Document 2
Document 2
Document 2
5 Laurent Maxit
Univ Lyon, INSA-Lyon, LVA EA677
Laboratoire Vibrations- Acoustique
Bât. St. Exupéry, 25 bis av. Jean Capelle
F-69621 Villeurbanne Cedex, France
10 laurent.maxit@insa-lyon.fr
Abstract
15 This paper investigates the modelling of a vibrating structure excited by a turbulent boundary
layer (TBL). Although the wall pressure field (WPF) of the TBL constitute a random excitation,
the element-based methods generally used for describing complex mechanical structures
consider deterministic loads. The response of the structure to a random excitation like TBL is
generally deduced from calculations of numerous Frequency Response Functions. The result is
20 that the process requires costly computational resources. To tackle this issue, an efficient
process is proposed for generating realizations of the WPF corresponding to the TBL. This
process is based on a formulation of the problem in the wave-number space and the
interpretation of the wall pressure field as uncorrelated wall plane waves. Once the WPF have
been synthesized, the local vibroacoustic responses are calculated for the different realizations
25 and averaged together in the last step. A numerical application of this process to a plate beneath
a TBL is used to verify its efficiency and ability to reproduce the partial space correlation of
the excitation. Finally an application on a stiffened panel modelled with the finite element
method is proposed to illustrate the interest of the proposed process.
1
I. INTRODUCTION
35 Structures excited by the turbulent boundary layer (TBL) are very common in practical
applications. Cars, airplanes, trains and submarines may be excited by pressure fluctuations due
to the turbulent flow caused by their movements. In order to reduce the noise radiated from or
transmitted by these structures, it is important to understand at the design stage how the
structure reacts to TBL excitation. It is then necessary to develop numerical tools to predict the
40 vibration or the pressure radiated from the complex panels excited by the turbulent flow. This
topic has been the purpose of many works in the literature and it remains an important research
topic. To be convince, the reader can find a recent state of the art on this topic in the book [1]
published after the FLINOVIA symposium held in 2013 at Roma.
Many studies were carried out in the past to develop this type of calculation process for
60 predicting the vibro-acoustic response of structures excited by fully developped TBL. The
former ones concerned generally simple plates excited by a turbulent flow. In the end of the
60th, Strawderman [11] gave a review of existing models (at this period) of finite and infinite
plates under turbulence. Although neither the finite nor infinite model agrees wholly with the
experimental results, he indicated that the vibration statistics computed from the finite plate
65 model are in better agreement with the experimental results than those computed from the
2
infinite panel model. He also investigated with Christman [12] the effect of heavy fluid loading
on the vibratory response. Davis [13] proposed a space integration method including the light
fluid loading effect to estimate the power density functions of the displacement of the finite
plate and of the radiated acoustic pressure. Graham [14] reviewed statistical models of the
70 boundary layer and investigated the more specific case of aircraft statistical models of TBL.
The response of a finite panel under boundary layer excitation was determined by using the
modal superposition method and a wavenumber integration technique. For naval applications,
Ko and Schloemer [15, 16] proposed a method for evaluating the transmitted flow noise
received by a rectangular hydrophone embedded in an infinite extended viscoelastic layer. The
75 wavenumber filtering effects of both the elastomer layer and the rectangular hydrophone were
highlighted by their approach. Mazzoni [17] proposed a deterministic model to approximate the
response of an elastic rectangular plate at a low Mach number. The approximation was based
on the observations of numerical studies showing that the subconvective region of the turbulent
excitation power spectrum contributes significantly to the response of the panel. Rumerman
80 [18-20] derived different expressions giving broad band estimations of the acoustic power
radiated from a ribbed plate excited by TBL. He assumed a wavenumber-white pressure
excitation and that the ribs radiated independently, that leads the formulations to be more
accurate in the high frequency domain. Recently, Ciappi and al. [21] studied numerically and
experimentally the response of two composite panels under TBL excitation for nearly subsonic
85 flow conditions. They use literature empirical models for the WPF with the input date obtained
from the analyses of experimental wall pressure data. The comparison between finite element
and experimental results showed a good agreement between the different results.
For the prediction of the vibratory response of complex panels under TBL excitation, the
90 element-based methods considering deterministic harmonic excitations are generally
considered for describing the vibro-acoustic behavior of the panel. Finite Element Modeling
(FEM) can be used for a pure structural problem whereas FEM coupled with Perfectly Matched
Layers (PMLs) [22], the Boundary Element Method (BEM) [23], or the Infinite Element Model
(IEM) [24] can be used for an acoustic radiation problem. The coupling between the statistical
95 model used to describe the wall pressure fluctuations and the deterministic vibroacoustic model
represent a difficulty in the calculation process described above. Generally, this coupling is
established thanks to a formulation of the random excitation problem in the frequency-space
domain [25]. The ASD function of the system response (i.e. structure acceleration, acoustic
pressure) at a receiving point is then linked to the CSD function of the wall pressure fluctuations
3
100 through Frequency Response Functions (FRFs) [24]. These FRFs are defined between the
receiving point and a set of points distributed on the excited surface. In order to correctly
describe the partial correlation of the excitation, it is necessary to consider a large number of
points on the excited surface and compute a large number of FRFs [24-25]. A study can be
found in [26] which highlights the issues induced by the transformation of the pressure
105 distribution into discrete locations (i.e. nodes), in particular the aliasing effect. Hong and Shin
[25] examined in details the maximum mesh size required for reliable finite element analysis.
They showed that the mesh size should be defined under consideration of the spatial distribution
of the CSD function of the WPF in addition to the dynamic characteristic of the considered
structure. This may lead to consider a very fine finite element mesh and it results that the finite
110 element calculations are both time and memory consuming.
Different alternatives have been proposed for overcoming these drawbacks. Ichchou et al.
[27] were developed an equivalent “rain on the roof” excitation model for the high frequency
range which largely simplified the Finite Element calculations. Hong and Shin [25] proposed
115 an uncorrelated loading model of the WPF which was based on the compensation of the wall
pressure correlation lost due to the coarse mesh. A good accuracy with an exact solution was
obtained with this approach on a simply supported beam. The proposed loading model can also
be applied on more complex structures. In the same time, De Rosa and Franco [28] proposed a
scaling procedure in order to reduce the computation cost which can be induced by a high modal
120 density. It consists to reduce the dimensions not involved in the energy transmission whereas
the damping is increased in order to keep the same dissipated energy. The same authors and
others [29] have also analyzed scaling laws from experimental data involving four plates in air
or in water flow. More recently, the same team [30] proposed a frequency modulated pseudo-
equivalent deterministic excitation. This approach named PEDEM is derived from the pseudo-
125 excitation method [31-32]. This latter involves a modal decomposition of the load matrix related
to the CSD function of the WPF and it converges to the exact response if all the eigensolutions
are considered. Different PEDEM approximations were studied for overcoming the drawback
of the modal decomposition of the load matrix at each frequency step. The approximations
depended on the considered frequency ranges (i.e. low, mid or high) which could be identified
130 with a general criterion given by the authors. They were validated on a chain of linear oscillator
and on a flexural plate. Another type of approaches [24, 33] which has been developed recently
consists in synthetizing realizations of the wall pressure fluctuations corresponding to the CSD
function in the frequency-space domain. The process is based on a Choslesky decomposition
4
of the wall pressure CSD matrix [34]. Once the pressure fields corresponding to the different
135 realizations have been obtained, the vibroacoustic model is used to calculate the system
response at the receiving point for each realization, separately. The response to the TBL
excitation is finally deduced by an ensemble average on the different realizations. With this
process, the number of load cases considered in the vibroacoustic calculations corresponds to
the number of realizations, which is generally much lower that the number of FRFs considered
140 with the standard approach described above. This is the advantage provided by this approach.
