Skip to main content

Full text of "Measurement of Separated Flow Structures Using a Multiple-Camera DPIV System"

See other formats

i '£>»/'■ / 

Langley Research Center 

Measurement of Separated 
Flow Structures Using a 
Multiple-Camera DPIV System 

William M. Humphreys, Jr. 
Scott M. Bartram 
NASA Langley Research Center 
Hampton, VA 23681 

Presented at the 19th International Congress 

on Instrumentation in Aerospace 

Simulation Facilities (ICIASF) 

27-30 August 2001 / Cleveland, Ohio 


William M. Humphreys, Jr.* 
Scott M. Bartranrf 

NASA Langley Research Center 
Hampton, Virginia 23681 


A novel multiple-camera system for the 
recording of digital particle image velocimetry 
(DPIV) images acquired in a two-dimensional 
separating / reattaching flow is described. The 
measurements were performed in the NASA 
Langley Subsonic Basic Research Tunnel as part 
of an overall series of experiments involving the 
simultaneous acquisition of dynamic surface 
pressures and off-body velocities. The DPIV 
system utilized two frequency-doubled Nd:YAG 
lasers to generate two coplanar, orthogonally- 
polarized light sheets directed upstream along 
the horizontal centerline of the test model. A 
recording system containing two pairs of 
matched high resolution, 8-bit cameras was used 
to separate and capture images of illuminated 
tracer particles embedded in the flow field. 
Background image subtraction was used to 
reduce undesirable flare light emanating from the 
surface of the model, and custom pixel alignment 
algorithms were employed to provide accurate 
registration among the various cameras. Spatial 
cross correlation analysis with median filter 
validation was used to determine the 
instantaneous velocity structure in the 
separating / reattaching flow region illuminated 
by the laser light sheets. In operation the DPIV 
system exhibited a good ability to resolve large- 
scale separated flow structures with acceptable 
accuracy over the extended field of view of the 
cameras. The recording system design provided 
enhanced performance versus traditional DPIV 
systems by allowing a variety of standard and 
non-standard cameras to be easily incorporated 
into the system. 


d camm , camera pixel spacing, um 

* Research Engineer, Advanced Measurement 

and Diagnostics Branch 
t Engineering Technician, Advanced 

Measurement and Diagnostics Branch 







physical spacing of dots on 
calibration dot card, mm 

particle image displacement, 
uncorrected, um 

particle image displacement 
with zero pulse separation, (am 

particle image displacement, 
corrected, um, see equation (4) 

amplitude of i th , j th pixel on 
camera / in receiver k 

amplitude of background 
subtracted i" 1 , j ch pixel on 
camera / in receiver k, see 
equation (2) 

amplitude of minimum image 
i*, j* pixel on camera 1 in 
receiver k, see equation (1) 

magnification of camera 
system measured from dot card 
image, see equation (3) 

number of measured pixels 
between dots on dot card 

particle image / flow velocity, 

laser pulse separation, usee 

standard deviation of 

parameter X, see equations (5) 
and (6) 

a 2 variance of parameter X, see 

equations (5) and (6) 


The study of incompressible flow separation and 
reattachment is classic to fluid dynamics 
research, and the understanding of such flows is 
of great importance in many applications. 
Traditionally these flow structures have been 
investigated using a variety of probes, surface- 
mounted sensors, and flow visualization 
techniques, and an enormous body of data for 
various model geometries now exists in the 
literature.'" 2 Historically, most of the data 
characterizing the global structure of separating 
flows was of a qualitative nature. 3 However, the 
development of Particle Image Velocimetry 
(PIV) in the 1980's followed by Digital Particle 
Image Velocimetry (DPIV) in the 1990's 
introduced new techniques for quantitative 
investigation of complex flow fields. These new 
techniques have allowed detailed databases to be 
established, with the work of Grant 
presented as one example. 4 PIV is now 
considered a mature technology for 
instantaneous planar velocity measurements, and 
several good review articles appear in the 
literature which describe the technique in 
detail. 5 " 8 

For the past two years the authors, in conjunction 
with researchers at Michigan State University 
(MSU), have been conducting simultaneous 
measurements of dynamic surface pressures and 
off-body velocities in simplified separating 
flows. As part of this investigation a series of 
measurements were conducted using a two- 
dimensional model tested at low Reynolds 
number in the NASA Langley Subsonic Basic 
Research Tunnel (SBRT). The model generated 
a well-defined separation bubble which was 
interrogated using a combination of instruments. 
Dynamic surface pressures were acquired using 
an 80-element microphone array system while 
off-body velocities were acquired using a custom 
DPIV system. An extensive set of data was 
obtained in SBRT from which new flow 
measurement techniques and low dimensional 
models of the flow field are currently being 

