Background Oriented Schlieren: Experimental Aerodynamics Division

BACKGROUND ORIENTED
SCHLIEREN
L Venkatakrishnan
Experimental Aerodynamics Division
Lecture Course on
Advanced Flow Diagnostic Techniques
National Aerospace Laboratories
Bangalore
17-19 Sept. 2008
Background Oriented Schlieren 1
CONTENTS
1. INTRODUCTION .............................................................................................................2
2. BACKGROUND ORIENTED SCHLIEREN ...................................................................4
3. FILTERED BACK PROJECTION TOMOGRAPHY ......................................................7
4. VALIDATION.................................................................................................................11
5. APPLICATION NOTES .................................................................................................21
REFERENCES ........................................................................................................................26
Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

1. INTRODUCTION
Development of new flow diagnostic tools that are non-intrusive, quantitative as well as
applicable to real life full-scale flows is an important area of experimental aerodynamics
research. There is considerable research activity to integrate new technological developments
in various fields and to evolve new techniques of flow quantification. While all optical
techniques to study density fields in transparent media (usually gases or liquids) depend on
variation of the index of refraction in the medium and the resulting effects on a light beam
passing through the test region, quite different quantities are measured with each one.
Techniques like Schlieren and shadowgraph provide qualitative information on first and
second derivative of density respectively. Interferometry on the other hand, provides
quantitative information of density field but setting up such instrumentation requires
tremendous care.
The choice of lasers as a source of illumination, improvements in CCD technology and fast
computers for image acquisition and analysis have resulted in a resurgence of quantitative
optical methods. In recent years there have been some attempts (McMackin et al. 19951,
Dalziel et al., 19982, 20003, Meier 19994, Sutherland et al., 19995, Onu et al., 20036) to
quantify the density or density gradient field using broadly the Schlieren principle. Some of
the common steps in such a procedure include: imaging (with a camera) of a background
through a flow of interest and software analysis (which can differ in details) providing finally
the density or density gradient field. McMackin et al. (1995) 1, who used a Hartmann sensor.
The Hartmann sensor uses a lenslet array to split a single laser beam into a number of sub
apertures and detects their individual focal spot intensity pattern changes which occur by the
optical path differentials of the phase object. Thereafter tomographic techniques are used to
reconstruct the three-dimensional density field. A method called ‘synthetic Schlieren’ was
proposed by Dalziel et al., (1998, 2000 2,3 )and applied to measure the amplitudes of waves
generated by an oscillating cylinder in a stratified flow and agrees very well with theoretical
values (Dalziel et al., 20003 and Sutherland et al., 19995). Recently, Onu et al.,6 (2003) used
simple inverse tomographic techniques to adapt synthetic Schlieren to measure the amplitude
of axisymmetric internal waves generated by an oscillating sphere and compare it with
vertical displacements obtained from a conductivity probe.

Meier4 (1999) proposed the Background Oriented Schlieren (BOS) technique, which has the
potential to determine the density field. BOS can be described as a simple Schlieren
technique, based on image displacements of some photographic background by Schlieren in
the transfer channel (Meier 19994). A major advantage of this technique is that it requires
only a digital still camera with adequate resolution. The principle of Background Oriented
Schlieren is the same as that of synthetic Schlieren in exploiting the bending of light rays as
they pass through a density gradient. After obtaining the displacement due to the bending of
light rays, the line-of-sight integrated density field is calculated and tomographic techniques
used to obtain density field in a plane of interest. This differs from the work of Onu et al., 6
(2003) in the filtered back-projection tomographic technique used here which, does not need
assumption of axisymmetry and obtains density field instead of wave amplitudes.
Earlier studies (Richard et al. 20007, Raffel et al. 20008, Meier 20029), have demonstrated
several possible applications of BOS; these include density fields of helicopter generated
vortices and supersonic jets.

2. BACKGROUND ORIENTED SCHLIEREN
The image of an object is the convolution of the object function and the transfer channel
function. Thus deconvolution will describe the transfer channel function, if the object and
image are given.
Two images of a deliberately structured background are obtained. The first is through the
undisturbed transfer channel and the second is through the phase object of interest. The
“difference” between these two images provides information about the phase object. The
field gradients in the path of the imaging rays cause the deflection of the light rays leading to
shifts in the image details.
The principle of the technique is the refractive index variation due to density gradients. This
relation between the refractive index n and the density ρ is given by the Gladstone-Dale
equation (Merzkirch 198710, Goldstein 198311):
n −1 (1)
= G (λ )
ρ
where G(λ) is called the Gladstone-Dale constant and depends on characteristics of the gas
(medium) and weakly on the frequency of light used (Merzkirsch 198710). Values for G(λ)
can be found from handbooks of physical chemistry. Here, a value of 0.23 x 10-3 m3/kg has
been used.
In the first step (Meier 19999), an image is taken of a random dot pattern with the air (phase
object) at rest. The second image is taken in the presence of the flow containing density
gradients. Cross-correlation of the two images yields the displacements of the particles in the
x and y directions. Here, x is along the freestream direction -also the axis of symmetry for the
present case – (with origin at model tip) and perpendicular to the line-of-sight direction (z), y
is the vertical axis (origin at model centerline).
Since the deflection of a single beam contains information about the spatial gradient of the
refractive index integrated along the axial path, the image deflection ε is defined as
Z D +ΔZ D
1 δn
ε=
no ∫ δy
dz
(2)
Z D −ΔZ D
with the assumption that the half-width of the region of density gradient ΔZD << ZD.
BACKGROUND PLANE
DENSITY GRADIENT
Δy’ LENS
IMAGE PLANE
θ ε
θ=0
ZD Δy
Z=0
ZB
Zi
Figure 1. Optical path for density gradient measurements by light deflection (Meier 1999).
Further, it is seen from geometry (see Figure 1) that the virtual image displacement Δy ' is
related to the image displacement Δy by the lens distance from the background ZB and the
image distance from the lens Zi which can be replaced for large ZB by the focal length f of the
lens (Meier 2002):
Δy ' Δy Δy
= =
ZB Zi f
For small deflection angles, ε can be approximated as

