BOREE Presentation

Optical methods for the measurement of
constant and variable density flows.

J. BOREE *, L. BEN **, R. BAZILE **
*Laboratoire dtudes Arodynamiques, ENSMA, Poitiers, France
**Institut de Mcanique des Fluides de Toulouse, France
Int. Summerschool on variable density turbulent flows .
Barcelona Sept. 1 5, 2003
LEA
1 Introduction. Goals, constraints and challenges
2 Laser Doppler Anemometry
3 (Digital) Particle Image Velocimetry
4 Optical methods to measure spatial density fields
Outline
Introduction (1) Why do we measure ?
In the industry
Technical studies
Validation and improvement of performances of devices. Often now after a
first computation of the product.
In the laboratories
Strong interaction between experiments ; theory ; physical modelling
Development and validation of new concepts / models
Creation of data bases, often in model simplified situations for validation and
development of physical models.
The use of complex measurement techniques imply a strong interaction
between laboratories and industry
Introduction (2) Physical quantities of interest
In a mono-constituant medium :
If M < 0,3 : Measurement of velocity and pressure (and Temperature if heat transfer)
If M > 0,3 : Measurement of velocity, pressure, density or temperature
In a poly-constituant medium :
Density, Mass fraction of species
Droplet diameters, film thickness
Mach number
1 0 Subsonic Supersonic
Incompressible
flow
Compressible
flow
M=0,3
Mach : M=V/c (c: speed of sound)
Introduction (3) Ideal transducers
The output signal is proportional to magnitude of physical quantity
The physical quantity is measured at a point in space
The output signal represents the input without frequency distortion
Low noise on output signal
Transducer does not interfere with the physical process
Output is not influenced by other variables
t
s
s t ( )
u t ( )
Transducer
S(t)
U(t)
Introduction (4) Real transducers
Static transfer function (calibration curve)
Spatial resolution (finite size of measuring volume)
Temporal resolution (frequency response)
Signal-to-noise ratio
Need to understand the interferences between the sensor and the physical phenomena
The measurement of a quantity can be influenced by other physical variables
Introduction (5) Constraints and challenges in turbulent flows
A wide range of scales in a high Reynolds number turbulent flow (Chassaing 2000)
l : Integral scale ; u : Their characteristic velocity on the order of the turbulent intensity
Direct effect of viscosity negligible at these scales if
Rate of dissipation determined by large-scale
dynamics :
Energy cascade :
1 >> = v l . u R
l
k u 3 2 ~
( ) l
u
u l
u
3 2
= ~ c
Small scales properties depend on

[m
2
s
-1
] and [m
2
s
-3
]
Size
4 1 -
4 1
.
l
R
l u
u
w
=
|
.
|
\
|
~
v
4 3 -
4 3
.
l
R
l u
l
=
|
.
|
\
|
~
v
q
Velocity
Wavenumbers
Temporal and spatial resolution .
Example of a turbulent jet with D
0
=10 mm; U
0
= 50 m/s
Need of unsteady measurements with small measurement volumes
20D
0
downstream the exit :
 15 m/s
half width 20 mm
Turbulent Reynolds number : R
l
= 5000
Large scales :
l 20 mm ; u 3.75 m/s ; f /l= 750 Hz
Kolmogorov scales :
34 m ; w 0.4 m/s ;f / = 500 kHz
o
(
x
)
U
a
(
x
)
U
0
x
D
0
New needs to match the development of interpretation/prediction tools.
Analyse, Predict and eventually Control the role of Coherent structures
on
(i) entrainment and mixing of external fluid,
(ii) sound generation,
(iii) transport of a dispersed solid or liquid phase
Wake
Mixing
Layer
From : Pope (2000) Turbulent flows
Two Typical
Free shear flows
Coherent structures :
* Organised character
* Large Scale
Identification of CS embedded in a random velocity field.
(See Bonnet & Delville 2001 for free shear flows ; Adrian et Al 2000 for near wall flows)
Multi-point statistics (Hot Wire rakes ; PIV ; )
Conditional versus Non Conditional CS identification methods
Two important stochastic approaches based on two-point statistics
Linear Stochastic Estimation (LSE Adrian 1975)
Proper Orthogonal Decomposition (POD Lumley 1967)
Link with new simulation strategies (Large Eddy Simulation ; Unsteady
Reynolds Averaged Navier Stokes ; )
Validation using the same tools
Generation of initial conditions
Low dimensionnal approaches for control strategies
Principles of the technique
Spatial resolution : Measurement volume
About seeding
On the use of LDA data for statistical analysis.
Outline
Optical measurements of the velocity of Particles that seed the flow. (see
recent book of Albrecht et al 2003 )
Measurement of 1,2 or even 3
components of the velocity
Small measurement volume at
the intersection of Laser beams
Non intrusive methods but one
needs optical access
Absolute measure
(No calibration needed)
Coherent, monochromatic,
linearly polarised light
Important : Tracers must follow the flow
(Low pass filters)
Laser Doppler Anemometry (1) Principles of the technique
(Source Dantec)
Laser
P
P
.
M
.
U
G
i
e
G
d
e
G
( )
( )
d d d d d
d d
i i i i i
i
k k e k
e f
k k e k
f
G
G
G
G
G
G
G
= =
= =
; 2
P.M. the toward points ; : Detection
; 2
: Beam Incident
t
t
The Photo-multiplicator (P.M.) points toward point P.
The frequency of the incident light seen by the moving particle
is modulated :
The frequency of the light transmitted by the particle and detected by
the photo-
detector is modulated :
Finally :
( ) U . k f f
i i seen
G G
= t 2
( ) U . k f f
d seen d
G G
+ = t 2
( ) ( )U . k k f f
i d i d
G G G
= t 2
Laser
P
P
.
M
.
U
G
i
e
G
d
e
G
( )
( )
d d d d d
d d
i i i i i
i
k k e k
e f
k k e k
f
G
G
G
G
G
G
G
= =
= =
; 2
P.M. the toward points ; : Detection
; 2
: Beam Incident
t
t
Finally :
Orders of magnitude : f
i
is of order 10
14
Hz while (f
d
-f
i
) spans from a few kHz (for
low speeds) to a MHz (Supersonic flows).
One cannot measure directly f
d
but the interferences of two frequencies on the
detector is used : The differential mode is presented next.
( ) ( ) ( )U . e e U . k k f f
i d
i
i d i d
G
G G
G G G
= =
t
t
2
2
i d
e e U
G G
G
= << and light of speed As
( ) ( ) ( )
i d i i d d i d i i d
e e k k k e e e k k k
G G G G G
G G
~ + =
Differential mode currently used :
Thus :
The measure does not depend on the angle of observation.
( ) ( ) ( ) ( )U e e f f U e e f f
i d
i
i d i d
i
i d
G
G G
G
G G
.
1
et .
1
2 2 2 1 1 1
= =

