2 2 2

0 0 0 0
Reactance [−]

−2 −2 −2 −2

−4 −4 −4 −4

−6 −6 −6 −6

−8 −8 −8 −8
1 2 3 1 2 3 1 2 3 1 2 3
Frequency [kHz] Frequency [kHz] Frequency [kHz] Frequency [kHz]
Fig. 9. Impedance eduction result for liner B and upstream acoustic source using the single-frequency mode matching method (black) and the multiple-frequency mode
matching with the following impedance models: EHR (blue), rational function (green) and multipole (red). (For interpretation of the references to color in this figure legend,
the reader is referred to the web version of this article.)

No flow Mach 0.10 Mach 0.20 Mach 0.25

5 5 5 5
4 4 4 4
Resistance [−]

3 3 3 3
2 2 2 2
1 1 1 1
0 0 0 0
−1 −1 −1 −1
1 2 3 1 2 3 1 2 3 1 2 3
2 2 2 2

0 0 0 0
Reactance [−]

−2 −2 −2 −2

−4 −4 −4 −4

−6 −6 −6 −6

−8 −8 −8 −8
1 2 3 1 2 3 1 2 3 1 2 3
Frequency [kHz] Frequency [kHz] Frequency [kHz] Frequency [kHz]
Fig. 10. Impedance eduction result for liner B and downstream acoustic source using the single-frequency mode matching method (black) and the multiple-frequency mode
matching with the following impedance models: EHR (blue), rational function (green) and multipole (red). (For interpretation of the references to color in this figure legend,
the reader is referred to the web version of this article.)

in incorrect impedances. The multiple-frequency mode matching is entirely representative of the impedance behaviour, suggesting
method better estimates the impedance at these frequencies. a limitation in its use for high Mach numbers. The grazing flow
None of the models account explicitly for grazing flow effects, effect in the models should be further investigated, which can be
which is corrected by the Ingard-Myers boundary condition [31]. achieved implicitly in the parameters or explicitly in correction
The parameters are set free to vary with Mach number, and in terms. In the latter care must be taken to prevent the violation of
some cases a clear trend can be observed, e.g. the damping e in the fundamental conditions. The optimization routine could
liner B with the EHR model, or the parameter r7 in the rational include not only all frequencies but also all Mach numbers, thus
function model. The results at Mach 0.25 indicate that no model resulting in a single optimization problem.
332 A.M.N. Spillere et al. / Applied Acoustics 129 (2018) 322–334

Acknowledgements The velocity in the z-direction in a given position z can be calcu-

lated from
!
This work was supported by FINEP (Funding Authority for Stud- ð1Þ
ai jkð1Þ z að1Þ ð1Þ
r jkzr z
ies and Projects) and CNPq (National Council for Scientific and uz ðzÞ ¼ e zi  e ; ðA:3Þ
Z0 Z0
Technological Development) and it is a partnership with Embraer
S.A. under the project ’Silent Aircraft’. or, by comparison with Eq. (A.1),
1 1
uz ðzÞ ¼ p ðzÞ  pr ðzÞ: ðA:4Þ
Appendix A. Overdetermined wave decomposition Z0 i Z0
From Eq. (A.1), it is possible to write the pressure at a position z ¼ z1
This appendix describes a multiple-microphone wave decom- as a function of the pressure at another position z ¼ z2 as
position method. It takes into account as many microphones as ð1Þ ð1Þ
the user wants, to build an overdetermined system of N equations pðz1 Þ ¼ pi ðz2 Þejkzi ðz1 z2 Þ
þ pr ðz2 Þejkzr ðz1 z2 Þ
: ðA:5Þ
(N being the number of microphones) and 2 variables (the ampli-
If the acoustic pressures in two positions z1 and z2 are known, one
tudes of the opposite travelling waves) that can be easily expanded
could use Eq. (A.5) to form a system of 2 equations and 2 unknowns,
to include the effect of flow and to support more or less micro-
the pressure amplitude at, for instance, z1 ; pi ðz1 Þ and pr ðz1 Þ. This sys-
phones. The minimum number of microphones is obviously two.
tem could also be extended to an overdetermined system of N mea-
The inclusion of more microphones has the advantage of helping
surements in different positions, resulting in
suppress random errors like aerodynamic fluctuations in the test
2 3y 8 9
data [42]. It is derived as follows. ð1Þ ð1Þ
pðz2 Þ >
ejkzi ðz2 z1 Þ
ejkzr ðz2 z1 Þ >
> >
Assuming that all higher order modes have decayed, only plane 6 7> > >

