effect of tip vortex aperiodicity on measurement uncertainty
TRANSCRIPT
RESEARCH ARTICLE
Effect of tip vortex aperiodicity on measurement uncertainty
Mahendra J. Bhagwat • Manikandan Ramasamy
Received: 22 December 2011 / Revised: 4 June 2012 / Accepted: 25 June 2012 / Published online: 24 July 2012
� Springer-Verlag (outside the USA) 2012
Abstract Vortex aperiodicity introduces random uncer-
tainty in the measured vortex center location. Unless cor-
rected, this may lead to systematic uncertainty in the vortex
properties derived from the measured velocity field. For
example, the vortex core size derived from averaged or mean
flow field appears larger because of aperiodicity. Several
methodologies for aperiodicity correction have been devel-
oped over the past two decades to alleviate this systematic
uncertainty. However, these do not always reduce the
accompanying random uncertainty. The current work shows
that the analysis methods used to derive the vortex properties
from the measured velocity field play an important role in
the resultant random uncertainty in these properties; perhaps,
even more important role than the aperiodicity correction
methodology itself. It is hypothesized that a class of methods
called global methods, which use a large extent of measured
data, yield a smaller measurement uncertainty compared to
local methods. This hypothesis is verified using a newly
proposed global method based on a planar least-squares fit.
The general applicability of the method is demonstrated
using previous particle image velocimetry measurements of
rotor tip vortices. The results clearly demonstrate a reduced
random uncertainty in the vortex core properties, even in the
presence of secondary vortical structures. Furthermore, the
results are independent of the choice of aperiodicity cor-
rection methodology.
1 Introduction
Over the last decade, particle image velocimetry (PIV) has
found increasing use in rotor flow field and tip vortex mea-
surements, and with great success—see, e.g., Raffel et al.
(2007). The applications have spanned a wide range from
very large-field (e.g., Wadcock et al. 2011) to microscopic
(e.g., Ramasamy et al. 2010). As the PIV measurement res-
olution increases, attention is again focused on understanding
the associated measurement uncertainties. The two types of
uncertainties are systematic uncertainty, which refers to
accuracy, bias, offsets, etc., and random uncertainty, which
refers to variability, repeatability, precision, etc. Systematic
uncertainty is typically quantified using calibration and spe-
cially designed measurements.
A unique contributor to uncertainty in tip vortex mea-
surements is their wandering or meandering motion; this is
a seemingly random movement of the vortex center loca-
tion normal to the vortex axis. For rotor tip vortices, this is
often referred to as aperiodicity because the wandering
motion appears as variation in vortex center location from
one rotor revolution (period) to another. While tip vortices
exhibit some inherent wandering motion, the blade aperi-
odicity, i.e., the variation in blade location from one period
to another (see, e.g., van der Wall and Schneider 2007),
also contributes to the perceived wandering motion of the
tip vortex. This vortex aperiodicity appears as the random
uncertainty in the measured tip vortex location. However, it
also results in additional systematic uncertainty called
‘‘Type II bias,’’ which is a result of transformation of the
measurement variable(s) through a nonlinear model. For
example, the vortex aperiodicity leads to a larger perceived
vortex core size. Several methods have been developed to
correct this uncertainty due to vortex aperiodicity. How-
ever, these methods do not always ensure that the random
M. J. Bhagwat (&)
US Army Aeroflightdynamics Directorate (AMRDEC)
Ames Research Center, M/S 215-1,
Moffett Field, CA 94035, USA
e-mail: [email protected]
M. Ramasamy
UARC/AFDD, NASA Ames Research Center,
M/S 215-1, Moffett Field, CA 94035, USA
123
Exp Fluids (2012) 53:1191–1202
DOI 10.1007/s00348-012-1348-7
uncertainty does not grow in the process of removing or
minimizing the systematic uncertainty. In this paper, the
aperiodicity correction methodologies are reviewed in the
context of their effect on the overall resulting measurement
uncertainty in the vortex properties.
Vortex wandering has been a particularly challenging
problem for point measurement techniques like hot-wire
anemometry (HWA) or laser Doppler velocimetry (LDV),
where an averaged flow field can only be constructed by
independently averaging measurements at each point. One
of the earliest attempts to assess the effect of wandering on
tip vortex core properties is reported by Baker et al. (1974).
Using this aptly called ‘‘simple average,’’ the wandering
motion also contributes to velocity fluctuations at a fixed
location, which may be mistaken for turbulence. This effect
can be eliminated by conditional sampling (Takahashi and
McAlister 1987) or must be corrected using some measure
of the vortex wandering. Devenport et al. (1996) present an
analytic correction procedure based on measured wander-
ing magnitude using an assumed velocity profile. Leishman
(1998) presents a similar numerical correction procedure
using the measured vortex velocity profiles. Planar mea-
surement techniques like PIV present other alternatives to
help mitigate the effect of vortex aperiodicity.