X
F
D
L
Y
Z
L
a
s
e
r
P
U
G
1 i
e
G
P
.
M
.
d
e
G
2 i
e
G
( ) ( )
( )
X
i
D i i
i
d d D
U f U e e f f f
2 sin 2
ou .
1
1 2 2 1
= = =
G
G G
(Graph from Dantec)
LDA Fringe Model
Focused laser beams intersect at the beam waist and form the measurement volume
Interference in the plane of intersection results in a pattern of bright and dark
stripes/planes
This can be considered as a practical way of visualisation
In a variable density flow, the index of refraction varies. One has to choose short
focal length to have a chance of getting a stable control volume ! The resulting
fringe distortion can induce a noise significant in low turbulence flows (Zhang &
Eisele 1998)

LDA Fringe Model
When the particle traverses this fringe pattern, the scattered light fluctuates in
intensity with a frequency equal to the velocity of the particle divided by the fringe
spacing.
Particles moving in either the forward or reverse direction will produce identical
signals and frequencies.
With frequency shift in one beam relative to the other, (use of a Bragg cell) the
interference fringes appear to move at the shift frequency (typically 40 MHz).
Negative velocities can then be distinguished
U
G
Ux
Backscatter configuration- Source : Dantec
Flow with particles
d (known)
t (measured)
Signal
Time
Laser
Bragg
Cell backscattered light
measuring volume
Detector
Processor
On the left : Doppler burst.
After High-pass filtering, one obtains the signals on the right
The Doppler frequency can be computed by using different techniques :
Counter (on N zero crossing)
Spectral analysis
The quality of the signal must be optimised (Laser power, particles, optical parameters )
The SNR depends on the measurement volume domain crossed by the particle
Important : ONE PARTICLE ONLY in the measurement volume
This fixes the maximum measurable frequency
About seeding
Outline
Laser Doppler Anemometry (9) Spatial resolution : Measurement volume
The transmitting system
generates the measurement
volume
The measurement volume
has a Gaussian intensity
distribution in all 3
dimensions
The measurement volume is
an ellipsoid
Dimensions/diameters
x,
y
and
z
are given by the 1/e
2
intensity points
f
D
L
Y
Z
X
Transmitting
system
Measurement
volume
Intensity
distribution
0
1/e
2
1
y
X
Z
Y
Source : Dantec
Length :
Width : Height :
x
X
Z
f
( ) 2
4
u t
o
cos D
f
L
x
=
L
y
D
f
t
o
4
=
( ) 2
4
u t
o
sin D
f
L
z
=
Number of fringes :
Fringe separation :
( ) 2 2 u
o
sin
f
=
( )
L
f
D
tan f
N
t
u 2 8
=
Source : Dantec
Volume :
( ) ( ) 2 2 3
4
3
u u
o
t
cos sin
V
x
d
=
Typical values : Parameters of a Phase Doppler Anemometer system
(Bore, Ishima, Flour - 2001)
Transmitting Optics
Colours of the beams : Green for U velocity / Blue for V velocity
Gaussian beam diameter : 1,4 mm
Focal length of the front lens 600 mm
Beam separation 38 mm
Diameter of the measurement volume G :281 m / B : 266 m
Length of the measurement volume G :8,9 mm / B : 8,4 mm
Fringe number 35
Fringe spacing G :8,1 m / B : 7,7 m
Shift frequency 40 MHz
Note the very long measurement volumes which would imply
a significant spatial averaging
Shorter measurement volume by using a pinhole and a Side scatter (off-
axis) configuration
But difficult to align and vibration sensitive
Effective measurement volume in the example
Off axis angle -64
Front lens focal length 310 mm
Width of the slit 100 m
Magnification of receiving optics 2
Effective length of the measurement volume 200 m
Transmitting
optics
Flow
R
e
c
e
i
v
i
n
g