Δy '
ε=
ZD
hence,
ΔyZ B
ε= (3)
ZD f
From equations 2 and 3 it follows that for a given flow field, the shift of the background and
the sensitivity ( ( Δy grad n ) ~ Z D Z B f ) increases with increasing distance of the background
from the density gradient. Increasing focal length f is compensated by increasing the distance
between the camera and the background.

It thus follows that the obtained displacements are the density gradients at each point in the
field. The gradients of this displacement field yield an elliptic partial differential equation
also known as Poisson equation (see equation 17), solution of which yields the line-of-sight
integrated density distribution, which is projection of the three-dimensional density field in
the direction of viewing. The distribution of density in a given plane is then calculated by
means of a transformation from this projection.

3. FILTERED BACK PROJECTION TOMOGRAPHY

When a test field cannot be approximated as being infinite in the line-of-sight direction, a
mathematical method called tomography can determine two-dimensional density fields from
one-dimensional projected data sets. The test field for asymmetric or generally 3-D situations
must be interrogated from many different directions yielding many projections that must be
simultaneously deconvoluted. Stacking up these two-dimensional conversions will create a
fully three-dimensional image from the projected two-dimensional information. Usage of the
optical tomography technique in flow-visualization can be traced back over three decades
(e.g., Matulka and Collins, 197112). Progress in reconstruction methods (Hertz 198613, Faris
and Byer 198714, Cha and Sun 198915) has aided in new applications of the technique. The
most difficult problem in optical tomography is the lack of sufficient data for reconstruction,
which occurs when the views are truncated or inadequate in number (for non-axisymmetric
three-dimensional flows).
The two most popular methods for performing this transform involve either Fourier domain
(Fourier Convolution) methods or the algebraic solution to a non-unique system of equations,
termed Algebraic Reconstruction Technique (ART) (Muralidhar 200016). The Fourier domain
method implemented as convolution back projection method is simple but suffers from
reduced reconstruction accuracy when the amount of projection data is limited or restricted.
If the experimental test field is axisymmetric, only one projection is required and hence the
convolution is greatly simplified and termed as the Abel Conversion (Walsh and Kim
199517). While algebraic reconstruction methods are iterative and work better in cases of
limited projections, a major disadvantage is that they are time-consuming and the errors for
distributions may increase after a certain number of iterations, while the errors for projections
keep on reducing.
Although the Abel conversion seems suited for the data in the present work, it is highly
susceptible to noise, in addition to being unsuitable for non-axisymmetric flows. This is the
case with all back projection methods as back projection alone results in a blurred
reconstruction image. This is because a pure back projection adds a 1/r convolution to the
image (Ramachandran et al., 197118). Filtering must be applied to correct for this and obtain
an accurate image of the object. There are a number of choices in the type of filter to use. The
simplest and most mathematically rigorous one is the ramp filter:

H (ω ) = π ω (4)
However, this filter is very sensitive to noise. For this reason, other filters are chosen for
practical applications. The most commonly used filter is the Shepp-Logan filter, which
combines a sinc function (which is the continuous inverse Fourier transform of a rectangular
pulse of width 2π and height unity) with the ramp filter (Shepp and Logan 197419):
⎧⎪π ω sin c (ω ω max ) ω ≤ ω max

H (ω ) = ⎨ (5)
⎪⎩ 0 ω ≤ ω max
This filter results in a small amount of blurring, but is much less sensitive to noise.
Hence, a filtered back projection (FBP) algorithm has been used with a Shepp-Logan filter,
which in addition to noise suppression makes the method suitable for both axisymmetric and
non-axisymmetric flows.
l'
Figure 2. Configuration of projection on one view direction at θ (Feng et al20

 )
This algorithm reconstructs an image from parallel beam projections. In parallel beam
geometry, each projection is formed by combining a set of line integrals through an image at
a specific angle. An approximation to the image is then formed based on the projections.
In FBP, a planar cross-section (x,y) (see Figure 2), at certain z is used. For one view direction
at θ (θ =0 is z axis), the projection Pθ(l) is calculated from the displacements obtained from
the correlated images (Feng et al. 200120):
nλ
Δρ ( x, y ) dl
K ∫ray
Pθ (l ) = = (6)
here, l =x cos θ + y sin θ (see Figure 2)

∞ ∞
Pθ (l ) = ∫ ∫ Δρ ( x, y ) δ ( x cosθ + y sin θ − l )dxdy
−∞ −∞
(7)
Its Fourier transform is

∞
Sθ (ω ) = ∫ Pθ (l )e − jωl dl
−∞
∞ ∞
∫ ∫ Δρ ( x, y ) e
− j ( xω cosθ + yω sin θ )
= dxdy (8 & 9)
−∞ −∞
Assume
∞ ∞
∫ ∫ Δρ ( x, y ) e
− j 2π (ω1 x +ω2 y )
G (ω1 , ω2 ) = dxdy
−∞ −∞
Using the Projection Slice theorem (Radon, 1917 21 )

