Vous êtes sur la page 1sur 57

Water Research Laboratory

Never Stand Still Faculty of Engineering School of Civil and Environmental Engineering
Non-stationary Flow Downstream of an Ahmed Body: Frequency
Analysis and Preliminary Assessment of the Triple
Decomposition Technique

WRL Research Report 262


May 2017

By
V Bouillaut, F Murzyn, A Mehel and S Felder
Bibliographic Data Sheet

Report Title Non-stationary Flow Downstream of an Ahmed Body: Frequency Analysis and
Preliminary Assessment of the Triple Decomposition Technique
Report Authors V Bouillaut, F Murzyn, A Mehel and S Felder
Report No. WRL 262
Report Status Final
Date of Issue May 2017
ISBN EAN: 9780733437250
978-0-7334-3725-0
Number of pages 49
Keywords Ahmed body; frequency analysis; Strouhal number; triple decomposition technique;
wake flow

Document Status

Version Reviewed By Approved By Date Issued


Draft Dr Hang Wang
Final Dr Stefan Felder Prof Ian Turner 05 May 2017

Abstract
Air quality is a major concern for people living in cities. Pollutant exposure is dangerous, harmful and
associated with huge costs in terms of public health. Ground vehicles are an important source of fine and
ultrafine particle emissions in urban areas. Reducing exposure to such particles requires an improvement of
knowledge of their dynamics when released in the atmosphere from the exit of the tailpipe. Since the
behaviour of these particles is influenced by the surrounding flow, a better understanding of the flow
properties is required. Herein, the wake flow properties developing downstream of a simplified car model
(Ahmed body) at a reduced scale had been studied in a wind tunnel at ESTACA, France. In the
experiments, the upstream flow velocity was constant corresponding to a model Reynolds number of
Re=50,000. A 2D LDV system was used to investigate the velocity fields. In the present study, the
experimental data were analysed with focus on the non-stationary flow that developed in the near wake of
the Ahmed body. The most representative frequencies were assessed using the Fast Fourier Transform.
Characteristic Strouhal numbers were observed and compared with the literature showing that scale effects
cannot be neglected. A first assessment of the triple decomposition technique was undertaken which aimed
at identifying the contributions of the turbulent structures to the global dynamics. The main results of the
study showed that: i) low and high frequencies could be identified in the wake of the Ahmed body; the
corresponding Strouhal numbers were consistent with the Benard-Von Karman vortex street; ii) high
frequencies were mostly found in the lower and upper shear layers where the turbulence activity was
strongest; iii) good agreement was found with previous experiments with similar flow conditions but
different measurement techniques; iv) application of the triple decomposition technique to the wake flow
was possible assuming that both lower and upper cut-off frequencies could be identified; v) contributions of
turbulent structures of low, mean and high frequencies to the root mean square of the velocity were
presented and discussed. These present results provide new flow features which will improve the
understanding of the complex wake flow behind Ahmed bodies.
Contents

1. Introduction 1
1.1 Context of the present study 1
1.2 Basic concepts of interactions between particles and wake flow 3
1.3 Dimensionless numbers 7
1.4 Principle of the triple decomposition technique 9
2. Experimental set up, instrumentation and flow conditions 10
2.1 Wind tunnel 10
2.2 Ahmed body 11
2.3 Velocity measurements using LDV 12
2.4 Collected data 13
3. Signal processing 15
3.1 From the raw data to an interpolated signal 15
3.2 Fast Fourier transform (FFT) and Power Spectral Density (PSD) of the velocity 16
4. Results 18
4.1 Spectral analysis and characteristic frequencies 18
4.1.1 Spectral analysis of velocities in flow direction 18
4.1.2 Spectral analysis of velocities in upwards direction 20
4.1.3 Spectral analysis of velocities in transverse direction 21
4.2 Summary of characteristic frequencies and Strouhal numbers in the wake flow 22
4.3 Application of the triple decomposition technique downstream of an Ahmed body 24
4.3.1 Introduction 24
4.3.2 Preliminary results: influence of dimensionless longitudinal distance on the
root mean square velocity 25
5. Discussion 27
5.1 Limitation of sampling frequency of raw velocity data 27
5.2 Accuracy of the LDV measurements 28
5.3 Wind tunnel measurements and scale effects 28
6. Conclusion and future works 30
7. Acknowledgments 32
8. References 33
9. Appendices 37
9.1 Matlab code 37
9.1.1 Step 1: Reading data from Excel files to Matlab 37
9.1.2 Step 2: Performing a Fast Fourier Transform and triple decomposition of the
velocity signals 38
9.1.3 Step 3: Turbulence analysis of the decomposed signal components 41
9.2 Results of the FFT analysis for all velocity data in the present study 43

-i-
List of Tables

Table 2-1: Optical properties of the LDV system 13


Table 2-2: Data acquisition positions (Rodriguez 2016) 14
Table 4-1: Summary of the characteristic frequencies and Strouhal numbers in the
present study 24
Table 4-2: Cut-off frequencies for the triple decomposition technique 25

List of Figures

Figure 1-1: Schematic of human body with pathways of exposure to nanoparticles


(Buzea et al. 2007) 1
Figure 1-2: Bimodal distribution of particle diameters emitted from a Diesel engine
(Kittelson et al. 2004) 2
Figure 1-3: Sketch of a typical car model defined as Ahmed body (Kahn and Umale
2014) 5
Figure 1-4: Drag coefficient of an Ahmed body as a function of the rear slant angle 
(Thacker 2010) 5
Figure 1-5: Flow topologies in the wake of an Ahmed body for different rear slant
angles  (Franck et al. 2009) 6
Figure 1-6: Topology of the wake flow for an Ahmed body for =12.5° (Franck et al.
2009) 6
Figure 1-7: Benard-Von Karman vortex street (Van Dyke 1982) 7
Figure 1-8: Kelvin-Helmholtz instability in shear flows (Van Dyke 1982) 8
Figure 2-1: Wind tunnel and 2D LDV system at ESTACA (flow direction from right to
left) 10
Figure 2-2: Sketch of experimental test section in the wind tunnel (Rodriguez 2016) 11
Figure 2-3: Homogeneous region in the test section (side view, flow from right to left) 11
Figure 2-4: Dimensions of the Ahmed body model (scale 1:5.3) in the experiments of
Rodriguez (2016). Lengths in mm 12
Figure 2-5: Ahmed body installed in the wind tunnel (=35°) in the study of
Rodriguez (2016) 12
Figure 3-1: Example of instantaneous raw horizontal velocity U in the wake of the
Ahmed body (x/H=0.463; y/H=0.463; z/H=0.452) 15
Figure 3-2: Instantaneous raw velocity U and the corresponding linearly interpolated
signal (x/H=0.463; y/H=0.463; z/H=0.452) 16
Figure 3-3: FFT of the interpolated signal for U at (x/H=0.463; y/H=0.463;
z/H=0.452) 17
Figure 3-4: PSD of the interpolated signal for U at (x/H=0.463; y/H=0.463;
z/H=0.452) 17
Figure 4-1: FFT of the interpolated horizontal velocity signal U at same elevation
(y/H=0.093), same transverse position (z/H=0.226) and various
longitudinal positions (x/H) in the wake of the Ahmed body 20
Figure 4-2: FFT of the interpolated horizontal velocity signal U at same longitudinal
position (x/H=1.574), same transverse position (z/H=0.226) and various
elevation (y/H) in the wake of the Ahmed body 21

- ii -
Figure 4-3: FFT of the interpolated horizontal velocity signal U at same longitudinal
position (x/H=0.093), same elevation (y/H=0.093) and various transverse
positions (z/H) in the wake of the Ahmed body 22
Figure 4-4: Triple decomposition applied to the root mean square of the horizontal
component of the velocity. Influence of x/H 26
Figure 5-1: Particle size distribution of oil droplets released from the fog generator 27
Figure 9-1: FFT of the interpolated horizontal velocity signal U at same elevation
(y/H=0.093), same transverse position (z/H=0.226) and various
longitudinal positions (x/H) in the wake of the Ahmed body 45
Figure 9-2: FFT of the interpolated horizontal velocity signal U at same longitudinal
position (x/H=0.833), same transverse position (z/H=0.226) and various
elevation (y/H) in the wake of the Ahmed body 46
Figure 9-3: FFT of the interpolated horizontal velocity signal U at same longitudinal
position (x/H=0.093), same transverse position (z/H=0.226) and various
elevation (y/H) in the wake of the Ahmed body 46
Figure 9-4: FFT of the interpolated horizontal velocity signal U at same longitudinal
position (x/H=1.574), same transverse position (z/H=0.226) and various
elevation (y/H) in the wake of the Ahmed body 47
Figure 9-5: FFT of the interpolated horizontal velocity signal U at same longitudinal
position (x/H=3.796), same transverse position (z/H=0.226) and various
elevation (y/H) in the wake of the Ahmed body 48
Figure 9-6: FFT of the interpolated horizontal velocity signal U at same longitudinal
position (x/H=0.093), same elevation (y/H=0.093) and various transverse
positions (z/H) in the wake of the Ahmed body 49

- iii -
List of Symbols

Cx Drag coefficient in x-direction


d Diameter of the struts [m]
dacq Duration of data acquisition [s]
f Frequency [Hz]
hs Height of the struts [m]
H Height of the Ahmed body [m]
Hs Height of the test section (wind tunnel) [m]
Ix Turbulence level (Ix=u’/U)
Lf Focal length of the lens [m]
Lm Length of the Ahmed body [m]
lm Width of the Ahmed body [m]
Ls Length of the test section (wind tunnel) [m]
ls Width of the test section (wind tunnel) [m]
n Constant
Re Reynolds number (Re=U×H/)
St Strouhal number (St=f×H/U)
U Horizontal component of the velocity [m/s]
U0 Upstream horizontal component of the velocity [m/s]
u Instantaneous velocity [m/s]
u’ Slow fluctuating component of the horizontal velocity [m/s]
u’’ Fast fluctuating component of the horizontal velocity [m/s]
V Vertical component of the velocity [m/s]
x Longitudinal coordinate [m]
X* Dimensionless distance in the flow direction (X*=x/H)
y Vertical coordinate [m]
Y* Dimensionless distance in the vertical direction (Y*=y/H)
z Transverse coordinate [m]
Z* Dimensionless distance in the transverse direction (Z*=z/H)
φ Rear slant angle
 Boundary layer thickness [m]
 Kinematic viscosity of air [m2/s]

- iv -
Acronyms

2D Two-Dimensional
AT Arrival Time
CO Carbon monoxide
ESTACA Ecole Supérieure des Techniques Aéronautiques et de Construction Automobile
FFT Fast Fourier Transform
LDA1 Laser Doppler Anemometry 1 (corresponding to the horizontal component of the
velocity)
LDA2 Laser Doppler Anemometry 2 (corresponding to the vertical component of the velocity)
LDV Laser Doppler Velocimetry
N/A Not available
NOx Nitrogen Oxides
OECD Organization for Economic Co-operation and Development
O3 Ozone
PIV Particle Imagery Velocimetry
PM Particle Matter
PM1 Particle Matter (diameter below 1 µm)
PM2.5 Particle Matter (diameter below 2.5 µm)
PM10 Particle Matter (diameter below 10 µm)
PNC Particle Number Concentration
PSD Power Spectral Density
rms Root mean square
SO2 Sulphur dioxide
TT Transit Time
UFP Ultra Fine Particle
WHO World Health Organization

-v-
1. Introduction

1.1 Context of the present study


In the past few years, the air quality has become a major issue in terms of public health and
environmental protection. In 2016, a French parliamentary report assessed its cost between 68
and 97 billion Euros per year, considering only the mortality and the morbidity linked with it
(Roumegas and Saddier 2016). Further costs, including those associated with the deterioration
of public buildings (for example) were not considered in this amount. Likewise, other
international studies have shown that across the world the pollution is responsible for the deaths
of many people every year as well as for the aggravation of allergies and many cardiac,
pulmonary and respiratory diseases (Erba et al. 2015; OECD 2016; European Environment
Agency 2016; Xia et al. 2016). The World Health Organization (WHO) estimated that throughout
the world 3.7 million people die every year caused by inhalation of ultrafine particles (WHO
2015). In a recent report, the cost of air pollution could reach up to 1% of the Gross Domestic
Product by 2060 (OECD 2016). The World Bank and the Institute for Health Metrics and
Evaluation (2016) wrote that “an estimated 5.5 million lives were lost in 2013 (one in ten
deaths) to diseases associated with outdoor and household air pollution, causing human
suffering and reducing economic development”. Altogether, these elements provide partial
explanation to the increasing interest for this issue across the world. Pulmonologists and heart
specialists agree that ultrafine particles (UFP) and nanoparticles are the most harmful. These
carbonaceous particles, also known as "soot particles" can penetrate deeply the respiratory
system and inside cells causing appearance and/or aggravation of respiratory and cardiovascular
diseases, cancer (e.g. lungs, bronchi). Figure 1-1 illustrates the pathways of exposure to
nanoparticles for a human body (Buzea et al. 2007).

Figure 1-1: Schematic of human body with pathways of exposure to nanoparticles


(Buzea et al. 2007)

WRL Research Report 262 FINAL May 2017 1


Pope et al. (2013) and Semmler-Behnke et al. (2014) have shown that UFPs can penetrate into
the blood system and cells increasing the risk of lung cancer and cardiovascular diseases
significantly. Working with rabbits, Valentino et al. (2016) have shown that the effects of
pollution related to diesel engines can be transmitted from generation to generation. They also
showed that nanoparticles are able to move across the placenta infecting the unborn child
(Valentino et al. 2016). UFP have undoubtedly major health and financial consequences by
increasing medical related absences, death and hospitalizations.