o
p
t
i
c
s
w
i
t
h

D
e
t
e
c
t
o
r
Projected pin-hole
Measurement
volume
Laser
Signal
processing
Transmitting
optics
Receiving optics
with detector
Signal
conditioner
Flow
HeNe
Ar-Ion
Nd:Yag
Diode
Beamsplitter
(Freq. Shift)
Achrom. Lens
Gas
Liquid
Particle
Achrom. Lens
Spatial Filter
Photomultiplier
Photodiode
Spectrum analyser
Correlator
Counter, Tracker
Amplifier
Filter
PC
( Source Dantec )
3 Components LDA Fibre optical system
(2C systems are more common and much easier to set-up)
About seeding
Outline
Laser Doppler Anemometry (14) About seeding
LDA measures the velocity of single particles
One needs to detect the particles (!)
Dynamical behaviour of the particles in an unsteady flow ?
The time sequence of the velocity signals (particles flowing through the
control volume) depends on the spatial repartition of seeds in the
turbulent flow.
Homogeneity of seeding is an important condition. i.e. NO correlation between
particle density and flow velocity.
Inhomogeneously seeded flows are frequent : Simple example of the seeded free jet
issuing in a non seeded ambient !
Exemple of an inhomogeneous
seeded jet for flow visualisation
Scattered light intensity (Intensity of the scattered light -Logarithmic scale - as a function
of the relative angle between the Laser beam and the receptor)
Forward scatter configuration leads to the best SNR
The intensity is a minimum near 90 and 270
The backscattered intensity is 100 to 1000 times smaller
180 0
90
270
210
150
240
120
300
60
330
30
180 0
330 210
240 300
270
150
120
90
60
30
180 0
210
150
240
120
270
90
60
300
30
330
d
p~
0.2
d
p~
1.0
d
p~
10
Argon Laser : Green l=514,5 nm ; Blue l=488,3 nm / He-Ne Laser : l=632,8 nm

Dynamical behaviour of the particles in an unsteady flow ?
For particles much denser than air (
f
<<
p
), under the hypothesis of a Stokes flow,
the dynamical equation becomes :
The particle response function to the dragging by the fluid turbulence leads to a
filtering effect of the turbulence energy density spectrum (Hinze 1975) with :

p
as to be compared with the kolmogorov time scale
For example in air (standart conditions) :
Silicone Oil (
p
= 900 kg/m
3
) : d
p
=1 m f(-3dB) = 57 kHz
Glycerin droplets (
p
= 1260 kg/m
3
) : d
p
=1 m f(-3dB) = 41 kHz
TiO
2
particles (
p
= 4260 kg/m
3
) : d
p
=1 m f(-3dB) = 12 kHz
( )
d
U U
d
U
dt
d
p p
p f p
p p
p
18
18
2
2
= =
( )
( )
( )
p
p
f
p
dB f
) f (
f E
f E
2 1 3
2 1
1
2
=
+
=
t t
( )
2 1 -
l k
R u l ~ t
Spatial repartition of seeds :
V
d
: Volume of the measurement volume
P(k) : Probability to have exactly k particles in V
d
C
p
: Averaged particle concentration (number/m
3
)
The mean number of particles in V
d
is = C
p
. V
d
For a statistically homogeneous seeding, P(k) is described by a Poisson distribution
:
For LDA measurements, no more than one particle should be present in the
detection volume at one time. To keep this probability below 0,5%, one note that :
This condition is an upper bound for C
p
( )
( )
= e ! k k P
k
( ) ( ) 005 0 1 1 1 1 0 , e k P . V . C
d p
< + = > < =

Inter-arrival time statistics :
For a constant convective velocity component 
x
perpendicular to the fringes,
the mean number of particle N
p
flowing through the detection volume per unit time
is : N
p
= C
p
. 
x
.A
d
A
d
is the surface of the measurement volume perpendicular to X axis.
The arrival time probability density distribution is then exponential with :
Note however that LDA only detect data for (one particle in detection volume) :

x
is the diameter of the measurement volume along X axis.
According to Benedict et al (2000), this poisson arrival time distribution is valid
even for high turbulent intensity (up to 40%). Note however that it should be
distorded in zero mean velocity turbulence.
( )
( )
i
t
j
t
p
N
p ij
e N t P

= A
( )
| | time dead Processor + > = A
x
x
i j ij
U
t t t
o
Inter-arrival time statistics :
In a flow with reasonable turbulent intensity, a plot of P(t) is a good test of the
homogeneity of the seeding.
On the right hand side, one sees clearly a cut-off due to the finite size of the
measurement volume and to the processor dead time.
This cut-off is a strict upper-bound for spectral analysis using LDA data.
=17m/s ; u= ; Courtesy of C. Hoarau - LEA
D.D.P de Poisson
10
100
1000
10000
0 0,0005 0,001 0,0015 0,002 0,0025 0,003
t (s)
D
.
D
.
P

(
1
/
s
)
MESUREE
THEORIQUE
D.D.P des temps d'attente
0
1000
2000
3000
4000
5000
6000
7000
8000
1,0E-06 1,0E-05 1,0E-04 1,0E-03 1,0E-02
t (s)
d
.
d
.
p