This paper examines the construction, 
calibration, and operation of the DPIV system 
which was employed to collect off-body velocity 

measurements for the SBRT separated flow 
characterization studies. In particular, the design 
of the unique multiple-camera recording system, 
similar in structure to designs demonstrated by 
Kahler and Kompenhans, 9 is described in detail. 
Methods of calibration and relevant error sources 
associated with the measurements are discussed. 
Finally, representative data obtained in the 
facility using the DPIV system is presented. 


Tunnel Facility 

The NASA Langley Subsonic Basic Research 
Tunnel (SBRT) is an open-circuit wind tunnel 
containing a rectangular test section measuring 
0.84 meters high by 0.57 meters wide by 1.85 
meters long. The front portion of the test section 
contains port and starboard glass windows 
measuring approximately 0.75- by 0.75 meters, 
as well as smaller window inserts in the floor 
and ceiling. The tunnel employs a 6:1 
contraction and is capable of a maximum 
freestream velocity of approximately 60 
meters/sec through the use of an axial fan driven 
by a 200-horsepower variable frequency motor. 
Tunnel data acquisition hardware includes a 48- 
channel electronically scanned static pressure 
monitoring system as well as a series of fixed 
test section static taps and thermocouples to 
monitor freestream conditions in real-time. 
Figure 1 shows a three-dimensional rendered 
view of the tunnel. 


The model employed for this study was designed 
and built by Laura Hudy in MSU's Department 
of Mechanical Engineering as part of a NASA 
Graduate Student Research Program project and 
consisted of a splitter plate containing a fence 
attached perpendicularly to the flow at the 
leading edge. 10 The model was mounted in the 
center of the SBRT facility such that it bisected 
the tunnel test section vertically. The model was 
constructed of aluminum and measured 1.6 
meters long by 0.36 meters wide. Aluminum 
extensions were attached to the model such that 
the total width of the splitter plate matched the 
width of the test section. The leading edge fence 
total height was 34.9 mm, with a step height of 
7.9 mm. The fence generated a separated flow 
region covering approximately 13 percent of the 
chord length of the splitter plate. This separation 
zone was maintained as two-dimensional flow 

through the use of thin interior side walls 
attached at the sides of the splitter plate inside 
the test section - these secondary walls 
prevented introduction into the separation region 
of spanwise flow generated by the test section 
side-wall boundary layers. Flush-mounted glass 
windows were incorporated into the interior side 
plates to allow viewing of the centerline flow 
from outside of the tunnel test section. 

The model was populated with an array of 80 
flush-mounted, low-cost electret microphones 
acting as hydrodynamic pressure sensors. The 
array consisted of 28 microphones mounted 
along the top centerline of the splitter plate just 
behind the leading edge fence with two rows 
each containing 13 microphones located on 
either side of the centerline. A total of 40 static 
pressure taps were also located on the top of the 
splitter plate - 28 taps located parallel to the 
centerline row of microphones and four other 
rows each containing 3 taps. A photograph of 
the model mounted in SBRT showing the 
dynamic pressure sensors, static taps and interior 
side wall DPIV windows is given in Figure 2. 
The acquisition and analysis of the microphone 
and static tap data is beyond the scope of this 
paper - reference 10 gives a detailed overview of 
these topics. 


The DPIV instrument consisted of five sub- 
elements, namely a laser optics system, particle 
seeding system, image recording system, 
synchronization timing system, and data analysis 
system. The first four of these sub-elements is 
described in this section with the data analysis 
system described subsequently. 

Laser Optics 

The laser optics system main components 
included two 600-millijoule, frequency-doubled 
Nd:YAG pulsed lasers for illumination and a 
series of optical components for beam combining 
and light sheet generation. The 532-nm 
wavelength beams emanating from each laser 
were aligned to a common optical axis using a 
half-wave retardation plate and a thin-film 
polarizing beamsplitter, shown in the schematic 
of Figure 3. The combined beams were then 
directed through a series of cylindrical and 
spherical lenses to generate a long focal length, 
converging light sheet with minimum thickness 
of approximately one millimeter and maximum 

extent of 75 millimeters. The light sheet was 
directed into the tunnel test section through an 
access window positioned at the far downstream 
end of the test section. The sheet was then 
reflected upstream in the test section along the 
horizontal centerline of the model and adjusted 
such that it just grazed the top surface of the 
splitter plate. This beam path was chosen to 
minimize flare light reflected from the top 
surface of the model, allowing measurements to 
be obtained deeper in the separated flow region 
boundary layer. Figure 4 shows a photograph of 
the laser / light sheet system installed in SBRT. 