Among the major contributors to air pollution are the transport modes in the air and on roads.
Car engines and other vehicles emit several toxic gases (NOx, CO, SO2, O3) and solid particles
known as fine, ultrafine particles depending on their diameter. Particles with diameter below 10
µm, 2.5µm and 1µm are defined as PM10, PM2.5 and PM1, respectively. Particles with diameter
below 0.1µm and 0.05µm are referenced as ultrafine and nanoparticles, respectively (Figure 1-
2). All these pollutants are released in the atmosphere from the tailpipe of motor vehicles
causing significant air pollution (Buseck and Adachi 2008). Particles emitted from a Diesel engine
have a bimodal distribution (Figure 1-2). Although the heaviest particles are the largest ones,
the smallest (UFP) are the most abundant and of biggest environmental concern. These particles
are associated with the most harmful effects (dotted line in Figure 1-2). Special attention must
be paid to UFPs including their fundamental motion, the associated dispersion processes in the
wake of motor vehicles and their infiltration into the in-cabin microenvironment of the vehicles.

Figure 1-2: Bimodal distribution of particle diameters emitted from a Diesel engine
(Kittelson et al. 2004)

As an example of an industrialised country, the French road industry is accountable for the
emission of up to 54%, 16% and 19% for NOx, PM10 and PM2.5, respectively (Roumegas and
Saddier 2016). Likewise, light commercial vehicles emit approximately 30% of the PM2.5
whereas trucks using gasoil release roughly 37% of the NOx. At the continental scale, the
European Environment Agency mentions that the automotive industry contributes up to 23%,
30% and 12% for the emission of CO2, NOx and PM2.5 respectively (European Environmental
Agency 2016). Although cars, buses and trucks are responsible for a significant part of those
emissions, some significant improvements have been made in the last two decades in engine

WRL Research Report 262 FINAL May 2017 2


performances and reduction in vehicle mass using composite materials leading to a reduction of
fuel consumption and thus a decrease in pollutant emissions. This trend has also been fostered
by new and more restrictive regulations such as the EURO standard for vehicle emissions. As an
example, between EURO3 from 2000 and EURO6b from 2014, the emission thresholds of
released particles (in mass) and NOx have been reduced by factors of 11 (from 50 to 4.5
mg/km) and 6 (from 500 to 80 mg/km), respectively.

Further reductions of emission levels are still required, particularly regarding the finest particles
(PM10, PM2.5 and PM1). In recent years, two critical threshold levels (advisory and alert) for
both gaseous pollutants and particles have been defined. As an example, in France, for PM10s,
these are set at 50 (advisory) and 80 (alert) mg/m3 per hour per day. When higher levels are
recorded, recommendations are given to the public including limiting outdoor sport activities and
precautionary measures for pregnant women, young and old people. The use of vehicles on inner
city roads is limited and the use of public transportations is recommended instead (free of
charge). Reducing the emissions of the finest particles into the atmosphere is one of the key
issues which may be achieved through trapping of particles (including at the source) with
efficient filters (High Efficiency Particulate Air) preventing their spread into the atmosphere and
their infiltration in vehicle cabins. A better assessment of the rates at which human beings are
effectively subjected to these particles is also of major concern as well as the identification of
infiltration factors. To achieve these outcomes, it is important to improve the current knowledge
of the particle dynamics of the particles emitted from the tailpipe of motor vehicles.

Recent studies have proven that UFPs are also able to infiltrate into buildings including schools
through micro-cracks and windows seals. Consequently, people are exposed to UFPs not only
outside but also in indoor environments (Liu and Nazaroff 2003; Jeng et al. 2007). UFPs
contribute to the deterioration of both outdoor and indoor air quality increasing the interest of
the scientific community for improvement of air quality. New strategies target to reduce pollution
levels from tailpipes of vehicles, to improve our knowledge of the UFP dispersion processes in
the wake of vehicles or to upgrade filter efficiency. Other solutions may be linked with driving
recommendations helping people to reduce their exposure to UFPs (fan settings, AC on/off,
recirculation on/off, windows closing/opening…). An important decrease of exposure to UFPs can
also be achieved by limiting the infiltration of pollutants into the car cabin. A better
understanding of the non-stationary wake flow developing downstream of a vehicle is crucial
including the interactions between particles and the flow turbulence. In terms of application, it is
expected that this fundamental understanding of the fluid dynamic processes could be linked
with the formulation of new recommendations to automotive manufacturers regarding driving
practices. In addition, this would lead to an improvement of the numerical model efficiency
through a better definition of the boundary conditions and lead to realistic assessments of
particle concentrations in wake flows of vehicles. The description of the non-stationary flow
developing downstream of a car is the major goal of the present study. The research project at
UNSW Water Research Laboratory is part of a large research project undertaken at ESTACA
(France) focusing on air quality in transportation systems with special interest in ground
vehicles.

1.2 Basic concepts of interactions between particles and wake flow


When particles are emitted from the tailpipe of a vehicle, some of the particles will infiltrate the
surrounding buildings through micro cracks and will thus expose their occupiers. The remaining
particles will either scatter in the ambient atmosphere or penetrate into car cabins through air
inlets, windows or leakages (Lee, 2013). Gaseous pollutants such as CO and NOx and solid
carbonaceous particles disperse or accumulate into the ambient air contributing to global

WRL Research Report 262 FINAL May 2017 3


warming. When particles infiltrate into the cabin of a vehicle, this provides a micro-environment
where pollutants can quickly accumulate with negative impacts on health and safety of
passengers through inhalation. The ability of particles to infiltrate into the cabin depends upon a
range of different parameters linked with the driving practices, that is the ventilation settings or
the use of air conditioning for instance. Further parameters are linked to the speed of the
vehicle, the surface leakage, the surrounding traffic, the type of road including highways,
tunnels, bridges, buildings in urban environments as well as the distance between vehicles and
weather conditions. It is worth noting that particle dynamics is not only related to the initial
condition of the emission but also to the interaction between particles and large scale unsteady
turbulent motions in the wake of a vehicle, such as swirls and wake recirculation. A recent study
by Mehel and Murzyn (2015) has shown that high Particle Number Concentration (PNC) levels
can be found in the centreline of the near-wake or on both sides of the vehicle in the far wake.
Particles are typically caught in the recirculation region in the near-wake of the vehicle. The
near-wake region spreads over a dimensionless distance 0 < x/H < 2 downstream of the vehicle,
where x is the streamwise distance downstream of the vehicle and H is the height of the vehicle.
In this work, the upstream flow velocity is typically constant. Farther downstream (2 < x/H < 5),
longitudinal vortices generated from a vehicle’s lateral edges become predominant. These large
scale turbulent structures entrap particles on both sides of the vehicle leading to dissymmetrical
distribution of recorded PNC (Mehel and Murzyn 2015). Interestingly most of numerical models
consider Gaussian distribution of PNC rather than dissymmetrical distributions found during wind
tunnel experiments. Thus numerical results predicting PNC should be considered with caution
and improvements are possible based on wind tunnels investigations. Some recent studies
demonstrated interactions between particles and flow turbulence strengthening the need for a
deepening understanding of the fundamental flow processes (Carpentieri et al. 2012; Mehel and
Murzyn 2015). Further experimental research is needed to improve the understanding of the
time dependent and non-stationary particle distributions. The purpose of the present study is to
improve the understanding of the dynamics of the wake flow motions considering non-stationary
aspects.

Studies at the prototype scale are difficult due to the associated costs and the multitude of
relevant parameters including wind properties, yaw angle, roughness, boundary layer, vehicle
shape. Small scale experiments in wind tunnels are a suitable alternative to undertake
experiments in a well-controlled environment with the ability to modify different relevant
parameters. However scale effects may limit the extrapolation to the real-scale since a full
dynamic similitude cannot be achieved simultaneously. Herein the most important dimensionless
number is the Reynolds number. It is therefore important to systematically investigate potential
scale effects by repeating experiments at different scaling ratio. Considering the non-stationary
flow, a further relevant dimensionless number is the Strouhal number.

Understanding the aerodynamics around and downstream of a moving car is complex due to the
extreme variety of shapes and geometry of cars. Ahmed et al. (1984) designed a generic
simplified car geometry which retains most important features of real ground vehicles. Although
it does not take into account the flow through the engine compartment, the cabin and around
wheels and mirrors, an Ahmed body allows simulation of the three main topologies of wake flow
depending on the rear slant angle φ (Figure 1-3). Figure 1-3 illustrates a typical Ahmed body
highlighting the definition of the rear slant angle.

WRL Research Report 262 FINAL May 2017 4


Figure 1-3: Sketch of a typical car model defined as Ahmed body (Kahn and Umale 2014)

Considering the Ahmed body, the drag coefficient of a vehicle depends on the rear slant angle as
shown by Thacker (2010) (Figure 1-4). In Figure 1-4, Cx represents the drag coefficient in x
direction.

Figure 1-4: Drag coefficient of an Ahmed body as a function of the rear slant angle 
(Thacker 2010)

Depending on , Lahaye (2014) defined three different flow regimes:


 φ < 12.5°: The flow is considered as bidimensional in the symmetrical plan. Two
counter-rotating vortices with a toric shape can be found (Figure 1-5a and b, structures
A and B in Figure 1-6). On both sides of the Ahmed body, two small counter-rotating
vortices in the transversal plan develop and dissipate further downstream (Structure C in
Figure 1-6). In this case, the drag is mostly due to a pressure drop and the detachment
of the flow. The corresponding Ahmed body is usually called “squareback”.
 12.5° <  < 30°: The flow becomes highly three-dimensional. On the rear slant, a
detached bubble can be observed. Energetic longitudinal structures appear on the lateral

WRL Research Report 262 FINAL May 2017 5


surface of the slant (Figure 1-5c). The drag is mostly linked to pressure drops and to the
tridimensional nature of the flow. This configuration is called “fastback”.
  > 30°: The vortices created from the detachment of the rear slant dissipate in the
wake (Figure 1-5d). With increasing φ, the flow features from the initial configuration
are found again leading to a sudden decrease in drag. The drag reduction can also be
observed in Figure 1-4. The flow is comparable to the initial “squareback” situation
(Figure 1-5d).

Those three scenarios represent the fundamental behaviours of a wake flow which can be
observed for most car shapes. More details regarding these flow topologies can be found in
Hucho (1987), Wang et al. (2013), Lahaye (2014), Aljure et al. (2014) and Rodriguez (2016).

(a) Counter-rotating vortices with a toric shape (b) Counter-rotating vortices with a toric shape
for φ < 12.5° are affected by increasing angle φ < 12.5°

(c) Longitudinal structures appear on the (d) Vortices created from the detachment of
lateral surface of the slant 12.5° <  < 30° the rear slant dissipate in the wake  > 30°:

Figure 1-5: Flow topologies in the wake of an Ahmed body for different rear slant angles 
(Franck et al. 2009)

Figure 1-6: Topology of the wake flow for an Ahmed body for =12.5° (Franck et al. 2009)

WRL Research Report 262 FINAL May 2017 6


Ahmed bodies have been used in many studies because they are easy to build and of low cost
(Krajnovic and Davidson 2004; Gosse 2005; Leclerc 2008; Thacker 2010; Grandemange 2013;
Wang et al. 2013; Lahaye 2014; Aljure et al. 2014). Most of these studies provided experimental
data. Studies have focused in particular on the characterization of the wake flow (e.g. Krajnovic
and Davidson 2004; Grandemange 2013; Wang et al. 2013; Lahaye 2014), on the drag force
(Leclerc 2008), on the dispersion of the temperature (passive scalar) in the wake flow (Gosse
2005) and on the effects of the aspect ratio (i.e. the ratio between width of the Ahmed body and
the length of the rear slant) on the wake (Venning et al. 2015). Experiments in water channels
and in wind tunnels as well as numerical investigations were undertaken for a wide range of
Reynolds numbers (104 < Re < 106). Only a few studies focused upon the dispersion of the
particles (Mehel and Murzyn 2015), the gaseous pollutants in the wake of vehicles (Gosse 2005;
Carpentieri et al. 2012) and on the non-stationary flow aspects (Vino et al. 2005; Wang et al.
2013; Tunay et al. 2014). Herein the present study aims to provide further insights into the non-
stationary flow aspects which are needed for a better understanding of particle dispersion
processes in the wake of an Ahmed body. To date, very few previous studies characterized the
frequencies and the related Strouhal number in the wake of a car which are associated with the
turbulence structures in the near-wake (Thacker 2010; Lienhart and Pego 2012; Tunay et al.
2014). Using a 2D LDV system, the velocity fields in the wake flow of an Ahmed body were
recorded at ESTACA, France. The collected data were post-processed in the present study
including the application of a Fast Fourier Transform to identify the characteristic frequencies.
These frequencies are used for the triple decomposition of the raw velocity signals into its mean,
fast and slow fluctuating velocity components. The triple decomposition yielded the contribution
of each frequency range to the global dynamics of the flow. It is worthwhile to note that the
triple decomposition technique has already been applied successfully to multiphase flows (Felder
2013; Felder and Chanson 2014; Wang et al. 2014). The present study is the first application of
the triple decomposition technique to a wake flow.

