You are on page 1of 42

CRANFIELD UNIVERSITY

Analysis of CFD and experimental results of a mixing layer over a


deep cavity at a high-subsonic speed

Unai Urrutia (S373769)

SATM
Aerospace Computational Engineering

Academic Year: 2021 - 2022


ABSTRACT
Comparisons between experimental and numerical results for a high-subsonic
(M=0.85) open cavity with a length-depth ratio of L/D=0.42 have been carried out.
It is the adequate flow and cavity structure to analyse two-dimensional unsteady
simulations because this geometry combination leads to a two-dimensional
organization of the flow. The computational results show very regular self-
sustained oscillations and oscillatory modes caused by the feedback mechanism
are predicted correctly, with a fundamental frequency at 1938 Hz and higher
discrete frequencies that are close to the harmonics. Experimental results agree
with Rossiter modes, especially for the first mode and computational modes
predict with an acceptable error, both the amplitude and frequency of modes. It
has also been identified that the amplitude error is caused by the overestimation
of the computational values rather than a specific error related to the modes, and
amplitudes match after consideration of the inherent error. As seen in the
literature for similar cases, a hybrid method that combines a common RANS
model for the boundary layer and an LES Sub-Grid Scale model for the rest of
the flow has been proposed to solve the frequency mode gap between
experimental results and computational results. Moreover, the possibility of the
inclusion of an object in the cavity has also been considered, as it is the main
case in the industry application of open cavities such as in the case of bays where
weapons are stored. An attempt has been made to evaluate possible changes of
conditions both in the flow and in the geometry, considering cases in which Mach
number, Reynolds, flow velocity or the cavity measures are changed.

i
TABLE OF CONTENTS
ABSTRACT .......................................................................................................... i
LIST OF FIGURES ............................................................................................ iii
LIST OF TABLES ............................................................................................... iv
LIST OF EQUATIONS ........................................................................................ v
1 INTRODUCTION ............................................................................................. 1
2 LITERATURE REVIEW.................................................................................... 3
3 METHODOLOGY............................................................................................. 8
3.1 Period and sampling frequency ................................................................. 8
3.2 Filtering...................................................................................................... 9
3.3 Mean reduction.......................................................................................... 9
3.4 Zero Padding ........................................................................................... 11
3.5 Fourier transforms ................................................................................... 11
3.6 Windowing ............................................................................................... 12
3.7 Rectification ............................................................................................. 13
3.8 Conversion from Pressure to Sound Pressure Level............................... 14
3.9 Data Interpolation and Comparison ......................................................... 14
4 RESULTS AND DISCUSSION ....................................................................... 15
4.1 Aliasing .................................................................................................... 15
4.2 Pressure signal........................................................................................ 15
4.3 Spectrum ................................................................................................. 18
4.4 Sound Pressure Level ............................................................................. 21
5 CONCLUSION ............................................................................................... 32
REFERENCES ................................................................................................. 35

ii
LIST OF FIGURES
Figure 1- Raw CFD Data................................................................................... 15
Figure 2 - CFD Data Oscillations ...................................................................... 16
Figure 3 - Zero Meaned CFD Data ................................................................... 17
Figure 4 - Windowed CFD Data ........................................................................ 18
Figure 5 – Windowed Spectrum Before Rectification ........................................ 19
Figure 6 - Spectrum Without Windowing and Mean Reduction ......................... 19
Figure 7 – Windowed Spectrum After Rectification ........................................... 20
Figure 8 - Spectrum Comparison ...................................................................... 21
Figure 9 - CFD and Experimental SPL Comparison ......................................... 22
Figure 10 - CFD and Rossiter Values Comparison ........................................... 23
Figure 11 - Rossiter and Experimental SPL Comparison .................................. 24
Figure 12 - CFD and Experimental Error Evolution ........................................... 26
Figure 13 - Aligned Error Evolution ................................................................... 27
Figure 14 - Adjusted CFD and Experimental Results Comparison ................... 28

iii
LIST OF TABLES
Table 1 - Mode Frequency comparison ............................................................ 24
Table 2 - Mode Amplitude Comparison............................................................. 25

iv
LIST OF EQUATIONS
Equation 1 - Brown Equation .............................................................................. 3
Equation 2 - Rossiter Formula ............................................................................ 4
Equation 3 - Modified Rossiter Formula .............................................................. 5
Equation 4 - Period Calculation........................................................................... 8
Equation 5 - Sampling Frequency Calculation .................................................... 8
Equation 6 - Nyquist Equation............................................................................. 8
Equation 7 - Pressure Signal Components ......................................................... 9
Equation 8 - Pressure Signal Filtering................................................................. 9
Equation 9 - Mean Reduction............................................................................ 10
Equation 10 - Fourier Transform ....................................................................... 11
Equation 11 - Fast Fourier Transform ............................................................... 12
Equation 12 - Amplitude Rectification ............................................................... 13
Equation 13 - Conversion Pressure to SPL....................................................... 14