S (ω ) = G (ω cos θ , ω sin θ ) = G (ω , θ )
as well as
G (ω ,θ + π ) = G ( −ω , θ ) (10)
∞
h ( l ) = ∫ ω e jωl dω
(11)
−∞
where h(l ) is a filtering function to clip high frequency noise (described later),
we get
∞ ∞
Δρ ( x, y ) = ∫ ∫ G (ω , ω ) e 1 2
j 2π (ω1 x +ω2 y )
dxdy
−∞ −∞
π ∞
=∫ ∫ Sθ (ω ) ω e jωl dωdθ (12)
0 −∞
That is:
π (13)
Δρ ( x, y ) ≅ ∫ ⎡ ∫ Pθ (ζ ) h ( l − ζ ) dζ ⎤ dω
∞
⎢ −∞
0 ⎣ ⎦⎥
The distribution of Δρ ( x, y ) at a certain z level is calculated from the function h(l) and the
projection Pθ(l). Then the cross-sections can be stacked to form a 3-D distribution.
A more accurate result can be obtained by using more projections in the reconstruction. As
the number of projections (the number of views – for a steady flow simultaneous recording is

not necessary) increases, the reconstructed image more accurately approximates the original
image. In the present study however, the flow being axisymmetric, all the projections are
identical. The filtered back projection algorithm filters the projections using the Shepp-Logan
filter described earlier (equation 5) and then reconstructs the image using the filtered
projections.
In some cases, noise can be present in the projections. This noise, usually high frequency, has
to be clipped without affecting the density distribution. To achieve this, a method known as
reduced bandlimit for projection has been proposed and successfully used earlier for other
tomographic applications (Feng et. al, 200120). In this method, the function h ( l ) (equation
11) can be rewritten by making the bandlimited projection assumption, as follows:

∞
h ( l ) = ∫ H (ω )eiω l dω (14)
−∞
⎧ ω if ω < W
H (ω ) = ⎨ (15)
⎩ 0 otherwise
where W is the maximum frequency, i.e., bandlimit, which is usually half of Nyquist. The
influence of noise in the projections is assumed to belong mainly to the high frequency in the
Fourier domain. The bandlimit, W, is searched for each projection in the Fourier domain. The
W is determined at which the filtered projection [ h ∗ Pθ ] has less fluctuation. It works like a
low-pass frequency filter. On the other hand, the bandlimit W should be as large as possible.
Therefore, every projection (this test case being axisymmetric, a single projection suffices) is
assigned a different reduced bandlimit, hence termed “reduced bandlimit technique”. Further
details and the validation of the reduced bandlimit technique can be found in Feng et. al,
200120. During processing a Hamming window is employed for smoothing. The Hamming
window helps to avoid disconitinuities at the ends is to taper the signal to zero or near zero
and hence reduce the mismatch and is defined as (Oppenheim and Schafer, 198922):
⎪⎧0.54 − 0.46 cos ( 2π n ( N − 1) ) 0 ≤ n < N

ωn = ⎨ (16)
⎪⎩ 0 otherwise
where ω n is the frequency at the nth point in the time series.

Hence the influence of noise can be reduced by neglecting the high frequency, while the
density distribution may not be affected
The implementation of this method has been done using the inverse radon transform which is
well documented in literature (e.g., Kak and Slaney 198823).

4. VALIDATION
4.1. Experimental Setup
In this section, the application of BOS to determine first the density gradients and thereafter
the density, in a plane of an axisymmetric supersonic flow over a cone-cylinder is described.
The extraction of the plane of interest from the line-of-sight integrated density field has been
achieved by using a Filtered Back-Projection Tomography technique (outlined earlier). This
technique allows extraction of two-dimensional data sets from the projections of three-
dimensional flows when multiple views are obtained. The particular application in this paper
is to an axisymmetric flow – the axis of symmetry (x) being perpendicular to the line-of-sight
(z) - hence a single view is sufficient.
Validation of density measurements using the BOS technique is carried out (Venkatakrishnan
and Meier 200424) in a relatively simple flow at high speed. The method has been applied to a
cone-cylinder flow at Mach 2.0; this model configuration was chosen so that BOS
measurements could be compared with results available in Cone Tables (e.g. Sims, 1964).
The determination of the density field involves the following steps: a) calculation of
displacements in the background which is imaged through the flow of interest and these
displacements are the vectors of density gradient at each point; b) calculation of the line-of-
sight integrated density field by solution of the Poisson equation, which is the gradient of the
above displacement and c) use of optical tomography to determine the density field in the
actual plane of interest.
The experiments were carried out in the 0.3m x 0.3m supersonic test section of the trisonic
wind tunnel at NAL. In this tunnel, supersonic Mach numbers are produced using fixed
nozzle block liners. The model configuration was a 150 (semi-apex angle) cone-cylinder. This
was chosen since the density field behind the conical shock wave at zero incidence available
from Cone Tables (e.g., Sims 196425), can be utilized for comparison with the results
obtained by BOS. The model details are given in Figure 3a. Tests were made at a Mach
number of 2.0. The settling chamber total pressure was 2.1kg/cm2 and the Reynolds number
based on cylinder diameter (D=0.04m) was 2 x 104. Some of the broad features of the flow
like the shock wave angle, the features of expansion fan and surface Mach numbers are
included as well in Figure 3a.
For the BOS, a structured background to focus on was created by means of a normal random
number generator. This generated a matrix, 2000 x 2000 in size of random numbers whose
Expansion fan
500
Shockwave
26.50 M=2
o
33.9 ρ
Ms=2.0 Ms=2.25 =1
ρ∞
40 mm
ρ ρ Ms=1.72 15
0
=1
ρ∞ = 0.76
ρ∞ ρ
= 1.35
ρ∞
125 mm 74 mm
Figure 3a. Experimental Geometry of a cone-cylinder in the 0.3m tunnel
elements were normally distributed with zero mean, and unit variance and standard deviation.
The matrix was plotted as a binary image - black dots for all non-zero values. A normal
distribution was chosen to minimize the errors away from the optical axis. Since vibrations
often present in the wind tunnel environment can cause erroneous results on cross-
correlation, care should be taken to minimize vibrations of the back ground and camera.
Further, choosing a large enough exposure time (here 1/100th of a second) and averaging
several displacement fields can successfully address this problem. The printed out pattern
was sandwiched between two thin sheets of glass and mounted on a heavy stand fixed on a
concrete block to minimize vibrations (See Figure 3b). It was then illuminated to achieve
better signal to noise ratio by means of placing a large halogen lamp far enough behind so as
to approximate parallel beam conditions, thus eliminating the need for methods to resort sets
of fan beam projections into parallel beam projections (Kak and Slaney 198824).
The optimal location for the background, light source and camera were arrived at by using
equation 4 and keeping in mind that increasing sensitivity (displacement of image) meant
lower physical resolution as the interrogation size used in the correlation algorithm would
have to be correspondingly larger.