1.3 Dimensionless numbers


For wake flows, turbulence can be observed in two principal features, i.e. the Benard-Von
Karman vortex street and the Kelvin-Helmholtz instabilities (Eulalie 2014). When the flow ceases
to be stationary and the velocity becomes time- and space-dependent, vortex shedding occurs
periodically downstream of an Ahmed body. Vortex shedding forms a repeating pattern of
swirling vortices (Figure 1-7) called Benard-Von Karman vortex street. This is usually
encountered in flows downstream of a cylinder or other types of bluff bodies. It is caused by the
unsteady separation of fluid flow around bluff bodies. Even if the phenomenon largely disappears
in fully rough turbulent flows, its characteristic frequency can still be found in spectral analysis
(Eulalie 2014).

Figure 1-7: Benard-Von Karman vortex street (Van Dyke 1982)

WRL Research Report 262 FINAL May 2017 7


A second mechanism is linked to Kelvin-Helmholtz instabilities which occur when velocity
differences are present across the interface of two fluids, resulting in the creation of a wave
pattern (Figure 1-8). This instability occurs at higher frequencies when the Strouhal number St
exceeds 0.6 (Eulalie, 2014):
St=f × H/U (1)

where f is a characteristic frequency, U is a velocity, and H a characteristic length scale. When a


periodical vortex shedding occurs downstream of a body, the Strouhal number links the vortex
shedding frequency with flow velocity and geometry of the body.

Figure 1-8: Kelvin-Helmholtz instability in shear flows (Van Dyke 1982)

The turbulence behind the Ahmed body can be also described by the Reynolds number Re. The
Reynolds number characterises the nature of the flow including laminar, transient or turbulent
flows. It represents the ratio between inertia and viscous forces and it is defined as:

Re = U × H/ (2)

where  is the kinematic viscosity of the fluid. Both Strouhal and Reynolds numbers can be
based on different parameters depending on the flow situation and the relevant characteristic
time and length scales. For example, for an Ahmed body, the characteristic length scale may be
the height of the vehicle H. In the present study, the characteristic velocity U=U0 was the mean
velocity of the flow, measured upstream of the model in the wind tunnel and H was the height of
the car model. The Strouhal number was assessed through the frequency analysis. Identifying
the characteristic frequencies of the flow behind an Ahmed body is an important part of the triple
decomposition technique. Furthermore, a discussion about scale effects is required as the
experiments were conducted in a relatively small-scale wind tunnel compared to previous
experimental studies.

In wind tunnels, the exhaust of nanoparticle can be achieved through a tube located under the
vehicle. To limit scale effects as much as possible, it is important to accurately set the ejection
rate (and thus the velocity) of the carbon particles in the test section in such a manner that the
study reproduces faithfully the engine emissions at the prototype scale. This similitude must be
also taken into account in addition to the Reynolds similitude. In the present study only the
Reynolds similitude is considered as particles are not injected.

Prior to the investigation of the interaction between particles and air flow turbulence it is
necessary to understand the air flow in the wake of the Ahmed model. This is the main goal of
the present study. Mean and turbulent fields must be assessed to characterise the structure of
the wake flow and its time scales. In a next step, the concentration fields (PNC) will be
investigated. In a final stage the results of both turbulence and velocity fields and PNC will be

WRL Research Report 262 FINAL May 2017 8


correlated allowing a better identification of the interaction between particles and flow. The
present study is part of the first step mentioned above.

1.4 Principle of the triple decomposition technique


The triple decomposition technique aims at separating a raw instantaneous velocity signal into
three components, i.e. the mean flow velocity as well as the slow and fast velocity fluctuations of
the flow. The decomposition is undertaken using low pass, band pass and high pass filtering of
the raw signal. The decomposed parts correspond to the slow, mean and fast fluctuating
velocities, respectively. The fast fluctuating velocity component corresponds to the “true”
turbulent motion of the flow (Hussain and Reynolds, 1972; Lyn and Rodi, 1994; Fox et al. 2005;
Brown and Chanson, 2013). Such a decomposition can be used to assess the contribution of the
decomposed velocity components to the overall turbulence motion improving the accurate
understanding of the flow dynamics.

The triple decomposition technique was originally developed for mono-phase flow for which the
instantaneous velocity was given (Hussain and Reynolds, 1972; Lyn and Rodi, 1994). The triple
decomposition approach has been recently expanded to non-stationary air-water flows on pooled
stepped spillways (Felder 2013; Felder and Chanson 2014) and hydraulic jumps (Felder and
Chanson 2012; Wang et al. 2014). In air-water flows, it is not possible to measure an
instantaneous velocity and the triple decomposition technique was applied to the instantaneous
void fraction signal before the decomposition was applied. Further details can be found in Felder
and Chanson (2014).

In the present study, the application of the triple decomposition technique to the wake flow
downstream of a car model was tested to ascertain the contribution of the fast and slow velocity
fluctuations to the structure of the turbulence. After the introduction, the experimental set-up
and the measurement technique are introduced. Note that no experiments were conducted in
the present study, and that the raw velocity data were measured by Rodriguez (2016). Section 3
details the signal processing with special emphasis on data resampling methods and spectral
analysis. The results dealing with the spectral analysis, frequency identification and assessment
of the triple decomposition technique are presented and discussed in section 4. A discussion
about the accuracy of the results and possible scale effects is presented in section 5 before the
conclusion of the report. The appendix provides supplementary data.

WRL Research Report 262 FINAL May 2017 9


2. Experimental set up, instrumentation and flow conditions

2.1 Wind tunnel


All measurements were conducted in the subsonic wind tunnel at ESTACA Campus Ouest in Laval
(France). The experimental test section had a length of LS = 1 m, a width of lS = 0.3 m and a
height of HS = 0.3 m (Figure 2-1). The sidewalls of the experimental facility were made of
transparent Plexiglas allowing the use of optical measurement devices such as laser Doppler
velocimetry (LDV) and detailed flow visualizations (Figure 2-1). The bottom of the channel was
smooth and horizontal with 40 equally-spaced holes (Figure 2-2). These holes were used to
attach models such as cars, wings and other obstacles in the wind tunnel. When required, they
were plugged with home-made plastic caps to avoid any disturbance of the flow including in the
measurements by Rodriguez (2016). The roof of the test section included a removable gate
which facilitated the installation of experimental models in the test section (Figure 2-2). The air
flow was supplied through suction from the downstream end of the wind tunnel generated by an
engine with power of 3 kW. The velocity of air could be accurately controlled providing average
flow velocities between 0 and 40 m/s in the test section. The mean flow velocity was measured
with a Pitot tube in the inlet section. At the upstream end of the tunnel, a convergent ensured
the homogeneity of the flow which was overall one-dimensional, i.e. mean horizontal velocity U
was much higher than the mean vertical velocity V. A detailed and accurate calibration of the
wind tunnel without installed model was undertaken by Rodriguez (2016). Rodriguez (2016)
showed that the turbulence level in flow direction Ix (Ix=u’/U) outside of the turbulent boundary
layers was below 1%. The maximum boundary layer thickness was assessed as = 12 mm
(/H~0.22) at the downstream end of the test section. As a consequence, the flow was partially-
developed. This detailed survey identified a large region where the flow was homogeneous and
not disturbed by the presence of the walls (green area in Figure 2-3). The Ahmed body was set
up in this homogeneous region, so that the upstream flow was well-known and without wall
effects (that is out of the boundary layer and without any blockage effect). A smoke generator
was used for flow seeding for LDV measurements and flow visualization. It was placed on the
side of the wind tunnel to avoid any additional disturbance.

Figure 2-1: Wind tunnel and 2D LDV system at ESTACA (flow direction from right to left)

WRL Research Report 262 FINAL May 2017 10


Figure 2-2: Sketch of experimental test section in the wind tunnel (Rodriguez 2016)

Figure 2-3: Homogeneous region in the test section (side view, flow from right to left)

In the experiments used in this report, the upstream velocity of the flow was fixed at U0=U=14.5
m/s. The corresponding Reynolds number defined in terms of the height of the vehicle model
was Re~5.104 for all experiments. Figure 2-3 illustrates the definition of the flow direction x and
the vertical direction y in dimensionless terms as X* and Y* corresponding to dimensionless
distances X*=x/H and Y*=y/H respectively. The origin was located at the centreline of the test
section, on the rear bottom of the car model. It is worthwhile to note that the model was fixed
on four 15 mm high struts with circular section. The struts were installed to position the lower
part of the car outside of the boundary layer and the origin of axis was 15mm above the bottom
of the wind tunnel.

2.2 Ahmed body


For the experiments used in the present study, a car model with a rear slant angle of =35° was
used. It was made of aluminium with smooth surfaces to avoid any flow disturbances. A sketch
of the car is shown in Figure 1-3 at full scale as designed by Ahmed et al. (1984). Considering
the size of the experimental facility in the present study, the car was scaled in a ratio of 1:5.3 to
the original Ahmed model size. The model in the present study had a length Lm=0.196 m, width
lm=0.073 m and height H=0.054 m. The model was mounted on 4 cylindrical struts having a

WRL Research Report 262 FINAL May 2017 11


diameter d = 0.006 m and a height hs=0.015 m (Figure 2-4). The dimensionless ground
clearance was hs/H=0.278. The scale of the model had a scaling ratio of ~1:19 compared to a
prototype scale car. The front size area of the model car was small enough to limit the blockage
ratio to below 5% to avoid unwanted wall effects (Wang et al. 2013). For all measurements, the
car model was fixed at the entrance of the test section, with the upstream edge of the car
installed on the first row of holes in the channel bottom. Figure 2-5 shows the car model and the
test section in the present study from a different viewing angle.

Figure 2-4: Dimensions of the Ahmed body model (scale 1:5.3) in the experiments of Rodriguez
(2016). Lengths in mm

Figure 2-5: Ahmed body installed in the wind tunnel (=35°) in the study of Rodriguez (2016)

2.3 Velocity measurements using LDV


Velocity measurements were conducted using a 2D LDV system manufactured by DANTEC. LDV
is an optical and non-intrusive measuring technique that does not disturb the flow. The
measurement principle of the LDV system can be summarized as follows:
 An optical probe is mounted on a displacement system driven by a computer. It contains
an optical device from which two (2) pairs of laser beams are emitted with two (2)
different wavelengths (Table 2-1). The first pair of laser beams measures the horizontal
component of the instantaneous velocity vector while the second pair records the vertical
instantaneous velocity component.

WRL Research Report 262 FINAL May 2017 12


 The four (4) laser beams pass through a convergent lens which has a focal length of
Lf=500 mm. The point where all laser beams cross is called the measurement volume. It
is located in the flow and can be shifted for profiling of the flow velocities.
 The dimensions of the measurement volume are accurately defined by the optical
parameters of the system. Within this very small volume, a network of interference
fringes is created with equally-spaced dark and bright fringes. The size of the fringes is
also accurately defined (Table 2-1).
 When a seeding particle crosses the measurement volume, a light signal is emitted with
a frequency proportional to the velocity of the particle, according to Mie theory. The
signal is collected, analysed and converted into a velocity by a processor.
 The collected data are then processed through a software (BSAFlow) supplied by
DANTEC.
 Note that a Bragg cell is added in the optical device which allows the determination of
the direction of the velocity.

During the experiments, the “burst mode“ was chosen and both horizontal and vertical
components of the velocity vector were simultaneously recorded according to a coincidence
window. To improve the data rate, a seeding generator was used. Its position was defined after
a sensitivity analysis which ensured that it did not disturb the incoming flow. The average
particle size of the seeding droplets was around 1 µm. Several thousand instantaneous samples
were collected at each position to ensure data convergence. It is worthwhile to note that LDV is
one of the most accurate non-intrusive technique to measure velocity fields. It has been
intensively used in the past decades either in air or water flows including for investigations of the
wake flow downstream of an Ahmed body (Lienhart and Pego 2012; Wang et al. 2013). It has a
high spatial resolution (very small measurement volume) and fast response in time (high data
rate ranging from tens to thousands kiloHertz depending on the seeding). Furthermore, it does
not need any calibration. It is an easy-to-use system as the displacement table and data
acquisition are fully computer-controlled. The experimental set-up of Rodriguez (2016) is shown
in Figure 2.1.

Table 2-1: Optical properties of the LDV system

Measurement volume Number of Distance between


LDV Beam Wave length (nm) (mm3) fringes 2 fringes (m)

Beam 1 660 0.1684*0.1681*2.806 30 5.448


Beam 2 785 0.2003*0.1999*3.338 30 6.396

To ensure good quality of data, a preliminary calibration was conducted to define the minimum
duration (dacq) for data acquisition. Finally, dacq = 90 s. The repeatability of measurements was
also checked and further information can be found in Rodriguez (2016).

2.4 Collected data


For a good description of the wake flow, velocities were collected in four (4) different (x,y)
planes for z/H=0; 0.226; 0.452; 0.678 where z was the transversal direction. For each location,
195 (13×15) different positions (x,y) were investigated. The domain ranged from x/H=0.093 to
x/H=4.722 (15 positions) and y/H=0.093 to y/H=1.296 (13 positions). Table 2-2 summarizes
these positions (Rodriguez 2016). For each position, raw data were recorded comprising the
arrival time of the measured particles AT, the transit time of the recorded particle in the
measurement volume TT, the horizontal component of the velocity LDA1, the vertical component

WRL Research Report 262 FINAL May 2017 13


of the velocity LDA2, the root mean square of LDA1 (rms1) and the root mean square of LDA2
(rms2). These data were then post-processed in the present study.