v
1 INTRODUCTION
It was in the 19th century that sounds and resonant tones that are created when
a cavity or a surface cut is subject to airflow were discovered. However, this
feature is becoming increasingly popular and challenging as time goes by
because cavity flows are widely present in aerospace applications and they are
becoming popular with the purpose of avoiding radar detection and the pursuit of
stealth in the developments of future aircraft. The solution for this case lays in the
internal weapon bays, which are open when necessary. However, as the flow
over an open cavity is usually unsteady when the store of the weapon is opened
and the weapon is subtracted, a cavity is created inside the store and the flow
field in the cavity and around it depends on the geometry of the cavity and
conditions of the flow. This circulation of flow creates an aerodynamic effect that
is not beneficial as it creates resonances and important fluctuations in the
pressure that affect the walls of the cavity and also the aircraft itself. These self-
sustaining pressure fluctuations can create vibrations in the structure that can
lead to structural fatigue and also can make the store move.

Cavities can be subjected to different kinds of flows according to the velocity of


the flow circulating on them and they are categorized as subsonic, transonic or
supersonic depending on the Mach number, which is a ratio between the velocity
of the flow and the velocity of the sound. If the Mach number is lower than 0.8,
the flow is categorized as subsonic, if the Mach number is between 0.8 and 1 is
named as transonic and if it is higher than 1, supersonic. In the case that was
analysed by Forestier et all (Nicholas Forestier, 2002) and will be compared with
computational methods in this paper, the Mach number is 0.8 so it is categorized
as high-subsonic.

Cavity flows are also categorized as open cavity flows or closed cavity flows
depending on the aspect ratio, length over depth ratio, of the cavity. Although
many authors set the threshold in different levels, can be assumed that everything
above L/D=7 is a closed cavity and the cavities below that limit are open cavities.
Open cavities are the ones that are subjected to strong pressure oscillations that
can lead to very important noise radiation that could reach levels around 170dB.

1
Other important impacts can be related to structural damage due to vibrations,
movement of the structure and heat transfer.

Among the open cavities, there are depth and shallow open cavities. Shallow
cavities are ones that have an aspect ratio that is higher than 2 and depth cavities,
the ones that have a ratio lower than 2. However, the main difference between
the shallow and depth cavities is that for shallow ones, the random component of
the unsteady pressure predominates, while for the depth cavities, the
predominant component is the periodic one. This difference is analysed by
obtaining the spectrum of the pressure oscillations. For the shallow cavities, in
which the random oscillations are predominant, the spectrum covers all the
frequencies and there are no clear peaks and valleys. However, for deep cavities,
as oscillations are mainly periodic, these are concentrated in some frequencies
called modes ore stages and clear peaks and valleys appear in the spectrum of
the pressure.

With a Mach number of 0.8 and a length to depth ratio of 0.42, the combination
of cavity geometry and flow analysed in this paper and compared to the
experimental values obtained by Forestier et all is particular as it creates very
stable self-sustained oscillations with a two-dimensional organization of the flow.
The study of this case is very important to evaluate the nature behind this
difference and also to try to link this difference between two-dimensional and
three-dimensional organizations to the elliptic instability. This geometric and flow
combination is helpful to study this relationship because it has a flow that is very
organized.

To analyse the nature of this difference, the objective of this paper is to compare
results obtained with experiments developed by Forestier et all and Rossiter
(Rossiter, 1964) formula to the results obtained with the computational methods
and suggest explanations if significant differences are found. Different
alternatives that could happen in industrial applications such as the
concatenation of different cavities or the inclusion of objects in the cavities will
also be evaluated. Finally, minor changes in values will be induced to the problem
to evaluate how the results would change.

2
2 LITERATURE REVIEW
Sondhaus (Sondhaus, 1850) was the first one that analysed in 1854 that blowing
air against an edge could produce tones. In the following 80 years, it was
discovered that vortex motion was the key component when producing the sound
when blowing the air against the edge. However, many experiments done during
those years were not successful as were based on analogies with water jets that
did not give conclusive results. Moreover, investigations did not include any
systematic way or formula to deduce the frequencies and was assumed that
jumps in frequencies had a simple numerical relation as octaves.

It was not until 1937 that Brown (Brown, 1937), with his experiments on the edge
tone, published the results that led to an equation [Eq.1] that gave the sequence
of frequencies.

n = 0.466j(U − 40)(𝐼 ⁄ℎ − 0.07)

Equation 1 - Brown Equation

It was the first time that the frequency of tones was related to the mean velocity
of the flow and the distance of the edge h. However, this formula is limited
because it gives adequate frequencies within a 6 per cent range for frequencies
between 20 to 5000 c/sec and flows velocities between 120 to 2000 cm/sec.

In 1955, Krishnamurty (K.Krishnamurty, 1955) studied the flow of subsonic and


supersonic flows over a rectangular cavity, concluding that they emit strong
acoustic radiation. It is the first study in which frequency distributions are included
and for M=0.81, the results are similar to those obtained for this paper, where the
first tone is the strongest with 160dB, the second tone is close to harmonic and
the subsequent are much lower. The relation between the length of the cavity,
speed of flow and frequency was also stated. As airspeed increases the
frequency of tones increases and the higher the length of the cavity, the lower
the amplitude of the tones.