(
1
/
s
)
LDV
tdel = 17 micros
About seeding
Outline
Laser Doppler Anemometry (20) Statistical Analysis using LDA data
This important part is not developed in the present course.
A very detailed, recent and valuable course is available in Von Karman Institute
Lecture Serie 2002-04 Post-processing of experimental and numerical data April
22-26, 2002. See also Albrecht et al 2003
See in particular the contributions of C. Tropea ; D. Veynante ; A. Boutier in this
VKI volume
Statistical analysis for LDA data concerns mainly :
Estimation of velocity moments at a point
Correlation and spectral density estimation
Some recent work is devoted to Two-points spatio-temporal correlations.
To estimate velocity moments, caution is required to :
Get statistically converged data
The number of independent data required is very large for high order moments (see
Benedict and Gould 1996)
Minimise the bias on velocity statistics due to possible velocity/data rate correlation.
This bias has been known for a long time. The first step is to check if there exist a
correlation between velocity and inter-arrival times.
Such a correlation exist if the seeding is dense (high data rates). High velocities are then
sampled more often than low velocities.
Inter-arrival time weighting, re-sampling procedures are then correct, robust and valid
also for inhomogeneous seeding. Transit time weighting is valid for homogeneous
seeding only.
If there is no correlation between particle arrival time and velocity. Ensemble averaging
is better as most statistical weighting over-correct the data.
To estimate correlation or spectral density :
Difficulty due to randomly sampled signals
Well established theorems for regularly sampled data fail (Nyquist criterium).
As small inter-arrival times are the most probable, the signal contains information
about very high velocity fluctuations
A very large research activity (see review in Benedict et al 2000)
Modern techniques perform very well up to (or above !) the mean data rate.
Two main routes for spectral estimation :
Slotting : Direct estimation of the correlations where the correlation lag time is
descretised into bins
Regular resampling of the data and fourrier transforms : The Sample and Hold technique
(S+H) is the more frequently discussed because analytical corrections have been
proposed (Adrian & Yao 1987 ; Nobach et al 1998 ; Simon & Fitzpatrick 2003)
Image acquisition. Constraints
Image processing to educe particle displacement.
Spatial resolution, Dynamic ranges.
Introduction to advanced iterative methods
Outline
Particl Image Velocimetry (1) Principle of the technique
PIV is an Imaging technique.
Uses two frozen views ( at t & t+t) of a field of solid or liquid particles carried by a
flow
Estimate the corresponding
displacement field
Information about the average
velocity field along trajectories
traversed in the time t
Low pass filtered representation of the velocity field
Important : Tracers must follow the flow
( ) ( ) ( )
)
=
' '
'
, ' ' , ' ;
t
t
dt t t X U t t X D
G G G G
( ) ( ) t t t t X D t X U o o o + = , ; ,
G G G G
Typical optical
configuration :
Result : Instantaneous velocity field in a plane.
With one camera : Two components of velocity
With Two cameras : Three components of velocity
Recent progress with Holography
Important steps :
1 : Acquisition of two relevant images at t and t+t
2 : Processing of images to educe displacement
(J. Westerweel
for Dantec)
Ingredients (After J. Westerweel in www.dantecmt.com ) :
FLOW
RESULT
seeding
illumination
imaging
registration
sampling
quantization
enhancement
selection
correlation
estimation
validation analysis
Interrogation
Acquisition
Pixelization
Image acquisition.
Outline
Particl Image Velocimetry (4) Image acquisition
Optical parameters
Index matching often performed in variable density flows with high density
differences (see Alahyari & Longmire 1994, Ramaprabhu & Andrews, 2003)
Camera :
focal length f
Aperture D (f/D number)
magnification M
(J. Westerweel for Dantec)
Laser sheet :
Wave length
Intensity I
0
Thickness z
0
Important quantities (Vesterweel 1993, Adrian 1991) :
Object focal depth :
All observed particles are in focus if z
0
< z
Image diameter d
t
of a particle of real diameter d
p
:
d
s
is the diffraction limited spot diameter. (d
s
>> M d
p
for small tracers !)
For example : for =532 nm ; M 0.5 and f/D 5.6 one gets : d
s
11 m and z 600 m
The value of d
t
is a very important parameter of the technique.
The optical setting of the camera has to be optimised
2
2
2
1
1 4
D
f
M
z
o
|
|
.
|
\
|
+ ~
( )
2
2 2 2 2 2 2
1 44 2
(
+ + = + ~
D
f
M . d M d d M d
p s p t

Two non dimensionnal numbers (Adrian 1991) :
Source density :
N
s
is the number of particles in the volume of the object plane corresponding to one
particle image. C
p
is the tracer concentration (m
-3
)
If N
s
<<1, the average distance between particles is much larger than the particle-
image diameter and the image consist in isolated particle-images. PIV is performed
in this particle-image mode.
Image density :
N
I
is the number of particles in the volume of the object plane corresponding to one
interrogation region of size
I
.
N
I
is an important parameter for the processing techniques that is optimised on the
spot.
2
2
0
4
M
d
. z . C N
t
p s
t
A =
2
0 I p I
. z . C N A A =
Example : Experimental study of the breakdown of a Compressed Vortex (Maurel
2001, Bore et al 2002)
Two images at
t and t+t in a
noisy environment.
Use of background
substraction in this situation
Image acquisition.
Outline
Particl Image Velocimetry (8) Computing the particle displacement
Two situations depending on image density :
We focus here on PIV techniques.
Impossible to recognise individual tracers. One needs a statistical approach
A very important research activity concerning analysis methods (See review paper of
Scarano 2002 ; Stanislas 2002)
Basic intercorrelation method is presented first (well accepted reference Benchmark
method)
2
0 I p I
. z . C N A A =
High image density N
I
P.I.V.
Particle Image
Velocimetry
Low image density N
I
P.T.V.
Particle Tracking
Velocimetry