Table 2-2: Data acquisition positions (Rodriguez 2016)

x/H [-] y/H [-] z/H [-]


0.093 0.093 0
0.278 0.186 0.226
0.463 0.279 0.452
0.648 0.372 0.678
0.833 0.465
1.018 0.558
1.203 0.651
1.388 0.744
1.573 0.837
1.758 0.930
1.943 1.023
2.869 1.116
3.795 1.296
4.721
5.647

WRL Research Report 262 FINAL May 2017 14


3. Signal processing

3.1 From the raw data to an interpolated signal


The wind tunnel generated a flow with an incoming velocity U0=14.5 m/s corresponding to a
Reynolds number Re~5.104. Data were acquired for 90 s at each measurement position following
a sensitivity analysis made by Rodriguez (2016). As LDV acquisition depended on a random
process where seeded particles were measured in a fixed volume measurement, the acquisition
of velocity data was also random. This means that the particle arrival times were not equally-
spaced (Figure 3-1). Thus, the number of acquired samples within these 90 s measurement
interval differed from file to file. As a consequence, the raw signal had to be resampled to satisfy
an important criterion for the Fast Fourier Transform, i.e. 2n samples. Figure 3-1 presents an
example of a typical raw velocity signal over a period of 90 s at x/H=0.463; y/H=0.463;
z/H=0.452. The figure highlights the randomness of the data recording.

10
Raw Signal

5
Velocity (m/s)

-5

-10

-15
0 10 20 30 40 50 60 70 80 90
Time (s)

Figure 3-1: Example of instantaneous raw horizontal velocity U in the wake of the Ahmed body
(x/H=0.463; y/H=0.463; z/H=0.452)

A resampling of the raw signal was done with a linear interpolation of the raw data in Matlab.
The preliminary step was to define a number of intervals that matched the duration of the
acquisition. This number was set as the next exponent of two with respect to the number of
acquired samples. The interpolated data were compared to the raw data to ensure an accurate
superposition. This is shown in Figure 3.2 where raw and resampled data are superimposed for
the example mentioned above (x/H=0.463; y/H=0.463; z/H=0.452). Strong agreement between
raw and resampled signals was found. The resampled signal provided a constant time interval
between two points enabling a Fast Fourier Transform of the resampled signal.

WRL Research Report 262 FINAL May 2017 15


10
Raw Signal
Interpolated Signal

Velocity (m/s)
0

-5

-10

-15
0 10 20 30 40 50 60 70 80 90
Time (s)

Figure 3-2: Instantaneous raw velocity U and the corresponding linearly interpolated signal
(x/H=0.463; y/H=0.463; z/H=0.452)

3.2 Fast Fourier transform (FFT) and Power Spectral Density (PSD) of the
velocity
The interpolated signal (blue line in Figure 3-2) was processed with a Fast Fourier Transform
(FFT). A typical FFT result is presented in Figure 3-3 and the matching power spectral density
(PSD) in Figure 3-4. The "pwelch" function in Matlab was used with an overlap of 50% to satisfy
the Nyquist criterion. The same method was used by Tunay et al. (2014). It is worthwhile to
note that the experiments by Tunay et al. (2014) were conducted at similar geometric scale and
similar Reynolds number allowing a direct comparisons of results. The main difference between
the studies was the fluid (water versus air).

For the triple decomposition, two characteristic frequencies are required. Ideally, those are
expected to be clearly identifiable from the FFT. In the present study, for some measurement
positions, the characteristic peaks were hard to determine. This was mainly due to a data rate
which was sometimes too low. Although the seeding was conducted thoroughly, some regions of
the flow were still poor in seeding particles. To overcome this issue, the Savitzky-Golay filter was
used to smooth the signal and outline trends (Figures 3-3 and 3-4). This method preserved the
pertinent high frequency component of the signal allowing a comparison with previous studies
(Lienhart and Pego 2012; Tunay et al. 2014). In Figure 3-3, some typical peaks could be
identified around 3 Hz and 60 Hz, respectively. This information was important for the triple
decomposition technique.

All data were processed with a self-developed Matlab code. For more information, please refer to
the appendices.

WRL Research Report 262 FINAL May 2017 16


0
10
FFT

-1
10

FFT

-2
10

-3
10 0 1 2
10 10 10
Frequency (Hz)

Figure 3-3: FFT of the interpolated signal for U at (x/H=0.463; y/H=0.463; z/H=0.452)

3
10
PSD
2
10

1
10

0
10
PSD

-1
10

-2
10

-3
10

-4
10

-5
10 0 1 2
10 10 10
Frequency (Hz)

Figure 3-4: PSD of the interpolated signal for U at (x/H=0.463; y/H=0.463; z/H=0.452)

WRL Research Report 262 FINAL May 2017 17


4. Results

Data collected by Rodriguez (2016) were post-processed in the present study as outlined in
section 3 including the Fast Fourier Transform of the interpolated velocity signals. The data were
from various locations in the near-wake of the Ahmed body model with rear slant angle =35°.
The data analysis focused upon the horizontal velocity component U. For all measurement
locations, the spectral analysis provided the characteristic frequencies enabling the triple
decomposition technique of the raw velocity signal. In this section the results are presented
including the FFT of the raw velocity signals, as well as the turbulence properties of the
decomposed signals following the triple decomposition. The results of the FFT analysis are
presented for three scenarios: a) various dimensionless positions x/H downstream of the car for
constant y/H and z/H; b) various dimensionless distances above the wind tunnel bottom y/H for
constant x/H and z/H; c) various dimensionless transverse distances z/H for constant x/H and
y/H. All results are presented in the appendix. The FFT allowed the detection of characteristic
frequencies of the wake flow which were compared with the results of other experimental studies
of Ahmed bodies with various rear slant angles 25° ≤  ≤ 35° (Lienhart and Pego 2012; Tunay
et al. 2014). The section concludes with the presentation of the results of the triple
decomposition analysis.

4.1 Spectral analysis and characteristic frequencies


4.1.1 Spectral analysis of velocities in flow direction

Figure 4-1 shows the results of the FFT analysis for different dimensionless distances 0.093 <
x/H < 3.796 for constant vertical y/H and transverse positions z/H. In all figures, the vertical
elevation above the floor of the wind tunnel was y/H = 0.093 because in the proximity of the
floor the turbulent activity was maximum.

Figures 4-1a and b show similar frequencies in the area just downstream of the Ahmed body. No
characteristic frequency peaks were found despite some small disturbance around 60 Hz
corresponding to a Strouhal number St≈0.223 (Figure 4-1b). Some small peaks appeared also
around 3 Hz (Figure 4-1a) and 7 Hz (Figure 4-1b) corresponding to Strouhal numbers of 0.011
and 0.026, respectively. At the longitudinal position x/H=0.648 (Figure 4-1b), the characteristic
Strouhal numbers agreed well with the results of Tunay et al. (2014) who mentioned Strouhal
numbers between 0.08 (major peak in the FFT at 0.23 Hz) and 0.18 for x/H~0.8.

Between 0.833 ≤ x/H ≤ 1.759, frequency peaks could be observed around 60 Hz corresponding
to a Strouhal number of 0.223 (Figures 4-1c to 4-1e). This frequency appeared to be linked to a
Strouhal number that is often related to the Benard-Von Karman vortex street (Thacker, 2010;
Eulalie, 2014). As mentioned by Thacker (2010) and Lienhart and Pego (2012) the frequency
corresponded to a periodic vortex shedding. In the flow region closest to the rear slant and in
the wake region close to the ground, such high frequencies would be expected (Lienhart and
Pego 2012). Interestingly, Tunay et al. (2014) found the highest level of turbulent kinetic energy
in a similar region for 0.7 < x/H < 1 and 0 <y/H < 0.2 for =35°. Wang et al. (2013) observed
similar results in experiments with hotwire probes identifying the presence of quasi-periodical
structures. Wang et al. (2013) linked these large scale turbulent structures to the Benard-Von
Karman vortex street behind a 2D bluff body. The same conclusion was found by Vino et al.
(2005) based upon measurements of surface pressures behind a bluff body. In the present study
further frequency peaks were also observed between 1.5 and 6 Hz (Figure 4-1c), between 1.5
and 3 Hz (Figure 4-1d) and between 2 and 8 Hz (Figure 4-1e) respectively. The corresponding
Strouhal numbers were in a range of 0.006 ≤ St ≤ 0.03. For comparison, the lowest Strouhal

WRL Research Report 262 FINAL May 2017 18


number mentioned by Tunay et al. (2014) was St = 0.08 for a rear slant angle of =35°. Far
downstream of the Ahmed body, frequency peaks were reduced (Figure 4-1f) and for x/H=2.870
no frequency peaks were observed (Figure 4-1g). The results showed consistently low frequency
activities between 1 and 8 Hz corresponding to 0.004 ≤ St ≤ 0.03 for all dimensionless distances
x/H whereas the high frequency events were mostly found in a limited area downstream of the
rear part of the Ahmed body where the highest turbulent levels occur (St≈0.223). Additional
figures are presented in Appendix 9.2.

0 0
10 10
FFT FFT

-1 -1
10 10
FFT Amplitude

FFT Amplitude
-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(a) x/H=0.093 (b) x/H=0.648


0 0
10 10
FFT FFT

-1 -1
10 10
FFT Amplitude

FFT Amplitude

-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(c) x/H=0.833 (d) x/H=1.019


0 0
10 10
FFT FFT

-1 -1
10 10
FFT Amplitude

FFT Amplitude

-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(e) x/H=1.759 (f) x/H=1.944

WRL Research Report 262 FINAL May 2017 19


0
10
FFT

-1
10
FFT Amplitude

-2
10

-3
10 0 1 2
10 10 10
Frequency (Hz)

(g) x/H=2.870

Figure 4-1: FFT of the interpolated horizontal velocity signal U at same elevation (y/H=0.093),
same transverse position (z/H=0.226) and various longitudinal positions (x/H) in the wake of
the Ahmed body

4.1.2 Spectral analysis of velocities in upwards direction

Figure 4.2 presents the results of the FFT for different dimensionless distances y/H above the
bed of the wind tunnel for constant longitudinal x/H and transverse positions z/H respectively.
The longitudinal position was located within the area where the strongest turbulence activities
were observed x/H=1.574 (Rodriguez 2016). These results showed that the highest frequency
peaks of about 60 Hz were closest to the bottom where the turbulent kinetic energy was highest
(Figure 4-2a). As observed in Figure 4-1, no high frequency activity was visible for x/H < 0.648.
However, for x/H > 0.833, frequency peaks were observed close to the bed. This result is in
agreement with the findings of Lienhart and Pego (2012). Tunay et al. (2014) observed a
frequency peak in the upper part of the flow where a shear layer developed at the top of the
Ahmed body in the region closest to the car model for x/H~0.5 and y/H~0.8. In this region,
Tunay et al. (2014) found St~0.22. In the present study, Figure 4-1c shows a peak at around 40
Hz corresponding to a Strouhal number St≈0.15. The characteristic frequency may be related to
the development of the upper shear layer even if x/H was larger compared to Tunay et al.
(2014). Overall the frequency analysis revealed a good agreement with the results of Tunay et
al. (2014). Further corresponding results are presented in Appendix 9.2.

WRL Research Report 262 FINAL May 2017 20


0 0
10 10
FFT FFT

-1 -1
10 10
FFT Amplitude

FFT Amplitude
-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(a) y/H=0.093 (b) y/H=0.185


0
10
FFT

-1
10
FFT Amplitude

-2
10

-3
10

-4
10 0 1 2
10 10 10
Frequency (Hz)

(c) y/H=0.833

Figure 4-2: FFT of the interpolated horizontal velocity signal U at same longitudinal position
(x/H=1.574), same transverse position (z/H=0.226) and various elevation (y/H) in the wake of
the Ahmed body

4.1.3 Spectral analysis of velocities in transverse direction

Figure 4-3 shows the results of the FFT analysis for different dimensionless transverse positions
z/H while the longitudinal and transverse positions were constant at x/H=0.093 and y/H=0.093
respectively. These locations corresponded to the lowest position in the lower shear layer with
the largest turbulence levels. For all transverse positions, Figure 4-3 shows characteristic peaks
around 2 to 4 Hz. However no characteristic frequency peaks were identified in the higher
frequency range and for all transverse positions z/H. This finding was in agreement with
previous observations due to the close proximity downstream of the Ahmed body (Figure 4-1).
The behaviour tends to confirm a symmetrical behaviour in transverse direction. Further
investigations are required for larger dimensionless distances x/H to identify possible
dissymmetry effects.