One of the first PIV measurements of rotor tip vortices
by Raffel et al. (1998) showed that averaging several PIV
planar measurements (images) gives close agreement with
the corresponding LDV measurements, which inherently
include vortex aperiodicity effects in the measured aver-
aged flow field. Instantaneous planar PIV measurements
provide the ability to identify and correct for the aperio-
dicity effects. Yamauchi et al. (1999) present tip vortex
PIV measurements for the TRAM rotor and describe the
aperiodicity correction methodology in detail. The proce-
dure is similar in spirit to conditional sampling, where only
a small set of samples with the vortex centers close toge-
ther are chosen. The individual samples are aligned using
their respective vortex centers prior to averaging, i.e., the
measurements are ‘‘corrected’’ for the movement of vortex
center location before averaging—hence the name ‘‘cor-
rected average’’. Unlike point measurement techniques,
PIV allows one to obtain the vortex core properties from
each individual planar measurement (image) as well. The
average of these individual (instantaneous) measurements
would also be representative of mean vortex properties.
To better understand these aperiodicity correction
methodologies, examples of simple average and corrected
average are shown in Fig. 1. Measurements are shown as
velocity vector field colored with velocity magnitude, with
the red being the highest and the blue being the lowest. For
visualization purposes, the vortex center is shown with a
filled circle, and the approximate circular vortex core
boundary, corresponding to the local maxima in swirl
velocity, is shown with a dotted line. The vortex core
shows a data void because of the difficulties in seeding the
flow inside the core. Large variations in the vortex center
location are readily apparent even in these four instances.
When a number of such individual images are averaged to
obtain mean flow properties, the resulting flow field is
smeared because of the variation in the vortex center
location. This is clearly seen in the simple average flow
field, where the vortex appears larger in size, and with
lower swirl velocity, than any of the instantaneous mea-
surements (core boundaries shown by dotted lines).
An intuitively better methodology for obtaining the
mean flow characteristics is to first align the instantaneous
measurements using their respective vortex centers, as
shown in Fig. 1. This is the central idea of the corrected
average or conditional average for aperiodicity correction.
In this case, the averaging does not produce the smearing
effect seen with simple average. In the corrected average
flow field, the vortex looks nearly the same size as the
instantaneous measurements.
The third methodology is the individual average, which
does not ‘‘correct’’ the individual measurements for ape-
riodicity. Instead, the vortex properties are derived from
each instantaneous measurement and then averaged to give
mean vortex properties. This also readily provides an
estimate of measurement uncertainty in terms of mean and
standard deviation of the desired vortex property.
Ideally, any of these methodologies would be equally
effective. However, Yamauchi et al. (1999) found that the
average of individual vortex measurements was signifi-
cantly different from the ‘‘corrected’’ average obtained by
aligning the individual measurements. This was attributed
to large scatter in the individual vortex properties, i.e., to
measurement uncertainty. Later experiments on a model-
scale rotor (Ramasamy et al. 2007) also showed such dis-
crepancy between averaged individual measurements and
corrected average profiles, while for the HART II mea-
surements, the conditional average is considered manda-
tory (Lim and van der Wall 2010). One motivation for the
current work is to better understand and, eventually, to
eliminate these inconsistencies.
The underlying experimental uncertainty in the velocity
measurements itself is outside the scope of the current
work. However, the choice of analysis method is important
in controlling the propagation and amplification of this
uncertainty into derived quantities of interest like the
vortex core size and circulation. Determining the vortex
center from the velocity field is an important first step for
the corrected average process. van der Wall and Richard
(2006) suggest that a convolution-based method improved
the accuracy of vortex center identification and resulted in
reduced scatter in the derived vortex properties of interest.
This claim is explored further in this paper, and the
1192 Exp Fluids (2012) 53:1191–1202
123
improvement in the final measurement uncertainty in vor-
tex properties resulting from the vortex center identifica-
tion method is examined. Note that while vortex
aperiodicity can be a big contributor to the uncertainty in
the derived vortex properties, the methods used to derive
those properties can be even bigger contributors.
The current work briefly reviews various methods for
determining the vortex center and other properties of
interest, and how these methods can potentially affect the
final uncertainty. The aperiodicity correction removes a
systematic uncertainty in the vortex core radius (and peak
swirl velocity). However, the analysis method determines
how the random uncertainty propagates to the derived
quantity. It is important to ensure that the method does not
add/amplify random uncertainty which may mask the
benefits of aperiodicity correction. A new method based on
a planar least-squares fit is proposed for obtaining both the
vortex center and other vortex properties of interest. This
method is shown to give reduced uncertainty in derived
quantities like vortex core radius, circulation, etc. Fur-
thermore, with the planar fit method, the three aperiodicity
correction methodologies give self-consistent results. The
method is first demonstrated using the AFDD model-rotor
tip vortex measurements (Ramasamy et al. 2010) and then
extended to other measured flow fields including the
TRAM rotor measurements (Yamauchi et al. 1999) and the
HART II measurements (Lim and van der Wall 2010).