However, it always requires a fine finite element mesh of the panel for frequencies above the
hydrodynamic coincidence frequency and the Choslesky decomposition is time consuming. In
order to tackle these issues, this paper proposes an alternative process to the ones described in
[24, 33]. It is based on a formulation of the random excitation problem in the frequency-
145 wavenumber domain and the interpretation of the wall pressure field as uncorrelated wall plane
waves. Realizations of the WPF are directly generated from an analytical expression depending
on the wavenumber-frequency spectrum of the WPF. The Cholesky decomposition is then not
be required. For each realization, the panel response induced by the deterministic WPF (of the
considered realization) can then be estimated from a low-frequency deterministic vibro-
150 acoustic model of the panel. For instance, it can be achieved by using a finite element model
when dealing with a complex panel. The stochastic response of the panel is then obtained from
an ensemble average of the different panel responses. The interest of this type of approach is
that it requires a relatively small number of realizations for estimating the stochastic response
of the panel. This point will be studied on the basic case of a simply supported plate. Moreover,
155 we will highlight how in some situations, the well-known filtering effect of the panel [35, 36]
can be considered in the process in order to reduce the mesh size of the panel. The accuracy of
the proposed process will be studied in function of the WPF model (i.e. Corcos or Chase), the
convective velocity of the flow and the panel thickness. A stiffened plate modelled with the
finite element solver MSC/NASTRAN will also be considered to illustrate the interest of the
160 present approach for practical application.
5
- The numerical process for estimating the panel response to TBL excitation is
summarized in Sec. 4.
- Then, in Sec. 5, numerical applications are proposed for a basic case of studying the
170 influence of different parameters on the accuracy of the approach proposed.
- Finally, before the concluding remarks, an application on a stiffened plate is
proposed in Sec. 6.
Ly
Lx
Fig. 1. Baffled simply supported plate excited by a homogeneous and stationary TBL.
180
Let us consider a baffle panel of surface p excited by a TBL as shown in Fig. 1. We assume
that the TBL is fully developed, stationary, and homogeneous over p . Moreover, the panel and
the boundary layer are assumed to be weakly coupled. It is then assumed that the vibration of
the plate does not interfere with the wall pressure. The spectrum of the wall pressure
185 fluctuations over a rigid surface can then be considered for charactering the panel excitation.
This can be estimated from the parameters characterizing the turbulent boundary layer (i.e.
convective velocity, U c ; boundary layer thickness; wall shear stress), and one of the wall
pressure models proposed in the literature [2-7]. The space-frequency cross spectrum of the
6
pp x x , may be written on the specific form proposed by Graham
wall pressure fluctuations S TBL
S TBL
x x, S pp U c S ppTBL x x, , (1)
pp
where:
- S pp (in Pa2/Hz) is the Auto Spectrum Density (ASD) function of the WPF
- S pp x x, (in rad2/m2)) is the normalized Cross Spectrum Density (CSD) function
TBL
195 of the WPF depending on the spatial separation between two points, x x , and the
angular frequency.
This form can be used with different models for the ASD function and for the normalized
CSD function, independently one from each other. For example, the Goody [6] or Rozenberg
200 [7] models can be used for the ASD function whereas the Corcos [2] or Chase [3] models can
be considered for the normalized CSD functions.
It is assumed that the panel has a linear vibroacoustic behavior that can be represented by a
deterministic model (like FEM). It may be complex. That is to say, for example, that it could
205 be made of different layers of different materials. It could also be stiffened by ribs on the side
opposite the flow.
The goal for us consists in estimating the vibrations of this panel when it is excited by the
wall pressure fluctuations induced by the TBL. In the next section, we give the outlines of the
210 formulation of this problem in the frequency-wavenumber space. Details of the formulation can
be found in the literature [38-40].
Mathematical formulation
pb ~x, represents the wall-pressure fluctuations due to the TBL on the plate at point x as a
215 function of time t. The plate acceleration at point x due to wall-pressure fluctuations, x, t ,
can be expressed as the convolution product
7
x, t h x, x, t p x, ddx ,
~
b
~ ~ (2)
p
where h x, ~x, t is the acceleration impulse response at point x for a normal unit force at point
~x .
As the turbulent flow induces a random process, the plate response is characterized by the
220 auto correlation function of the acceleration, R . Assuming that the process is stationary and
ergodic (i.e. expectation replaced by the limit of a time average), R can be written as:
T /2
R x, t lim x, t x, t d .
1
T T
(3)
T / 2
By introducing (2) in (3) and taking the time Fourier transform of the result, we obtain the
Auto Spectrum Density (ASD) of the acceleration at point x (see details in [38-40]):
S x, H x, ~x, S TBL
* ~ ~~ ~~
~ ~~
pp x x, H x, x, dxdx , (4)
pp
where H x, ~x, h x, ~x, t e jt dt is the Frequency Response Function (FRF) in terms of
225 acceleration at point x for a normal force at point ~x and the asterisk denotes the complex
conjugate.
Now, let us consider the space Fourier transform of the wall pressure spectrum, pp k , .
TBL
This is related to the wall pressure spectrum in the physical space S pp x x, by
TBL
8
H x, k , is generally called the sensitivity function [41]. The interpretation of Eq. (7)
~
indicates that this quantity corresponds to the acceleration at point x when the panel is excited
by a unit wall plane wave of wavevector k (i.e. by a WPF p ~x e jkx , ~x ).
~
235 b p
In Eq. (5-6), improper integrals exist over the wavenumber space. In the following, it is
assumed that they can be approximated by considering the rectangular rule and by truncating
and regularly sampling the wavenumber space. The criterion for defining the cut-off
240 wavenumbers and the wavenumber resolutions will be discussed later. However, it should be
underlined here that the cut-off wavenumbers can be different between Eq. (5) related to the
wall pressure and Eq. (6) related to the panel vibration due to the well-known filtering effect of
the panel [35]. p and denote the sets of wavenumbers selected to estimate Eq. (5) and (6),
pp x x , TBL k , e
1 ik . x x
2 pp
S TBL k , (8)
4 p
The outlines of the formulation for estimating the vibratory response of the panel have been
presented here. One can emphasize that these developments can be easily adapted for evaluating
the noise radiated by the panel (see the details of the formulation in [38-40]). It is however
250 outside the scope of the present paper which focus on the synthetize of the WPF and the
prediction of the panel vibration.
Definition
Let us define the concept of uncorrelated wall plane waves. One recalls that the term wall
270 plane wave refers to the blocked pressure acting on a panel surface varying spatially as
e jkx , x p where k is the wavevector of the wave considered. It is assumed that this wave has
a stochastic amplitude. The blocked pressure induced by a wall plane wave of wavenumber
k can then be written
p x, t Re A t e
jk x
, (10)
where A t is a random variable.
275 This wall plane wave is clearly a surface wave in the sense that it is only defined at the surface
of the panel. Moreover, we underline that the wavevector k may be arbitrary in the 2-D real
space k ℝ2 . It does not depend on the acoustic propagation as it is the case for an acoustic
plane wave.
The pressure CSD function corresponding to this wall plane wave is therefore:
S p p x x, S A A e
jk xx
, (11)
280 where S A A is the ASD function of the wave amplitude.