WRL Research Report 262 FINAL May 2017 21


-1 -1
10 10
FFT FFT
FFT Amplitude

FFT Amplitude
-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(a) z/H=0 (b) z/H=0.226


-1 -1
10 10
FFT FFT
FFT Amplitude

FFT Amplitude

-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(c) z/H=0.452 (d) z/H=0.678

Figure 4-3: FFT of the interpolated horizontal velocity signal U at same longitudinal position
(x/H=0.093), same elevation (y/H=0.093) and various transverse positions (z/H) in the wake of
the Ahmed body

4.2 Summary of characteristic frequencies and Strouhal numbers in the


wake flow
Table 4-1 summarizes the characteristic frequencies for 22 locations scrutinised in the present
study. We were particularly interested in the close wake of the car model where the flow is
highly turbulent and fluctuating. Furthermore, even if 195 measuring points were investigated,
some of them were associated with non-homogeneous seeding conditions meaning that bursts of
particles were sometimes required to ensure a suitable data rate. Nevertheless, it was believed
that these non-homogeneous seeding conditions may affect the results by overweighting some
samples. This specific topic is the core of an in-depth investigation: a new and innovative data
analysis method, currently being developed, aims at providing reliable and repeatable results
whatever the seeding conditions are. Herein, we focused our investigations in measuring points
for which the seeding was homogeneous. Three different characteristic frequency peaks are
listed for each measurement location including a low range peak, a high frequency peak and an
intermediate peak. All frequencies were in a range between 1 Hz and 62 Hz. Typically the lowest
frequency peaks were observed for frequencies between 1 and 3 Hz independent of the

WRL Research Report 262 FINAL May 2017 22


measurement position (Table 4-1). The intermediate peaks were more diverse ranging between
2.5 and 11 Hz; no clear trend was recognised in terms of measurement locations. The upper
frequency peaks were in the range of 41 to 62 Hz. While there was no clear trend in terms of
measurement locations, for locations just downstream of the vehicle (i.e. 0.093 ≤ x/H ≤ 0.463)
and towards the far end of the measurement range (2.87 < x/H), no characteristic maximum
frequencies were found. This appeared to be linked with lower turbulence activities just
downstream of the Ahmed body and in the area far downstream. The limited frequency range in
the present study did not allow the identification of higher frequency peaks and experiments
with higher sampling rate are required. For some data points, the lower and the intermediate
peaks were very close and therefore only the lowest of the frequency peaks was considered as
the lower characteristic frequency in the triple decomposition technique.

The data in the present study showed a maximum Strouhal number St=0.231 corresponding to a
frequency of 62 Hz. This Strouhal number is similar to the results of Tunay et al. (2014) and
Lienhart and Pego (2012) although the latter results were for a smaller rear slant angle (=25°)
and for a Reynolds number one order of magnitude larger. A direct comparison of the present
data with Lienhart and Pego (2012) may be affected by scale effects. Tunay et al. (2014) found
Strouhal numbers between 0.08 and 0.22 for a rear slant angle of =35° within 0.5<x/H<0.8
and 0<y/H<0.8 and for Re~1.48.104, i.e. a Reynolds number of similar order of magnitude to
the present study. Tunay et al. (2014) were not able to measure larger frequencies because they
used a PIV system working at 15 Hz. The present Strouhal numbers associated with the lowest
frequency peaks 10-2<St<10-1 were about one order of magnitude smaller than those suggested
by Tunay et al. (2014) for comparable positions. Scale effects may explain these differences.
Lienhart and Pego (2012) found lowest characteristic frequency peaks at about 5 Hz which was
relatively close to the present results considering the different scaling. According to Lienhart
(2016), frequencies below 9 Hz would be expected for the present flume configuration even
though an accurate assessment of the lowest frequency peaks was sometimes difficult.

WRL Research Report 262 FINAL May 2017 23


Table 4-1: Summary of the characteristic frequencies and Strouhal numbers in the present study

Positions Lower Upper


Strouhal Intermediate Strouhal Strouhal
peak peak
x/H y/H z/H (lower) peak (Hz) (intermediate) (upper)
(Hz) (Hz)
0.093 0.093 0 2.0 0.007 N/A N/A N/A N/A
0.093 0.093 0.226 1.0 0.004 3.5 0.013 N/A N/A
0.093 0.093 0.452 1.0 0.004 2.5-5.5 0.093-0.0205 N/A N/A
0.093 0.093 0.678 2.0 0.007 N/A N/A N/A N/A
0.093 0.185 0.226 N/A N/A N/A N/A N/A N/A
0.093 0.370 0.226 2.0 0.007 3.0-5.0 0.011-0.019 N/A N/A
0.463 0.093 0.226 1.0 0.004 N/A N/A N/A N/A
0.648 0.093 0.226 3.0 0.011 7.0 0.026 57.0 0.212
0.833 0.093 0.226 1.5 0.006 5.5 0.021 59.0 0.220
0.833 0.185 0.226 0.5 0.002 11.0 0.042 60.0 0.223
1.019 0.093 0.226 1.0 0.004 2.5 0.093 57.0 0.212
1.204 0.093 0.226 3.0 0.011 N/A N/A 57.5 0.214
1.389 0.093 0.226 1.0 0.004 N/A N/A 62.0 0.231
1.574 0.093 0.226 2.0 0.007 2.5 0.093 60.0 0.223
1.574 0.185 0.226 1.0 0.004 2.5-4.0 0.093-0.015 59.5 0.222
1.574 0.833 0.226 1.0 0.004 1.5 0.006 41.5 0.155
0.007-
1.759 0.093 0.226 2.0-3.0 5.0-8.0 0.019-0.030 58.5 0.218
0.011
1.944 0.093 0.226 1.0 0.004 3.0-4.0 0.011-0.015 61.0 0.227
2.870 0.093 0.226 3.0 0.011 N/A N/A N/A N/A
0.004-
3.796 0.093 0.226 1.0-3.0 N/A N/A N/A N/A
0.011
3.796 0.278 0.226 2.0 0.007 4.0-6.0 0.015-0.022 N/A N/A
3.796 0.463 0.226 3.0 0.011 6.0 0.022 53.5 0.199

4.3 Application of the triple decomposition technique downstream of an


Ahmed body
4.3.1 Introduction

The triple decomposition technique aims to separate the instantaneous velocity signal U(t) into
three (3) components comprising the mean component U, a slow fluctuating component u' (t)
and a fast fluctuating velocity component u"(t). For the horizontal velocity, this can be expressed
as:

U(t) = U + u'(t) + u"(t) (3)

In this section, the triple decomposition technique is applied to a selected range of data in the
wake flow of the Ahmed body. The triple decomposition technique was based upon upper and
lower cut-off frequencies determined in the previous section (Table 4-1). The upper and lower
cut-off frequencies were used to decompose the raw instantaneous velocity signal into three (3)
components. In the present study, the triple decomposition technique was applied in terms of
the root mean square of the horizontal component of the velocity Urms. The root mean square of
the velocity represented a characteristic measure of the turbulence fluctuations of the flow, i.e. a
measure of the turbulence intensity. The technique was only applied to selected data points

WRL Research Report 262 FINAL May 2017 24


(Table 4-2). The selected data represented the best raw data signals and for several longitudinal
positions downstream of the Ahmed body. The data represented the near-wake region in the
lower shear layer which was characterised by strongest turbulent activities, i.e. best suited for
the triple decomposition technique. For all locations, the upper peak was clearly distinguishable
and the lowest cut-off peak corresponded to the minimum characteristic frequency peak.
Table 4-2 summarizes the lower and upper cut-off frequencies.

Table 4-2: Cut-off frequencies for the triple decomposition technique

Positions Lower cut-off Upper cut-off


frequency (Hz) frequency (Hz)
x/H y/H z/H

0.648 0.093 0.226 3.0 57.0


0.833 0.093 0.226 1.5 59.0
1.204 0.093 0.226 3.0 57.5
1.389 0.093 0.226 1.0 62.0
1.574 0.093 0.226 2.0 60.0
1.759 0.093 0.226 2.0 58.5

4.3.2 Preliminary results: influence of dimensionless longitudinal distance on the


root mean square velocity

The triple decomposition was applied to the root mean square of the instantaneous raw velocity
Urms for a range of longitudinal positions x/H. Figure 4-4 presents the results of the triple
decomposition technique as a function the dimensionless distance x/H. In Figure 4-4, the
"interpolated signal" refers to the instantaneous interpolated raw signal, while the "mean
component", the "slow fluctuation" and "high fluctuation" components represent the decomposed
velocity components. The comparison of the root mean square of the raw velocity signal U rms
(and thus the turbulence level in the flow direction) with the decomposed components revealed
that the turbulent motion was predominantly governed by the slow and mean components of the
flow. Fast fluctuations above 60 Hz did not significantly affect the turbulent flow motion.

With increasing distance downstream of the Ahmed body, the root mean square of the
interpolated raw velocity decreased which was linked with a decrease in turbulence activities
(Figure 4-4). Simultaneously the mean and slow fluctuating components decreased while the
contribution of the fast fluctuating components increased proportionally. Except for one position
(x/H=1.204), the contribution of the slow component seemed to be more important than the
mean component.

The present results are preliminary. The results confirm however that the triple decomposition
technique can be applied to instantaneous raw velocities downstream of an Ahmed body
providing a better understanding of the flow dynamics in wake flows. Further investigations are
required to build on these first results. In particular, further detailed measurements of flow
velocities downstream of the Ahmed body are needed. Furthermore the data processing should
be expanded to different turbulence parameters including Reynolds stresses and turbulent time
scales.

WRL Research Report 262 FINAL May 2017 25


4
Interpolated Signal
3.5 Mean Component
Slow Fluctuation
High Fluctuation
3
Urms (m/s)

2.5

1.5

0.5
0.8 1 1.2 1.4 1.6 1.8 2
x/H

Figure 4-4: Triple decomposition applied to the root mean square of the horizontal component of
the velocity. Influence of x/H

WRL Research Report 262 FINAL May 2017 26


5. Discussion

The present study confirmed some interesting features in the non-stationary flow downstream of
an Ahmed body. The results were based upon experiments with a 2D LDV measurement system.
Distinctive frequencies were found corresponding to different regions of the flow defined by their
turbulent activity. Furthermore, some agreements were found in terms of Strouhal numbers with
different previous studies which used different experimental techniques such as PIV and/or
sensors. Some limitations existed in the present experiments that should be addressed in future
work to expand the triple decomposition approach to further parameters. Three of these
recommendations are discussed in this section.

5.1 Limitation of sampling frequency of raw velocity data


Using LDV for velocity measurements implies flow seeding. In a wind tunnel, this is ensured by a
fog generator fed with oil. In the present study, one model was used for which the particle size
distribution was known (Figure 5-1). At the nozzle exit, oil droplets had a mean diameter of
1.068 µm while the majority of droplets had sizes below 1 µm (Figure 5-1). The sizes of the
droplets were sufficiently small compared to the size of the interference fringes of the laser
beam (5.448 µm) forming in the measurement volume (0.1684*0.1681*2.806 mm3). This fog
fluid was provided by DANTEC and manufactured by Safex.

Figure 5-1: Particle size distribution of oil droplets released from the fog generator

Depending on the position of the measurement within the wake of the Ahmed body, regions
could be observed where it was difficult for oil droplets to enter. This resulted in low data rates
making the frequency analysis meaningless. To overcome this issue, bursts of droplets were sent
from the fog generator to ensure a continuous supply of seeding. This procedure led to a sudden
and brief increase in the data rate for most of the investigated positions of up to several kHz.
The results in the present report were not significantly affected by the issue of droplet seeding
as the data rate was at least several tens of Hertz and reached hundreds of Hertz in many cases.
The present data rate was larger compared to data rates of a PIV system. Overall, for the
present rear slant angle =35° and the range of investigated positions, the rate of particles per
unit of time was sufficiently large (Rodriguez 2016). The present data quality was in agreement
with other LDV studies (e.g. Lienhart and Pego 2012).

WRL Research Report 262 FINAL May 2017 27


5.2 Accuracy of the LDV measurements
LDV is a non-intrusive and accurate optical technique. This means that it does not create any
unintended disturbances that might affect the description of the flow. Furthermore, it is not
sensitive to temperature fluctuations. Nevertheless, there are different factors that might have
affected the accuracy of the present results:
 Duration of the acquisition: If the acquisition time dacq is too short, the dynamics of the
turbulent structures might not be captured. In the present study, the acquisition time
has been defined according to a rigorous calibration which ensured a convergence of the
data. As a result dacq = 90 s in the present investigations. This was large enough
compared to the time scales of the flow and larger than in most other studies. A
sensitivity analysis showed that for 300 s > dacq > 90 s the deviations for both mean and
root mean square velocities were below 4% (Rodriguez 2016). For comparison, Tunay et
al. (2014) used an acquisition time dacq = 25 s for a comparable Reynolds number.
 Repeatability of measurements: The repeatability of data has also been assessed in five
(5) recordings which showed a maximum deviation of 10% for mean and root mean
square values independent of dacq (Rodriguez 2016).
 Data filtering: In the present study, no filtering was applied and the raw data were
resampled before FFT computations.
 When bursts of droplets were sent into the wind tunnel, some additional limitations may
occur for optical reasons. If more than one particle crossed the measurement volume at
the same time, the recording of mean and turbulent properties of the flow may be
biased. Rodriguez (2016) discussed this effect and proposed a new calculation method.
While this feature was outside of the scope of the present study, it is believed that this
point did not significantly affect the determination of the characteristic frequencies. Close
agreements between the present results and previous experimental works tended to
confirm this.
 The vicinity of a wall can lead to erroneous data due to light reflections. Similarly,
unclean or imperfect glue joints at the wall may affect the light path making
measurements impossible. This was not a concern in the present study and precautions
were taken to avoid these issues.