The contemporary Plumbee et all (H.E. Plumbee, 1962) al also investigated the
nature of the periodic pressure fluctuations and concluded that it is caused by the

3
unsteadiness in the turbulent boundary layer that approaches the cavity.
However, this conclusion was not in line with Krishnamurty’s investigation
because it was discovered that periodic pressure fluctuations also exist while the
boundary layer is laminar.

However, the investigations led by Krishnamurty were based on very small


cavities and did not allow the investigation of the flow inside the cavity. This is the
reason why in the same period, Roshko (Roshko, 1955) investigated the effect of
the depth of the cavity varying the breadth/depth ratio to analyse the periodic
high-intensity sound waves that were created when airflow passed the cavities.
It was found that as the ratio of breadth to depth decreases, the pressure
fluctuations increase. This increase in pressure fluctuations makes the amplitude
of the frequency modes increase but does not affect the value of the frequencies
where these modes are located.

Rossiter in 1966 found that pressure fluctuations contain a random component


and periodic component and that the random component is the predominant one
in shallower cavities while the periodic component is predominant for deeper
cavities, with length/depth ratios higher than 4. Another discovery was the jumps
on the frequency of the dominant mode when increasing the speed of the flow,
which seemed to be arbitrary.

Rossiter also retook the idea of finding an empirical equation that could find the
values of frequencies based on the feedback loop that acoustic waves create in
the downstream boundary from vortices that were at the beginning from the
upstream cavity boundary. As a result, obtained a semi-empirical equation [Eq.
2]:

𝑈 (𝑚 − 𝛼)
f=
𝐿 ( 1 + 𝑀)
𝐾

Equation 2 - Rossiter Formula

4
This equation was then modified by Heller et all (H.H. Heller, 1971):

𝑈 (𝑚 − 𝛼)
f=
𝐿 1 (𝛾 − 1) < >?/<
(𝐾 + 𝑀 91 + 2 𝑀 = )

Equation 3 - Modified Rossiter Formula

B
𝛼 = 0.062 AC D is a constant that is linked to the phase lag between shear layer

and upstream pressure wave. This equation enables to get values of frequencies
accurately in most cases but still, the problem with amplitude estimations stated
by Brown is not solved.

The values obtained with this equation match a lot of values, especially those
concerning narrow cavities with a ratio of length to depth greater than 2. However,
for deeper cavities, depending on the Mach number, when M>0.7, self-sustained
oscillations are closer to harmonics than in shallower cavities. As a result, the
parameter α has to be lowered, as Rossiter stated when establishing that α
depends on the length to depth ratio.

In 1979, Rockell & Naudascher (Naudascher, 1978) categorized the self-


sustaining oscillations, attributed to instability in the cavity and feedback
mechanisms, in three groups. First, fluid dynamic, when oscillations come from
the instability of the flow itself, second, fluid resonant, when oscillations are forced
by standing wave effects and third, fluid-elastic, where are combined with the
movement of a solid wall. Experiments with cavities that had different shapes we
also done, including different ramps, to analyse the effect in the pressure
oscillations. They concluded that pressure oscillations are very influenced by
shear instability, disturbance feedback, resonant wave conditions and structural
elasticity.

The impact of the sound resonances in the structure and stability of the object
has also been studied. The structural fatigue caused by the pressure oscillations,
which is especially important in cavities that have a low length to depth ratio, was
thoroughly studied by East (East, 1966). On the other hand, adverse pressure
distributions were described and investigated by Stallings et all (R.L. Stallings,

5
1995), who measured the forces and moments created in cavity flows with
subsonic and supersonic flows and also concluded that the effect of the Mach
number on the forces on the cavity was an indirect effect through the effect of
Mach number in the type of cavity flow field rather than a direct effect.

With the inclusion of the computational methods and the development of more
advanced computational techniques and facilities, several computational studies
have been carried out in the last years. However, many of the developed methods
are not successful in translating experimental results and features to
computational simulations. As stated by Colonious (Colonius, 2001), there are
two main problems that can lead to low-quality sound radiation results. The first
one is the excessive dissipation and dispersion that can be obtained when using
low order and upwind finite difference and finite volume discretization. The
second reason is the possibility of having an excessive reflection of disturbances
caused by artificial boundary conditions. They present a technical issue because
artificial boundary conditions have to permit the flow of vortical and acoustic
waves without reflecting them.

Moreover, the quality of the computational results depends significantly on the


computational method that is selected. As Loupy et all (Loupy, 2018) viewed in
the developed analysis with M=0.85, results with LES methods are far better than
with URANS models. Regarding frequency spectrum, the accuracy of both the
frequencies and amplitudes depend on the selected methods. However, for flows
with high Reynolds numbers, the cost of using LES methods is very expensive
with current facilities. Consequently, several authors have proposed hybrid
models that combine LES and URANS models to have better results in an
efficient way, including Spalart & Allmaras(Allmaras, 1997), who proposed a
method called Detached-Eddy Simulations.