Now, let us consider a set of Uncorrelated Wall Plane Waves (UWPW) of wavenumbers
k , . The total blocked pressure p x, t is given by:
As the wall plane waves are assumed to be uncorrelated, the CSD function between the
amplitudes of two different waves is null: S A A 0 if . Hence the CSD function of the
285 pressure induced by this set of uncorrelated wall plane waves is therefore:
10
S UWPW
pp x x, S A A e jk x x
. (13)
S A A
pp
TBL
k , k , . (14)
4 2
By introducing Eq. (14) in Eq. (13) and by comparing the results with Eq. (8), it can be seen
immediately that S pp x x, S pp x x, . This clearly demonstrates that the TBL
UWPW TBL
the panel acceleration in response to the set of UWPW may be written as:
300 When defining the wave amplitudes by Eq. (14), the direct comparison of Eq. (16) and Eq.
In conclusion, the TBL excitation can be represented by a set of UWPW when the wave
amplitudes are defined with Eq. (14).
11
p k x, S A A e e
j k jk x
, (17)
the terms S A A represent the wave amplitudes. The root square of the ASD function of the
wave amplitudes is used to counteract the second-moment related to the ASD function.
It can be easily demonstrated that the CSD function of the wall pressure corresponding to an
infinite number of these realizations, S pp x x, , corresponds well to the CSD function of the
S
315
uncorrelated wall pane wave field, S pp x x, . Indeed, S pp x x, can be written by
UWPW S
definition:
x x, E S A A e j e jk x S A A e j e jk x ,
k k
S
S pp (18)
k
where E k represents the ensemble average over the realizations and the upper bar denotes the
complex conjugate of the complex number.
320
By rearranging the different terms,
S
S pp x x, S A A S A A e
j k xk x
,
Ee
j k k
k
(19)
325 This demonstrates well that when considering an infinite number of realizations, the CSD
function of the WPF defined by Eq. (17) is equivalent to the CSD function of the uncorrelated
wall pane wave field.
To summarize this section 3, one can emphasize that: (a), the TBL excitation can be
330 represented by a set of uncorrelated wall plane waves when the wave amplitudes are defined
by Eq. (14); (b), Realizations of the random pressure field corresponding to a set of uncorrelated
wall plane waves can be obtained using Eq. (17); (c), The CSD function of the WPF
12
corresponding to these realizations converge to the CSD function of the set of uncorrelated wall
plane waves when the number of realization tends to infinity.
335 It results that realizations of the random pressure field corresponding to a TBL excitation
can be obtained by using Eq. (17) when the wave amplitudes are defined by Eq. (14). The CSD
function of the WPF corresponding to an infinite number of realizations is then equal to the
CSD function of the TBL excitation. An infinite number of realizations of the WPF as described
previously are then equivalent to the TBL excitation. In consequence, the response of the panel
340 under TBL excitation can be estimated from the response of the panel excited by the WPF
corresponding to these realizations. In practice, a finite number of realizations K will be
considered to approximate the TBL excitation. The accuracy of the calculation process
summarized in section 4 will be studied in section 5 in function of the number of realizations.
The following is a description of the numerical process for estimating the panel response to a
TBL excitation from the realizations of the uncorrelated wall plane wave field. This process is
350 directly derived from the previous section. One has seen that the realizations of the WPF defined
by Eq. (17) with Eq. (14) are representative of the WPF of the TBL excitation. The proposed
process consist then to estimate the panel response to the WPF of each individual realization
and then to average the obtained panel responses over the different realizations.
355 Let us consider a deterministic vibroacoustic model of the panel in order to estimate the
panel response to deterministic load cases. This model can be an analytical model for an
academic structure or an element-based model for a complex panel (as it will be illustrated in
section 6 for a stiffened plate represented by a finite element model within the MSC/NASTRAN
code).
360
The numerical process proposed can be decomposed into three steps:
- The first step consists in calculating the WPF of K realizations of the uncorrelated wall
plane wave field representing the TBL excitation. It is carried out by using the formula (17) and
considering the wave amplitudes defined by Eq. (14). When using an element-based method
365 for describing the vibro-acoustic behavior of the panel, the WPF should be applied to the nodes
13
of the mesh belonging to the interaction surface between the panel and the flow. For the node i
of this set of nodes, the pressure corresponding to the kth realization (deduced from Eq. (14) and
(15)) is given by:
k
p x , y ,
i i
k x , k y , k xk y e j k x k y ,
TBL
pp x
i
y
i k
(21)
4 2
where:
370 - x and y represent the axis in the streamwise direction and the crosswise direction,
respectively;
- k x , k y are the coordinates of the wavevector k ;
- k x and k y are the wavenumbers resolutions in the spanwise and streamwise directions,
- In the second step, the vibroacoustic model is used to estimate k x, , the panel
acceleration at point x when the panel is excited by the deterministic WPF, p k x, calculated
380 in the previous step. When an element-based model is considered, expression (21) is directly
used to prescribe the pressure on the nodes at the interface between the panel and the flow. A
direct frequency analysis can then be performed for example to estimate the panel acceleration
k . This calculation is repeated for the different realizations k 1,..., K . The number of load
cases considered in the vibroacoustic simulations therefore corresponds to the number of
385 realizations;
- Finally, in the last step, the ASD of the acceleration at point x is estimated by an
ensemble average of the acceleration responses, k x, , k 1,..., K estimated in the previous
step:
SS x, E k x, k x, k1,..., K . (22)
390
SS x, corresponds then to the ASD of the acceleration at point x induced by the WPF of
the K realizations. As the CSD function of the WPF of the K realizations converges to the CSD
14
function of the WPF of the TBL excitation when K , SS x, converge to the ASD
function of the plate acceleration at point x induced by the TBL excitation. In practice, a finite
395 number of realizations K will be considered to approximate the ASD function of the plate
acceleration. This will be studied in the next section.
V. NUMERICAL APPLICATIONS
400 For evaluating the numerical process described in the previous section, we are going to compare
its results on a basic application case with the results obtained by a direct calculation of Eq. (9).
In the following, the latter calculation is named the sensitivity function method whereas the
numerical process described in section 4 is called the sampling method.
15
Parameters Numerical value
Panel thickness h 3 mm
Panel length in the streamwise direction Lx 1.5 m
Panel length in the crosswise direction Ly 0.9 m
Two different models of WPF will be considered in order to study the influence of the model
on the accuracy of the results. From Eq. (1) and the definition of the space Fourier transform
(5), the CSD function of the WPF in the wavenumber space pp can be written:
TBL
430
2
TBL
k x , k y , S pp U c ppTBL k x , k y , , (23)
pp
where pp k x , k y , is the normalized CSD function of the WPF in the wavenumber space
TBL
The Corcos model is first considered because it provides an analytical expression of the CSD
435 function both in the space-frequency domain x x, [2] and in the wavenumber-frequency
domain k , . The Corcos model is a semi-empirical model. Although the Corcos model is
simple and it is frequently used in the literature, it has been pointed out by different authors that
it may be deficient to represent accurately the subconvective domain of the WPF [44]. One can
also notice that some improvements have been proposed recently to circumvent this issue [45].
440 We consider however the Corcos model in the following due its simplicity, without any
16
consideration on its ability to reproduce fairly the WPF induced by a TBL. The Corcos
normalized CSD function of the WPF used in the following depends on the convective velocity
given in Tab. 1 for the nominal case. It is given by ([10], [37]):
4 x y
ppTBL k x , k y , , (24)
2 Uckx 2 Uck y
2 2
x 1 y
with the Corcos’s parameters: x 0.11 , y 0.77 .
445
The second model of WPF considered for the numerical applications is the Chase model [10,
43]. It has been deduced from theoretical developments based on the Poisson equation and it
depends on numerous empirical parameters. It is supposed to be more accurate in the low
wavenumber region that the empirical models [45]. The Chase normalized CSD function of the
450 WPF is expressed by Eq. (3.18) to (3.20) of Ref. [10]. The empirical constants given after these
equations in this reference are also considered in the present paper. This model depends on the
TBL displacement thickness, the friction velocity and the convective velocity which are given
on Tab. 1 for the nominal case.