Flow Field Seeding 

To provide a uniform distribution of seed in the 
illuminated region of the flow above the splitter 
plate, the tunnel was seeded with the chemical 
bis(2-ethylhexyl) sebacate (commonly known as 
DEHS). This moderate viscosity, low toxicity 
oil was injected into the tunnel immediately 
upstream of the honeycomb section using a six 
jet atomizer. Figure 5 depicts a typical 
aerodynamic particle size distribution obtained 
from atomization of DEHS using a Laskin 
nozzle, with a mean aerodynamic size of 
approximately 0.9-1.0 micrometers. It was 
anticipated that the six jet seeder in SBRT would 
yield a similar distribution. Control of the 
seeding density in the tunnel test section was 
achieved by varying the number of jets operating 
in the atomizer as well as the pressure delivered 
to the jets. The seeding delivery system was 
optimally positioned in front of the tunnel 
honeycomb section to concentrate the seed in a 
vertical plane at the center of the tunnel. 

Image Recording System 

A novel feature of the DPIV system described 
here concerned the design of the image recording 
system. To capture illuminated particle images, 
four matched 1300- by 1030-pixel, 8-bit, 
progressive scan (full frame) cameras were 
connected to individual digitizers and frame 
buffers. The four cameras were arranged in two 
pairs with each pair forming an independent 
imaging system (referred to as receivers 1 and 2), 
each capable of viewing an illuminated planar 
area measuring 10 cm wide by 8 cm tall at a 
distance of approximately 1.5 meters. Each 
camera was attached to a 135-mm, f/2 lens. 
Receivers 1 and 2 were arranged side by side 
normal to the light sheet and positioned such that 
their individual fields of view slightly 

overlapped, generating an overall view 19 cm 
wide by 8 cm tall. Figure 6 contains a 
photograph showing the recording system 
installed in the SBRT facility, and Figure 7 
shows a photograph of a single receiver with 
individual components labeled. Note that one of 
the receivers in Figure 6 is shown mirrored with 
respect to the other to allow the hardware to be 
positioned directly adjacent to one another. 
Nevertheless, the functionality of each receiver 
was the same. 

Data acquisition was achieved for each receiver 
by means of polarization separation, depicted 
graphically in Figure 8. Scattered light from 
particles illuminated within the light sheet were 
directed to individual cameras in a receiver using 
a polarizing beamsplitter and flat mirror. The 
process began by generating a linearly polarized 
(p-polarization), 10-nanosecond pulsed light 
sheet at time t. Particles illuminated by the sheet 
scattered light which was captured by each 
receiver in the recording system. The scattered 
light, also p-polarized, passed directly through 
the beamsplitter to the line of sight camera in 
each receiver. This was followed by the 
generation of a second orthogonally polarized (s- 
polarization), 10-nanosecond pulsed light sheet 
at time t+Ar, where At represents the laser pulse 
separation. The s-polarized scattered light 
emerged from the beamsplitter at right angles to 
the incident light and was thus directed to the 
mirrored camera in each receiver. The 

polarization of the scattered light was preserved 
due to the size of the particles being of the same 
magnitude as the wavelength of the laser light. 
Preservation of polarization and proper 
alignment of the polarizing beamsplitters made it 
possible to capture each of the two illuminated 
particle images on separate cameras, allowing 
standard spatial cross correlation techniques to 
be applied to the data. 

This camera arrangement was similar to that 
described by Kahler and Kompenhans for 
imaging of two spatially-displaced, 
double-pulsed light sheets in reference 9. 
However, the operation of the recording system 
for this study was different in that two coplanar 
light sheet pulses were captured by the cameras. 
While the receiver configuration depicted in 
Figure 8 was more complex than for traditional 
DPIV systems, it allowed greater flexibility in 
construction of the camera geometry. The 
greatest benefit of this design was that it allowed 
a variety of rugged commodity cameras to be 

used in the system versus specialized cross 
correlation cameras. It also allowed hardware 
commonality to be achieved with related global 
velocimetry techniques in use at NASA Langley, 
in particular Doppler Global Velocimetry which 
utilizes a very similar camera arrangement. 