5.3 Wind tunnel measurements and scale effects


Experimental investigations in scaled wind or water tunnels may be affected by scale effects. The
most representative dimensionless number in the present study was the Reynolds number based
upon the height of the car model. Considering a prototype car with a height of H = 1 m moving
in an urban area with a velocity of about 8 < U < 14 m/s, the corresponding Reynolds number
would be about 5.105 < Re < 106. In previous experimental studies, the Reynolds numbers were
in a range of 104 < Re < 106 depending on the size of the experimental facilities and the fluid.
For a given scale, water is sometimes used to reach higher Re of approximately one order of
magnitude. As an example, Tunay et al. (2014) worked in a water channel with Re=1.48×104.
Wang et al. (2013) conducted an experimental campaign in a wind tunnel with Re=5.3×104.
Both have the same rear angle as the present study (=35°). Mc Arthur et al. (2016) performed
measurements in a water channel facility with Re=3.78×104 but the model was slightly different
compared to the present Ahmed body, i.e. they used a truck model without rear slant angle
(=0°). As a consequence, the flow topology differed from the present setup. In these studies,
the Reynolds numbers were similar to the present study (Re=5×104) despite some differences in
the rear slant angle and the shape of the model (McArthur et al. 2016). Other experimental
investigations were undertaken with larger Reynolds numbers, e.g. Re= 1.3×105 in Vino et al.
(2005) and Re=7.4×105 in Lienhart and Pego (2012). These studies were conducted for

WRL Research Report 262 FINAL May 2017 28


Reynolds numbers of one order of magnitude larger than the present study. Both studies were
conducted in wind tunnels.

For the test with the smaller Reynolds numbers, some interesting features were reported by
Tunay et al. (2014) in terms of Strouhal number. For =35°, Tunay et al. (2014) observed 0.08
< St< 0.22, which was in good agreement with the present findings. Considering the flow
downstream of a circular cylinder, the Strouhal number is a unique function of the Reynolds
number and independent of the Reynolds number when Re > 103 (St=0.21). According to Vino
et al. (2005), the vortex shedding downstream of square cylinders is well-defined up to
Re≈7.7×105. This finding indicates that no scale effect should be expected downstream of
Ahmed bodies for Reynolds number below 7×105.

Very few experimental investigations assessed the unsteady flow downstream of Ahmed bodies
and apart from Tunay et al. (2014) no other study considered a rear slant angle of =35°. It is
expected that an increase in Reynolds number would lead to an increase of Strouhal number.
This is in agreement with the results of other studies undertaken at larger scales. For instance,
Lienhart and Pego (2012) found St≈0.45 for Re=7.4×105 (=25°). Thacker (2010) observed
St≈0.42 for Re=5.5×105. For the same rear slant angle (=25°), Tunay et al. (2014) assessed
St≈0.31 for Re=1.48×104 while Vino et al. (2005) found St≈0.34 for 4.7×105 < Re < 6.62×105
(=25°). As a consequence, scale effects cannot be neglected for the range of investigated rear
slant angles.

Considering the ratio of the geometric scales between Lienhart and Pego (2012) and the present
experimental conditions, i.e. geometric scaling ratio of 1:5 and velocity scale ratio of 1:2.8,
would result in a frequency scale ratio of 1:1.8. Comparing the results of Lienhart and Pego
(2012) with the present data would result in a lower characteristic frequency of about 9 Hz.
While the present study did not have the same rear slant angle, only slightly lower characteristic
frequencies were observed. Considering potential scale effects, the results were in relatively
close agreement.

It is also important to note that results were sensitive to other external conditions. For instance,
the ground clearance should be taken into account as well as the development of the boundary
layer upstream of the Ahmed body. Comparing the present ground clearance with the studies of
Tunay et al. (2014) and Lienhart and Pego (2012) revealed some differences in ground
clearance, i.e. dimensionless ground clearance of 0.28, 0.17 and 0.60, respectively. These
differences could also explain some differences between flow features downstream of the Ahmed
bodies.

WRL Research Report 262 FINAL May 2017 29


6. Conclusion and future works

The present study was a preliminary study of the non-stationary wake flow downstream of a car
model (Ahmed body). The rear slant angle of the Ahmed body was =35°. The experiments
were conducted in a relatively small-sized wind tunnel at ESTACA, France. The upstream flow
was uniform, i.e. turbulence level below 1% outside of the boundary layers with a velocity
U0=14.5 m/s. The corresponding Reynolds number was about Re=5×104. A detailed series of
velocity measurements were conducted using a 2D LDV system for 0.093≤x/H≤4.722,
0.093≤y/H≤1.296 and 0≤z/H≤0.678. The aim of the present work was to assess the basic
properties of the non-stationary flow with special interest in the characteristic frequencies.
Detailed spectral analysis of the velocity fields was conducted to identify the representative
frequencies of the flow in the wake of the car. LDV data were resampled using a linear
interpolation and an FFT analysis of the resampled data was used to identify the characteristic
frequencies of the velocity signals. For most data, two (2) frequency peaks were identified
allowing the determination of Strouhal numbers and the application of the triple decomposition
technique. The triple decomposition technique was successfully applied to the root mean square
of the horizontal velocity component. The main conclusions are:
 Just downstream of the car model x/H<0.65, no characteristic high frequency peaks
were observed and low frequency activity tended to emerge below 7 Hz corresponding to
Strouhal numbers of 0.011 < St < 0.026.
 With increasing x/H, characteristic peak frequencies were found around 60 Hz
corresponding to St=0.223. The Strouhal number corresponded to periodic vortex
shedding related to Bernard-Von Karman vortex street.
 Low frequency activities of between 1.5 and 8 Hz were also found far downstream of the
Ahmed body. Far downstream, high frequency activity dampened for x/H>2.870.
 In vertical direction, the highest frequency activity was found close to the bottom where
the turbulent kinetic energy reached its maximum levels in the lower shear layer. A
similar trend was found in the upper shear layer in the vicinity of the roof of the car.
 In the transverse direction, no differences were found albeit the present data were
limited to the near-wake region.
 The present data were in agreement with the available literature for comparable
Reynolds numbers and rear slant angles. For larger scales and Reynolds numbers, the
present data differed from other studies indicating that possible scale effects cannot be
ignored above a critical Reynolds number.
 For selected locations, two characteristic frequency peaks were identified and the triple
decomposition technique was applied to the root mean square of the horizontal velocity
component. The results showed that the mean and slow fluctuating velocity components
contributed most to the turbulence flow motions.

Overall the present study highlighted some interesting features of the non-stationary flow
developing downstream of a car model in a wind tunnel. Characteristic Strouhal numbers of 0.2
were found in the close wake of the Ahmed body corresponding to a vortex street. The
preliminary investigations demonstrated that the triple decomposition can be applied to such
wake flows. Further experiments are required to strengthen these first results. Particularly, it
would be interesting to expand the work to different rear angles, to further measurement
positions downstream of the car as well as to different upstream velocities. Interactions between
two (2) vehicles should also be considered representing a more realistic situation. The present
contribution provided new information about the dynamics of the flow in the wake of a car
model. It will help to improve the understanding of the interactions between particles and
turbulence, i.e. key factors that may influence pollutant dispersion and/or infiltration in the wake
of a vehicle. From an experimental point of view, it will be useful to optimize sampling

WRL Research Report 262 FINAL May 2017 30


parameters, acquisition duration, meshing, size of sampling sensors for the two-phase flow
campaign. Considering discrepancies that appear in the literature regarding the spectral analysis,
it also brings practical information that will be used to calibrate new numerical models.

WRL Research Report 262 FINAL May 2017 31


7. Acknowledgments

The authors would like to thank Dr Hang Wang (The University of Queensland, Australia) for his
detailed review of the report and valuable comments. The first author would like to thank Dr
Stefan Felder and Professor Ian Turner (Director of WRL) from the Water Research Laboratory,
School of Civil and Environmental Engineering, UNSW Sydney for the invitation to undertake his
internship as part of his studies at ESTACA. He particularly thanks Stefan Felder for his patience
and availability during the internship. Romain Rodriguez (PhD Student at ESTACA, Laval, France)
is greatly acknowledged for providing his LDV data.

WRL Research Report 262 FINAL May 2017 32


8. References

Ahmed, S.R., Ramm, G., and Faltin, G. (1984). "Some salient features of the time-averaged
ground vehicle wake. " SAE Technical paper 840300, Detroit (USA).

Aljure, D.E., Lehmkuhl, O., Rodriguez, I. and Oliva, A. (2014). "Flow and turbulent structures
around simplified car models" Computers and Fluids, Vol. 96, pp. 122-135 (DOI:
10.1016/j.compfluids.2014.03.13).

Brown, R. and Chanson, H. (2013). "Turbulence and suspended sediment measurements in an


urban environment during the Brisbane river flood of January 2011" Journal of Hydraulic
Engineering, ASCE, Vol. 139, No. 2, pp 244-252 (DOI: 10.1061/(ASCE)HY.1943-7900.0000666).

Buseck, P.R. and Adachi, K. (2008). "Nanoparticles in the Atmosphere" Elements, Vol. 4, pp.
389-394 (DOI: 10.2113/gelements.4.6.389).

Buzea, C., Pacheco, I.I. and Robie, K. (2007). "Nanomaterials and nanoparticles: sources and
toxicity" Biointerphases, Vol. 2, No. 4, pp 17-71 (DOI: 10.1116/1.2815690).

Carpentieri, M., Kumar, P. and Robins, A. (2012). "Wind tunnel measurements for dispersion
modelling of vehicles wakes" Atmospheric Environment, Vol. 62, pp. 9-25 (DOI:
10.1016/j.atmosenv.2012.08.019).

Erba, S., Escande-Vilbois, S., Fellinger, F., Forray, N., Legrand, H. and Pinet, M. (2015). "La
gestion des pics de pollution de l’air", Rapport CGEDD – IAG, 133 pages (in French).

Eulalie, Y. (2014). "Etude aérodynamique et contrôle de la traînée sur un corps de Ahmed culot
droit. General mathematics" PhD Thesis, University of Bordeaux, France (in French).

European Environmental Agency (2016). "Explaining road transport emissions. A non-technical


guide" Publications Office of the European Union, 55 pages (DOI: 10.2800/71804).

Felder, S. (2013). "Air water flow properties on stepped spillways for embankment dams:
aeration, energy dissipation and turbulence on uniform, non-uniform and pooled stepped chutes"
PhD Thesis, School of Civil Engineering, The University of Queensland, Brisbane, Australia.

Felder, S. and Chanson, H. (2012). "Air-Water Flow Measurements in Instationary Free-Surface


Flows: a Triple Decomposition Technique." Hydraulic Model Report No. CH85/12, School of Civil
Engineering, The University of Queensland, Brisbane, Australia, 161 pages (ISBN
9781742720494).

Felder, S., and Chanson, H. (2014). "Triple decomposition technique in air-water flows:
application to instationary flows on a stepped spillway" International Journal of Multiphase Flow,
Vol. 58, pp 139-153 (DOI:10.1061/J.imultiphaseflow.2013.09.006).

Fox, J.F., Papanicolaou, A.N. and Kjos, L. (2005). "Eddy taxonomy methodology around
submerged barb obstacle with a fixed rough bed" Journal of Engineering Mechanics, ASCE, Vol.
131, No. 10, pp 1082-1101 (DOI: 10.1061/(ASCE)0733-9399(2005)131:10(1082).

Franck, G., Nigro, N., Storti, M. and D’Elia, J. (2009). "Numerical simulation of the flow around
the Ahmed vehicle model" Latin American Applied Research, Vol. 39, No. 4, pp 295-306.

WRL Research Report 262 FINAL May 2017 33


Gosse, K. (2005). "Etude expérimentale de la dispersion d’un scalaire passif dans le sillage
proche d’un corps d’Ahmed" PhD Thesis, The University of Rouen, France (in French).

Grandemange, M. (2013). "Analysis and control of three-dimensional turbulent wakes : from


axisymmetric bodies to road vehicles" PhD Thesis, Polytechnic School, ENSTA ParisTech, France
(in French).

Hucho, W.H. (1987). "Aerodynamics of road vehicles" Butterworths and Co, London, UK.

Hussain, A.K.M.F. and Reynolds, W.C. (1972). "The mechanics of an organized wave in turbulent
shear flow. Part 2: Experimental result" Journal of Fluid Mechanics, Vol. 54, No. 2, pp 241-261
(DOI: 10.1017/S002211207000667).

Jeng, C.J., Kindzierski, W.B. and Smith, D.W. (2007). "Particle Penetration through Inclined and
L -Shaped Cracks" Journal of Environmental Engineering, Vol. 133, No. 3, pp. 331-339 (DOI:
10/1061/(ASCE)0733-9372(2007)133:3(331)).

Kahn, R.S. and Umale, S. (2014). "CFD aerodynamic analysis of Ahmed body" International
Journal of Engineering Trends and Technology, Vol. 18, No. 7, pp 301-308.

Kittelson, D.B., Watts, W.F. and Johnson, J.P. (2004). "Nanoparticle emissions on Minnesota
highways" Atmospheric Environment, Vol. 38, No. 1, pp 9-19 (DOI:
10.1016/j.atmosenv.2003.09.037).

Krajnovic, S. and Davidson, L. (2004). "Large eddy simulation of the flow around an Ahmed
body" Proceedings of HTFED04, ASME Heat Transfer / Fluids Engineering Summer Conference,
July 11-15, Charlotte, North Carolina (USA), 10 pages.

Lahaye, A. (2014). "Caractérisation de l’écoulement autour d’un corps de Ahmed à culot droit"
PhD Thesis, The University of Orleans, France (in French).

Leclerc, C. (2008). "Réductin de la traînée d’un véhicule automobile simplifié à l’aide du contrôle
actif par jet synthétique" PhD Thesis, The National Polytechnic Institute of Toulouse, France (in
French).

Lee, E.S. (2013). "Passenger exposures to ultrafine particles and in-cabin air quality control" PhD
Thesis, The University of California at Los Angeles, USA.

Lienhart, H. (2016) "Private communication".