455 For the sake of simplicity, the vibratory responses of the panel are normalized by
2
U
S pp c , which makes it independent of the ASD of the WPF. In others words, this
normalized response corresponds to the vibratory response of the panel induced by the WPF
defined by the normalized CSD function, ppTBL k x , k y , (see Eq. (23)).
by using the modal expansion method. The modal angular frequency, m, n , the modal shape,
m, n and the modal mass, M m, n , are given for each couple of non-null integers m, n by:
hLx Ly
2
D m n m n
2
Eh 3
where D is the flexural rigidity given by D
12 1 2
.
17
465 For the modal expansion, the modes having a frequency in the extended angular frequency
band 0, 3/2 max are considered, where max represents the highest angular frequency of interest.
m, n ℕ∗ × ℕ∗ / m,n 0, 3/2 max denotes the set of modal order couples corresponding to
these modes.
470 To estimate the ASD function of the panel acceleration directly with Eq. (9), it is necessary
to estimate the sensitivity functions H at point xM , yM for each couple of wavenumbers
~
k , k
x y :
p 1p e jk Lx 1 p
2
, if k ,
L 2 p L
I p k
L
(28)
1
2 jL , otherwise.
475
On the other hand, to estimate the ASD function of the panel acceleration with the numerical
process described in Section 4, it is necessary to estimate k xM , yM , , the acceleration
response at point xM , yM induced by the wall pressure (17) corresponding to the kth realization:
Fmk,n S A A e m,n k x , k y .
j k
(30)
480 The latter expression is well adapted for the present case for which we can calculate an
analytical expression of m, n (i.e. Eq. (27-28)). For a complex panel, the forced response or the
mode shapes can be calculated by FEM and can be known at discrete points (i.e. the nodes of
18
the mesh). In the literature ([24-25], [28]), it has already been shown that the size mesh should
be defined carefully for describing correctly the spatial variations of the WPF and for avoiding
485 aliasing phenomenon [25]. In order to study the influence of this type of approximation in the
framework of the proposed approach, we also perform an approximation of the modal forces
by considering a spatial discretization of the mode shapes by S points along the streamwise
direction and R points along the crosswise direction. The modal forces for the kth realization
can then be approximated by using the rectangular rule:
R S
F p k rx, sy, m, n rx, sy xy .
k
m, n
(31)
r 0 s 0
Lx Ly
490 with x , y .
R S
The spatial resolutions x and y can be defined using a criterion based on the TBL
characteristics or on the plate characteristics. This point will be studied in section V.D.4.
510
19
Fig. 2a shows the normalized CSD function of the wall pressure spectrum of the Corcos
model. It exhibits the highest values for wavenumbers close to the convective wavenumber k c .
Furthermore, in Fig. 2b, the highest values of the sensitivity functions (calculated with Eq. (26))
can be observed for frequency and wavenumbers close to the modal frequencies and the modal
515 wavenumbers in the streamwise direction, respectively. To illustrate this, we have indicated the
modal frequencies and the modal wavenumbers of the first 10 modes of the plate in Tab. 2. The
frequency and wavenumber of the highest amplitudes in Fig. 2b correspond well to the modal
frequency f m,n and the modal wavenumbers k m of the plate modes with n=1. It should be
emphasized that only these particular modes have the most significant contributions in the
520 sensitivity functions shown in fig. 2b because the crosswise wavenumber is equal to 0 for this
figure (i.e. k y 0 ). Whatever the case, it can be however concluded that the highest values of
the sensitivity functions can be observed for wavenumbers below or close the natural flexural
(a)
(b)
Fig. 2. (a), The normalized CSD function of wall pressure spectrum given by the Corcos
model, pp kx ,0, (dB, ref. 1); (b), The sensitivity function of the plate at point xM , yM ,
TBL
525
20
m n -1
f m, n (Hz) m km (m-1) n kn (m )
Lx Ly
Table 2. Modal information for the first ten modes of the plate: f m, n , the modal frequency
530 m, n 2f m, n ; m, n, the modal orders in x and y directions, respectively; km , k n , the modal
For defining the set of wavenumbers p related to the wall pressure field (i.e. Eq. (8)), the
cut-off wavenumbers should be defined only from the characteristics of the wall pressure
535 spectrum (as the blocked pressures are independent of the panel). The truncation of the
wavenumber space should include the convective ridge in the streamwise direction. Then, one
defines the cut-off wavenumber by:
k xp kcmax , (32)
where k cmax is the convective wavenumber at the higher frequency of interest and is a margin
the wavenumber space should be done by considering, both, the excitation characteristics and
the panel characteristics. Fig. 3a shows the result of the product between and the wall pressure
spectrum (i.e. Fig 2a) and the sensitivity function (i.e. Fig 2b). It should be underlined that this
545 product appears in the summation of Eq. (9) to evaluate the ASD function of the plate
21
acceleration. It can be observed on Fig. 3a that the contribution of the convective domain is
negligible. This is due to the well-known filtering effect of the pressure fluctuations by the
panel [35, 36]. Thus, for this case, the truncation of the wavenumber space in the streamwise
direction can be achieved without considering the convective ridge. The cut-off wavenumber
550 in the streamwise direction used to define the set of wavenumbers in Eq. (9) or in Eq. (17)
k x 1 k max
f , (33)
where k max
f is the convective wavenumber at the higher frequency of interest.
(a)
(b)
Fig. 3. Values of the product between the sensitivity function and the wall pressure spectrum,
ppTBL kx ,0, H xM , yM , kx ,0, (i.e. integrand of Eq. (6), dB, ref. 1). Results for two models of
2
555
wall pressure spectrum: (a), Corcos model; (b), Chase model. Dashed line: convective
wavenumber; Solid line: flexural wavenumber.
It should however be mentioned that it is not a general result. The filtering effect of the
560 structure is not always enough important to vanish the contributions of the convective ridge. It
22
depends in particular on the frequencies of interest [35], the panel boundary conditions [36],
and the considered model of the WPF. This latter dependency is highlighted on Fig. 3b showing
the same type of results than Fig. 3a when considering the Chase model. One can observe that
the contributions of wavenumbers above the flexural wavenumber are more important with the
565 Chase model than for the Corcos model. This is due to a stronger decrease of the CSD function
in the low wavenumber domain of the Chase model compared the Corcos model. For the
considered case, the criterion defined by Eq. (33) may be in its limit of validity with the Chase
model. This will be verified in the section 5.D.1 In the case of the filtering effect of the structure
is not dominant, the criterion (34) based on the TBL characteristics should be applied to
570 estimate the panel response:
k x 2 k xp kcmax . (34)
575 wavenumber k yp related to the wall pressure field (i.e. Eq. (8)) has been fixed at 300 rad/m 1
with a trial-and-error process. For the panel response, the result of the product between the
sensitivity functions and the wall pressure spectrum is dominated by the wavenumbers below
or close to the natural flexural wavenumber of the plate (results not plotted here). The cut-off
wavenumber in the crosswise direction can then be defined as the one in the streamwise
580 direction:
k y k max
f . (35)
- Wavenumber resolutions
The wavenumber resolutions in the two directions should be defined such that they correctly
represent the spatial variations in the wavenumber space of the wall pressure spectrum and the
585 sensitivity function. The analytical expression of the sensitivity functions for the panel
considered (i.e. Eq. (26-28)) gives an order of magnitude of these spatial variations (inversely
proportional to the panel lengths) whereas the wall pressure spectrum varies relatively slowly
as a function of the wavenumbers. In the following, the wavenumber resolutions are then fixed
at 0.25 rad/m, independently of the frequency. For a more complex panel, a trial and error
590 process can be used to fix these parameters.