Light Source
Schlieren
Background Windows Tunnel Model Camera
Glass Plate
1.2m 0.7m 0.3m 0.7m
Figure 3a. Experimental Geometry of a cone-cylinder in the 0.3m tunnel
The direction of viewing was perpendicular to the background and through the flow by using
the Schlieren windows built into the tunnel. The windows were made of optical quality (λ/8)
glass and did not introduce any distortions. The imaging was achieved by means of a
commercially available Sony DSC F-707 digital still camera with 5.1 mega pixel resolution.
The camera was mounted on a heavy tripod, fixed to the ground, 0.7m away from the tunnel.
Figure 4a,b. The background dot pattern a) without the flow and b) with the flow.

Figures 4 a, b show the background dot pattern imaged with the cone-cylinder in the
foreground in the absence and presence of the flow respectively. They appear identical due to
the high density of the dots in the background image. Eight such image sets were taken both
with and without the flow. The displacements calculated from these data sets were then
averaged to eliminate any noise caused by vibration to obtain the output vector field as
described in the next section.
Figure 5. Conventional Schlieren of the flow over 150 cone-cylinder at Mach 2.0 in the 0.3m trisonic
tunnel.
Figure 5 shows the conventional Schlieren (using a horizontal knife-edge) of the flow which
can be used later for comparison.
4.2. Image Analysis

The images were analyzed by means of a cross-correlation algorithm developed for the
purpose. The sub-pixel accuracy of the calculated displacement was 0.1 pixels using a
Gaussian sub-pixel interpolation technique commonly used in Particle Image Velocimetry.
The maximum displacement for the present optical setup was found to be 3 pixels and hence
an interrogation window of 16 x 16 pixels was found to be more than adequate; this translated
into a physical resolution of 2mm. The output data was plotted on a 160 x 160 rectangular
mesh. Figure 6 shows the average (of eight images) displacements as a vector field the
magnitude and direction of which are the density gradients of the flow (the arrows point
towards lower density).
0.08
0.06
0.04
0.02
y (m)
-0.02
-0.04
-0.06
-0.08
-0.15 -0.1 -0.05 0

x (m)
Figure 6. The average displacement field: vectors of density gradients.
The derivative of the density gradients forms an elliptic partial differential equation of the
form:
∂2 ∂2
ρ ( x , y ) + ρ ( x, y ) = S ( x, y ) (17)
∂x 2 ∂y 2
i.e., ∇ 2 ρ = S
where S(x,y) is the source term, which is calculated at each mesh point from the
displacements obtained through correlation. Equation 17 is also commonly known as the
Poisson equation. Neumann boundary conditions are chosen in which normal
r
derivative n̂ • ∇ρ , n̂ being the outward normal to the surface is specified at the boundaries.
∂ρ ∂ρ
This implies that the value of and are specified at each mesh point on the mesh
∂x ∂y
boundary parallel to the y and x axes respectively. This is solved on the same rectangular

mesh using the Successive Overrelaxation method (Ehrlich, 1981266) as above and yields the
line-of-sight integrated density field, shown in Figure 7.
0.08
0.06
0.04
0.02
y (m)
-0.02
-0.04
-0.06
-0.08
-0.15 -0.1 -0.05 0
x (m)
Figure 7. The projected density field as computed from the Poisson equation.
This density field is then used as the projected data set for the filtered back projection. As
seen before, this flow being axisymmetric requires only a single data set and the viewing
angle (θ ) is thus set to zero. The contribution to the noise is largely from the limitations of
image resolution and is assumed to belong mainly to high frequency in the Fourier domain
(Feng et al., 2001). This is minimized using the Shepp-Logan filter discussed earlier.
4.3. Results and discussion

Figure 8 shows the back-projected normalized density field at one view angle. The value of
the freestream total density ( ρ∞ ) is used for normalization. The regions of low and high
density ratios corresponding to the shock and expansion fan are clearly seen. The patches of
low density ratios downstream of the Prandtl-Meyer expansion are due to the overexpansion
effect. As this is followed by a recompression effect, the density ratio is seen to return to the
freestream value. The slight patchiness observed in the freestream upstream of the shock is