Lienhart, H. and Pego, J.P. (2012). "Spectral density and time scales of velocity fluctuations in
the wake of a simplified car model" SAE Technical Paper 01-592, Detroit, USA, 10 pages.

Liu, D.L. and Nazaroff, W.W. (2003). "Particle penetration through building cracks " Aerosol
Science and Technology, Vol. 37, pp. 563-573 (DOI: 10.1080/02786820300927).

Lyn, D.A. and Rodi, W. (1994). "The flapping shear layer formed by flow separation from the
forward corner of a square cylinder" Journal of Fluid Mechanics, Vol. 267, pp 353-376 (DOI:
10.1017/S0022112094001217).

WRL Research Report 262 FINAL May 2017 34


McArthur, D., Burton, D., Thomson, M. and Sheridan, J. (2016). "On the near wake of a
simplified heavy vehicle" Journal of Fluids Engineering, Vol. 66, pp 293-314 (DOI:
10.1016/j.fluidsstructs.2016.07.011).

Mehel, A. and Murzyn, F. (2015). "Effect of air velocity on nanoparticle dispersion in the wake of
a vehicle model: wind tunnel experiments" Atmospheric Pollution Research, Vol. 6, No. 4, pp
612-617 (DOI: 10.5094/APR.2015.069).

OECD (2016). "The economic consequences of outdoor air pollution" OECD publishing, Paris,
(DOI: 10.1787/9789264257474-en).

Pope, C.A., Ezzati, M. and Dockery, D.W. (2013). "Fine particulate air pollution and life
expectancies in the United States: the role of influential observations " Journal of the Air and
Waste Management Association, Vol. 63, No. 2, pp. 129-132 (DOI:
10.1080/10962247.2013.760353).

Rodriguez, R. (2016). "Etude dans une soufflerie de la dispersion de nanoparticules issues des
gaz d’échappement d’un véhicule et de leur infiltration dans les habitacles " Internal report of
PhD Thesis, ESTACA, France (in French).

Roumegas, J.L. and Saddier, M. (2016). "Evaluation des politiques publiques de lutte contre la
pollution de l’air" Rapport d’information n°3772, Présidence de l’Assemblée Nationale, Paris,
France (in French).

Semmler-Behnke, M., Lipka, J., Wenk, A., Hirn, S., Schäffer, M., Tian, F., Schmid, G.,
Oberdörster, G. and Kreyling, W.G. (2014). "Size dependent translocation and fetal accumulation
of gold nanoparticles from maternal blood in the rat" Particle and Fibre Toxicology, Vol. 11, No.
33 (DOI: 10.1186/s12989-014-0033-9).

Thacker, A. (2010). "Contribution expérimentale à l’analyse stationnaire et instationnaire de


l’écoulement à l’arrière d’un corps de faible allongement " PhD Thesis, The University of Orleans,
France (in French).

Tunay, T., Sahin, B. and Ozbolat, V. (2014). "Effect of rear slant angles on the flow
characteristics of Ahmed body" Experimental Thermal and Fluid Science, Vol. 57, pp 165-178,
(DOI: 10.1016/j.expthermflusci.2014.04.016).

Valentino, S.A., Tarrade, A., Aioun, J., Mourier, E., Richard, C., Dahirel, M., Rousseau-Ralliard,
D., Fournier, N., Aubrière, M-C., Lallemand, M-S., Camous, S., Guinot, M., Charlier, M., Aujean.,
E., Al Adhami, H, Fokkens, P.H., Agier, L., Boere, J.A., Cassee, F.R., Slama, R. and Chavatte-
Palmer, P. (2016). "Maternal exposure to diluted diesel engine exhaust alters placental function
and induces intergenerational effects in rabbits" Particle and Fibre Toxicology, Vol. 13, No. 39,
14 pages (DOI: 10.1186/s12989-016-0151-7).

Van Dyke, M. (1982). "An album of fluid motion”, The parabolic press, 176 pages.

Venning, J., Lo Jacono, D., Burton, D., Thompson, M. and Sheridan, J. (2015). "The effect of
aspect ratio on the wake of the Ahmed body" Experiments in Fluids, Vol. 56, No. 126, 11 pages
(DOI: 10.1007/s00348-015-1996-5).

WRL Research Report 262 FINAL May 2017 35


Vino, G., Watkins, S., Mousley, P., Watmuff, J., and Prasad, S. (2005). "Flow structures in the
near-wake of the Ahmed model" Journal of Fluids Engineering, Vol. 20, pp. 673-695 (DOI:
10.1016/j.jfluidsstructs.2005.03.006).

Wang, H., Felder, S. and Chanson, H. (2014). "An experimental study of turbulent two-phase
flow in hydraulic jumps and application of a triple decomposition technique" Experiments in
fluids, Vol. 55, No. 7, paper 1775, 18 pages (DOI: 10.1007:s00348-014-1775-8).

Wang, X.W., Zhou, Y., Pin, Y.F. and Chan, T.L. (2013). "Turbulent near wake of an Ahmed
vehicle model" Experiments in Fluids, Vol. 54, No. 1490, 19 pages (DOI: 10.1007/s00348-013-
1490-x).

WHO (2015). "Economic cost of the health impact of air pollution in Europe: clean air, health and
wealth", Copenhagen: WHO Regional Office for Europe, 56 pages.

World Bank and Institute for Health Metrics and Evaluation (2016). "The cost of air pollution.
Strengthening the economic case for action" Washington DC: World Bank, Licence: Creative
Common Attribution CC BY 3.0 IGO.

Xia, T., Zhu, Y., Mu, L., Zhang, Z-F. and Liu, S. (2016). "Particulate matter induced pulmonary
diseases in 21st century – attention to ambient ultrafine and engineered nanoparticles" National
Science Review, Vol. 3, No. 4, pp. 416-429 (DOI: 10.1093/nsr/nww064).

WRL Research Report 262 FINAL May 2017 36


9. Appendices

9.1 Matlab code


9.1.1 Step 1: Reading data from Excel files to Matlab

This program is used to open Excel files and to store the data in Matlab. Two matrices are used
containing the file names and folder name. A loop allows the navigation through the folder
matrix. Each file contained in the matrix with file names is opened and the data are extracted to
Matlab including the time series and the instantaneous velocity. Furthermore for each file the
corresponding mean flow velocity, the sampling frequency and the coordinates of the
measurement points are extracted and stored in Matlab.

close all
clear all
clc

file_={'x25_y25_z-24.3.xlsx','x35_y5_z-12.2.xlsx','x45_y5_z-
12.2.xlsx','x65_y5_z-12.2.xlsx','x75_y5_z-12.2.xlsx','x85_y5_z-
12.2.xlsx','x95_y5_z-12.2.xlsx','x205_y25_z-12.2.xlsx'};
n_=length(file_);

Fs=zeros(n_);

folder_={'Ahmed_35'};
n_2=length(folder_);

m=1;
freq=[1.1 55.2;2 57;1.5 58.8;3 57.7;1.1 62;1.8 63.7;1 58.5;2.9 53.7];
%Concatenation gives the name of the global files without having to copy
%paste it
global_file=strcat('Global_',file_(1:n_));
folder_path=strcat('C:\Users\z5131390\Desktop\Project\',folder_(1:n_2));

Excel = actxserver('Excel.Application');

Excel.Visible = false;

for j=1:length(folder_);
for i=1:n_;

xlspath = folder_path{j} ;
xlsfile = file_{i} ;

Workbook = Excel.Workbooks.Open(fullfile(xlspath,xlsfile));

ActiveSheet = Excel.Worksheets.Item('BSA Export');

%For the velocity and time-series


r(1) = ActiveSheet.Range('B7');
r(2) = ActiveSheet.Range('B7').End('xlDown');
time_raw = ActiveSheet.get('Range',r(1),r(2)).Value;
time_raw = cell2mat(time_raw);%Converts the table from cell to matrix
time_raw = time_raw./1000;% Converts the time in seconds
r_(1) = ActiveSheet.Range('D7');

WRL Research Report 262 FINAL May 2017 37


r_(2) = ActiveSheet.Range('D7').End('xlDown');
velocity_raw = ActiveSheet.get('Range',r_(1),r_(2)).Value;
velocity_raw = cell2mat(velocity_raw);

Workbook.Close(false);

xlspath_ = folder_path{j}
xlsfile_ = global_file{i} ;

Workbook = Excel.Workbooks.Open(fullfile(xlspath_,xlsfile_));

ActiveSheet = Excel.Worksheets.Item('BSA Export');

%Take the sampling frequency and the point's coordinates


Fs(i)=ActiveSheet.Range('E7').Value;
absi=ActiveSheet.Range('B7').Value;
ord=ActiveSheet.Range('C7').Value;

[lin, P, pxx, ff, xq] = FFT(time_raw,


velocity_raw,file_{i},folder_{j},absi,ord,i,j,freq);

Workbook.Close(false);

end
end

Quit(Excel);

9.1.2 Step 2: Performing a Fast Fourier Transform and triple decomposition of the
velocity signals

This program is used to perform the spectral analysis for selected data points. First the code
interpolates the raw instantaneous velocity data to achieve equal time intervals. Then it does a
Fast Fourier Transform on the interpolated data and displays the FFT and Power Spectral Density
(PSD) functions. Based upon the characteristic frequencies the software performs the triple
decomposition analysis of the interpolated signal.

function [lin, P, pxx, ff, xq] = FFT( time_raw,


raw,file_,folder_,absi,ord,i,j,freq )

n = 2^nextpow2(length(time_raw));
x_=[5 10 15 20 25 30 35 40 45 50 55 60 70];
y=[5 15 25 35 45 55 65 75 85 95 105 155 205 255 305];
x=1;
%Determination of the new time-series
xq = time_raw(1):(time_raw(length(time_raw))-
time_raw(1))/n:time_raw(length(time_raw));

%Determination of the linear interpolation


lin = interp1(time_raw,raw,xq);
Fs_=length(lin)/xq(length(xq));
%Determination of the FFT
Y = fft(lin,n);
P = abs(Y/n);
f = Fs_*(0:(n/2))/n;%Application of Nyquist-Shannon criteria

WRL Research Report 262 FINAL May 2017 38


h(x)=figure; %Display of the mesh
for i_=1:length(x_)
t = 1:5:310;
line([t;t],repmat(x_(i_),1,numel(t)),'color','b','linestyle','-')
end
for i_=1:length(y)
t=0:5:75;
line(repmat(y(i_),1,numel(t)),[t;t],'color','b','linestyle','-')
end
hold on
plot(absi,ord,'or')
axis equal
title([folder_,' ', file_])
x=x+1;
smoo_fft=smooth(P,55,'sgolay',3);

% Display of the FFT


h(x)=figure;
loglog(f,P(1:n/2+1));%Again, application of the Nyquist-Shannon criteria
hold on
loglog(f,smoo_fft(1:n/2+1),'k-')
% loglog((Fs/16):(Fs/2),10.^(((-5/3)*log10((Fs/16):(Fs/2)))+0.08),'r-
')%Display of the -5/3 slope
xlabel('Frequency (Hz)')
ylabel('Amplitude')
legend('FFT','span 55 degree 3')
title([folder_,' ', file_])
x=x+1;
% Smoothing the signal using the Savitzky-Golay method
% smoo_signal=smooth(lin,500,'sgolay',5);

h(x)=figure;
plot(time_raw, raw, 'g-')%Display of the raw signal
hold on
plot(xq,lin,'b-');%display of the linear interpolation
hold on
% plot(xq,smoo_signal, 'r-'); %Display of the smoothing curve of the
interpolqted signal
legend('Raw Signal','Interpolated Signal')
ylabel('Velocity (m/s)');
xlabel('Time (s)');
title([folder_,' ', file_])
x=x+1;

%Power spectral Density function using welch approximation


[pxx,ff] = pwelch(lin,n,n/2,f,Fs_);%Power spectral Density function
St=ff.*0.0547/14;
St_s=Fs_*0.0547/14;

smoo_psd=smooth(pxx,55,'sgolay',3);
%
[pk,MaxFreq]=findpeaks(smoo_psd,'NPeaks',2,'SortStr','descend','MinPeakDi
stance',2000);

for l=1:length(smoo_psd)
if ff(l)>=0.1
pxx_2(l)=pxx(l) ;

WRL Research Report 262 FINAL May 2017 39


end
end
%
[pk2,MinFreq]=findpeaks(pxx_2,'NPeaks',1,'SortStr','descend','MinPeakDist
ance',500);

h(x)=figure;
% loglog(St,pxx);%Display of the PSD with the Strouhal number on the x-
axis
% hold on
% loglog(ff(MinFreq),pk2,'or')
% hold on
loglog(ff,pxx,'g-')%Display of the PSD with the frequency on the x-axis
hold on
loglog((St_s/16):0.005:(St_s/2), 10.^(((-
5/3)*log10((St_s/16):0.005:(St_s/2)))-4), 'r-')%Display of the -5/3 slope
hold on
loglog(ff,smoo_psd,'k-')
% hold on
% loglog(ff(MaxFreq(2)),pk(2),'or')
xlabel('Frequency (Hz)/Strouhal Number')
ylabel('Magnitude (dB)')
legend('PSD Frequ','-5/3 slope','span 55 degree 3','Peak')
title([folder_,' ', file_])
x=x+1;

%Triple decomposition
[x, raw_mean_,raw_highf_,raw_slowf_,h ] = Triple_dec( lin, freq,i,
Fs_,xq, folder_,file_,h,x );