Timing Synchronization 

To ensure proper synchronization of cameras and 
lasers in the DPIV system, the timing circuit 
shown in Figure 9 was employed. A pulse 
generator initiated a continuous 10-Hz TTL 
pulse train which acted as a master sync signal. 
This pulse train was connected to the laser 
timing controller which caused the lasers to fire 
at a continuous 10-Hz pulse repetition rate to 
maintain energy stability. The master TTL pulse 
train was also directed to a slave pulse generator 
configured for external triggering. Inverted and 
non-inverted TTL outputs from the slave 
generator were connected to camera and digitizer 
trigger inputs, respectively. When the external 
trigger on the slave generator was enabled, each 
camera and digitizer would begin acquiring 
images at 10 frames per second until the external 
trigger was disabled. Custom software was 
developed to control the acquisition of image 
data from the four cameras in the recording 
system, and allowed a slower acquisition rate to 
be obtained by providing an image "stride"; i.e., 
unwanted images were simply skipped and not 
written to disk during an acquisition cycle. 


Laser Pulse Separation 

The laser pulse separation was monitored by 
placing a fast rise time photodiode along the 
periphery of the laser optics system. The 
photodiode was adjusted to respond to secondary 
reflections of laser light from lenses, windows, 
etc. The output of the photodiode was attached 
to a high speed digital oscilloscope which 
provided a trace showing the relative amplitude 
of each laser pulse as well as the pulse 

Dot Card Recordings 

The DPIV recording system described here 
relied on a process whereby scattered light from 
particles at two instances of time were captured 
on separate cameras. While such an arrangement 
allowed general purpose cameras to be used, it 

also greatly increased the complexity of the 
calibration process. Since DPIV relies on the 
tracking of particle images from one laser 
exposure to another to obtain velocity data, 
alignment errors in the system shown in Figure 7 
(i.e., camera misalignment, polarizer skew, etc.) 
induce direct and sometimes substantial bias 
errors in these measured velocities. To help 
reduce these errors, a card containing a uniform 
series of dots with a horizontal and vertical dot 
spacing of 5.7 mm was placed in the plane of the 
light sheet. Figure 10 depicts one of these dot 
cards. Using white light illumination of the card, 
a manual alignment was performed by 
continuously capturing images from all four 
cameras. Pairs of images representing the data 
obtained from a particular receiver were 
subtracted from one another and a difference 
image displayed in real time on a monitor. 
Optimal alignment was achieved by causing as 
many dots as possible in the field of view of each 
receiver to "disappear". Once optimal alignment 
was achieved, a sequence of five images of the 
dot card was captured by each camera. These 
dot card images were then used in pixel 
alignment algorithms, described subsequently, to 
attempt to minimize spatial distortions and 
ensure accurate pixel alignment between the 
pairs of cameras constituting each receiver in the 
recording system. 

Zero Displacement Recordings 

As a secondary check on the dot card recordings 
to ensure the highest registration accuracy 
among all cameras, a series of zero displacement 
recordings were performed before data 
acquisition was initiated. This process consisted 
of setting the laser pulse separation to zero (i.e., 
firing both Nd:YAG lasers in unison) and 
seeding the flow while running the tunnel at a 
nominal speed (typically 10-15 m/sec). A 
series of images were then captured each 
containing a uniform spatial distribution of seed 
across the entire field of view of the recording 
system. The zero laser pulse separation was 
designed to remove all flow induced motion 
from the captured images, leaving only receiver 
mismatch and optical field distortions to account 
for movement of particle images. The captured 
images were processed as regular DPIV data. 
Note that perfect alignment of both cameras in 
each receiver of the recording system would 
have revealed a zero velocity vector in all 
processed interrogation regions. Such a zero 
velocity field was impossible to achieve in 

practice, and thus the analysis of the zero 
displacement recordings quantified bias errors 
introduced to the DPIV data due to camera 


Background Removal 

It was desired to obtain DPIV velocity data as 
close to the top surface of the splitter plate as 
possible, necessitating that unwanted flare light 
and background noise in acquired images be 
removed. The authors chose to use a technique 
described by Kuhn, Kompenhans, and Monnier 
to perform this background removal. The 
technique began with the generation of a series 
of four "minimum" images, one for each camera 
in the recording system. These minimum images 
were computed using sequences of DPIV images 
acquired by the cameras under normal operating 
conditions (i.e., seeded flow, nominal laser pulse 
energy and timing, etc.). Letting fu(ij) represent 
the amplitude of a pixel at position (i,j) for an 
image taken in a sequence by camera / in 
receiver k, the minimum image M u (ij) for the 
camera was computed via 

M u (i,j)' 

_//«('.» >.f fu( i J)< M u( i J) (1) 
Mu(i,j) otherwise 

Note that M u (ij) was computed over all pixels 
and over all images in a camera sequence. The 
resulting four minimum images, one for each 
camera, were subtracted on a pixel by pixel basis 
from each of the individual images in a 
sequence, i.e., 

f\UJ) = f u (iJ)-M k ,(iJ) 


The modified images were then used in all 
subsequent processing. 