ρ
ρ∞ 0.16 0.41 0.65 0.90 1.15 1.39 1.64 1.89 2.13
0.08
0.06
0.04
0.02
y (m)
-0.02
-0.04
-0.06
-0.08
-0.15 -0.1 -0.05 0
x (m)
Figure 8. The back–projected normalized density field as viewed from one direction.
probably due to a slight vibration of the background. Averaging of the displacement fields
ensured no directional density gradients in such regions, but resulted in a slight noise. No
attempt has been made to filter out the noise from the entire region, to avoid artifacts
generated by the filtering process. However the signal-to-noise ratio in the regions of interest
is adequate for drawing unambiguous conclusions.
In order to determine the integration constant in equation (2), we need to know and match to
a reference density at one point in the field. In this application, the freestream total density
( ρ∞ ) for the chosen Mach number is easily calculated by using isentropic relations, as the
tunnel settling chamber pressure and temperature are known. This value in the freestream is
used as the reference and determines the constant. The freestream total density ( ρ∞ ) is also
ρ
used to normalize the density field. The variation of the density ratio ( ) at a streamwise
ρ∞

location of 20 mm from the cone apex, as a function of angle (β) from the cone surface to the
shock is plotted in Figure 9. Again from isentropic relations, the shock wave angle (βshockwave)
is found to be .The computed values at the corresponding location from Cone tables (Sims
196425) are also plotted in the figure for comparison. The agreement may be seen to be very
good considering the resolution of the interrogation window (2mm) used for the calculation
of density gradients.
35
30
25
20
15
1 1.05 1.1 1.15 1.2 1.25 1.3 1.35 1.4
ρ/ρ
Figure 9. Comparison of angular variation of density ratio behind the shock.
Figure 10 shows density variations downstream of the expansion fan at two

streamwise locations of 2.5D ( ≅ 0.10m) and 5.0D ( ≅ 0.20m) from the cone apex. The
density at the surface as calculated from isentropic calculations is also shown for
comparison with BOS. The expansion fan at the cone-cylinder junction is locally two-
dimensional and leads to higher Mach number (compared to freestream value M∞)
immediately downstream as has also been reported by other studies (Coustols and
Cousteix, 199427). As the flow proceeds downstream along the cylinder further
recompression takes place, resulting in a value close to M∞. It was not possible to
obtain density values closer than 5mm from the cylinder surface because of the

limitation on the measurement resolution and the fact that it was very close to
0.025
BOS
0.02
X=2.5D X=5.0D
0.015
y (m)
0.01
From isentropic relations
0.005
Cylinder Surface
0
0.7 0.8 0.9 1 0.7
1.1 0.8 0.9 1 1.1
ρ/ρ ρ/ρ
∞ ∞
Figure 10. Density profiles at different streamwise locations.
the expansion fan at the junction. Downstream (x=5.0d) of the recompression waves, the flow
has returned to near freestream values, and the profiles show the expected pressure loss.
From the density gradients (particle displacements from correlation), we can obtain a
Schlieren image in the direction required (i.e. knife-edge vertical or horizontal). In this case,
the vertical gradient (horizontal knife-edge) is chosen as was the case in the conventional
Schlieren (Figure 5). Figure 11 shows the computed Schlieren image. Figure 5 may be seen
for a comparison. It is seen that the shock and it’s interaction with the expansion has been
captured and the wave angles are as expected. Despite the higher noise compared to the
conventional Schlieren, the weaker reflected shock (from the boundary) has also been
captured. However, the width of the shock is relatively larger compared to that in the
conventional Schlieren, because of the resolution chosen for BOS image analysis (2 mm).
The shock region can be reproduced thinner in BOS images if a higher resolution cross-
correlation algorithm is chosen for interrogation of images.

0.08
0.06
0.04
0.02
y (m)
-0.02
-0.04
-0.06
-0.08
-0.15 -0.1 -0.05 0
x (m )
Figure 11. Computed Schlieren from particle displacements.
The effect of the camera resolution and sensitivity (bits/pixel) was also assessed on the
calculated density field by BOS. For this purpose a scientific grade Kodak ES 1.0 cross-
correlation camera (normally used for PIV measurements) with a 1.0 mega pixel was also
used in the present experiment for identical flow conditions. It was found that the SONY
digital camera (with 5.0 mega pixel resolution) provided significantly enhanced image details
indicating the importance of a high resolution camera for the BOS technique. Most off-the-
shelf digital still cameras nowadays offer such resolution, which benefits the general
application of this technique.

5. APPLICATION NOTES
There are many different applications of BOS in aerodynamic research, and more yet to
come. Some examples carried out at NAL are given below.
5.1. Density Measurements in an Axisymmetric