Finally, it was in 2002 that Forestier et all investigated a particular combination of


cavity geometry and flow features that leads to results that are different from
those for shallower cavities. Considering a Mach number of 0.8 and a length to
depth ratio of 0.42, concluded that this combination had a two-dimensional flow
organization that differed from the ones with three-dimensional organizations and

6
that it could be useful to understand the difference between them and relate it to
the elliptic instability.

7
3 METHODOLOGY
3.1 Period and sampling frequency
First, the period is calculated as the average of all the periods in the time vector
[Eq.4] and it is the average time between two pressure samples in the pressure
signal.

LM
1
𝑇= I(𝑇JK? − 𝑇J )
𝑁H
JN?

Equation 4 - Period Calculation

The sampling frequency Fs is the inverse of the period [Eq.5].

1 1
𝐹P = =
𝑇 1 LM

𝑁H JN?(𝑇JK? − 𝑇J )

Equation 5 - Sampling Frequency Calculation

This sampling frequency will then be compared to the frequencies of the pressure
signal to verify if the sampling frequency is high enough to avoid aliasing. To test
this condition, the Nyquist equation [Eq.6] will be used, which establishes that the
sampling frequency needs to be at least two times higher than the bandwidth of
the signal.

𝐹P ≥ 2𝐹SJTUVW

Equation 6 - Nyquist Equation

8
3.2 Filtering
The pressure signal has three components [Eq.7] that are the mean component
(p’), the periodic component (p’’) and the random component (p’’’).

𝑝 = 𝑝′ + 𝑝ZZ + 𝑝′′′

Equation 7 - Pressure Signal Components

To subtract the random component and have only the mean and the periodic
components [Eq.8], a filter can be applied that will minimize the impact of those
random components.

𝑝 = 𝑝′ + 𝑝ZZ + 𝑝′′′ − 𝑝′′′

𝑝 = 𝑝′ + 𝑝′′

Equation 8 - Pressure Signal Filtering

Considering that the random component has a very high frequency, low
passband filters could be applied using for example Butterworth equations.
However, as the frequencies are known and MATLAB has several filtering
functions, it is recommendable to use the default ones. Two of them are the
SmoothData and Savitzky-Golay filters. After comparing, the effect of the filters
for this geometry and flow conditions, SmoothData function has been applied
because of its simplicity and cleanness.

3.3 Mean reduction


The next step is to reduce the mean value of the pressure signal because only
the oscillatory component of the signal is interesting for the analysis. The main
purpose of the reduction is that the direct component of the values is a component
with frequency 0Hz and if it is not deducted, will appear in the spectrum and can
cause confusion.

The pressure signal is composed of the direct component and alternating


component, being the last one that is interesting for the analysis. The resulting

9
pressure is the sum of both components so if the direct component is reduced,
only the alternating component will appear. If this reduction is applied [Eq.9], the
value of the spectrum at 0Hz will be zero and no other value as would be without
this reduction. In this report, direct subtraction of the mean value of the mean has
been applied because it is the simplest method and as the mean is
homogeneous, there is no need to apply different mean reduction for each phase
of the signal.

𝑝 = 𝑝′ + 𝑝′′

𝑝 = 𝑝Z + 𝑝ZZ − 𝑝′′

𝑝 = 𝑝′

Equation 9 - Mean Reduction

However, this reduction of the mean can be done in several ways. Another
alternative would be to apply a high pass filter that would get rid of the frequencies
below the threshold. As the frequency of the direct component is 0Hz, it would be
reduced to zero or a minimum value depending on the order of the filter and the
cut-off frequency. For this specific case, as the difference between the frequency
that is to be deleted (0Hz) and the first harmonic (1938Hz) is very big, it would be
sufficient with a first-order filter.

It is important to point out that this filter should have a null gain for the passband
because otherwise, it would modify the pressure oscillation values and another
rectification step would necessary. For example, if a Butterworth filter is applied,
it should be designed so that the gain is one.

10
3.4 Zero Padding
Zero padding is a technique that is used to add values to an array and make it a
value that is equal to a power of two. The values that are added are zeros to avoid
modifying the information in the signal and are added just before the last
component of the original array.

The objective of this process is to make the Fourier transform faster as it is


computationally more efficient to operate the fast Fourier transform with lengths
of the time domain that are a power of 2. However, does not add higher resolution
to the time-domain signal and results could appear smoother as the length of the
plot in the x-axis would be greater.

Zero padding is also used to ensure that the obtained results when circular
convolution is applied are the same as when applying linear convolution,
assuming that linear convolution is the desired method. This is applied to ensure
that the convolution results are not mixed and the information introduced in the
Fast Fourier Transform is as desired.

3.5 Fourier transforms


As the objective of the analysis is to obtain the periodic components of the
pressure signal and classify them according to amplitude and frequencies, the
Fourier transform has to be applied to convert the pressure oscillations signal
from the time domain to the frequency domain.

As every signal, the pressure signal is a sum of signals in the time domain that
have different frequencies. The function of the Fourier transform [Eq.10] is to
separate these signals into different signals and classify them according to
frequencies and amplitudes.