2 Methodology
This section gives an overview of tip vortex aperiodicity
correction, followed by methods to determine the vortex
properties. These latter methods are categorized into local
and global methods based on the spatial extent of the data
used relative to the vortex core. The newly proposed
method, a planar least-squares fit, is then described in detail.
2.1 Aperiodicity correction
As described earlier in the introduction, there are three
categories of aperiodicity correction methodologies: the
simple average (SA), the corrected average or conditional
average (CA), and the individual average (IA). The SA can
Fig. 1 Schematic showing simple average and corrected average using a few sample PIV vector fields. The simple average gives a larger core
size than the original samples, whereas the corrected average gives nearly the same core size
Exp Fluids (2012) 53:1191–1202 1193
123
be traced back to fixed-point measurement techniques like
LDV/HWA and is often combined with some empirical or
semi-analytic aperiodicity correction—see Devenport et al.
(1996) and Leishman (1998). In the current work, the
analytic expression given by Devenport et al. (1996) is
used to correct the core radius obtained using SA. The
corrected average and individual average are possible only
with a field measurement technique like PIV, and can be
computationally intensive. However, these are applicable
even with arbitrarily large aperiodicity magnitude, unlike
the aperiodicity correction applied to the SA. The third
method, IA, has its advantages in simplicity; no flow field
averaging is required, and only the vortex properties
derived from each individual (instantaneous) measurement
are averaged. This also provides an estimate of measure-
ment uncertainty in the form of mean, l, and standard
deviation, r, of each vortex property of interest, i.e., the
measured property has a 68 % confidence level in
the l ± r interval.
All aperiodicity correction methodologies belong to one
of these three categories and differ only in their imple-
mentation details like the method for determining the
vortex center and/or other vortex properties. These meth-
ods are divided into two categories: ‘‘local’’ methods and
‘‘global’’ methods. As the names would suggest, the local
methods are mainly based on the flow field in close vicinity
of the vortex center while the global methods are based on
a much larger flow field surrounding the vortex. In other
words, the local flow field near the vortex center is the most
important for the local methods. It should be noted that the
global/local categorization is not very rigid; it is simply a
means to explain the observed differences in these meth-
ods. It seems intuitively obvious that a global method that
uses a large amount of measured data would be better than
local methods that use a very small subset confined to the
vicinity of a local maxima (often including measurements
with lower signal-to-noise ratio and excluding those with
possibly higher signal-to-noise ratio). However, this
hypothesis remains to be carefully examined.
2.2 Methods to determine vortex center
The most common methods to calculate the vortex center
from measured flow field are based on some flow invariant
property like vorticity, k2, the Q-criterion, etc. These flow
quantities exhibit a local extremum at the vortex center—
hence the name ‘‘local methods.’’ Commonly used local
methods include the measurement grid node that carries the
maximum value of vorticity (e.g., Vogt et al. 1996; Hei-
neck et al. 2000; Yamauchi et al. 1999; Ramasamy et al.
2007a), helicity, axial velocity (e.g., Ramasamy et al.
2009), or Q-criterion (e.g., Ramasamy et al. 2009; van der
Wall and Richard 2006). Centroid or area center of
vorticity (e.g., Ramasamy et al. 2009; Lim and van der
Wall 2010) has also been used to determine the vortex
center. Ideally, the vortex center defined by any of these
criteria is the same. However, the vortex center estimates
do not always agree (e.g., Lim and van der Wall 2010;
Ramasamy et al. 2007a). Cucitore et al. (1999) present a
summary of the effectiveness and limitations of such local
vortex identification criteria.
One short-coming of all of these methods is that they
rely mostly on measurements near the vortex core. The
measurements near the vortex core have a lower signal-to-
noise ratio and correspondingly larger statistical errors
because of the challenges in seeding the flow around the
vortex core. This often results in a seed void near the
vortex center and a corresponding, but often smaller, data
void. In fact, the seed void is a fairly reliable indicator of
vortex center location (e.g., Leishman 1996; McAlister
2004) and has even been used for corrected averaging
(McAlister 2004). For the AFDD model-rotor vortex
measurements, data near the vortex core with low signal-
to-noise ratio were eliminated and were not replaced using
interpolated vectors, rendering local methods inapplicable.