Underexpanded Jet using Background Oriented Schlieren
Technique
This study attempted mapping the density field of a complex flow like that of an
underexpanded jet (Venkatakrishnan, 2004, 200528,29). The measurements were made on an
axisymmetric converging sonic nozzle operating at moderately and highly expanded values
of nozzle pressure ratios.
This effort involved two major steps:
a) validation of the filtered back-projection tomographic technique which was successfully
carried out using a multi-jet configuration in the 0.5m base flow facility wind tunnel at NAL.
Multiple view angles were used to obtain the central plane of the configuration; the technique
is validated when the centre plane can be constructed from two views which show the exit
diameter of each of the two jets on the centre plane to be 36mm. The jets were run at an
underexpanded condition of Poj/Pa = 7 (Pe/Pa = 5.32).
b) application to a sonic jet having moderate and high degrees of expansion, to obtain density
contours. This was run at pressure ratios corresponding to moderately underexpanded (Poj/Pa
~ 3, Pe/Pa ~ 1.7), and highly underexpanded (Poj/Pa = 6, Pe/Pa = 3.17). The axial extent of
the imaging was x = 5D.
The validation results show that the technique is capable of obtaining a desired plane of
interest from a line-of-sight integrated data set, successfully extracting the central plane of the
jets with the correct exit diameter and spacing.
Figure 12a shows the density contours in the central plane of the underexpanded jet. While
the jet exit density is used to provide the integration constant required for matching the
density at a point, it is seen that the ambient density is correctly captured. The structures in
the shear layers are also educed; emphasizing the sensitivity of the technique. The density
field obtained shows all the familiar features of an underexpanded jet like the curved shock at
the lip, internal weak shock system etc. Figure 12b which plots the centerline variation of

density, shows the expected oscillation of density through the shock cells with reduction in
amplitude downstream. From literature this oscillation subsides 8-10 nozzle diameters
downstream of nozzle exit which is nearly double the distance imaged here. However the
high unsteadiness of the jet resulted in a slight averaging at the current sample rate. Shorter
exposure times by either increasing shutter speed or reducing illumination pulse width can
successfully address the problem. Further, the results are less prone to noise for flows with
large density gradients. However, this study shows that meaningful quantitative density data
can be extracted using minimal hardware with this methodology.
3
Density (kg/m ): 0.0 1.0 2.0 3.0 4.0
1.0
2
35000 1.
0
1.0
1.
1.0
0
30000
1.0
1
0
1.
0
1. 1.0 2.0
25000 1.0 2.
0
2.0
Density
0.0 4.
0
2.0
2.0
y/D
0
3.0
2.0
1.0
20000
3.0
0.0 2.0 1.0

2.0 2.10.0
1.20.0
15000
-1
1.0 0
1.
1.
0
1.0
10000
1.0
1.0
-2
0
5000
1.
0-0.15 1 -0.1 2 -0.05 3 0 4

x/D
x
Figure 12a Density contours in the underexpanded jet (Pe/Pa = 3.17) using BOS.

3.5 From isentropic relations
2.5
ρ/ρ
a
1.5
0.5
0
0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 4.50
X/D
Figure 12b Axial variation of normalized centerline density.
5.2. BOS Density Measurements in Afterbody Flows with

Shock and Jet Effects
In this study (Venkatakrishnan 200730), the mean density field is documented using the BOS
technique in a supersonic shock separated flow at the aft end of an axisymmetric body
including jet effects. This would involve tomographic reconstruction to extract the desired
plane of the density field after obtaining the line-of-sight integrated density field using BOS.
The afterbody was a boat-tailed configuration of circular arc geometry with a boat-tail angle
of 12o. For the freestream Mach number of 1.34, the boundary layer thickness was 8mm. For
the jet-on case, the jet was run at an underexpanded condition of Poj/P∞ ~ 6, which
corresponded to a degree of expansion Pe/P∞ = 2.8. Density measurements were made for the
free base (without the central jet) and also with an underexpanded sonic jet run at a JPR (Jet
Pressure Ratio) of 6. The measurements carried out also included surface pressure
measurements on the afterbody and flow visualization with spark Schlieren and surface flow
with oil film.
Figures 13a, b show two instantaneous conventional Schlieren images (horizontal knife edge)
of the jet-off and jet-on cases. Examination of multiple Schlieren images show that there is
considerable movement of the shock. The distance of the shock excursion for the jet-on case
is about 8mm (~δo) - significantly lower than that for the jet-off case (14mm ~ 2δo).
A mild asymmetry seen in the oil flow visualizations is also seen in Figure 13a with the
shock location on the lower surface relatively closer to the nozzle lip. This is reduced
significantly in the jet-on case due to the fact that the axisymmetric jet plume plays the
dominant role in determining the flow topology on the afterbody. For the jet-on case the point
of reattachment is on the jet shear layer at a distance between 18mm - 20mm downstream of
nozzle exit. The reattachment shock is seen at this location (Figure 13b). The flow separation
occurs at about 18mm (2.2δo) upstream of the nozzle exit for the jet-off case and about 36mm
for the jet-on case, due to the positive pressure field imposed by the underexpanded jet plume
on the afterbody boundary layer.
Figure 14a shows the centre-plane of the mean density field for the jet-off case after
tomography. The figure clearly shows the boundary layer separation occurring around 18mm
as was obtained from the surface flow visualization results. The rise in density across the
shock (ρ2/ρ1 =1.78) matches that calculated from pressure data using isentropic relations
(ρ2/ρ1 =1.73). Further, the mild non-axisymmetry is found to be captured in the density field
as well, as is observed by an examination of Figure 2a. The density behind the afterbody is
seen to be quite uniform as expected and is about 40% of freestream density.
Figure 2b presents the mean density field in the central plane for the jet-on condition. The
rise in density across the shock (ρ2/ρ1 =1.56) matches that calculated from pressure data
using isentropic relations (ρ2/ρ1 =1.6). The flow topology is dominated by the pressure field
of the underexpanded jet. For the present degree of underexpansion, the jet density
normalized by the freestream value (ρjet/ρ∞) is 1.14 from the BOS data The separated
boundary layer reattaches on the jet shear layer resulting in a reattachment shock
accompanied by a density rise as is seen in Figure 2b at x =30mm downstream of exit.
Based on the measurements a mean flow topology has been drawn for both the jet-off and jet-
on cases and is shown in Figure 3a, b.
The data present a clear picture of the density field in this complex flow where other density
measurement techniques are extremely difficult and hence would be extremely useful for
understanding and CFD code validation.