Camera Pixel Alignment 

As discussed previously, accurate pixel 
registration among cameras in each of the 
recording system receivers was of paramount 
importance if velocity field bias errors were to be 
minimized. The pixel alignment algorithm 
employed for this study utilized sequences of dot 
card images (Figure 10) acquired during system 
calibration. There are numerous techniques for 
pixel alignment which are described in the 
literature, including geometric back projection 

and second order nonlinear fits (reference 5). 
For this study the authors chose a piecewise 
bilinear dewarping technique devised for 
Doppler Global Velocimetry by Meyers to 
remove perspective and optical distortions and 
"straighten" the card images. 13 The technique 
began by identifying all card dots in the field of 
view of each camera and computing the centroid 
location of each dot to subpixel accuracy. 
Located dots were then grouped into four- sided 
polygon regions with a dot located at each 
vertex. Bilinear dewarping was applied to each 
polygon to square the region. Each pixel in the 
squared region was then mapped to its 
corresponding location (a non-integer quantity) 
in the original image. The value of the pixel was 
obtained by applying a weighted average 
obtained from the four adjacent pixels in the 
mapped location. This technique was applied to 
both zero displacement and actual data 

Note that this dewarping procedure maps pixels 
from one spatial domain to another, causing a 
corresponding change in apparent camera 
magnification to occur. To measure the new 
camera magnification the dot card images were 
dewarped and the pixel to pixel spacing of 
adjacent dots (with a physical spacing of 5.7 mm 
on the card) were measured. The average 
magnification was then easily computed via 

M = 

(N p )(d cl 



where d ctml is the measured dot card dot spacing, 
N p is the average number of pixels between 
adjacent dots on the imaged card, and d camera is 
the physical distance between adjacent pixels in 
the camera. Measurements of M were taken both 
horizontally and vertically to check for 
consistency in the magnification readings. The 
two values of M for this study were identical at 
0.088 since the cameras employed in the 
recording system contained detectors consisting 
of square 6.7-micrometer pixels with 100% fill 
between pixels. 

Image Analysis 

Spatial cross correlation analysis was performed 
for all image pairs obtained from receivers 1 and 
2 comprising the recording system. The analysis 
routines were written by the authors and are 
based on classical DPIV techniques as described 

by Raffel in reference 5. The relevant 
processing parameters employed for the analyses 
are tabulated in Table 1 . 

Table 1. DPIV Processing Parameters 

Analysis Method 

Multiple Frame Cross 

Interrogation Region 


64 Pixels Square 

Interrogation Region 

50 Percent 

Correlation Plane 
Peak Detection 

3-point Parabolic Fit 

Image Threshold 


For each interrogation region examined in a pair 
of acquired images, the fully corrected (i.e., pixel 
aligned) average particle image displacement 

vector, D , was obtained via subtraction of the 
vector obtained by analysis of the interrogation 
region in the dewarped zero displacement 

image, d , , from the vector obtained by analysis 
of the identical interrogation region in the 

dewarped data image, d . The velocity vector 
was subsequently computed using the laser pulse 
separation. The calculation is represented via 




V = 




Given an image size of 1300- by 1030-pixels and 
the processing parameters in Table 1, a velocity 
field containing a maximum of 40 horizontal and 
32 vertical vectors could be generated from the 
analysis. In many cases the velocity field was 
restricted to a subset of the maximum available 
depending on regions of interest in the flow. 

One of the difficulties encountered in 
measurement of separating and reattaching flows 
with DPIV concerns the effect on the cross 
correlation function of high velocity gradients 
present in interrogation regions bounding areas 
of high and low speed flow. For instance, the 
boundary region between a free stream flow and 
separation bubble is typically more difficult to 
accurately measure using cross correlation 
techniques unless gradients are taken into 
account. For the results presented in this paper, 
no corrections for gradients were performed. 
However, the validation of a gradient correction 
technique implemented by Dr. Ahmed Naguib at 
MSU is currently being conducted by the authors 

and will be incorporated into the analysis of the 
SBRT dataset in the near future. 