pi ðz1 Þ
 6 ejkð1Þ ðz3 z1 Þ
ejkzr ðz3 z1 Þ 7
ð1Þ
< pðz3 Þ >
=
waves propagate in a hard-wall duct. Therefore the acoustic field, 6 zi
7 ðA:6Þ
¼6 . .. 7 . ;
given by Eq. (3), can then be rewritten to include only propagating pr ðz1 Þ 6 .. >
7 > .. > >
4 . 5> > >
>
waves in the form. ð1Þ
jkzi ðzN z1 Þ ð1Þ : ;
e ejkzr ðzN z1 Þ pðzN Þ
ð1Þ ð1Þ
ð1Þ
pðzÞ ¼ pi ðzÞ þ pr ðzÞ ¼ ai ejkzi z þ arð1Þ ejkzr z : ðA:1Þ and the y represents the Morre-Penrose pseudo-inverse [44]. With
Eq. (A.6), it is possible to calculate the incident and reflected waves
In this case, the wave numbers in directions x and y are zero, and at any axial position within a hard duct, given that the pressure is
ð1Þ
the dispersion relation, Eq. (4), can be solved for kz , so that known on at least two positions from, for instance, microphone
measurements. It is also worth mentioning that from pi and pr it
ð1Þ k0 is possible to calculate reflection coefficients.
kz ¼ ; ðA:2Þ
1M
ð1Þ
Appendix B. Impedance model coefficients
where the plus sign is used for the incident wave (kzi ) and the
ð1Þ
minus, for the reflected wave (kzr ). When using data from experi- This appendix summarizes the coefficients found for each impe-
ments, Eq. (A.2) is modified to include the effects of visco-thermal dance model.
damping using the model by Dokumaci [43]. See Tables B.5,B.6,B.7,B.8,B.9,B.10.

Table B.5
EHR parameters found in the optimization for liner A.

Flow velocity R m b Tl e
Upstream No flow 0.0714 4.4832  105 1.5203 1.6956  104 0.1217
Mach 0.10 0.0973 5.6074  105 0.9892 1.1274  104 0.0879
Mach 0.20 0.2091 4.6138  105 1.2081 1.2513  104 0.3531
Mach 0.25 0.2717 2.0852  106 1.8215 1.8853  104 0.6973
Downstream No flow 0.1389 8.9807  105 0.3407 3.8225  105 0.0233
Mach 0.10 0.0729 7.8446  105 0.8167 7.6516  105 0.1710
Mach 0.20 0.5970 7.1030  105 0.4823 4.4499  105 0.1003
Mach 0.25 0.7024 6.7898  105 0.3913 3.4040  105 0.1187

Table B.6
EHR parameters found in the optimization for liner B.

Flow velocity R m b Tl e
Upstream No flow 0.1365 6.7866  105 0.9261 7.8233  105 0.0739
Mach 0.10 0.0212 7.0848  105 0.6529 5.1165  105 0.1172
Mach 0.20 0.3116 2.6687  105 0.8567 8.9611  105 0.2541
Mach 0.25 0.0596 3.8359  105 1.2886 9.5513  105 0.6235
Downstream No flow 0.0890 7.3733  105 0.9441 7.5842  105 0.0882
Mach 0.10 0.0076 7.7194  105 1.3724 8.8592  105 0.1512
Mach 0.20 0.1856 5.0576  105 1.5309 1.0087  104 0.2448
Mach 0.25 0.1793 5.2848  105 1.4084 8.8402  105 0.3050
 A.M.N. Spillere et al. / Applied Acoustics 129 (2018) 322–334 333

Table B.7
Rational function parameters found in the optimization for liner A.

Flow velocity r1 r2 r3 r4 r5 r6 r7
Upstream No flow 0.1723 184.2342 0.0099 0.0044 0.0398 4.9488  10 6
9.6219  105
Mach 0.10 0.3558 176.8062 0.0099 0.0048 0.0518 4.7189  106 7.6965  105
Mach 0.20 0.5258 7.9251 3.8723  104 0.0050 0.0500 4.9993  106 7.3182  105
Mach 0.25 0.7449 9.0632 4.4415  104 5.0282  104 0.0050 5.0293  105 6.9209  105
Downstream No flow 0.1937 119.9105 0.0066 0.0047 0.0773 3.8236  106 9.2826  105
Mach 0.10 0.2006 15.0745 7.2127  104 0.0051 0.0503 4.8078  106 8.5666  105
Mach 0.20 0.6175 10.5427 4.6016  104 0.0050 0.0500 5.0005  106 7.4447  105
Mach 0.25 0.7124 7.3195 2.8778  104 4.9910  104 0.0050 4.9964  105 7.0015  105

Table B.8
Rational function parameters found in the optimization for liner B.