L>?
<aJ
𝑋 [𝑘 ] = I 𝑥U 𝑒 > L bU

UNc

Equation 10 - Fourier Transform

11
As sampling converts data into discrete values, the Fourier transform will not work
in this case. However, the Discrete Fourier Transform can be applied, which
analyses data in discrete values and provides frequency domain components
also in discrete values. Moreover, the Fast Fourier transform [Eq.11] is a modified
version of the DFT that needs less computational steps to perform.

L>?
<aJ
𝑋[𝑘] = I 𝑋c [𝑛]𝑒 > L bU

UNc

Equation 11 - Fast Fourier Transform

Finally, as the Fourier transform provides complex numbers in which the real part
is the amplitude of the pressure spectrum and the imaginary part is the phase,
this information needs to be separated to further analysis. In MATLAB, the real
part of the number is obtained applying the ABS function and the imaginary part
with the function IMAG.

As a difference in the computational and experimental mode frequencies has


been observed, the Welch method has been used to confirm the frequency values
obtained with the Fast Fourier Transform match those obtained with the Welch
method and then, are correct.

3.6 Windowing
The following step is to apply a window, which is a convolution of two functions
that leads to a third function, so that leakage is avoided. This step needs to be
applied before the Fourier Transform but has been included after it for a better
understanding. Sometimes the measured signal does not have an integer
number of periods and the finiteness introduced when applying FFT may
introduce transition shapes into the signal.

Spectral leakage is a consequence of the discontinuities in the original signal,


which has a non-integer number of periods and this effect can be reduced by
applying a window, which helps reduce the amplitude of the mentioned
discontinuities in the boundaries. The undesired discontinuities appear as high-

12
frequency components of the signal that are added to the original signal, whose
frequency can be much higher than the Nyquist frequency and thus the sampling
frequency wouldn’t be sufficient so they would be aliased in frequencies included
in the bandwidth. Then, the spectrum that is obtained when applying FFT is not
the correct one but a modified version that includes errors.

Indeed, the Fast Fourier Transform itself applies a rectangular window that does
not have any impact on the pressure values. For this case, the selected window
is the Hanning window, which is a sine wave or a combination of them, because
it has good results in 95% of the cases and reduces spectral leakage.

However, the Hanning window, unlike the rectangular, applies a reduction factor
of 0.5 in the amplitude that must further be rectified to obtain desired pressure
levels.

3.7 Rectification

As mentioned above, when applying window functions, some of the window types
apply a gain that needs to be rectified to obtain the desired values (Brandt, 2010)
applying a correction factor Aw [Eq.12].

𝑁
𝐴f =
∑L>?
LNc 𝑤(𝑛)

Equation 12 - Amplitude Rectification

For the Hanning window, the values that are obtained after the Fourier transform,
have to be multiplied to rectify the effect of the window in the amplitude of the
pressure. This amplitude rectification has to be applied after the Fourier transform
and just before the conversion to Sound Pressure Levels.

13
3.8 Conversion from Pressure to Sound Pressure Level

As the Fast Fourier transform converts from the time domain to frequency domain
but does not modify Y axes values, the pressure unit in pascal is maintained. To
obtain the Sound Pressure Signal in decibels, the actual pressure is compared
with the threshold of hearing for pressure variations P0, which is 2 𝑥 10>i pascal
and a conversion is applied [Eq.13]. A relevant aspect of this conversion is that
as the conversion is logarithmic, small variations in decibels represent large
differences in pressures. Consequently, the results of the amplitudes of the sound
frequencies need to be very accurate because otherwise, the gap in pressure
levels can be very relevant.

𝑝
𝑆𝑃𝐿 (𝑑𝐵) = 20𝐿𝑜𝑔 p q
𝑝c

𝑝
𝑆𝑃𝐿 (𝑑𝐵) = 20𝐿𝑜𝑔 A D
2 𝑥 10>i

𝑝 <
𝑆𝑃𝐿 (𝑑𝐵) = 10𝐿𝑜𝑔 A D
2 𝑥 10>i

𝑝 <
𝑆𝑃𝐿 (𝑑𝐵) = 10𝐿𝑜𝑔 A D
2 𝑥 10>i

Equation 13 - Conversion Pressure to SPL

3.9 Data Interpolation and Comparison


Finally, once the pressure signal spectrum is converted to decibels, as the length
of computational and experimental arrays is not equal, data needs to be
interpolated to compare both functions. This step can be done in MATLAB with
the function Interp. It is important to set the grid size of the interpolation equal or
lower than the lowest period of the signals because otherwise information could
be lost.

14
4 RESULTS AND DISCUSSION
4.1 Aliasing
First of all, the quality of the data has been analysed, to conclude if it is good
enough to complete the following analysis and compare it to the experimental
data. To complete this task, data have been analysed according to aliasing.

To do so, the sampling frequency has been compared to the Nyquist frequency
[Eq. 5]. In this case, as the highest harmonics of the signal is around 8,000 Hz,
the sampling frequency must be higher than 16,000Hz. As the sampling
frequency 𝐹𝑠 is 1.8778x106 Hz, which is clearly bigger than the Nyquist
frequency, lack of aliasing can be confirmed.