A global method is clearly necessary. One global method
found in the rotorcraft literature is the convolution-based
method used by van der Wall and Richard (2006); this
showed improved accuracy in determining the vortex
center. Although the measurements reported by van der
Wall and Richard (2006) did not show a data void near the
center, the results definitely demonstrated the benefit of
using a larger measured flow field rather than only a few
measurement points inside the vortex core.
2.3 Methods to determine vortex properties
Having determined the vortex center, the next step is to
determine key vortex properties like vortex core radius,
peak swirl velocity, circulation, etc. The choice of methods
for determination of vortex core properties from either the
averaged or the instantaneous flow field is also independent
of the aperiodicity correction. For the CA method, the
vortex center for each individual measurement must be
determined prior to determining the vortex core properties
from the averaged flow field. For SA, the subsequent cor-
rection, like the Devenport correction, also needs the vor-
tex center locations (aperiodicity magnitude). In fact, the
accurate estimation of the vortex center is a necessary
prerequisite for the accurate estimation of the vortex core
properties (van der Wall and Richard 2006).
The methods to determine vortex properties can also be
divided into local and global categories. In this case, the
local methods are usually based on measurements along
straight-line cuts through the vortex center, perhaps using
the analysis techniques developed for LDV/HWA
1194 Exp Fluids (2012) 53:1191–1202
123
measurements. The peak swirl velocity along these cuts
defined the vortex core radius and asymptotic value of
circulation defined the vortex strength (Yamauchi et al.
1999). These methods are considered local methods
because the core radius and peak swirl velocity are deter-
mined from a local maximum in velocity. A variation of
this uses a linear curve fit (based on an assumed vortex
model) to the data along a straight line (e.g., Lim and van
der Wall 2010; Bhagwat and Leishman 2000). A global
method based on a fit to the entire measurement plane (or a
suitably large enough subset of that) was used by van der
Wall and Richard (2006) with good success.
2.4 New method: planar least-squares fit
The newly proposed method is based on a planar least-
squares fit of the measured velocities to those given by an
assumed vortex model. The center of the vortex (xc, yc) is
defined in the measurement coordinates, x; y; zð Þ. The
vortex convection velocity or background velocity is rep-
resented by uc; vcð Þ; also in the measurement coordinates.
In general, the measurement plane is not normal to
the vortex axis. Two rotation angles are required for the
transformation from the measurement coordinates to the
vortex coordinates, xv; yv; zvð Þ. This is schematically shown
in Fig. 2 where the PIV measurement window is shown
along with the measurement and vortex coordinates. A first
rotation of hy around the y-axis results in an intermediate
coordinates x1; y1; z1ð Þ and a subsequent rotation hx around
the x1-axis gives the vortex coordinates. The transforma-
tion from the measurement coordinates to the vortex
coordinates is then given in terms of the transformation
matrices Rx and Ry as
Rx ¼1 0 0
0 cos hx sin hx
0 � sin hx cos hx
264
375 Ry ¼
cos hy 0 � sin hy
0 1 0
sin hy 0 cos hy
264
375
xv
yv
zv
8><>:
9>=>;¼ Rx½ � Ry
� � x� xc
y� yc
z
8><>:
9>=>;
r ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffix2
v þ y2v
qð1Þ
where r is the vortex radial coordinate. The vortex induced
radial, axial and swirl velocities Vr;Vz;Vhð Þ are
transformed to the Cartesian vortex coordinates as
uv; vv;wvð Þ through the transformation matrix Rc as
uv
vv
wv
8<:
9=; ¼ RC½ �
Vr
Vz
Vh
8<:
9=; RC½ � ¼
xv
r 0 � yv
ryv
r 0 xv
r0 1 0
24
35 ð2Þ
These velocities are then transformed back to the
measurement coordinates using the inverse transform.
The results shown in current work use the Lamb-Oseen
model for vortex induced swirl velocity, i.e.,
Vh ¼Cv
2pr1� exp �a r=rcð Þ2
� �� �ð3Þ
where Cv is the vortex circulation, rc the core radius and
a = 1.25643 is a constant parameter. The resulting
function for the vortex velocity field is given by
f rc;Cv; xc; yc; uc; vc; hx; hy
� �¼ Ry
� �TRx½ �T RC½ �
Vh
0
0
8><>:
9>=>;
þuc
vc
0
8><>:
9>=>;
ð4Þ
The planar least-squares fit uses the above function with
the eight independent variables, viz., rc;Cv; xc; yc; uc; vc; hx;
hy; to obtain the best match with measured velocities.