Flow velocity r1 r2 r3 r4 r5 r6 r7
Upstream No flow 0.2607 181.0212 0.0075 0.0053 0.0602 6.1840  106 8.2270  105
Mach 0.10 0.3295 207.4059 0.0084 0.0047 0.0531 5.6022  106 7.4471  105
Mach 0.20 0.4923 23.7979 0.0013 0.0079 0.0533 5.8759  106 4.1753  105
Mach 0.25 0.5065 6.3727 2.9559  104 0.0050 0.0500 4.9998  106 4.3706  105
Downstream No flow 0.2461 244.6259 0.0100 0.0048 0.0497 4.9910  106 8.3460  105
Mach 0.10 0.3638 263.8840 0.0090 0.0035 0.0268 4.9391  106 8.7698  105
Mach 0.20 0.4959 28.9700 0.0010 0.0055 0.0477 18.4035 6.7259  105
Mach 0.25 0.3119 10.1286 3.1138  104 0.0050 0.0500 4.9577  106 7.0833  105

Table B.9
Multipole parameters found in the optimization for liner A.

Flow velocity R0 Xþ A k e
C ~
n

Upstream No flow 0.1667 9.1013  105 3.1332  105 3.3571 (165.38  7.5838i)  103 6.1288 + 419.6895i
Mach 0.10 0.0570 6.4970  105 7.0836  103 1.1459  104 (9.8292  9.4900i)  103 (53.3169 + 456.8983i)
Mach 0.20 0.0452 4.8212  105 1.9458  104 2.2837  104 (9.3400  158.45i)  103 (31.4579 + i75.3818i)
Mach 0.25 0.0257 5.3303  105 4.5328  104 5.7323  104 (8.7646  411.32i)  103 (1.9979 + 45.2902i)
Downstream No flow 0.0886 8.5994  105 5.0323  103 2.9762  104 (8.8646  370.70i)  103 (0.8002 + 0.2814i)
Mach 0.10 0.1953 8.5500  105 113.0906 9.3933  104 (1.0011  62.347i)  104 (77.9459 + 24.9803i)
Mach 0.20 0.5856 6.6559  105 1.2966  103 8.8988  103 (1.0189  46.694i)  104 (1.6817 + 145.0195i)
Mach 0.25 0.6068 2.5425  105 6.9732  103 9.5955  103 (1.0670  40.000i)  104 (4.0346 + 60.6768i)

Table B.10
Multipole parameters found in the optimization for liner B.

Flow velocity R0 Xþ A k e
C ~
n
5 4 3
Upstream No flow 0.1467 8.1362  10 2.5225  10 997.6840 (24.449  3.3389i)  10 (6.1317 + 171.0575i)
Mach 0.10 0.2530 6.9103  105 8.8214  104 242.9306 (5.6074  295.30i)  104 (80.2677 + 0.0683i)
Mach 0.20 0.0044 2.1551  105 1.6245  104 1.3155  104 (1.1470  4.8189i)  104 (90.8022 + 237.7685i)
Mach 0.25 0.5464 3.7546  105 1.1242  104 126.0300 (1.4233  23.379i)  104 (0.4613 + 109.9599i)
Downstream No flow 4.0928  104 8.1589  105 1.0013  104 2.5745  104 (13.272  7.4031i)  103 (0.1811 + 2.1739i)
Mach 0.10 0.3425 8.1382  105 4.2169  103 7.1975  104 (14.143  4.5318i)  103 (4.5800 + 1.9833i)
Mach 0.20 0.3615 6.8748  105 4.6169  105 114.2080 (24.476  2.7842i)  104 (0.4443 + 88.8374i)
Mach 0.25 0.1878 5.7927  105 8.5272  103 7.2823  104 (1.3577  95.982i)  104 (22.2515 + 52.2174i)