FIGURE 13A. INSTANTANEOUS SCHLIEREN FIGURE 13B. INSTANTANEOUS SCHLIEREN

IMAGE AT M∞ = 1.34 FOR JET-OFF CONDITION IMAGE AT M∞ = 1.34 FOR JET-ON CONDITION
SHOWING MAXIMUM UPSTREAM LOCATION OF SHOWING MAXIMUM UPSTREAM LOCATION OF
SHOCK.
SHOCK.
ρ/ρ∝: 0.20 0.42 0.64 0.86 1.08 ρ/ρ∝: 0.20 0.42 0.64 0.86 1.08
4 0.59 1.08 0.81 1.03 1.12
0.99
0.6 1.03
0.68 0.55 0.68 0.73 0.64 0.68
40 40
0.38 0.81
0.42 0.20
0.46 0.42 0.42 0.68 0.59 0.59

20 20
0.42 0.42 0.99 1.12

0.38 0.38
y (m)
0.90 0.77
y (m)
0 0
0.42 0.99
0.33 1.16
-20 0.33 -20
0.42 0.33 0.77 0.55 0.59
0.55 0.29
0.77 0.94
-40
-40
0.90 0.81
0.72 0.64 0.68
0.55
0.59 0.68 0.46 0.68
-60 0.81 0.94 1.08 0.90
-40 -20 0 20 40 -60
x (m) -40 -20 0 20 40
x (m)
FIGURE 14A. CENTRAL PLANE OF MEAN DENSITY FIGURE 14B. CENTRAL PLANE OF MEAN
FIELD AT M∞ = DENSITY FIELD OBTAINED USING BOS
1.34 FOR JET-OFF CONDITION. TECHNIQUE FOR JET-ON CONDITION.
Freestream Acceleration zone
Oblique Shock
M∞ = 1.34 M = 1.74
Mlocal = 1.5
Freestream Acceleration zone Reattachment Shock
Reattachment Oblique Shock

Separation point point – beyond
Schlieren view
18mm Separated shear layer
Reattachment point
Separation point
ρbase/ρ∞ = 0.4 Shock Cell
2 ρ /ρ = 1.78
1 2 1
FIGURE 15. FLOW TOPOLOGY FOR JET-OFF AND 7.1.

JET-ON CONDITIONS.