Velocity Field Validation 

Incorrect velocity vectors introduced to the 
output data by the cross correlation analyses 
were identified and removed using magnitude 
difference algorithms contained in the Clean Vec 
validation system developed by Soloff and 
Meinhart at the Laboratory for Turbulence and 
Complex Flow at the University of Illinois - 
Urbana. 14 No velocity interpolation was 
performed for this study. 


A generalized error analysis for the DPIV 
measurements constituting this study requires 
that the accuracy of individual particle image 
displacements obtained from the cross 
correlation analysis be quantified. This is not an 
easy task due to the complexity of the correlation 
algorithm, and the literature contains numerous 
DPIV uncertainty investigations. Early work 
concentrated on the identification of spurious 
vectors within computed velocity fields and on 
optimization of the instrument to minimize these 
errors. 15 " 16 More recently, Huang and 
others have provided techniques for minimizing 
errors in location estimates of correlation plane 
peaks. 17 

The authors previously derived generalized 
expressions for measured particle displacement 
and velocity uncertainties as part of an 
application of DPIV to an acoustically excited, 
zero-mean flow. 18 Using a derivation similar to 
that given in reference 18, it can be shown for 
the present study via a Taylor series expansion 
that the precision errors associated with 
calculation of individual DPIV displacement and 
velocity vectors can be expressed by 




tfW-) 2 ^ +(77) <: + (-77F-) X, 

-2(-j-) 2 <r,,<x 4 + 2{^-){^-)G d a KI 

1 d-d 

+ 2( M )( ^ )<T <^' 


.d.-d., , „, 1 

A/ 2 Ar 

A/At A/At 

A/At M At 

-, 1 ^ d ~ d -^ 

+ 2 ^r )( 77TT" )cJ ^ tT «' 
A/At A/ At 



where the a terms represent the precision errors 
of the variables appearing in equation (4). Note 
that equations (5) and (6) represent only first 
order approximations to the actual precision 
errors; nevertheless, they are instructive in terms 
of computing approximate bounds on the errors 
expected in this study. To perform the 
computation , the following mean values and 
standard deviations were assumed: 

Table 2. Precision Error Parameters 








40 usee 

100 nsec 


20 -67 urn (3 -10 
pixel displacement) 

0.67 urn 
(0.1 pixel) 

d z 

0-13 urn (0-2 
pixel displacement) 

0.67 urn 
(0.1 pixel) 

The standard deviations in Table 2 were chosen 
based on conservative estimates of expected 
precision errors in the measurement of each 
variable. For instance, the standard deviation in 
the measured displacement, d, and the measured 
zero pulse separation displacement, d z , were 
based on an assumption that the correlation peak 
could be located to 0.1 -pixel accuracy using the 
3 -point parabolic fit routine used in the analysis. 
The standard deviation in the measured 
magnification was based on an assumption that 
the centroids of dots identified on the dot card 
could be located to at least ±1 -pixel accuracy. 
Finally, the standard deviation in the pulse 
separation was based on an assumption that the 
pulse generators used to control the laser timing 
had internal jitters no larger than 
100 nanoseconds. 

Figure 1 1 shows some expected velocity 
uncertainties using equation (6) with the 
parameters shown in Table 2. The uncertainty is 
plotted as a percentage of the measured velocity, 
with these velocities computed using equation 
(4). A family of curves is shown for three 

different zero pulse separation displacements. 
As one would expect, the percent uncertainty 
increases as the measured velocity decreases 
since the measured velocity decreases faster than 
the standard deviation of the measurement. 
Nevertheless, for the velocity range shown in 
Figure 1 1 for reasonable values of d z , the 
precision error is of the order of 5% or less. 

Note that this analysis addresses precision errors 
only. Predominant bias errors are represented by 
the magnitude of any residual displacements 
which remain after the dewarped and pixel 
aligned zero displacement measurements d : are 
subtracted from the data via application of 
equation (4). The full quantization of these 
residual bias errors is part of an on-going 
investigation by the authors. 


A total of 32,800 individual image frames 
(representing 16,400 DPIV data sets) were 
acquired over four days of testing using the MSU 
separated flow model in SBRT. The majority of 
the data was acquired at a freestream velocity of 
15 m/sec and a Reynolds number of 8000 based 
on the total fence height. Figure 12 shows one 
instantaneous velocity vector field derived from 
cross correlation analysis of a typical set of 
image data. The vector field shows reattachment 
of the separated flow region at approximately 1 3 
percent of the chord length of the splitter plate. 
The vector fields acquired by the two receivers 
in the recording system are denoted in the plot. 
Note that overlapped vectors are not shown in 
this set of data. The separated flow region is 
clearly visible in the right half of the vector field, 
with reattachment occurring midway along the 
field of view of the downstream receiver. 