The two images are divided in interrogation regions :
The particle displacement is estimated locally by means of spatial cross-
correlation.
( Instant t )
( Instant t+t )
Principle of spatial cross-correlation : (after J. Westerweel for Dantec)
level grey : ; elsewhere 0 n in window 1
: n correlatio - Cross Spatial
2 2 1 1
) x ( I , i ) x ( W
x d ) s x ( I ) s x ( W ) x ( I ) x ( W ) s ( R
i i
K K
K K K K K K K K
= =
+ + =
)
: Particles at instant t
: Particules at instant t+t
A
Computation for A particle in
Window 1
Integration over the whole
window
(x and y : components of the displacement vector s)
Spatial cross-correlation :
Often computed by using
FFT algorithms on 32x32
pixels windows
Normalised cross-correlation are
sometimes preferred (see Scarano 2002)
Displacements are not multiples
of Pixels !
Evaluation of the maximum to reach a sub-pixel accuracy
Sub-pixel interpolation.
Often performed by using 3 x 3 pixels near the maximum to limit noise effect and using
either : (Westerweel 1993, Lecordier 1997)
Centre of mass determination (less accurate)
2x 1D parabolic interpolation
2x 1D Gaussian interpolation (best method)
The result is an estimation of the displacement on an individual window.
(From Maurel 2001)
Spatial cross-correlation :
1/e
2
width (w) of the correlation
peak ?
Mathematical analysis on a Dirac field :
The peak is a Dirac (westerweel 1993)
Optical response of the system
Pixelisation Convolution of the peak with a 2 base pyramid ( is
the pixel width.)
Induces a non-symetrical deformation of the peak which
is a maximum for sub-pixels displacements of (1/2). Peak-
Locking Bias effect particularly important if d
t
(Westerweel 1998)
2
t
d w ~
(From Maurel 2001)
w
?
Study of mean Bias and noise of the estimator for example on synthetic
images. (Image : Courtesy of B. Lecordier CORIA Rouen
Computations by Maurel 2001 on 32 x 32 interrogation windows)
The Bias and rms increases because more particles leave the window for increasing
displacement (In plane Loss of pairs)
Note the strong Peak locking phenomena
Typically d0.1 Pixels for displacements lower than of window size for the basic
intercorrelation method.
-1.00E-01
-8.00E-02
-6.00E-02
-4.00E-02
-2.00E-02
0.00E+00
2.00E-02
4.00E-02
6.00E-02
8.00E-02
1.00E-01
0 2 4 6 8 10 12 14 16
Dplacement impos (pixels)
E
r
r
e
u
r
s

(
p
i
x
e
l
s
Biais
Ecart-type
In a real turbulent field for standard
analysis :
In-plane loss of pairs
Could be optimised by shifting
the interrogation window (see after)
Out of plane loss of pairs
Particles leaving the laser sheet
because transverse fluctuations (w)
Limit t to w. t < z
0
/4
Local deformation and rotation leads to a broadening of the correlation peak. It can
even induce multiple peaks for large deformations
By requiring that the relative broadening is lower than the peak width, (Westerweel 1993
, Keanes & Adrian 1990) obtain the criterium
(M : Magnification, U : velocity variation over the window size due to local gradients)
0.01 0.02 0.03 0.04 0.05 0.06
0.01
0.02
0.03
0.04
0.05
0.06
10 m/s
(From Maurel 2001, 2001a)
t
d t . U . M < A o
Dynamical ranges
The spatial response of the correlation analysis is similar to a top hat filter having
same dimensions as the interrogation window.
The number of pixels of the camera fixes the dynamical range for the resolved length
scales.
Typically for a 1280 x 1024 PCO camera and a 32 x 32 pixels
2
interrogation window : D
l
40.
More vectors can of course be obtained by window overlapping but D
l
remains the same
Increasing D
l
is the goal of super-resolution techniques
The dynamical range for the resolved velocities D
U
is the ratio of maximum to
minimum displacement
Typically for s
max
= 32/4 = 8 and s
min
= 0.1 pixel : D
U
= 80
This low value can be increased (by a factor of 2 or more) by using modern analysing
techniques. Mind however out of plane losses.
These values should be compared with the range of length , velocity and gradient
scales in a turbulent flow.
Spurious vectors (or outliers)
Always probable for the best
experimental set-up.
The spatial repartition of the particles
follow a Poisson distribution
Always a finite probability NOT to
have enough pairs in a window
From experience, models and Monte-carlo
simulation, Spurious vectors are very sensitive to :
Particle mean concentration C
p
. Has to be optimised in the experiment
Loss of pairs
Classical elimination procedures are based on :
Signal/noise ratio of the correlation peak (>1.1)
Norm of the vectors (above a given threshold)
Median filtering : Comparison between the vector and the median (NOT the mean) of its
neighbours ; then elimination above a given threshold.
(From Maurel 2001)
100 200 300 400 500
V1
100
200
300
400
500
V
2
Image acquisition.
Outline
Particl Image Velocimetry (17) Introduction to advanced methods
PIV is a rapidly improving technique
Rely on non-specialised components
Widely applied in the laboratories and in the industry
On the technological side, improvement of :
Laser technology
High resolution, low noise CCD captors
One challenge for turbulent flows is time-resolved PIV
On the image processing side. Goals :
Enhance velocity and spatial resolution
Important problem of velocity gradient compensation
Find reliable and precise algorithms for stereoscopic PIV. This aspect is out of the
scope of this course but refer to (Prasad 2000, Calluau & David 2003)
Note the recent worlwide project for the assessment of PIV performances (Stanislas
2002, 2003 ; http://www.pivchallenge.org/)
Iterative Schemes. No window deformation
Multi step cross correlation with a discrete (Westerweel 97) or sub-pixel (Lecordier
97) window offset.
Correspond to a central difference differentiation.
Can also be used for further improvement of the resolution by decreasing the size after
the first convergence (Scarano & Riethmuller 1999)
First pass : computation
of the displacement using
standart method.
Iterative shift of the interrogation
window down to zero
(+/- !) residual displacement
Iterative Schemes. Sub-pixel window offset
Bias and noise of the estimator for example on synthetic images. Example of a
uniform displacement along the second bissectrice (Image : Courtesy of B. Lecordier CORIA
Rouen Computations by Maurel 2001 on 32 x 32 interrogation windows)
Very significant improvement of :
Biais problem because in plane loss of pairs is minimised
Peak-locking problems (for sub-pixel shift only) because the correlation peak appears
exactly at the origin after the convergence
-1.00E-01
-8.00E-02
-6.00E-02
-4.00E-02
-2.00E-02
0.00E+00
2.00E-02
4.00E-02
6.00E-02
8.00E-02
1.00E-01
0 2 4 6 8 10 12 14 16
Dplacement impos (pixels)
E
r
r
e
u
r
s