References [7] Murray PB, Ferrante P, Scofano A. Manufacturing process and boundary layer
influences on perforate liner impedance. In: 11th AIAA/CEAS Aeroacoustics
Conference (26th AIAA Aeroacoustics Conference), No. AIAA 2005-2849,
[1] Smith MJ. Aircraft Noise. Cambridge University Press; 2004.
Monterey, California, May 23–25 2005.
[2] Richter C. Liner impedance modeling in the time domain with flow, Ph.D.
[8] Jones MG, Watson WR, Nark DM. Effects of flow profile on educed acoustic
thesis. Technische Universität Berlin; 2009.
liner impedance. In: 16th AIAA/CEAS Aeroacoustics Conference, No. AIAA
[3] Drevon E. Measurement methods and devices applied to A380 nacelle double
2010-3763, Stockholm, Sweden, June 7–9 2010.
degree-of-freedom acoustic liner development. In: 10th AIAA/CEAS
[9] Bodén H, Zhou L, Cordioli J, Medeiros A, Spillere A. On the effect of flow
aeroacoustics conference, No. AIAA 2004-2907. Manchester (Great Britain);
direction on impedance eduction results. In: 22nd AIAA/CEAS aeroacoustics
2004.
conference, No. AIAA 2016-2727, Lyon, France, May 30–June 1 2016.
[4] Jones MG, Tracy MB, Watson WR, Parrott TL. Effects of liner geometry on
[10] Goldman A, Panton R. Measurement of the acoustic impedance of an orifice
acoustic impedance. In: 8th AIAA/CEAS aeroacoustics conference and exhibit,
under a turbulent boundary layer. J Acoust Soc Am 1976;60(6):1397–404.
No. AIAA 2002-2446, Breckejridge, Colorado, June 17–19; 2002.
[11] Murray P, Astley RJ. Development of a single degree of freedom perforate
[5] Kabral R, Bodén H, Elnady T. Determination of liner impedance under high
impedance model under grazing flow and high SPL. In: 18th AIAA/CEAS
temperature and grazing flow conditions. In: 20th AIAA/CEAS aeroacoustics
aeroacoustics conference (33rd AIAA Aeroacoustics Conference), No. AIAA
conference, No. AIAA 2014-2956, Atlanta, GA, 16–20 June 2014.
2012-2294, Colorado Springs, CO, 04-06 June 2012.
[6] Aurégan Y, Starbinski R, Pagneaux V. Influence of grazing flow and dissipation
[12] Zhang Q, Bodony DJ. Numerical investigation of a honeycomb liner grazed by
on the acoustic boundary condition at a lined wall. J Acoust Soc Am 2001;109
laminar and turbulent boundary layers. J Fluid Mech 2016;792:936–80.
(1):59–64.
334 A.M.N. Spillere et al. / Applied Acoustics 129 (2018) 322–334