The laminar Lamb-Oseen model is perhaps the most
commonly used viscous vortex core model, and has also
been employed to identify vortex structures in separated
flows (e.g., Schram et al. 2004; Cierpka et al. 2008). The
proposed method, however, is not restricted to the laminar
Lamb-Oseen model and can readily incorporate other
vortex models. It was shown by Ramasamy et al. (2010)
that the choice of vortex models did not affect the vortex
center location—each model gave a slightly different core
radius corresponding to different values of the core to total
circulation ratio. Therefore, the vortex model should not be
predetermined, but should rather be chosen based on the
actual measured flow field. All the results shown in the
current work correspond to relatively low vortex Rey-
nolds numbers Cv=mð Þ; where m is the kinematic viscos-
ity. Therefore, the Lamb-Oseen model was found to be
x1
z1
y1=y
xy
z
y
x
Measurement window
Vortex coordinates
Measurement coordinates
Vortex axis
Vortex center xc, yc
xv
yv
zv
xv =x1
yv
zv
Fig. 2 Inclination of vortex axis relative to measurement plane
Exp Fluids (2012) 53:1191–1202 1195
123
satisfactory. Furthermore, the methodology can easily
include all three vortex induced velocity components,
although only the vortex induced swirl velocity is used
here.
3 Results and discussions
In this section, the results obtained using the planar fit are
shown for three different rotor tip vortex measurements.
The first set of measurements is the model-scale rotor tip
vortex measurements made by Ramasamy et al. (2010) at
AFDD using microscopic-PIV using highly twisted tilt-
rotor blades. The second set of measurements are the
TRAM rotor measurements originally reported by Yama-
uchi et al. (1999), and the final set of measurements is the
HART II measurements at position 17, studied in detail by
all the HART II partners, and summarized by Lim and van
der Wall (2010). The HART II measurements in hover
reported by van der Wall and Richard (2006), and a small-
scale rotor measurements from University of Maryland
(UMD) by Ramasamy et al. (2007b) are also shown for
comparison.
3.1 AFDD model-rotor tip vortices
The measurements presented here were performed on a
three-bladed model-scale rotor system with a highly twisted
rotor blade that was similar to the XV-15 blade. The blade
had a radius R = 656 mm and chord c = 49 mm and was
operated at a rotational speed of X ¼ 800 rpm, i.e., tip speed
of XR ¼ 55 m/s. Results at a rotor thrust coefficient
CT = 0.0076 are shown at two early vortex ages of 4� and
15�. The thrust coefficient is defined as CT ¼ T=qpR2ðXRÞ2;where T is the rotor thrust and q is the air density. The vortex
age, wv; is the azimuth behind the blade; 4� and 15� corre-
spond to a downstream distance of 0.9 chords and 3.5 chords
or a non-dimensional time based on tip speed and chord of
0.0008 and 0.003, respectively.
Figure 3 shows a summary of all the vortex properties
for CT = 0.0076 and wv ¼ 15�. A map of vortex center
locations and the representative core size are shown in
Fig. 3a. The grid lines correspond to the measurement grid
and show the high resolution achieved using a microscope.
Recall that the corrected average procedure used in the
current work does not re-mesh and interpolatemeasure-
ments after aligning the vortex centers. As a result, the
corrected average does not completely eliminate the ape-
riodicity, but only reduces it to a value less than one grid
cell. The residual aperiodicity can be seen in Fig. 3a from
the corrected vortex center locations. The distribution of
vortex center locations xc and yc shown in Fig. 3b is sug-
gestive of a Gaussian distribution. The Devenport
correction is based on the deconvolution of a Gaussian
wandering motion; the observed distribution suggests that
such an analytic correction would indeed be applicable.
Figure 3c and d show the results for the vortex axis
orientations, hx and hy, and the mean vortex convection
velocities, uc and vc, respectively. The vortex core radius
and circulation are shown in Fig. 3e and f, respectively. All
three aperiodicity correction methodologies give results
consistent with each other for all vortex properties. For the
core radius, in Fig. 3e, the SA method gives a larger core
radius as expected; the subsequent Devenport correction
gives a core radius that closely agrees with the CA and IA
results. Notice that in all cases, the vortex properties show
a roughly Gaussian distribution. The normalized standard
deviation in circulation and core radius is about 5 %, which
is indicative of fractional random uncertainty. The differ-
ences in vortex properties obtained using the three aperi-
odicity correction methodologies are smaller than one
standard deviation.
The measurements are compared with three previously
reported rotor PIV measurements: the UMD measurements
from Ramasamy et al. (2007b) of small-scale rotor tip
vortices; the TRAM measurements from Yamauchi et al.
(1999) for the high thrust case; and the HART II mea-
surements from Fig. 12 of van der Wall and Richard
(2006), for the wv ¼ 44� hover case. The corresponding
vortex aperiodicity magnitudes (standard deviation of
vortex location normalized by core radius) are also shown
for comparison. Interestingly, the two measurements with
higher aperiodicity magnitudes showed lower resulting
measurement uncertainties. It appears then that uncertainty
must result from other factors like the measurement
resolution.