(
p
i
x
e
l
s
Biais
Ecart-type
-1.00E-01
-8.00E-02
-6.00E-02
-4.00E-02
-2.00E-02
0.00E+00
2.00E-02
4.00E-02
6.00E-02
8.00E-02
1.00E-01
0 2 4 6 8 10 12 14 16
Dplacements imposs (pixels)
E
r
r
e
u
r
s

(
P
i
x
e
l
s
Biais
Ecart-type
First pass on 64x64 pixel window.
Iteration on 32x32 window. Stop for residual
displacement lower than 0,005 pixels
Standart algorithm
Iterative Schemes. Linear displacement predictor takes into account the
linear terms in the deformation
The linear terms of the displacement spatial distribution are taken into account in
the correlation
Increase by a factor greater than 2 the pixel/pixel maximum acceptable deformation rate
(more than 1 pixel/pixel) (see Huang, Fiedler, Wang 1993, Scarano 2002)
First pass : computation
of the displacement using
standart method.
Iterative shift + deformation of the
interrogation window down to zero
(+/- !) residual displacement
A wealth of new ideas / techniques for image processing.
Other approaches based on :
Correlation image velocimetry (CIV) (for example Fincham & Delerce 2000)
Optical flow computation (Quenot et al 1998)
To address the problem of size resolution :
When decreasing the window size, the image density decreases. The signal/noise
ratio decreases (random correlation peaks and random errors in su-pixel
interpolation)
Other techniques combine first a standart PIV approach and then a particle tracking
approach on refined windows. Usual name : Super-resolution PIV (Stitou & Riethmuller
2000, 2002)
Methods for the measurement of concentration
Laser Induced Fluorescence (LIF)
Theory and application
Post-processing of LIF signal
PLIF for mixing measurements of a jet submitted to unsteady cross-flow
Temperature measurement using LIF
Outline
Concentration measurements methods
Intrusive measurements
-Ionisation probe
-chromatography
Optical techniques (Stepowski 1992)
Elastic scattering (
signal
=
laser
)
- Rayleigh scattering
- Mie scattering (Sautet & al 1994)
Inelastic scattering (
signal

laser
)
-Laser Induced Fluorescence (LIF)
-Spontaneous Raman scattering
Optical methods to measure spatial density fields
Advantage of LIF for concentration measurements
-
LIF