A novel multiple-camera system for the 
recording of digital particle image velocimetry 
(DPIV) images acquired in a two-dimensional 
separating / reattaching flow was successfully 
deployed for the first time at NASA Langley. 
The recording system contained two pairs of 
matched high resolution, 8-bit cameras which 
were used to separate and capture images of 
illuminated tracer particles embedded in the flow 
field. The data analysis system employed 
several custom algorithms which were applied to 
the acquired data. Background subtraction based 
on generation of minimum images was 

performed to reduce undesirable flare light 
emanating from the surface of the model. Image 
dewarping and pixel alignment algorithms using 
information derived from an examination of dot 
cards placed in the plane of the light sheet were 
performed to provide accurate pixel registration 
among the various cameras. Spatial cross 
correlation analysis with median filter validation 
was used to determine the instantaneous velocity 
structure in the separating / reattaching flow 
region illuminated by the laser light sheet. In 
operation the DPIV system exhibited a good 
ability to resolve large-scale separated flow 
structures with acceptable accuracy over the 
extended field of view of the recording system. 
A simple first-order precision error propagation 
using conservative estimates of the standard 
deviations derived from camera magnification 
and laser timing measurements and cross 
correlation analysis yielded expected velocity 
precision errors of 5% or less. The authors 
believe the multiple camera system described in 
this paper is a good alternative for applications 
where cross correlation cameras are either not 
available or are impractical. Further research is 
required, however, to fully quantify pixel 
alignment bias errors associated with the use of 
separate cameras for cross correlation analysis. 


The authors gratefully acknowledge the 
contributions of Dr. Ahmed Naguib and Laura 
M. Hudy of the Department of Mechanical 
Engineering at Michigan State University for 
their design and construction of the model used 
in this study, and for their numerous suggestions 
and advice regarding the analysis of the data 
obtained with the DPIV system. 


1) Cherry, N.J., Hillier, R., and Latour, M., 
"Unsteady Measurements in a Separated and 
Reattaching Flow", Journal of Fluid 
Mechanics, Volume 144, pp. 13-46, 1984. 

2) Ruderich, R., and Fernholz, H.H., "An 
Experimental Investigation of a Turbulent 
Shear Flow with Separation, Reverse Flow, 
and Reattachment", Journal of Fluid 
Mechanics, Volume 163, pp. 283-322, 1986. 

3) Smits, A.J., "A Visual Study of a Separation 
Bubble", Proceeding from International 
Symposium on Flow Visualization, Bochum, 

4) Grant, I., Owens, E., and Yan, Y., "Particle 
Image Velocimetry Measurements of the 
Separated Flow Behind a Rearward Facing 
Step", Experiments in Fluids, Volume 12, 
pp. 238-244, 1992. 

5) Raffel, M., Willert, C, and Kompenhans, J., 
Particle Image Velocimetry: A Practical 
Guide, Springer- Verlag, New York, 1998. 

6) Stanislas, M., Kompenhans, J., and 
Westerweel, J. (Eds.), Particle Image 
Velocimetry: Progress Towards Industrial 
Application, Kluwer Academic Publishers, 
Boston, 2000. 

7) Samimy, M, and Wernet, M.P., "Review of 
Planar Multiple-Component Velocimetry in 
High-Speed Flows", AIAA Journal, Volume 
38, Number 4, pp. 553-574, 2000. 

8) Westerweel, J., "Fundamentals of Digital 
Particle Image Velocimetry", Measurement 
Science and Technology, Volume 8, pp. 
1379-1392, 1997. 

9) Kahler, C.J., Kompenhans, J., 
"Fundamentals of Multiple Plane Stereo 
Particle Image Velocimetry", Experiments 
in Fluids (Supplement), pp. S70-S77, 2000. 

10) Hudy, L.M., "Simultaneous Wall-Pressure 
Array and PIV Measurements in a 
Separating/Reattaching Flow Region", 
Master's Thesis, Michigan State University, 
August, 2001. 

11) Meyers, J.F., "Development of Doppler 
Global Velocimetry for Wind Tunnel 
Testing", AIAA Paper 94-2582, 1994. 