4.2 Pressure signal


Figure 1 clearly shows that as expected, the pressure signal has a mean value
higher than zero with a frequency that is 0Hz, what will lead to a peak in the
frequency spectrum if it is not subtracted as stated in 3.3.

Figure 1- Raw CFD Data

15
In more detail, figure 2 shows clear self-sustaining oscillations of the pressure. It
can be observed also that the peaks are always very similar in value but not
exactly equal. The shape is also very similar to that provided by Forestier et all
for the experimental data for the low-passed filtered signal provided by
transducers in the upstream wall.

Figure 2 - CFD Data Oscillations

After reducing the mean value of the pressure oscillations to zero and filtering to
eliminate the random component, can be observed in Figure 3 that the mean
value for the pressure signal is zero. At this stage, only the periodic component
of the pressure data remains in the signal, which will later be introduced and
analysed by the Fast Fourier Transformation.

16
Figure 3 - Zero Meaned CFD Data

After applying the window, the pressure signal acquires the shape of the window
itself because of the properties of the convolution. The objective that was pursued
by the application of the window is completed because in both extremes’ values
are zero or close to zero and the values of the pressure in the windowed region
are maintained. The shape of the windowed pressure signal depends on the form
of the window. For example, as in the FFT, the applied window is the rectangular
window, the form of the resulting signal is not modified.

17
Figure 4 - Windowed CFD Data

4.3 Spectrum
If compared figures 5 and 6, can be observed that applying a Hanning window
reduces the amplitude of the pressure spectrum. It can also be observed that the
values for the windowed signal are close to half of those of the signal without
applying any window function.

18
Figure 5 – Windowed Spectrum Before Rectification

Figure 6 - Spectrum Without Windowing and Mean Reduction

19
However, after the application of the rectification, the values in Figure 7 match
those in Figure 6 so the objective that was pursued when applying the rectification
factor proposed in 3.7 is completed.

Figure 7 – Windowed Spectrum After Rectification

In figure 8 can be visualized that the peaks for the windowed pressure signal are
narrower, as the leakage that was contained in the signal has been eliminated
when applying the window. On the other hand, the peaks for the signal that has
not been windowed are wider because there is an overlap between neighbour
frequency values. Then, can be concluded that the application of the Hanning
window is effective against leakage for this case.

20
Figure 8 - Spectrum Comparison

4.4 Sound Pressure Level


After converting the frequency values to sound pressure levels, the first step is to
evaluate how accurate are the computational results compared to the
experimental ones. It is known from Forestier et all that the dominant frequency
is 1975 Hz and 155 dB the second one being close to harmonic.

For the computational values, as expected and predicted by Rossiter, being a


deep cavity, the periodic component is predominant over the random component
and the spectrum shows two clear modes. However, against Rossiter’s statement
that “For the deeper cavities there is usually one peak amplitude spectra which
is very much larger than the others, whereas for the shallower cavities there are
usually two or more peaks of almost equal magnitude”, which is not true in this
case, there are several peaks in the spectrum and the first one is the dominant

21
frequency, the second one being close to harmonics. This contradiction between
results and Rossiter’s evaluation was already identified by Forestier et all
because the experimental results showed patterns very similar to those with
computational ones.

In figure 9 can be observed that the amplitude of the computational values in the
peaks is slightly higher than that of the experimental values. Moreover, the
frequencies where the modes are located are lower for the computational results
than for the experimental results. Exact values are provided in Table 1.

Figure 9 - CFD and Experimental SPL Comparison

Figure 10 provides information about the difference between computational and


Rossiter equation values. For this case, the frequencies for the first mode are
relatively similar but for the second mode the difference is significative.

22
Figure 10 - CFD and Rossiter Values Comparison

However, can be observed in Figure 11 that as indicated in Forestier et all,


experimental and Rossiter values agree, especially for the first mode. For the
second mode, the gap in the frequencies, although not critical, increases
significantly when compared with the first mode.

23
Figure 11 - Rossiter and Experimental SPL Comparison

In Tables 1 and 2, frequencies and amplitudes are compared. According to


frequencies, as stated previously, for the first mode, Rossiter and experimental
values agree very well while the computational value has a gap that although
acceptable, should be considered. However, for the second mode, for both the
computational and Rossiter values, the difference is significant.

Experimental Computational Rossiter

Mode 1 1990 Hz 1938 Hz 1972.21 Hz

Mode 2 3991 Hz 3873 Hz 4049.42 Hz

Table 1 - Mode Frequency comparison

24
On the other hand, when comparing amplitudes, the difference is similar for both
the first and the second modes between experimental and computational values.
It is also important to consider that the values are higher for the computational
model, with a difference of 6.75 dB for the first mode and 7,8 for the second mode.