[13] Watson WR, Jones MG. Impedance eduction in ducts with higher order modes [28] Motsinger R, Kraft R. Design and performance of duct acoustic treatment. In:
and flow. In: 15th AIAA/CEAS aeroacoustics conference (30th AIAA Aeroacoustics of flight vehicles: theory and practice, Noise control, vol. 2;
Aeroacoustics Conference), No. AIAA 2009-3236, Miami, Florida, May 11–13 1991.
2009. [29] Elnady T, Musharrof M, Bodén H, Elhadidi B. Validation of an inverse analytical
[14] Santana LD, Roeck WD, Desmet W, Ferrante P. Two-port indirect acoustic technique to educe liner impedance with grazing flow. In: 12th AIAA/CEAS
impedance eduction in presence of grazing flows. In: 17th AIAA/CEAS aeroacoustics conference and exhibit, No. AIAA 2006-
aeroacoustics conference (32nd AIAA Aeroacoustics Conference), No. AIAA 2643. Massachusetts: Cambridge; 2006. May 8–10.
2011-2868, Portland, Oregon, 05–08 June 2011. [30] Munjal ML. Acoustics of ducts and mufflers with application to exhaust and
[15] Aurégan Y, Leroux M, Pagneux V. Measurement of liner impedance with flow ventilation system design. John Wiley & Sons; 1987.
by an inverse method. In: 10th AIAA/CEAS aeroacoustics conference, No. AIAA [31] Myers MK. On the acoustic boundary condition in the presence of flow. J Sound
2004-2838, Manchester, Great Britain; 2004. Vib 1980;71(3):429–34.
[16] Elnady T, Bodén H, Elhadidi B. Validation of an inverse semi-analytical [32] Rienstra SW, Darau M. Boundary-layer thickness effects of the hydrodynamic
technique to educe liner impedance. AIAA J 2009;47(2):2836–44. instability along an impedance wall. J Fluid Mech 2011;671:559–73.
[17] Jones MG, Watson WR, Parrott TL. Benchmark data for evaluation of [33] Brambley E. Well-posed boundary condition for acoustic liners in straight
aeroacoustic propagation codes with grazing flow. In: 11th AIAA/CEAS ducts with flow. AIAA J 2011;49(6):1272–82.
aeroacoustics conference (26th AIAA Aeroacoustics Conference), No. AIAA [34] Gabard G, Astley R. A computational mode-matching approach for sound
2005-2853, Monterey, California, May 23–25 2005. propagation in three-dimensional ducts with flow. J Sound Vib 2008;315
[18] Jing X, Peng S, Sun X. A straightforward method for wall impedance eduction (4):1103–24. http://dx.doi.org/10.1016/j.jsv.2008.02.015.
in a flow duct. J Acoust Soc Am 2008;124(1):227–34. [35] The MathWorks, Inc. Natick, Massachusetts, USA. MATLAB optimization
[19] Watson W, Jones M, Gerhold C. Implementation and validation of an toolbox user’s guide; 2015.
impedance eduction construction and validation of a broadband time [36] Levenberg K. A method for the solution of certain non-linear problems in least
domain impedance boundary condition technique. In: 17th AIAA/CEAS squares. Quart Appl Math 1944;2(2):164–8.
aeroacoustics conference (32nd AIAA Aeroacoustics Conference), No. AIAA [37] Özyörük Y, Long LN, Jones MG. Time-domain numerical simulation of a flow-
2011-2868, Portland, Oregon, June 5–8 2011. impedance tube. J Comput Phys 1998;146(1):29–57.
[20] De Roeck W, Desmet W. Indirect acoustic impedance determination in flow [38] Tam CK, Auriault L. Time-domain impedance boundary conditions for
ducts using a two-port formulation. In: 15th AIAA/CEAS aeroacoustics computational aeroacoustics. AIAA J 1996;34(5):917–23.
conference (30th AIAA Aeroacoustics Conference), No. AIAA 2009-3302, [39] Reymen Y, Baelmans M, Desmet W. Efficient implementation of tam and
Miami, Florida, May 11–13 2009. auriault’s time-domain impedance boundary condition. AIAA J 2008;46
[21] Richter C, Thiele FH, Li XD, Zhuang M. Comparison of time-domain impedance (9):2368–76.
boundary conditions for lined duct flows. AIAA J 2007;45(6):1333–45. [40] Troian R, Dragna D, Bailly C, Galland M-A. Broadband liner impedance
[22] Busse-Gerstengarbe S, Richter C, Thiele FH, Lahiri C, Enghardt L, Roehle I, eduction for multimodal acoustic propagation in the presence of a mean
Ferrante P, Scofano A, et al. Impedance eduction based on microphone flow. J Sound Vib 2016;392:200–16.
measurements of liners under grazing flow conditions. AIAA J 2012;50 [41] Spillere AMN, Medeiros AA, Serrano PG, Cordioli JA. Cross-validation of a new
(4):867–79. grazing flow liner test rig using multiple impedance eduction techniques. In:
[23] Rienstra SW. Impedance models in time domain, including the extended The 22nd international congress on sound and vibration, Florence, Italy, 12–16
helmholtz resonator model. In: 12th AIAA/CEAS aeroacoustics conference and July 2015.
exhibit, No. AIAA 2006-2686. Massachusetts: Cambridge; 2006. May 8–10. [42] Holmberg A, Åbom M, Bodén H. Accurate experimental two-port analysis of
[24] Özyörük Y, Long LN. A time-domain implementation of surface acoustic flow generated sound. J Sound Vib 2011;330(26):6336–54.
impedance condition with and without flow. J Comput Acoust 1997;5 [43] Dokumaci E. A note on transmission of sound in a wide pipe with mean flow
(03):277–96. and viscothermal attenuation. J Sound Vib 1997;208(4):653–5.
[25] Li X, Li X, Tam CKW. Improved multipole broadband time-domain impedance [44] Penrose R. A generalized inverse for matrices. Mathematical proceedings of the
boundary condition. AIAA J 2012;50(4):980–4. Cambridge philosophical society, vol. 51. Cambridge Univ Press; 1955. p.
[26] Guess AW. Calculation of perforated plate liner parameters from specified 406–13.
acoustic resistance and reactance. J Sound Vib 1975;40(1):119–37.
[27] Kooi JW, Sarin SL. An experimental study of the acoustic impedance of
Helmholtz resonator arrays under a turbulent boundary layer. In: 7th
aeroacoustics conference, Palo Alto, CA, October 5–7 1981.