For tip vortex measurements, resolution is not compared
as an absolute length scale, but relative to the vortex core
being measured. The ratio of the measurement length,
lm, or the PIV measurement window size, to the core
radius, i.e., lm/rc, has been used in the past as a measure of
relative PIV resolution. This is schematically shown at the
top of Fig. 4 with the squares representing the PIV pro-
cessing window relative to the vortex core size shown by
the circle. The AFDD model-rotor measurements have a
significantly lower lm/rc ratio owing to the use of a
microscopic lens. However, this appears to have played
only a small part in the resulting measurement uncertainty.
The UMD measurements have a small lm/rc ratio but a
larger uncertainty, while the HART II measurements have
a larger lm/rc ratio but a smaller uncertainty.
It must be reiterated that the uncertainty in derived
quantities such as the vortex core radius, or circulation, is
not the same as measurement uncertainty. Instead, it is the
compound result of how the underlying measurement
uncertainty in the velocity propagates into the derived
1196 Exp Fluids (2012) 53:1191–1202
123
(a) (b)
(d)(c)
(e) (f)
0
5
10
15
20
0.03 0.04 0.05 0.06 0.07 0.08
No.
of s
ampl
es
Vortex core radius, rc /c
SA=0.0746308SA+D=0.05717CA=0.0548237IA=0.056778±0.00333122
Fig. 3 Example of vortex properties obtained using the planar fit for the
AFDD model-rotor tip vortex measurements, CT ¼ 0:0076;wv ¼ 15�.Distribution of individual samples is shown as histograms along with the
simple average, corrected average, and individual values (mean and
standard deviation) are shown. a Vortex center locations before and after
center alignment, b Vortex convection velocities, uc, vc, c Vortex center
locations, xc, yc, d Vortex axis orientation angles, hx, hy, e Vortex core
radius, rc, and f Vortex circulation, Cv
Exp Fluids (2012) 53:1191–1202 1197
123
vortex properties (like core radius/circulation), along with
any additional uncertainty associated with the derivation
method itself. Therefore, it is important to examine the
analysis methods and to ensure that their contribution to
the resulting uncertainty, over and above the underlying
measurement uncertainty, is minimal. For example, gradi-
ent-based criteria can be severely contaminated by mea-
surement noise and lead to incorrect vortex center
identification (Kindler et al. 2011), and larger uncertainty
in the derived vortex properties. One advantage of the
current planar least-squares fit is that it directly uses the
measured flow field and does not require numerical com-
putation of flow derivatives.
One key difference between the four results shown in
Fig. 4 lies in the analysis methods—clearly, they have a
strong effect on the final perceived measurement uncer-
tainty. The UMD and TRAM experiments used local
methods for obtaining the vortex center and other vortex
properties. The HART II results, in particular, the results
presented in this figure from van der Wall and Richard
(2006), used global methods. The current work also uses a
global method as described earlier. Figure 4 strongly sug-
gests that using global methods for data analysis result in
significantly lower uncertainty in derived vortex properties.
This working hypothesis is tested here in two ways: first by
reanalyzing the AFDD measurements using a local method,
and then by reanalyzing previous measurements using a
global method (the proposed planar fit).
3.2 Local versus global method
The AFDD model-rotor measurements were analyzed
using a local method are shown in Fig. 5 for the case of
CT ¼ 0:0076;wv ¼ 4�. In this case, the vortex center was
first identified using the planar least-squares fit. The vortex
properties were, however, not obtained from the fit.
Instead, horizontal and vertical cuts were made through the
vortex center, and the vortex core was identified using the
peak swirl velocity along these cuts, the distance of the
peak swirl location from the vortex center giving the core
radius. These results are shown as a scatter plot in Fig. 5a
with the core radius for each measurement sample along
with the averaged values. The peak swirl velocity location
is not interpolated to a sub-grid level, but is restricted by
the grid resolution. This introduces some additional
uncertainty in the vortex core size. The results show some
inconsistencies—the SA core radius is much smaller than
0 5
10 15 20 25 30 35 40
UMDRamasamy et al
(2007b)
TRAMYamauchi et al
(1999)
HART-IIvan der Wall &
Richards (2006)
Currentwork
Unc
erta
inty
, σ/
μ (%
)
Core radiusCirculation
0
10
20
30
40
50
60A
perio
dici
ty,
σ/r c
(%
)
Fig. 4 Measurement uncertainty in the vortex core radius and
circulation compared for various experiments including the AFDD
model-rotor measurements. The vortex aperiodicity magnitude (stan-
dard deviation normalized by vortex core radius) is also shown for all
cases along with a schematic representation of the vortex core and the
relative PIV window size
(a) (b)
Fig. 5 Vortex core radius for the AFDD model-rotor derived from the measured velocity field using a a local method (based on peak swirl
velocity along horizontal/vertical cuts through the center), and b a global method (the planar fit). CT ¼ 0:0076;wv ¼ 4�
1198 Exp Fluids (2012) 53:1191–1202
123
the CA result and the IA result. The IA results show a
significantly large scatter of about 23 %, compared to those
obtained using the planar least-squares fit shown in Fig. 5b.