12) Kuhn, W., Kompenhans, J., Monnier, J.C., 
"Full Scale PIV Test in an Industrial 

Facility", Particle Image Velocimetry: 
Progress Towards Industrial Application, 
Kluwer Academic Publishers, Boston, pp. 
91-150, 2000. 

13) Meyers, J.F., "Doppler Global Velocimetry - 
The Next Generation?", AIAA Paper 92- 
3897, 1992. 

14) Soloff, S.M., and Meinhart, CD., CleanVec: 
PIV Vector Validation Software, Version 
1.13 Build 41, Released 1999. 

15) Keane, R.D., and Adrian, R.J., 
"Optimization of Particle Image 
Velocimeters, Part I: Double Pulsed 
Systems", Measurement Science and 
Technology, Volume 1, pp. 1202-1215, 

16) Host-Madsen, A., and McCluskey, D.R., 
"On the Accuracy and Reliability of PIV 
Measurements", Seventh International 
Symposium on Applications of Laser 
Anemometry to Fluid Mechanics, Lisbon, 
Portugal, paper 26.4, 1994. 

17) Huang, H„ Dabiri, D., and Ghanb, M., "On 
Errors of Digital Particle Image 
Velocimetry", Measurement Science and 
Technology Volume 8, pp. 1427-1440, 

18) Humphreys, W.M., Bartram, S.M., Parrott, 
T.L., and Jones, M.G., "Digital PIV 
Measurements of Acoustic Particle 
Displacements in a Normal Incidence 
Impedance Tube", AIAA Paper 98-2611, 

Figure 1. NASA Langley Subsonic Basic Research Tunnel. 

Figure 2. MSU Separated Flow Generator Model in SBRT. 
Note Microphones, Pressure Taps, and DPIV Windows. 

Beam-Combining Optics 

:Nd:YAG*1 > 


Sheet Forming Optics 

Light Sheet 

Rmax @ 45° 


Rmax @ 0° 

1/2 X. 

57° BS 

To Light 
Sheet Optics 



From Beam M * 

Figure 3. DPIV Laser Optics System. 

1 10 

Aerodynamic Particle Size, urn 

Figure 4. DPIV Laser Optics System in SBRT. Figure 5. DEHS Nominal Particle Size Distribution. 

Figure 6. DPIV Recording System in SBRT. 


S and P Polarized 

Light Sheet 




S Polarized 




Camera 2 


Camera 1 

Figure 7. Recording System Receiver Configuration. 

Figure 8. Polarization Separation Concept. 

Master Sync J Laser Timing 
Generator I Controller ; 

■; Flash Lamps 
; Q-Switches 

Slave Pulse 

► Camera Sync 
. Digitizer Sync 

Figure 9. Laser / Camera Synchronization. 


Figure 10. Pixel Alignment Dot Card. 

1 ' 1 


1 1 


1 ' 1 ' 


d z = 0.0 urn 




d z = 6.7 |im 
d 2 = 13.4 nm 






T • t 



» ' 




' ' J 





1 1 

1 1 



5 10 15 20 25 

Measured Velocity, m/sec 
Figure 11. Velocity Uncertainty Estimates. 


60 i— 

50 •- 

1 40 









© 20 

10 T| 

— Region of 


Interest j 



^>. <-/ //</< / / ,,,,,,, • ,• ,,,, 

, , , , t , • , ' ' 

*,,, ,,,,,/ 1,1 liliitli'il.l Ii/l'l ■ 

, ,, , ,, - 

'tti ,,,,,, //,,,//J,,t,,,,,i>, ,,,.,,,, 

,,,,,, ' 

.Oil, t<>, lit,!/, J, ,1,1,1, ,,,,,',',/, 

, i , , , 

7 U.,I,,,<I ///>•/, /,,.,,,, //,./', •,,,, 

■,<.„>,.<,;;,<, , "<<- 

.,,,, ,,,' 

'l,,,/,,,,uM,'nJ,.<'. ,,,,,*>. <<.,,,/,> 

L'„..',it,l,'t„l,.t. j/j, ,<>,.. ,</</., 
til" i , ltl , 

i.i,,,. ,//,, (,..<!„.•<■ > •-, 

1i„.,t u nii^ l i,"'itii > •!!> >•> 

It .Imlt <U*l> huuw n. «i.W' 

S , • ' ' 

Left Camera 

Right Camera 


200 1 50 

Streamwise Coordinate, mm 


Figure 12. Representative Separated Flow Data Obtained Downstream of 
Splitter Plate Leading Edge Fence.