Experimental Computational

Mode 1 155 dB 161.755 dB

Mode 2 153.21 dB 161.03 dB

Table 2 - Mode Amplitude Comparison

Besides, to have a better understanding of the nature of the error, all the vector
of error has been plotted in n Figure 12, where can be seen that most of the error
component between computational and experimental values is in the peaks as
they are not aligned. This is because the computational model does not
accurately predict the frequency modes. Moreover, it is clear that most of the
error component is positive because computational values are higher than
experimental ones. As seen in previous assignments, some computational
methods tend to underestimate the values and have negative errors and other
ones overestimate them and have positive errors. Without further information
about the methods used in this case is difficult to evaluate the causes of this
tendency to have higher values than experimental ones.

25
Figure 12 - CFD and Experimental Error Evolution

Quantitative analysis of this error would be biased because most of the


component of the errors comes from the difference of the peaks. However, to
have a better understanding of this error, mode peaks have been aligned and
error has been obtained again. In this case, figure 13 shows that the error is
random, with an absolute error of 7.3%. Again, it is clear that most components
of the error vector are positive. Exactly, considering the sign of the errors, the
average value of the error is 7.16 dB, which is an important overestimation
considering the logarithmic nature of the magnitude.

26
Figure 13 - Aligned Error Evolution

If both the experimental and computational values are again compared after
subtracting 7.16 dB to all the computational values to delete the positive error of
the model, can be concluded that the peaks are almost at the same level.
Consequently, can be considered in Figure 14 that the before mentioned
difference in table 2 is caused by the error that the computational method includes
in all the values inherently, not particularly to the mode peaks.

27
Figure 14 - Adjusted CFD and Experimental Results Comparison

All in all, the results obtained from the computational model represent accurately
the significance of the periodic component of the pressure oscillations against the
random component for deep cavities as the one analysed by Forestier et all. The
spectrum of the signal clearly shows that the self-sustained oscillations have
established periodic components and not random values. This is the reason why
it has a clear difference between peaks and values because if random
components were predominant, the spectrum would be

After the thorough analysis can be concluded that the computational model
accurately captures frequencies and amplitudes of the periodic components
within a reasonable range of error.

However, if necessary, computational values could be improved using other


techniques as this gap in the values of the modes could be caused by the

28
computational method used to get the computational results. Loupy et all in their
experiments with different computational methods concluded that obtained
results are significantly better with Detached-Eddy and Large-Eddy simulations
than with URANS simulations, especially when obtaining frequencies and velocity
distributions. The investigation resulted that the computational method not only
affects the accuracy of the frequency of the mode but also the accuracy of the
amplitude.

In Knowles, Lawson et all (Lawson, 2009), in the results obtained with M=0.85,
although the investigated cavity was shallower than that of Forestier et all, a
similar frequency gap was also identified when comparing the experimental and
computational values. This simulation was carried out with a URANS code, with
a 𝑘 − 𝜀 model. The proposed reason for the difference was that the URANS
model was not capable of representing accurately the effect of the thickness of
the boundary layer. A hybrid LES/RANS solution was proposed to solve this
difference, which was solved later in Khanal et all (Bidur Khanal ). In this case,
as anticipated by Loupy et all, the accuracy of results was far better and the
frequency gap between experimental and computational modes disappeared.

Although the only values that have been compared are those corresponding to
the problem proposed by Forestier et all, other slight condition changes have
been considered to expand the analysis of the case. The first condition is that if
parameters of the problem were changed, would be a three-dimensional and
periodic organization of the mixing layer instead of a two-dimensional as in
Forestier et all. The other three variables that have been considered are the flow
of the velocity, the width of the cavity and the Reynolds number. First, the change
of the flow velocity would change the mode frequencies jumping from one value
to the other. The impact of this change in velocity can be predicted with the
Rossiter formula and an increase in the velocity would suppose an increase in
the frequencies of the modes. Second, increasing the width of the cavity would
impact the amplitude of the modes but not on the frequencies. The change in the
width has little or no impact on the random component of the fluctuation but has
a significant impact on the periodic component. Finally, the change in the

29
Reynolds number of the fluid could affect the results either changing the thickness
of the boundary layer or changing the nature of the mixing in the shear layer. The
second effect was found to be negligible by Rossiter, but in the second case,
although the shapes of the frequency curves are similar, an important decrease
in the random and periodic components was detected and consequently, the
amplitudes of the modes were lower.

Another alternative that has been evaluated is the inclusion of several cavities
one right after the other. Tuna & Rockwell (Rockwell, 2014) concluded that each
of the cavities has its resonant state at the same moment, having a global
resonant effect. This coupled effect to happen, two conditions must be achieved.
The first one is that there has to be a face shift among the successive resonant
waves. The second condition is that a phase difference must exist between
separation and impingement borders. If these two conditions are fulfilled, the self-
sustained oscillation will maintain above the successive cavities.

Finally, the possibility of having objects stored in the cavities has also been
considered, as could happen with weapons that could be stored inside the
weapon bays. This modification of the problem is very important because, in the
aerospace industry, one of the major problems that cavity oscillations can cause
are related to cavities that store object inside them. In Khanal et all, both a cavity
with a store grid and an empty cavity were evaluated and compared for an
M=0.85. Although this experiment was done for shallower cavities, it could also
be valid for this geometry and flow combinations. The comparison was carried
out among computational results, experimental results and Rossiter modes.