The local method clearly introduces a much larger uncer-
tainty in the vortex core radius as compared to the global
method.
This observation reinforces the intuition that the analysis
method used to derive the vortex properties adds to the
perceived measurement uncertainty, with the local methods
giving significantly larger uncertainty than the global
methods. Some possible contributing factors are asym-
metric data void and presence of spurious data points near
the vortex core—see Ramasamy et al. (2010). Having
confirmed the hypothesis that global methods for deter-
mining vortex center and vortex properties are better, the
planar fit is applied to two of the previous measurements
shown in Fig. 4 for further reinforcement.
3.3 TRAM measurements
The planar least-squares fit was applied to the TRAM
results from Yamauchi et al. (1999). Two cases were
considered: a high thrust case and a low thrust case that has
two counter-rotating vortices. These are forward flight
measurements, and the PIV flow field shows large mean
vortex convection velocities and larger vortex axis orien-
tation angles. Another complication was that the PIV
measurement window often contained two vortices in close
proximity. Yamauchi et al. (1999) overcame this by simply
making smaller analysis windows surrounding each indi-
vidual vortex. In the current work, however, the planar
least-squares fit can be easily modified to include two
vortices.
The vortex core radii derived using this two-vortex
planar fit are summarized in Fig. 6. The SA, CA and IA
results are compared with those reported by Yamauchi
et al. (1999). Note that only one vortex for the high thrust
case was reported in that work, whereas the current work
includes results for both the vortices. As expected, the SA
results are similar for the two methodologies but the CA
results show a consistent behavior with the planar fit. The
IA results for the planar fit show a much smaller standard
deviation than the previous results. The resulting fractional
uncertainty is about 8 % using the proposed new method,
compared to about 25 % as reported by Yamauchi et al.
(1999). This again confirms the hypothesis that a global
method results in a much smaller measurement uncertainty
than a local method.
3.4 HART II measurements
The planar fit was also applied to the HART II measure-
ments at position 17. This position has been studied in
great detail by the HART II partners, and a comparative
summary of analyses performed by US Army AFDD, DLR
and ONERA are presented by Lim and van der Wall
(2010). In this case, the flow field was characterized by the
presence of strong vortical flow structures in close vicinity
of the vortex core. An example flow field in Fig. 7 shows a
relatively strong vortex sheet in close vicinity of the vortex
core being examined. In this case, the contours represent
the velocity magnitude, and at least a part of the vortex
core boundary can be clearly identified by the large swirl
velocity region near the top-right corner. The derived core
0.02
0.04
0.06
0.08
0.1
0.12
0.14
CCW
low
CWCT
Yamauchi et al (1999)
CCWhigh CT
CCW
low
CWCT
current
CCW-1
high
work
CCW-2CT
Vor
tex
core
rad
ius,
rc
/c
IA ± std.dev.SASA+DevenportCA
Fig. 6 Core radius calculated from the flow field using the planar fit
compared with results presented by Yamauchi et al. (1999), where
only one vortex was reported for the high thrust case. The results are
shown as IA with standard deviation together with CA and SA values.
Current work include the Devenport correction with SA
localglobal
Vortex core given by different methods
Vortex sheet coordinates
xs
ys
Fig. 7 Simple average flow field for the HART II position 17 shown
as velocity vectors along with contours of velocity magnitude, blue is
the lowest and red is the highest. The vortex center location and size
identified using four different methods are also shown
Exp Fluids (2012) 53:1191–1202 1199
123
centers and sizes using four different methods are also
shown in the figure with different colors.
The planar fit can be easily modified to account for the
presence of a vortex sheet. The sheet appears like one half
of a vortex with the other half stretched along the sheet
length. A quick representation of such vortex sheet was
devised using the Lamb-Oseen vortex profile at the sheet
edge (semicircular region), combined with the tangential
velocity along the length of the sheet also given by the
Lamb-Oseen profile as a function of normal distance from
the sheet. In the vortex sheet coordinates, xs; ysð Þ; as shown
in Fig. 7, with the vortex sheet extending along the nega-
tive xs-axis, the sheet induced velocity field given by
xs� 0 Lamb-Oseen vortex (Eq. 3)
xs� 0 us ¼ Cs
2py 1� expð�a y2
r2s
� �; vs ¼ 0
ð5Þ
The planar least-squares fit now includes the sheet location,
circulation strength (Cs), core radius (rs) and orientation
axes as additional parameters. Note that the sheet repre-
sentation requires one extra orientation angle to account for
in-plane rotation of the sheet (orientation of the sheet axes
xs, ys). Such a simple representation is, perhaps, sufficient
to eliminate the influence of the vortex sheet on the tip
vortex core properties. However, a more appropriate rep-
resentation must be chosen to model the vortex sheet if the
vortex sheet properties are also of interest.