23
Analysis of results
1. The sensitivity method
595 The ASD function of the panel acceleration at the receiving point M has been evaluated with
the sensitivity method for the nominal test case. Calculations have been performed for the two
cut-off wavenumber criterions (33) and (34) (i.e. k x 1 24.3 rad/m and k x 2 45.2 rad/m,
respectively) and for the two models of WPF described in section 5.A (i.e. Corcos and Chase
models). The results are plotted on Fig. 4 in function of the frequency. For the Corcos model, a
600 very good agreement between the two calculations are observed on the whole frequency band
of interest. This confirms that the structure filters sufficiently the convective ridge of WPF in
order that this latter can be neglected. For the Chase model, the agreement between the two
calculations is very good up to around 200 Hz. Above this frequency, the calculation
considering the criterion (33) underestimates slightly the panel response. A difference of 2.5
605 dB can be observed at 300 Hz. This can be explained from the observations made on Fig. 3b in
the previous section. These results highlights well that the criterion (33) should be used with
carefully. We reach its limit of validity for the present case with the Chase model. However for
the present case, the prediction remains globally a correct estimation of the plate response.
610 In the following, the results of the sensitivity method using the criterion (34) will be used as
a point of comparison in order to evaluate the accuracy of the sampling method.
24
(a)
(b)
Fig. 4. ASD function of the panel acceleration at the receiving point M as a function of the
frequency, 10 log 10 S M (dB, ref. 1 m/s2/Hz0.5). Comparison of results obtained with the
615 sensitivity function method with different criterion of the truncation of the wavenumber
space: Full line, criterion (34) based on the TBL characteristics; dash line, criterion (33) based
on the plate characteristics. Results for two models of wall pressure spectrum: (a), Corcos
model; (b), Chase model.
The realizations of the wall pressure field are achieved using Eq. (17). By way of illustration,
the WPF of two realizations at 300 Hz are given in Fig. 5 considering the Corcos model:
- The first one (Fig. 5a) has been obtained when the wavenumber set is defined with the
625 TBL characteristics (i.e. p ). Spatial variations due to wave propagations in the
25
streamwise direction appear. The wavelength of these waves is around 0.2m, which correspond
roughly to the convective wavelength ( c 2 / kc 0.16 rad/m at 300 Hz).
- The second one (Fig. 5b) considers the wavenumber set defined from the panel
characteristics ( ). In particular, the criterion (33) is applied to define the cut-off
630 wavenumber in the streamwise direction. One can notice that the spatial variations present
higher wavelengths and the amplitudes are lower than in Fig. 5a. This is directly due to the
truncations of the wavenumber space which is more restrictive when considering the panel
characteristics than the TBL characteristics. The WPF of Fig. 5b does not represent the
convective ridge of the TBL. It explains why the amplitudes are lower. However, it represents
635 the part of the pressure field induced by the TBL that contributes to the panel vibration.
(a)
(b)
Fig. 5. Two realizations of the WPF at 300 Hz obtained with Eq. (17): (a), using the
wavenumber set defined with the TBL characteristics (i.e. p ); (b), using the
640
In order to study the WPF synthetized with Eq. (17), one will compare the spatial coherence
estimated from the WPF of K realizations with the one given by the analytical expression of the
26
Corcos model. The spatial coherence between point x and x' can be estimated from K
realizations by:
E p k x, p k x' , k1,..., K
x, x' , . (36)
E p x, E p x' ,
k 2 k 2
k1,..., K k1,..., K
Fig. 6. Spatial coherence of the TBL pressure field as a function of the spatial separation in
the streamwise direction (Upper) and in the crosswise direction (Lower). Full line, analytical
formula of the Corcos model; dotted line, numerical estimation considering 30 realizations;
660 dash line, numerical estimation considering 900 realizations.
27
3. Results of the sampling method on the nominal test case
The process described in section 4 is applied to evaluate the ASD function of the panel
acceleration at the receiving point. The nominal test case with the Corcos model and 30
665 realizations are considered. The cut-off wavenumber criterion based on the TBL characteristics
(34) is applied. For each realization, the panel acceleration has been obtained using the modal
expansion Eq. (29-30). One recalls that the modal forces resulting of the WPF is then calculated
analytically. To illustrate the process, one has plotted on Fig. 7 the results of the 30 realizations
(grey line). One can observe a relatively large dispersion of the plate response in function of
670 the realizations. The ensemble average of these acceleration responses (i.e. Eq. (22)) is then
calculated in order to estimate the ASD function of the panel acceleration. The result (dash line)
has been plotted on Fig. 7.
675
Fig. 7. 10 log 10 S M (dB, ref. 1 m/s2/Hz0.5). Calculations with the sampling method: Grey
lines, results of 30 realizations; dashed-dotted line, Results obtained with averaging on the 30
realizations. Calculation parameters: Corcos model, cutoff wavenumber criterion (34) based
on the TBL characteristics.
680 For studying the accuracy of the sampling method, one compares the previous result (dash
line) with the result of the sensitivity method (full line) on Fig. 8. A very good agreement
between the two calculations can be observed. This indicates that although the panel response
of the 30 realizations exhibits an important dispersion, an average over only 30 realizations is
sufficient to give a correct estimation of the ASD function of the panel acceleration.
28
685
Fig. 8. 10 log 10 S M in function of frequency (dB, ref. 1 m/s2/Hz0.5). Comparison between
three calculations: full line, the sensitivity function method (used as reference); dash line: the
sampling method with the modal forces calculated analytically; dashed-dotted line, the
sampling method with the modal forces estimated using the fine mesh (25 x 9). Calculation
690 parameters: Corcos model, cutoff wavenumber criterion (34) based on the TBL
characteristics.
For this first result of the sampling method, Eq. (17) were used to synthetize the WPF and
the modal forces were calculated analytically with Eq. (30). This is appropriate for academic
695 cases. For more complex cases, the WPF will be defined on a point mesh and it will be
introduced in the numerical model of the panel. In the literature, different authors considering
different spatial methods ([24], [25], [28]) already showed that the mesh of a finite element
model should be defined carefully in this case for describing both, the structure behavior and
the aerodynamic field. For studying the influence of the definition of the WPF on a discretized
700 mesh in the framework of the proposed approach, let us considered a first mesh of points on the
plate. This will be called the fine mesh and it is defined by the spatial resolutions x and y
defined by (as proposed in Ref. [25]):
x and y , (37)
k x 2 k y
where k x 2 and k y are defined by the criteria (34) and (35), respectively.
29
For the nominal test case, the mesh is composed of 25 points in the streamwise direction and
705 9 points in the crosswise direction. The modal forces used to estimate the plate acceleration
(29) can then be approximated using Eq. (31). The result of the sampling method considering
the modal forces estimated on the fine mesh has been plotted (dashed-dotted line) on Fig. 8.
One can observe that the discretization of the WPF using this fine mesh does not introduce
significant discrepancy which is consistent with the works proposed in the literature.
710
Now, the accuracy of the sampling method is studied in function of the cut-off wavenumber
criterion and the mesh size. To do that, a second mesh called the coarse mesh is considered with
715 the spatial resolution in the streamwise direction defined by:
x ,. (38)
k x 1
725 When the cut-off wavenumber criterion (34) based on the convective wavenumber is
considered, one notices that the discrepancies are generally less than 2 dB when the fine mesh
is considered, whereas large discrepancies above around 180 Hz can be observed when the
coarse mesh is used. This is observed for the two WPF models. Although the contributions of
the convective ridge are filtered by the panel and may be neglected as shown in section V.C, a
730 fine description of them is required in order to obtain good convergence of the calculation. In
contrary, when considering the coarse mesh and the cut-off wavenumber criterion (33) based
on the flexural wavenumber, a good accuracy is observed on the whole frequency of interest,
for both WPF models. This result confirm that the effect of the convective ridge is negligible
on the panel vibration for the present case. Moreover, the coarse mesh is sufficient for
30
735 estimating the panel response because the pressure field of each realization (as shown on Fig.