The investigation concluded that for the first mode, an agreement between modes
for clean cavity and Rossiter formula results was almost but those for the cavity
with the stored weapon differed significantly. On the other hand, for the second
mode, the frequency for the cavity with the weapon was very similar to that of the
Rossiter formula but differed significantly from the one from the empty cavity. It
was proposed the possibility of a superposition between the lateral and
longitudinal instabilities as the reason that the mode at 212 Hz in the empty cavity,
which was significantly lower than the Rossiter mode at 279 Hz.

30
Moreover, the empty cavity only included the first two modes while the cavity with
the weapon included several peaks. It was supported that as the cavity with a
stored weapon was divided into two with little movement of airflow from one half
to the other in the store, reducing the flow fluctuations, several modes could be
at the same time without coupling.

31
5 CONCLUSION

The study and analysis provided in the report have compared the computational
results with those obtained experimentally for a Mach number of 0.8 and a length
to depth ratio of 0.42. According to the quality of the data, it has been concluded
that for the used sampling frequency there is no aliasing and spectrum leakage
has been reduced applying a Hanning window.

On the other hand, it has been confirmed that the computational model captures
correctly the periodic self-sustained oscillations as seen when comparing the
pressure oscillations for both the computational and experimental results. It has
also been confirmed that for deep cavities as the analysed one, the predominant
component is the periodic one, as the spectrum has clear peaks that are related
to frequencies at which the pressure oscillates. It has also been concluded that
the computational method accurately captures the periodic oscillations of the
pressure for this cavity and flow characteristics within an acceptable error margin.

However, the computational model tends to result values that are higher than the
experimental ones and if this inherent error is rectified, the amplitudes for
computational and experimental results agree with a minimum error.

A better understanding of the nature of the phenomenon and more accurate


frequency and amplitude results may require more detailed analysis with hybrid
LES and RANS methods such as DES, which has been demonstrated to obtain
better results for Mach numbers similar to the analyzed one. However,
computational requirements should also be considered when dealing with high
Reynolds numbers.

Finally, it has been concluded that if problem conditions were changed, an


increase in the flow velocity would led to increasing jumps in mode frequencies
while changing the width of the cavity would have a significant impact on the
amplitude but no in the frequencies. The inclusion of an object in the store could
also lead to unpredicted values that would make the computational values differ
significantly from the experimental ones.

32
33
34
REFERENCES
Allmaras, P. S. (1997). Comments on the feasibility of LES for wings and on
hybrid RANS/LS approach. International conference on DNS/LES.
Ruston, Lousiana.

Bidur Khanal , K. K. (n.d.). Computational study of cavity flowfield at transonic


speeds. Defence Academy of the United Kingdom.

Brandt, A. (2010). Noise and Vibration Analysis: Signal Analysis and


Experimental Procedures. John Wiley & Sons, Ltd.

Brown, B. (1937). The vortex Motion Causing Edge Tones .

BURNISTON BROWN. (1937). THE VORTEX MOTION CAUSING EDGE


TONES.

Colonius, T. (2001). AN OVERVIEW OF SIMULATION, MODELING, AND


ACTIVE CONTROL OF FLOW/ACOUSTIC RESONANCE IN OPEN
CAVITIES. AIAA.

East, L. (1966). Aerodinamically induced resonance in rectangular cavities.


Sound and Vibration, 277-287.

H.E. Plumbee, J. G. (1962). A theorical and experimental investigation of the


acoustic response of cavities in an aerodynamic flow .

H.H. Heller, D. H. (1971). Flow induced pressure oscialtions in shallow cavities .


Sound and Vibration, 545-553.

K.Krishnamurty. (1955). Acoustic Radiation From Two-Dimensional Rectangular


Cutouts in Aerodynamic Surfaces. National Advisory Committee for
Aeronautics.

Lawson, K. K. (2009). Experimental and computational investigation of an open


transonic cavity flow. Journal of aerospace engineering.

35
Loupy, G. B. (2018). UNDERSTANDING TRANSONICWEAPON BAY FLOWS.
7th European Conference on Computational Fluid Dynamics. Glasgow,
UK.

Naudascher, D. a. (1978). Self-sustaining oscillations of flow past cavities. ASME.

Nicholas Forestier, L. J. (2002). The mixing layer oeder a deep cavity at high-
subsonic sped. Cambridge University Press, 101-145.

R.L. Stallings, E. P. (1995). Measurements of store forces and moments and


cavity pressures for a generic store in and near a box cavity at subsonic
and transonic speeds. NASA.

Rockwell, B. T. (2014). Self-sustained oscillations of shallow flow past sequential


cavities. Cambridge University Press.

Roshko, A. (1955). Some measurements of the flow in a rectangular cutout.


N.A.C.A Tech.

Rossiter, J. (1964). Wind-Tunner Experiments in the Flow over Rectangular


Cavities at Subsonic and Transonic Speeds.

Sondhaus, C. (1850). Ueber dei Schallschwingun- gen der Luft in erhitzten


Glasröhren und in gedeckten Pfeifen von ungleicher Weite. Annalen der
Physik und Chemie , 1–34.

36

You might also like