laser
rejection of elastic scattering after spectral filtering
(wall reflection, Mie scattering )
- Molecular seeding : tracers follow the flow
- Quantitative measurements after calibration
- Local and instantaneous measurements
Application fields of LIF
-Concentration measurements
liquids (spray, pollutant, mixing, )
gas (mixing, density)
flame (radicals OH CH NOx)
-Temperature measurements
PLIF for mixing measurements of a jet submitted to unsteady cross-
flow
Outline
Camera lens
pass-band filter
Intensified
CCD camera
Variable density jet
(seeded with fluorescent gas)
Pulsed or continuous LASER
Argon
YAG laser
Excimer Kr-Fl
Laser sheet
Optical methods to measure spatial density fields Typical experimental setup
Choice of gaseous acetone as fluorescent tracer
- Small variation of the fluorescent yield with T and P.
- High value of P
sat
(27.5 Kpa at 20C)
- Non toxic
- Low boiling point (53C)
- Absorption spectrum adapted to laser wavelength
Optical methods to measure spatial density fields fluorescent tracer
2 states model
E1
E2
Absorption
B
12
.u
Stimulated
emission
B
21
.u
Quenching
Q
21
Fluorescence
A
21
Ground state
Excitated
state
o
Quenching
(collision)
1> o
Fluorescence
o
Stimulated emission
Fluorescence energy ~ 1e-3 Excitation energy
Optical methods to measure spatial density fields Theorical considerations
0
( , ) ( , , )
. ( , , ). . ( , ). ( , , ). ( , , , ).
x
LIF
T C y z d
S Io x y z T C x y z P T C dV
e
o k k
o o |
=
)
=
S
LIF
proportionnal to concentration of the excited molecules in dV if:
- absorption is neglected
- and are constant
- I
o
is small enough to avoid saturation of fluorescence transition
: Optical efficiency (lens, filter, CCD sensitivity,)
Io: Incident energy flux for one pulse (J/m
2
)
: Absorption coefficient
: molecule cross section ( m
2
)
C: Seeding molar concentration (Mole/m
3
)
: fluorescence yield
dV : measurement volume: pixel section * sheet thickness (m
3
)
x: direction of laser beam
Optical methods to measure spatial density fields LIF signal expression
Expression of LIF signal
Absorption of incident laser beam
Absorption coefficient : (Beer-Lambert law)
Caracteristic length : attenuation of e
-1
(63%) => l=1/(C)
attenuation of 5% => l=1/(3C)
Gaseous acetone (P
v
=P
sat
(20c)) l
e-1
=2.82m ; l
5%
=90 cm
Small absorption can be corrected using the Beer Lambert law
0
( , ) ( , , )
x
T C y z d
e
o k k
)
=
concentration field
Axis concentration
Optical methods to measure spatial density fields Validation
Validation : axisymetric turbulent air jet
Evolution of the LIF signal as a function
of acetone concentration
Evolution of the LIF signal as a function
of the energy of the laser beam
Linearity of LIF signal
Optical methods to measure spatial density fields Validation
Effect of pressure on the acetone
fluorescence signal (Thurber 1999)
Effect of temperature on the acetone
fluorescence signal (Thurber 1999)
Optical methods to measure spatial density fields fluorescence signal
Choice of the laser wavelength
cross section (Thurber 1999)
fluorescence yield (Thurber 1999)
Optical methods to measure spatial density fields fluorescence signal
I) Signal acquisition in ICCD S
f
=S
LIF
+S
noise
Image acquisition with maximum signal intensity but whitout saturation
II) Laser sheet energy correction
Laser sheet energy difference (wing/center) less than 30%
III) C
m
= .S
f
with
Calibration with the same optical system and homogeneous mixture
with C
ref
known and C
ref
in the same order of C
m
measured
IV) C
m
=> Y
f
at variable density
ref
ref
C
S
| =
! Spacial cut-off frequency of the intensificator
Optical methods to measure spatial density fields Post- processing
S
f
=S
LIF
+S
noise
=S
LIF
+ (S
background noise
+S
equipment noise)
Image noise
Instantaneous image
Spacial correction ri
o
Laser sheet energy, optical aberration,
non uniformity ICCD
Instantaneous - noise
/
Mass concentration field
*
f
ref
ref
S
C
S
flow
Outline
Optical methods to measure spatial density fields Jet in unsteady cross flow(Ben 2001)
Camera sensicam 1280*1024, 12 bit, dynamic(NG
max
/NG
noir
) = 40
Intensificator fragment ( * 120 intensification, gate 100ns)
Laser Yag : 4 266 nm, 40 mJ/pulse, 10 Hz, 7 ns
Pass band filter : BG25 (cut green and UV scattering)
Objectif : NIKON Nikkor, 58, f1.2
Syntetic quartz window for UV transmission (suprasil)
Optical methods to measure spatial density fields Acquisition system
Error source
-Laser sheet energy fluctuation (laser +/- 5%)
-Seeding
-Reception (Intensification)
-Laser sheet multiscattering (532 nm/ 266 nm)
-Fluorescence of other molecule
Optical methods to measure spatial density fields Measure accuracy
Validation
-Mean mass flow conservation (PIV-LIF coupling)
Injected mass flow rate
4,8 g/s
Calculated mass flow rate
4,51 g/s
Air/fuel(CNG) mixture evolution in intake manifold
100
75
50
25
Richesse=1
0
- Continous injection
- Pulsed injection
Impinging jet ( Fujimoto & al 1997)
Optical methods to measure spatial density fields Jet behavior
3D reconstruction
flow
Outline
Optical methods to measure spatial density fields Temperature measurement
Simultaneous measurement of concentration and temperature
If 2 parameter (,T, C) are varying => 2 fluorescence signals
(Sakakibara & al 1999)
Optical methods to measure spatial density fields Temperature measurement
Principle: 2 fluorescent dyes (rhodamine B et rhodamine 110) and one laser wavelength (488) =>
1 fluorescent dye (acetone) with two laser wavelengths (308 et 248) =>
1
2
( )
c
c
Sf
f T
Sf
=
1
2
( )
Sf
f T
Sf
=
(Sakakibara & al 1999) (Thurber 1999)
Validation in a stratified flow
application: convection
Optical methods to measure spatial density fields Temperature measurements
Adrian, R. J. (1975). On the role of conditional averages in turbulence theory. Proc. 4th biennal symp. on turbulence in liquids.
Patterson, G. and Zakin, J. (Eds), Science Press, Princeton, NJ.
Adrian, R. J. (1991). Particle Imaging technics for experimental fluid mechanics. Ann. Rev. Fluid Mech. 22: 261-304.
Adrian, R. J., Christensen, K. T. and Liu, Z. C. (2000). Analysis and interpretation of instantaneous turbulent velocity fields. Exp. in
Fluids 29: 275 - 290.
Adrian, R. J. and Yao, C. S. (1987). Power spectra of fluid velocities measured by laser Doppler velocimetry. Exp. in Fluids 5: 17-28.
Alahyari, A. and Longmire, E. K. (1994). Particle image velocimetry in a variable density flow: application to a dynamically evolving
microburst. Exp. in Fluids 17: 434-440.
Albrecht, Borys, Damaschkes and Tropea, C. (2003). Laser Doppler and Phase Doppler measurement techniques, ISBN 3-540-67838-