5b) varies slowly when the criterion (33) is considered (contrary to the rapid variations of the
pressure field when the criterion (34) is considered, as shown in Fig. 5a). This explains why the
results converge with a coarse mesh when the criterion (33) is considered rather than the
criterion (34).
(a)
(b)
740
Fig 9. 10 log 10 S M in function of frequency (dB, ref. 1 m/s2/Hz0.5). Results with two
different models of WPF: (a), Corcos; (b), Chase. Full line, the sensitivity function method
(used as reference); dash line, the sampling method considering the coarse mesh (14 x 9) and
the cutoff wavenumber criterion (34) based on the TBL characteristics; dashed-dotted line:
the sampling method considering the coarse mesh (14 x 9) and the cutoff wavenumber
745 criterion (33) based on the plate characteristics.
31
Thus, when the panel filtering effect is dominant, there are two advantages in reducing the
wavenumber space:
- first, it reduces the number of wall plane waves to be considered (i.e. card ) in the
process of synthetizing the WPF (17). As the computing time of this process is directly
750 proportional to the number of wall plane waves considered, the time saving can be easily
deduced when using (33) instead of (34). However, it should be pointed out that this process,
which consists in performing a summation, consumes relatively little time;
- second, as it was observed previously, it permits to consider a coarser mesh than if the
wavenumber space were not reduced. In the case of a panel represented by a Finite Element
755 model, the size of the elements can be defined with a criterion on the panel characteristics and
not the TBL ones. The number of degrees of freedom, and by consequence, the FEM computing
times, can be significantly reduced.
We should however underline that this panel filtering effect is not always dominant and it
760 should be considered carefully. It is for instance the case when the frequency range of interest
is close to the hydrodynamic coincidence frequency. The sampling method can however always
be applied using the criterion (34) and a fine mesh described with (37).
32
(a)
(b)
(c)
(d)
775
Fig. 10. Difference of 10 log 10 S M between the sensitivity function method and the
6. Computation times
780
The calculations presented in this paper have been performed using the MATLAB software on
a standard personal computer (Intel Core i5 3.2 GHz, 8 Go Ram). When the cut-off wavenumber
criterion (34) were applied, 17738 points in the wavenumber space (or wall plane waves) were
considered. For the sensitivity method, 815 seconds were used to calculate the entire spectrum
785 whereas 331 seconds were used by the sampling method with the fine mesh. When the cut-off
wavenumber criterion (33) were applied, 9604 points in the wavenumber space were
33
considered. The computation time are reduced to 689 seconds for the sensitivity method and
114 seconds for the sampling method with the coarse mesh. This computing time have been
given for indication but they should be relativized. It can be strongly depend on the
790 vibroacoustic model used to describe the panel. When a commercial software is used, an
important parameter can be the number of load cases being considered. In the case of the
sensitivity method, this one corresponds to the number of wall plane waves whereas for the
sampling method, it is the number of realizations. It is an important advantage of the sampling
method compared to the sensitivity approach. Moreover, more than 90% of the computation
795 time of the sampling method corresponds to the synthetize of the WPF with (17). The
summation which appears in this equation is performed with a FOR loop in the MATLAB
program, which is time consuming. The use of a programming language such as FORTRAN or
C could certainly save computing time.
Even if the numerical process related to the sampling process is not fully optimized, the
800 computing times given previously clearly show that the sampling method permits to save
computing times compared to the sensitivity method.
805 To verify the accuracy of the sampling approach in function of different physical parameters,
we are going to modify two parameters of the nominal case: the convective velocity and the
panel thickness. The others parameters will remain unchanged. The Chase model and 30
realizations are considered in this section. The modification of these parameters will change the
hydrodynamic coincidence frequency and the interaction between the pressure fluctuation of
810 the TBL and the panel response.
830
Fig 11. 10 log 10 S M in function of frequency (dB, ref. 1 m/s2/Hz0.5). Results for the
convective velocity, Uc=25 m/s. Chase Model. Full line, the sensitivity function method
(reference); dash line, the sampling method considering the fine mesh (45 x 9) and the cutoff
wavenumber criterion (34) based on the TBL characteristics; dashed-dotted line: the sampling
method considering the coarse mesh (14 x 9) and the cutoff wavenumber criterion (33) based
835 on the plate characteristics.
35
again a good agreement between the different calculations. When using the coarse mesh, the
differences are slightly more important for the 6 mm thick panel than for the 1.5 mm thick
845 panel. It is always an effect of the weakness of the panel filtering effect. It should however
notice that the size of the coarse mesh (i.e. 90 points) is significantly lower than the one of the
fine mesh (i.e. 405 points). For this case, it can be accepted to lose slightly in accuracy in order
to decrease the number of degree of freedom of the vibro-acoustic model.
850
h=1.5 mm
h=6 mm
Fig 12. Same results than Fig. 11 with two different plate thickness: upper curves, h=1.5 mm;
lower curves, h=6 mm.
Now let us illustrate one interest of the sampling approach with an application on a complex
panel. Indeed, one has shown that the number of load cases which should be considered in the
sensitivity method corresponds to the number of wall plane waves whereas for the sampling
860 method, it corresponds to the number of realization which is small compared to the number of
wall plane waves. The number of forced responses which should be calculated with the vibro-
acoustic model is then relatively small when using the sampling method. This can be of high
interest when the vibro-acoustic calculations are performed by a commercial software for which
36
the calculation process cannot be easily modified. For highlighting this interest, one considers
865 a complex panel composed by a rectangular plate orthogonally stiffened by ribs regularly
spaced. The rectangular plate corresponds to the one of the nominal case (see Tab. 1). We recall
that the plate is assumed to be simply-supported at its four edges and it is excited by a turbulent
air flow with the flow direction parallel to the longest edges of the plate. The rib cross-section
is rectangular, 3 mm thick and 60 mm high. The rib spacing in the direction of the longest plate
870 edges is 500 mm whereas it is of 300 mm in the other direction. The plate and the ribs are both
made of aluminum (see Tab. 1). The characteristics of the flow considered for this application
are those of the nominal case (see Tab. 1) and the Chase model is used to describe the WPF.
The stiffened panel is modelled using the finite element method and the MSC/NASTRAN
875 software. The plate and the ribs are modelled with 2D shell elements (i.e. CQUAD4 elements
with PSHELL properties) as shown on Fig. 13. A criterion of six elements by flexural
wavelength at 300 Hz were considered. A direct analysis (i.e. SOL 108) in the MSC/NASTRAN
software allows us estimating the forced response of the stiffened panel excited by a
deterministic load. The sampling method is then used to estimate the panel response to the WPF
880 induced by the TBL. The receiving point of interest is M’ of coordinates
xM , yM 0.14 m, 0.22 m which corresponds to the node of the mesh closest the point M of
the nominal case of Sec. V.
60 mm
885
Fig. 13. Finite element mesh of the stiffened plate (view opposite side to the flow). 4216
nodes, 4095 quadrilateral elements.