REFERENCES
1. McMackin, L., Masson, B., Clark, N., Bishop, K., Pierson, R., Chen, E., “Hartmann
wave-front sensor studies of dynamic organized structure in flowfields”, AIAA
Journal, 33, pp2158-2164 (1995).
2. Dalziel, S.B., Hughes, G.O., Sutherland, B.R., “Synthetic Schlieren”, Proc. 8th
International Symposium on Flow Visualization, Sorrento (Italy), September 1-4
(1998).
3. Dalziel, S.B., Hughes, G.O., Sutherland, B.R., “Whole field density measurements by
‘synthetic schlieren’”, Experiments in Fluids, 28, pp322-335 (2000).
4. Meier G.E.A., “New Optical Tools for Fluid Mechanics’, Proc. 8th International
Symposium on Flow Visualization, Paper 226, Sorrento (1999).
5. Sutherland, B.R., Dalziel, S.B., Hughes, G.O., Linden, P.F., “Visualization and
measurement of internal gravity waves by ‘Synthetic Schlieren’. Part I. Vertically
oscillating cylinder”, Journal of Fluid Mechanics, vol 390, pp93-126 (1999).
6. Onu, K., Flynn, M.R., Sutherland, B.R., “Schlieren Measurement of Axisymmetric
Wave Amplitudes”, Experiments in Fluids, Experiments in Fluids, Vol.35, No 1,
pp.24-31 (2003).
7. Richard H., Raffel M., Rein M., Kompenhans J., Meier G.E.A., “Demonstration of
the applicability of a Background Oriented Schlieren (BOS) Method”, Proc. Of the
10th International Symposium on Applications of Laser Techniques to Fluid
Mechanics, 10-13 July 2000, Lisboa, Portugal.
8. Raffel M., Richard H., Meier G.E.A., “On the Applicability of Background Oriented
Optical Tomography for Large Scale Aerodynamic Investigations”, Experiments in
Fluids, Vol.28, pp.477-481 (2000).
9. Meier G.E.A., “Computerized Background Oriented Schlieren”, Experiments in
Fluids, Vol. 33, No 1, pp 181-187 (2002).
10. Merzkirsch, W., Flow Visualization, Academic Press (1987).
11. Goldstein, R.J., Fluid Mechanics Measurements, Hemisphere Publishing Corporation
(1983).
12. Matulka, R.D. and Collins, D.J., “Determination of three-dimensional density fields
from holographic interferograms”, J. Appl. Phys., 42, 3, pp1109-1119 (1971).
13. Hertz, H.M., “Kerr effect tomography for non-intrusive spatially resolved
measurements of asymmetric field distribution”, Applied Optics, 25, pp 914-921
(1986).
14. Faris, W. and Byer, R.L., “Quantitative three-dimensional optical tomographic
imaging of supersonic flows”, Science, 238, pp1700-1702 (1987).
15. Cha, S.S. and Sun, H., “Interfereometric tomography of continuous fields with
incomplete projections”, Opt. Lett., 14, 6, pp299-301 (1989).
16. Muralidhar, K., “Temperature field measurement in Bouyancy driven flows using
interferometric tomography”, Ann. Rev. of Heat Transfer, Vol. 12, pp265-375 (2000).
17. Walsh, T.E. and Kihm, K.D., “Tomographic deconvolution of laser speckle
photography for axisymmetric flame temperature measurement”, Journal of Flow
Visualization and Image Processing, vol. 2, pp299-310 (1995).
18. Ramachandran, G.N., Lakshminarayanan, A.V., “3D reconstructions from
radiographs and electron micrographs: Appl. of convolution instead of Fourier
transforms”, Proc. Nat. Acad. Sci., vol.68, pp.2236-40 (1971).
19. Shepp, L. A. and Logan B. F., “The Fourier Reconstruction of a Head Section” , IEEE
Trans. on Nuclear Science, NS-21, pp. 21-43 (1974).
20. Feng, J., Okamoto, K., Tsuru, D., Madarame, H., Fumizawa, M., “Visualization of
3D gas density distribution using optical tomography”, Chemical Engineering
Journal, 3865, pp1-9 (2001).
21. Radon, J., “On the determination of functions from their integrals along certain
manifolds”. Berschte Saechsische Akademi der Wissenschaften, 29, pp 262-277
(1917).
22. Oppenheim, A.V., and R.W. Schafer, Discrete-Time Signal Processing, Prentice-Hall,
pp. 447-448 (1989).
23. Kak, A.C. and Slaney, M., “Principles of computerized tomographic imaging”, IEEE
Press, New York (1988).
24. Venkatakrishnan, L and Meier G. E. A., “Density Measurements using Background
Oriented Schlieren Technique”, Experiments in Fluids, Vol 37, No 2, pp. 237 – 247
(2004).
25. Sims, Joseph L., “Tables for supersonic flow around right circular cones at zero angle
of attack”, NASA SP-3004, 1964.
26. Ehrlich, L., “An Ad-Hoc SOR method”, J. Comput. Phys., 43, pp31-45 (1981).
27. Coustols, E. and Cousteix, J., “Performance of riblets in the supersonic regime”,
AIAA J, vol 32, No. 2, pp431-433 (1994).
28. Venkatakrishnan, L., “Density Measurements in an Axisymmetric Underexpanded Jet
Using Background Oriented Schlieren Technique”, Paper Number: AIAA-2004-2603
in 24th AIAA Aerodynamic Measurement Technology and Ground Testing
Conference at Portland, Oregon, USA, 28 June – 1 July 2004.
29. Venkatakrishnan, L., “Density Measurements in an Axisymmetric Underexpanded Jet
Using Background Oriented Schlieren Technique”, AIAA J Vol 43, No 7, pp 1574-
1579. July 2005.
30. Venkatakrishnan, L., Suriyanarayanan P and Mathur N. B., “BOS Density
Measurements in Afterbody Flows with Shock and Jet Effects”, Paper No AIAA
2007-4223, 37th AIAA Fluid Dynamics Conference and Exhibit 25 - 28 June 2007,
Miami, FL.

Background Oriented Schlieren: Experimental Aerodynamics Division

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Background Oriented Schlieren: Experimental Aerodynamics Division

Загружено:

Авторское право:

Доступные форматы

BACKGROUND ORIENTED

2. BACKGROUND ORIENTED SCHLIEREN ...................................................................4

3. FILTERED BACK PROJECTION TOMOGRAPHY ......................................................7

5. APPLICATION NOTES .................................................................................................21

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

2. BACKGROUND ORIENTED SCHLIEREN

For small deflection angles, ε can be approximated as

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

3. FILTERED BACK PROJECTION TOMOGRAPHY

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

⎧⎪π ω sin c (ω ω max ) ω ≤ ω max

Figure 2. Configuration of projection on one view direction at θ (Feng et al20

here, l =x cos θ + y sin θ (see Figure 2)

Its Fourier transform is

Using the Projection Slice theorem (Radon, 1917 21 )

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

11) can be rewritten by making the bandlimited projection assumption, as follows:

⎪⎧0.54 − 0.46 cos ( 2π n ( N − 1) ) 0 ≤ n < N

where ω n is the frequency at the nth point in the time series.

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

Figure 3a. Experimental Geometry of a cone-cylinder in the 0.3m tunnel

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

1.2m 0.7m 0.3m 0.7m

Figure 3a. Experimental Geometry of a cone-cylinder in the 0.3m tunnel

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

4.2. Image Analysis

-0.15 -0.1 -0.05 0

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

4.3. Results and discussion

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

Figure 10 shows density variations downstream of the expansion fan at two

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

5.1. Density Measurements in an Axisymmetric

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

0.0 2.0 1.0

0-0.15 1 -0.1 2 -0.05 3 0 4

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

3.5 From isentropic relations

Figure 12b Axial variation of normalized centerline density.

5.2. BOS Density Measurements in Afterbody Flows with

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

FIGURE 13A. INSTANTANEOUS SCHLIEREN FIGURE 13B. INSTANTANEOUS SCHLIEREN

0.46 0.42 0.42 0.68 0.59 0.59

0.42 0.42 0.99 1.12

Reattachment Oblique Shock

FIGURE 15. FLOW TOPOLOGY FOR JET-OFF AND 7.1.

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

Lecture Course On Advanced Flow Diagnostic Techniques, NAL, Sept 2008

Вам также может понравиться