% %Turbulence Analysis
[x, h,
Tu_lin_,Tu_raw_mean_,Tu_raw_slowf_,Tu_raw_highf_,T_xx_lin_,T_xx_raw_mean_
,T_xx_raw_slowf_,T_xx_raw_highf_ ] = Turbulence_analysis(
h,raw,lin,raw_mean_,raw_slowf_,raw_highf_,xq,folder_,file_,freq,i,x );

%Sensitivity Analysis
[x,h, w, w_mean, w_slowf,
w_highf,R_xx_lin,R_xx_raw_mean,R_xx_raw_slowf,R_xx_raw_highf
]=sensitivity_analysis( lin, Fs_ ,xq,file_,folder_,h,x);

% [x, raw_mean_,raw_highf_,raw_slowf_,h ] = Test_freq( lin, freq,i,


Fs_,xq, folder_,file_,h,x );
% [x, raw_mean_,raw_highf_,raw_slowf_,h ] = Test_freq_r_xx( lin, freq,i,
Fs_,xq, folder_,file_,h,x );

% [x, h,lin_zoom_1, lin_zoom_2,xq_zoom_1, xq_zoom_2 ] = Test_autocorr(


xq,lin,h,Fs_,file_,folder_,x );

% n_zoom_2 = 2^nextpow2(length(lin_zoom_2));
% Fs_zoom_2=length(lin_zoom_2)/(xq_zoom_2(length(xq_zoom_2))-
xq_zoom_2(1));
% %Determination of the FFT
% Y_zoom_2 = fft(lin_zoom_2,n_zoom_2);
% P_zoom_2 = abs(Y_zoom_2/n_zoom_2);
% f_zoom_2 = Fs_zoom_2*(0:(n_zoom_2/2))/n_zoom_2;%Application of Nyquist-
Shannon criteria

WRL Research Report 262 FINAL May 2017 40


% smoo_fft_zoom_2=smooth(P_zoom_2,55,'sgolay',3);
% % Display of the FFT
% h(x)=figure;
% loglog(f_zoom_2,P_zoom_2(1:n_zoom_2/2+1));%Again, application of the
Nyquist-Shannon criteria
% hold on
% loglog(f_zoom_2,smoo_fft_zoom_2(1:n_zoom_2/2+1),'k-')
% % loglog((Fs/16):(Fs/2),10.^(((-5/3)*log10((Fs/16):(Fs/2)))+0.08),'r-
')%Display of the -5/3 slope
% xlabel('Frequency (Hz)')
% ylabel('Amplitude')
% legend('FFT','span 55 degree 3')
% title([folder_,' ', file_])

%Save the figures


fig_file_=strcat(file_,'.fig');
fig_file_2=strcat(folder_,fig_file_);
savefig(h,fig_file_2)
close(h)

end

9.1.3 Step 3: Turbulence analysis of the decomposed signal components

This program analyses the turbulence properties for the interpolated raw data as well as for the
decomposed signal components. The analysis includes the mean flow velocities, the standard
deviation of flow velocities and the turbulence intensities. The program writes the results to a file
and produces result plots.

function [x, h,
Tu_lin_,Tu_raw_mean_,Tu_raw_slowf_,Tu_raw_highf_,T_xx_lin_,T_xx_raw_mean_
,T_xx_raw_slowf_,T_xx_raw_highf_ ] = Turbulence_analysis(
h,raw,lin,raw_mean_,raw_slowf_,raw_highf_,xq,folder_,file_,freq,i,x )
%UNTITLED3 Summary of this function goes here
% Detailed explanation goes here
s1=1;
s2=1;
s3=1;
s4=1;
% raw_meanvalue_=mean(raw);
lin_mean_=mean(lin);
raw_mean_mean_=mean(raw_mean_);
raw_slowf_mean_=mean(raw_slowf_);
raw_highf_mean_=mean(raw_highf_);

% raw_rms_=rms(raw);
lin_rms_=std(lin);
raw_mean_rms_=std(raw_mean_);
raw_slowf_rms_=std(raw_slowf_);
raw_highf_rms_=std(raw_highf_);

% Tu_raw_=raw_rms_/raw_meanvalue_;
Tu_lin_=lin_rms_/lin_mean_;
Tu_raw_mean_=raw_mean_rms_/lin_mean_;
Tu_raw_slowf_=raw_slowf_rms_/lin_mean_;
Tu_raw_highf_=raw_highf_rms_/lin_mean_;

WRL Research Report 262 FINAL May 2017 41


% R_xx_raw=autocorr(raw);
R_xx_lin_=autocorr(lin,length(xq)-1);
R_xx_raw_mean_=autocorr(raw_mean_,length(xq)-1);
R_xx_raw_slowf_=autocorr(raw_slowf_,length(xq)-1);
R_xx_raw_highf_=autocorr(raw_highf_,length(xq)-1);
for q=2:(length(xq)-1)
if sign(R_xx_lin_(q-1))~=sign(R_xx_lin_(q))
w_(s1)=q;
s1=s1+1;
end
if sign(R_xx_raw_mean_(q-1))~=sign(R_xx_raw_mean_(q))
w_mean_(s2)=q;
s2=s2+1;
end
if sign(R_xx_raw_slowf_(q-1))~=sign(R_xx_raw_slowf_(q))
w_slowf_(s3)=q;
s3=s3+1;
end
if sign(R_xx_raw_highf_(q-1))~=sign(R_xx_raw_highf_(q))
w_highf_(s4)=q;
s4=s4+1;
end
end

dt=xq(2)-xq(1);

% T_xx_raw(o)=trapz(R_xx_raw)*dt;
T_xx_lin_=trapz(R_xx_lin_(1:w_))*dt;
T_xx_raw_mean_=trapz(R_xx_raw_mean_(1:w_mean_(1)))*dt;
T_xx_raw_slowf_=trapz(R_xx_raw_slowf_(1:w_slowf_(1)))*dt;
T_xx_raw_highf_=trapz(R_xx_raw_highf_(1:w_highf_(1)))*dt;

h(x)=figure;
% plot(freq(i,1),raw_meanvalue_,'dc');
% hold on
plot(freq(i,1),lin_mean_,'or');
hold on
plot(freq(i,1),raw_mean_mean_,'ob');
hold on
plot(freq(i,1),raw_slowf_mean_,'ok');
hold on
plot(freq(i,1),raw_highf_mean_,'og');
legend('Interpolated Signal Mean Value with Sfreq','Mean Component Mean
Value with Sfreq','Slow Fluctuation Signal Mean Value with Sfreq','High
Fluctuation Signal Mean Value with Sfreq');
ylabel('Velocity (m/s)');
xlabel('Lower Cut-Off Frequency (Hz)');
title([folder_,' ', file_])
x=x+1;

h(x)=figure;
% plot(freq(i,1),Tu_raw_,'dc');
% hold on
plot(freq(i,1),Tu_lin_,'or');
hold on
plot(freq(i,1),Tu_raw_mean_,'ob');
hold on
plot(freq(i,1),Tu_raw_slowf_,'ok');

WRL Research Report 262 FINAL May 2017 42


hold on
plot(freq(i,1),Tu_raw_highf_,'og');
legend('Interpolated Signal Turbulence Intensity Value with Sfreq','Mean
Component Turbulence Intensity Value','Slow Fluctuation Signal Turbulence
Intensity Value','High Fluctuation Signal Turbulence Intensity Value');
ylabel('Velocity (m/s)');
xlabel('Lower Cut-Off Frequency (Hz)');
title([folder_,' ', file_])
x=x+1;

% h(x)=figure;
% % plot(sensi,T_xx_raw,'or');
% % hold on
% plot(freq(1),T_xx_lin_,'or');
% hold on
% plot(freq(1),T_xx_raw_mean_,'ob');
% hold on
% plot(freq(1),T_xx_raw_slowf_,'ok');
% hold on
% plot(freq(1),T_xx_raw_highf_,'og');
% legend('Interpolated Signal Integral Time Scale Value with Sfreq','Mean
Component Integral Time Scale Value with Sfreq','Slow Fluctuation Signal
Integral Time Scale Value','High Fluctuation Signal Integral Time Scale
Value');
% ylabel('Velocity (m/s)');
% xlabel('Lower Cut-Off Frequency (Hz)');
% title([folder_,' ', file_])
% x=x+1;

h(x)=figure;
% plot(freq(i,1),raw_rms_,'dc');
% hold on
plot(freq(i,1),lin_rms_,'or');
hold on
plot(freq(i,1),raw_mean_rms_,'ob');
hold on
plot(freq(i,1),raw_slowf_rms_,'ok');
hold on
plot(freq(i,1),raw_highf_rms_,'og');
legend('Interpolated Signal RMS Value with Sfreq','Mean Component RMS
Value','Slow Fluctuation Signal RMS Value','High Fluctuation Signal RMS
Value');
ylabel('Velocity (m/s)');
xlabel('Lower Cut-Off Frequency (Hz)');
title([folder_,' ', file_])
x=x+1;

end

9.2 Results of the FFT analysis for all velocity data in the present study
This appendix presents all results of the FFT analysis. The results in this section are
complementary to the data presented in Section 4.1 of the report. Figures 9-1 to 9-6 present the
complete results of the FFT analysis of velocities in flow direction (Figure 9-1), in vertical
direction (Figures 9-2 to 9-5) and in transverse direction (Figure 9-6) respectively.

WRL Research Report 262 FINAL May 2017 43


0 0
10 10
FFT FFT

-1 -1
10 10
FFT Amplitude

FFT Amplitude
-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(a) x/H=0.093 (b) x/H=0.463


0 0
10 10
FFT FFT

-1 -1
10 10
FFT Amplitude

-2 FFT Amplitude -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(c) x/H=0.648 (d) x/H=0.833


0 0
10 10
FFT FFT

-1 -1
10 10
FFT Amplitude

FFT Amplitude

-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(e) x/H=1.019 (f) x/H=1.204


0 0
10 10
FFT FFT

-1 -1
10 10
FFT Amplitude

FFT Amplitude

-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(g) x/H=1.389 (h) x/H=1.574

WRL Research Report 262 FINAL May 2017 44


0 0
10 10
FFT FFT

-1 -1
10 10
FFT Amplitude

FFT Amplitude
-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(i) x/H=1.759 (j) x/H=1.944


0 0
10 10
FFT FFT

-1 -1
10 10
FFT Amplitude

-2 FFT Amplitude -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(k) x/H=2.870 (l) x/H=3.796

Figure 9-1: FFT of the interpolated horizontal velocity signal U at same elevation (y/H=0.093),
same transverse position (z/H=0.226) and various longitudinal positions (x/H) in the wake of
the Ahmed body

WRL Research Report 262 FINAL May 2017 45


0 0
10 10
FFT FFT

-1 -1
10 10
FFT Amplitude

FFT Amplitude
-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(a) y/H=0.093 (b) y/H=0.185

Figure 9-2: FFT of the interpolated horizontal velocity signal U at same longitudinal position
(x/H=0.833), same transverse position (z/H=0.226) and various elevation (y/H) in the wake of
the Ahmed body

0 0
10 10
FFT FFT

-1 -1
10 10
FFT Amplitude

FFT Amplitude

-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(a) y/H=0.093 (b) y/H=0.185


0
10
FFT

-1
10
FFT Amplitude

-2
10

-3
10 0 1 2
10 10 10
Frequency (Hz)

(c) y/H=0.370

Figure 9-3: FFT of the interpolated horizontal velocity signal U at same longitudinal position
(x/H=0.093), same transverse position (z/H=0.226) and various elevation (y/H) in the wake of
the Ahmed body

WRL Research Report 262 FINAL May 2017 46


0 0
10 10
FFT FFT

-1 -1
10 10
FFT Amplitude

FFT Amplitude
-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(a) y/H=0.093 (b) y/H=0.185


0
10
FFT

-1
10
FFT Amplitude

-2
10

-3
10

-4
10 0 1 2
10 10 10
Frequency (Hz)

(c) y/H=0.833

Figure 9-4: FFT of the interpolated horizontal velocity signal U at same longitudinal position
(x/H=1.574), same transverse position (z/H=0.226) and various elevation (y/H) in the wake of
the Ahmed body

WRL Research Report 262 FINAL May 2017 47


0 0
10 10
FFT FFT

-1 -1
10 10
FFT Amplitude

FFT Amplitude
-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(a) y/H=0.093 (b) y/H=0.278


0
10
FFT

-1
10
FFT Amplitude

-2
10

-3
10 0 1 2
10 10 10
Frequency (Hz)

(c) y/H=0.463

Figure 9-5: FFT of the interpolated horizontal velocity signal U at same longitudinal position
(x/H=3.796), same transverse position (z/H=0.226) and various elevation (y/H) in the wake of
the Ahmed body

WRL Research Report 262 FINAL May 2017 48


-1 -1
10 10
FFT FFT
FFT Amplitude

FFT Amplitude
-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(a) z/H=0 (b) z/H=0.226


-1 -1
10 10
FFT FFT
FFT Amplitude

FFT Amplitude
-2 -2
10 10

-3 -3
10 0 1 2
10 0 1 2
10 10 10 10 10 10
Frequency (Hz) Frequency (Hz)

(c) z/H=0.452 (d) z/H=0.678

Figure 9-6: FFT of the interpolated horizontal velocity signal U at same longitudinal position
(x/H=0.093), same elevation (y/H=0.093) and various transverse positions (z/H) in the wake of
the Ahmed body

WRL Research Report 262 FINAL May 2017 49

Vous aimerez peut-être aussi