37
The numerical process were decomposed in 3 steps:
890 - the WPF corresponding to 30 realizations were synthetized using Eq. (21) and a
MATLAB program. The cut-off wavenumber criterion (33) was considered for this application
as the ribbed panel was stiffer than the panel of the nominal case. For each realization, the WPF
was exported in an ASCII file using the MSC/NASTRAN input data file format. The size of
each file was 128 Mo;
895 - A direct analysis was performed with the MSC/NASTRAN solver to calculate the
forced responses induced by the WPF of the different realizations. To do this, multi-load cases
were managed in the software by defining different SUBCASE. The 30 files containing the
WPF of the 30 realizations were read in 170 seconds. The calculations were then performed in
90 seconds on the computer described previously. The forced responses at the receiving point
900 M’ were exported in a PCH ASCII file;
- The PCH file of 8 Mo were read with a MATLAB program and the ensemble average
over the 30 realizations (i.e. Eq. (22)) were performed to evaluate the ASD function of the panel
acceleration at point M’ induced by the TBL excitation.
905 In order to validate this numerical process, the calculations were performed first considering
the unstiffened panel (which corresponds to the nominal case of Sec. V). The results are
compared on Fig. 14 with the results of the sensitivity method. A good agreement between the
two calculations can be globally observed on the whole frequency range that validate the
numerical process. Above 250 Hz, slight shifts of the peaks can be noticed that can be attributed
910 to the finite element discretization of the vibro-acoustic problem.
38
Fig. 14. 10 log 10 S in function of frequency (dB, ref. 1 m/s2/Hz0.5) for the panel without
stiffeners. Full line, the sensitivity method at point M; dashed-dotted line, the sampling
915 method at point M’ using MSC/NASTRAN FEM code (with averaging on 30 realizations).
Grey lines, results of each one of the 30 realizations. Calculation parameters: Chase model,
Uc=50 m/s, cutoff wavenumber criterion (33) based on the plate characteristics.
920 The calculations were then achieved for the stiffened plate. We can underline that it was not
necessary to perform again the first step of the numerical process described above because the
characteristics of the flow are the same for the two cases. The WPF are then unchanged. The
results of the sampling method are plotted on Fig. 15. Compared to the unstiffened plate, we
observe that the vibratory levels are globally lower and the first peak appears at a higher
925 frequency. The fact that the static stiffness of the ribbed plate is higher than the one of the bare
plate explains this behavior. Moreover, some groups of peaks appears on the spectrum of the
stiffened panel. These groups can be related to the behavior of an orthotropic plate [46]. Indeed,
as the plate is stiffened by two ribs in one direction and three ribs in the other direction, the
flexural stiffness is lower in one direction than the other one. The panel has then a behavior
930 equivalent to an orthotropic plate in the low frequency range [46]. In higher frequencies, more
complex phenomena like the propagation of Bloch-Floquet waves [40] would influence the
panel behavior. It is however outside the scoop of the present application which has been
proposed for highlighting the interest of the sampling approach.
39
935
Fig. 15. 10 log 10 S M ' in function of frequency (dB, ref. 1 m/s2/Hz0.5) for the stiffened
panel. Dashed-dotted line, the sampling method using MSC/NASTRAN FEM code (with
averaging on 30 realizations). Grey lines, results of each one of the 30 realizations.
Calculation parameters: Chase model, Uc=50 m/s, cutoff wavenumber criterion (33) based on
940 the plate characteristics.
The numerical process proposed in this paper is based on two main characteristics:
945
- (a) on the formulation of the random excitation problem expressed in the frequency-
wavenumber domain. This permits using the analytical expression of the wall pressure spectrum
directly in the wavenumber of the well-known models in the literature (Corcos [2], Chase [3],
Smol’yakov [4], etc.). Moreover, the truncation of the wavenumber space can be achieved
950 easily with cut-off wavenumbers defined from the panel characteristics when the panel filtering
effect is predominant. This permits taking advantage of this well-known effect to optimize the
numerical process. Indeed, when the cut-off wavenumber is based on the panel characteristics,
it is not necessary to consider a fine spatial description of the pressure field, as discussed in Sec.
V. We should however emphasize that the filtering effect of the structure is not always enough
40
955 important to vanish the contributions of the convective ridge. This depends in particular on the
frequencies of interest, the panel boundary conditions, and the WPF model. In this case, the
criterion (34) based on the TBL characteristics and a fine spatial description of the pressure
field (see Eq. ((37)) should be considered;
- (b) on the interpretation of the WPF as uncorrelated wall plane waves. This provides
960 a simple way of synthetizing the realizations of the WPF corresponding to the CSD function.
Indeed, Eq. (17) can be applied directly by considering the ASD functions of the wave
amplitudes defined by Eq. (14). The WPF is then simply obtained by a summation of the wall
plane waves defined from the CSD function of the wall pressure. The process is then easily
implemented in a computer program and is not very time consuming.
965
The numerical application showed than the sampling method gives a good estimation of the
panel response even when a relatively small number of realizations (typically 30) are
considered. This means that only a few deterministic load cases should be considered in the
vibroacoustic calculations. This permits to save computing times compared to the classical
970 approaches like the sensitivity method. As it has been highlighted on a stiffened panel, the
process can be used with any commercial vibroacoustic software applications based on an
element-based method. In this situation, a pre-processing program can be used to synthetize the
pressure field on each node of the mesh for each load case (corresponding to one realization)
and to export them in an input datafile adapted for the vibroacoustic software. As the number
975 of load cases is relatively small, the size of this input datafile should also remain relatively
small. After running the vibroacoustic code, a post-processing program read the output datafile
containing the system responses and performed the ensemble averaging for the different load
cases (i.e. realizations). The process proposed is therefore non-intrusive in the vibroacoustic
software.
980
The proposed numerical process has been applied to estimate the vibratory response of a
vibrating panel. From a theoretical point of view, there is no obstacle for extending it to predict
the pressure radiated by the panel excited by a TBL. Indeed, the formulation in the frequency-
wavenumber domain of the radiated pressure can be achieved similarly to the panel vibration
985 [39]. It is then easily to show that the ASD function of the radiated pressure can be estimated
from the superposition of the uncorrelated wall pressure plane waves as for the panel vibration
(see Sec. III). From a practical point of view, it remains however to study the convergence of
41
the process in function of the number of realization and to define cut-off wavenumber criteria
adapted to the evaluation of the radiated noise.
990
ACKNOWLEDGMENTS
This work was carried out in the framework of the LabEx CeLyA ("Centre Lyonnais
d'Acoustique", ANR-10-LABX-60).
995
REFERENCES
[1] Flinovia – Flow induced noise and vibrations issues and aspects: A focus on
measurement, modeling, simulation and reproduction of the flow excitation and flow
induced response (Springer International Publishing Switzerland, 2015, pp. 357)
[2] G. M. Corcos, “Resolution of pressure in Turbulence,” J. Acoust. Soc. Am. 35, 192-
199 (1963).
[3] D.M. Chase, “The character of the turbulent wall pressure spectrum at subconvective
wavenumbers and a suggested comprehensive model,” J. Sound Vib. 112, 125-147
(1987).
[4] A.V. Smol’yakov and V.M. Tkachenko, “Model of a field of pseudosonic turbulent
wall pressures and experimental data,” Sov. Phys. Acoust. 37, 627-631 (1991)
[5] Y. Hwang, W. Bonness, and S. Hambric, “Comparison of semi-empirical models for
turbulent boundary layer wall pressure spectra,” J. Sound Vib. 319, 199-217 (2009).
[6] M. Goody, “Empirical spectral model of surface pressure fluctuations,” AIAA J. 42,
1788‑1794 (2004).
[7] Y. Rozenberg, G. Robert, et S. Moreau, “Wall-Pressure Spectral Model Including the
Adverse Pressure Gradient Effects,” AIAA J. 50, 2168‑2179 (2012).
[8] E. Ciappi, F. Magionesi, S. De Rosa, F. Franco, “Hydrodynamic and hydroelastic
analyses of a plate excited by the turbulent boundary layer,” J. Fluids Struc. 25, 321-
342 (2009).