7, Springer.
Ben, L. (2001) : Etude de l'injection du carburant gazeux de substitution GNV dans les moteurs pistons. Imagerie laser, bilans
nergtiques et environnemental. Thse de doctorat, Institut National Polytechnique de Toulouse
Benedict, L. H. and Gould, R. D. (1996). Uncertainty estimates for any turbulence statistics. Eigth International Symposium on
applications of Laser techniques to Fluid Mechanics, Lisbon.
Benedict, L. H., Nobach, H. and Tropea, C. (2000). Estimation of turbulent velocity spectra from laser Doppler data. Meas. Sci.
Technol. 11: 1089 - 1104.
Bonnet, J. P. and Delville, J. (2001). Review of coherent structures in turbulent free shear flows and their possible influence on
computational methods. Flow, Turbulence and Combustion 66: 333 - 353.
Bore, J., Ishima, T. and Flour, I. (2001). The effect of mass loading and inter-particle collisions on the development of the
polydispersed two-phase flow downstream a confined bluff body. J. Fluid Mech. 443: 129 - 165.
Bore, J., Maurel, S. and Bazile, R. (2002). Disruption of a compressed vortex. Phys. Fluids ( Prof. J. Lumley 70th birthday Symp.
Papers) Vol. 14, Issue 7: pp. 2543-2556.
Calluaud, D. and David, L. (2003). Stereoscopic particle image velocimetry measurements of the flow around a surface-mounted
block. to appear in Exp. in Fluids.
Bibliography
Chassaing, P. (2000). Turbulence en mcanique des fluides. Analyse du phnomne dans une perspective de modlisation l'usage de
l'ingnieur, Editions Cepadues.
Eckbreth, A.C.: Laser diagnostics for combustion temperature and species Abacus press
Fincham, A. M. and Delerce, G. (2000). Advanced optimization of correlation imaging velocimetry algorithms. Exp. in fluids 29: 13-
22.
Hinze, J. O. (1975). Turbulence, McGRAWHill Company.
Huang, H. T., Fiedler, H. F. and Wang, J. J. (1993). Limitation and improvement of PIV, part II. Particle image distorsion, a novel
technique. Exp. in Fluids 15: 263-273.
Keane, R. D. and Adrian, R. J. (1990). Optimisation of particle image velocimeters. Part I: Double pulsed systems. Meas. Sci.
Technol. 1: 1202-1215.
Lecordier, B. (1997). Etude de l'interaction de la propagation d'une flamme prmlange avec le champ arodynamique par
association de la tomographie laser et de la vlocimtrie par image de particules., Facult des Sciences de l'Universit de Rouen.
Lumley, J. L. (1967). The structure of inhomogeneous turbulence. Atmospheric turbulence and radio wave propagation.
Maurel, S. (2001). Etude par imagerie laser de la gnration et de la rupture dun coulement tourbillonnaire compress. Situation
modle pour la validation de simulations aux grandes chelles dans les moteurs, PhD Thesis I.N.P. Toulouse N1780.
Maurel, S., Bore, J. and Lumley, J. L. (2001). Extended proper orthogonal decomposition : Application to jet/vortex interaction.
Journal of Flow, Turbulence and Combustion 67: 125-136.
Nobach, H., Mller, E. and Tropea, C. (1998). Efficient estimation of power spectral density from laser Doppler anemometer data.
Exp. in Fluids 24: 499-509.
Prasad, A. K. (2000). Stereoscopic particle image velocimetry. Exp. in Fluids 29: 103-116.
Quenot, G. M., Pakleza, J. and Kowalewsky, T. A. (1998). Particle image velocimetry with optical flow. Exp. in fluids 25: 177-189.
Ramaprabhu, P. and Andrews, M. J. (2003). Simultaneous measurements of velocity and density in buoyancy-driven mixing. Exp. in
Fluids 34: 98-106.
Sakakibara, J. and Adrian, R. J. (1999). Whole field measurement of temperature in water using two-color laser induced fluorescence.
Exp. in Fluids 26: 7-15.
Bibliography
Sautet, J.C. and Stepowski, D. (1994) : Single shot laser mie scattering measurements of the scalar profiles in the near field of
turbulent jets with variable densities Experiments in fluids P353-367
Scarano, F. (2002). Iterative image deformation methods in PIV. Meas. Sci. Technol. 13: R1-R19.
Scarano, F. and Riethmuller, M. L. (1999). Iterative multigrid approach in PIV image processing. Exp. in Fluids 26: 513-523.
Simon, L. and Fitzpatrick, J. (2003). Cross-power spectra estimation using laser Doppler annemometry measurements. to appear in
Exp. in Fluids.
Stanislas, M., Okamoto, K. and Khler, C. (2003). Main results of the first international PIV Challenge. To appear in Meas. Sci.
Technol.
Stanislas, M. (2002). Principaux rsultats du premier challenge international sur la PIV. 8me congrs Franais de vlocimtrie laser,
Orsay, 17-20 septembre.
Stepowski, D. (1992): Laser measurements of scalar in turbulent diffusion flames. Progres Energy Combustion Science Vol 18
Stitou, A. and Riethmuller, M. L. (2000). Extension of PIV to super resolution using PTV. 10th International symposium on
applications of Laser techniques to fluid mechanics, Lisbon, july 10-13.
Stitou, A. and Riethmuller, M. L. (2002). Progress in PIV-PTV methods: application to concentration measurements. 11th
International symposium on applications of Laser techniques to fluid mechanics, Lisbon, july 8-11.
Thurber, M.C. (1999) Acetone laser induced fluorescence for temperature and multiparameter imaging in gaseous flows. Technical
report Standford university
Von-Karman-Institute (2002). Lecture Serie 2002-04 : Post-processing of experimental and numerical data, ISBN 3-540-67838-7,
Springer.
Westerweel, J. (1993). Digital particle Image Velocimetry, PhD dissertation, Delft University Press.
Westerweel, J. (1998). Effect of sensor geometry on the performance of PIV interrogation. Ninth International symposium on
applications of Laser techniques to fluid mechanics, Lisbon.
Westerweel, J., Dabiri, D. and Gharib, M. (1997). The effect of a discrete window offset on the accuracy of cross-correlation analysis
of digital PIV recording. Experiments in fluids 23: 20-28.
Zhang, Z. and Eisele, K. (1998). The effect of astigmatism due to beam refractions in the formation of the measurement volume in
LDA measurements. Exp. in Fluids 20: 466-471.
Bibliography

BOREE Presentation

Uploaded by

Copyright:

Available Formats

BOREE Presentation

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

BOREE Presentation

Uploaded by

Copyright:

Available Formats

Optical methods for the measurement of

constant and variable density flows.

Small scales properties depend on

Laser Doppler Anemometry (6) Principles of the technique

Argon Laser : Green l=514,5 nm ; Blue l=488,3 nm / He-Ne Laser : l=632,8 nm

Particl Image Velocimetry (6) Image acquisition

Particl Image Velocimetry (9) Computing the particle displacement

You might also like