42
[9] W. Bonness, D. Capone, and S. Hambric, “Low-wavenumber turbulent boundary layer
wall-pressure measurements from vibration data on a cylinder in pipe flow,” J. Sound
Vib. 329, 4166-4180 (2010).
[10] W. R. Graham, “A comparison of models for the wavenumber-frequency spectrum of
turbulent boundary layer pressures,” J. Sound Vib. 206, 541-565 (1997).
[11] W. A. Strawderman, “Turbulence-induced plate vibrations: An evaluation of finite-
and infinite-plate models,” J. Acoust. Soc. Am. 46, 1294-1295 (1969).
[12] W. A. Strawderman and R. A. Christman, “Turbulence-induced plate vibrations: Some
effects of fluid loading on finite and infinite plates,” J. Acoust. Soc. Am. 52, 1537–
1552 (1971).
[13] H. G. Davis, “Sound from turbulent boundary layer excited panel,” J. Acoust. Soc.
Am. 49, 878–889 (1971).
[14] W. R. Graham, “Boundary layer induced noise in aircraft. Part I: The flat panel model,”
J. Sound Vib. 192, 101–120 (1995).
[15] S. H. Ko and H. H. Schloemer, “Flow noise reduction techniques for a planar array of
hydrophones,” J. Acoust. Soc. Am. 92, 3409–3424 (1992).
[16] S. H. Ko and H. H. Schloemer, “Calculations of turbulent boundary layer pressure
fluctuations transmitted into a viscoelastic layer,” J. Acoust. Soc. Am. 85, 1469–1477
(1989).
[17] D. Mazzoni, “An efficient approximation for the vibro-acoustic response of a turbulent
boundary layer excited panel,” J. Sound Vib. 264 , 951–971 (2003).
[18] M. L. Rumerman, “Estimation of broadband acoustic power due to rib forces on a
reinforced panel under turbulent boundary layer-like pressure excitation. I. Derivations
using strong model,” J. Acoust. Soc. Am. 109, 563–575 (2001).
[19] M. L. Rumerman, “Estimation of broadband acoustic power due to rib forces on a
reinforced panel under turbulent boundary layer-like pressure excitation. II.
Applicability and validation,” J. Acoust. Soc. Am. 109, 576–582 (2001).
[20] M.L. Rumerman, “Estimation of broadband acoustic power radiated from a turbulent
boundary layer-driven reinforced finite plate section due to rib and boundary forces,”
J. Acoust. Soc. Am. 111, 1274–1284 (2001).
[21] E. Ciappi, S. De Rosa, F. Franco, P. Vitiello, M. Miozzi, “On the dynamic behavior of
composite panels under turbulent boundary layer excitations,” J. Sound Vib. 364, 77-
109 (2016).
43
[22] I. Harari, M. Slavutin, and E. Turkel, “Analytical and Numerical Studies of a Finite
Element PML for the Helmholtz Equation,” J. Comp. Acoust. 8, 121-137 (2000).
[23] D. W. Herrin, T. W. Wu, and A. F. Seybert, “Chapter 8 - Boundary element
modelling,” Handbook of Noise and Vibration Control (Ed. M. J. Crocker, John Wiley
and Sons, New York, NJ, 2007)
[24] “ACTRAN 12.2 User’s Guide. Vol. 1: Installation, operations, theory and utilities.”
(Free Field Technologies S.A, Belgium, 2012).
[25] C. Hong, K.K. Shin, “Modeling of wall pressure fluctuations for finite element
structural analysis,” 329, 1673-1685 (2010).
[26] F. Franco, S. De Rosa, E. Ciappi, “Numerical approximations on the predictive
responses of plates under stochastic and convective loads,” J. Fluids Struc. 42, 296-
312 (2013).
[27] M.N. Ichchou, B. Hiverniau, B. Troclet, “Equivalent ‘rain on the roof’ loads for
random spatially correlated excitations in the mid-high frequency range,” J. Sound
Vib. 322, 926-940 (2009).
[28] S. De Rosa, F. Franco, “Exact and numerical responses of a plate under a turbulent
boundary layer excitation,” J. Fluids Struc. 24, 212-230 (2008).
[29] E. Ciappi, F. Magionesi, S. De Rosa, F. Franco, “Analysis of the scaling laws for the
turbulence driven panel responses”, J. Fluids Struc. 32, 90-103 (2012).
[30] S. De Rosa, F. Franco, E. Ciappi, “A simplified method for the analysis of the
stochastic response in discrete coordinates,” J. Sound Vib. 339, 359-375 (2015).
[31] J. Lin, W. Zhang, J. Li, “Structural responses to arbitrarily coherent stationary random
excitation,” Comp. and Struct., 50, 629-633 (1994).
[32] C.C. Caprani, “Application of the pseudo-excitation method to assessment of walking
variability on footbridge vibration,” Comp. and Struct., 132, 43-54 (2014).
[33] A. Hekmati, D. Ricot, and P. Druault, “Numerical synthesis of aeroacoustic wall
pressure fields over a flat plate: generation, transmission and radiation analyses J.
Sound Vib. 332, 3163-3176 (2013).
[34] L.E. Wittig and A.K. Sinha, “Simulation of multicorrelated random processes using
the FFT algorithm,” J. Acoust. Soc. Am. 58, 630-634 (1975).
[35] W.K. Blake, Mechanics of Flow-Induced Sound and Vibration. Vol. II: Complex Flow-
Structure Interaction (Academic press, Inc. Orlando, FL, 1986. pp. 497)
44
[36] S. Hambric, Y. Hwang, and W. Bonness, “Vibrations of plates with clamped and free
edges excited by low-speed turbulent boundary layer flow,” J. Fluid Struct. 19, 93-110
(2004).
[37] T. Miller, J. Gallman, M. Moeller, “Review of turbulent boundary layer models for
acoustic analysis,” Proceedings of the 49th AIAA Aerospace Sciences meeting,
Orlando, Florida, January 2011.
[38] W. A. Strawderman, “Wavevector-Frequency Analysis with Applications to
Acoustics,” Technical Report No. 8209, NUSC (1988), pp. 244.
[39] C. Maury, P. Gardonio, and S. J. Elliott, “A wavenumber approach to modelling the
response of a randomly excited panel, Part I : General Theory,” J. Sound Vib. 252, 83
‑113 (2002).
[40] L. Maxit and V. Denis, “Prediction of flow induced sound and vibration of periodically
stiffened plates,” J. Acoust. Soc. Am. 133, 146-160 (2013).
[41] F. Birgersson, N.S. Ferguson and S. Finnveden, “Application of the spectral finite
element method to turbulent boundary layer induced vibration of plates,” J. Sound Vib.
259, 873–891 (2003).
[42] A.D. Pierce, Acoustics: An Introduction to Its Physical Principles and Applications
(Acoustical Society of America, Woodbury, NY, 1989, pp.678).
[43] D.M. Chase, “Modeling the Wavevector-Frequency Spectrum of Turbulent Boundary
Layer Wall Pressure,” J. Sound Vib. 70, 29-67 (1980)
[44] N.C. Martin, P. Leehey, “Low wavenumber wall pressure measurements using a
rectangular membrane as a spatial filter,” J. Sound Vib. 52, 95-120 (1977)
[45] A. Caiazzo, R. D’Amico, W. Desmet, “A generalized corcos model for modelling
turbulent boundary layer wall pressure fluctuations,” J. Sound Vib. 372, 192-210
(2016).
[46] D.A. Bies, C.H. Hansen, Engineering Noise Control: Theory and Practice (Spon Press
Fourth ed., Abingdon, UK, 2009, pp. 737)
45