Figure 7 shows that the vortex center location and core
radius obtained using the proposed planar fit are in good
agreement with prior calculations. The global methods
show a similar behavior and are, perhaps, least affected by
the presence of the vortex sheet in the flow field. The
vortex core radius and peak swirl velocity are shown as
scatter plots in Fig. 8a and b, respectively. At this position,
the tip vortex exhibited a moderate aperiodicity magnitude
and, therefore, the Devenport correction applied to SA
results gave a smaller core radius and larger peak swirl
velocity. For the core radius, the corrected SA and the IA
results were close to the previous calculations by AFDD,
ONERA and DLR reported by Lim and van der Wall
(2010). There was a small variation in peak swirl velocity
among these different calculations. Contrary to the com-
monly held belief, the simple average, albeit with the
Devenport correction, gave as good results as the other
methods. The CA and IA results are also in overall good
agreement and show that the uncertainty in core radius or
peak swirl velocity is about 20 %.
Another aspect hidden in the HART II results is the
uncertainty in vortex center location. As mentioned earlier,
the AFDD calculations shown by Lim and van der Wall
(2010) used a local method for vortex center based on
centroid of vorticity. The ONERA calculations also used a
local method based on the Q-criterion, while the DLR
calculations used a global method based on convolution of
k2. The uncertainty in vortex center location (that is, the
standard deviation in the vortex center location) is shown
in Fig. 9 for these different methods along with the planar
fit. The distinction between local and global methods is
again apparent with the global methods (convolution of k2
and the planar least-squares fit) resulting in significantly
lower uncertainty.
4 Concluding remarks
Vortex aperiodicity, or the random variation in vortex
center location, results in a larger perceived core radius in
the averaged flow field measurements. This is often recti-
fied using methodologies that fall in three general types:
(a)
(b)
Fig. 8 Scatter plots for the vortex properties for HART II position 17
shown with the SA, CA and IA results using the planar fit to include a
vortex sheet. Results from the HART II partners reported by Lim and
van der Wall (2010) are also shown for comparison. a Vortex core
radius, rc, b Peak swirl velocity, Vhmax=XR
1200 Exp Fluids (2012) 53:1191–1202
123
the simple average (SA) followed by an analytical/empir-
ical correction, the corrected average or conditional aver-
age (CA), and the individual average (IA). However, the
analysis methods used to derive the vortex properties of
interest also play a crucial role in how the underlying
measurement uncertainty is transformed into the final
uncertainty in the derived vortex properties. Global meth-
ods that use a larger area of measured data were shown to
give less uncertainty in the vortex properties compared to
local methods that use only a small subset of data in close
proximity to the vortex core. The newly proposed planar fit
method is one such global method, and it was shown to
give lower resulting random uncertainty in vortex proper-
ties. In fact, all three types aperiodicity correction meth-
odologies gave essentially the same result when used with
the proposed planar fit. The method was also applied to
previous tip vortex measurements for the TRAM and the
HART II rotors. In all cases, the results showed the same
behavior, thus confirming the wide applicability and ver-
satility of this method. Specific conclusions from this study
are summarized below.
1. The analysis method used to derive vortex properties,
like the core radius, plays a significant role in the
resulting uncertainty, perhaps a more significant role
than the aperiodicity correction or measurement res-
olution. It was hypothesized that global methods,
which use all measured data, result in lower uncer-
tainty in the derived vortex properties than local
methods, which use only a small sub-set of data close
to the vortex core. The hypothesis was confirmed using
the AFDD model-rotor measurements as well as the
TRAM and HART II measurements.
2. The HART II measurements at position 17 analyzed
using the proposed new method showed that the global
methods indeed resulted in smaller uncertainty in
vortex center location as compared to the local
methods.
3. Using a global method for deriving the vortex center
and core properties, the results were insensitive to the
choice of aperiodicity correction methodology. The
mean vortex core properties estimated using all three
aperiodicity correction methodologies (SA, CA and
IA) were self-consistent. The same behavior was
observed in previous measurements reanalyzed using
the proposed new method.
4. Contrary to commonly held belief, the simple average,
which is both the simplest and the fastest, gave a good
estimation of vortex core properties, provided the core
radius and peak swirl velocities were later corrected
using the inverse convolution method developed by
Devenport et al. The TRAM measurements showed
that the simple average results were not sensitive to the
choice of local or global derivation method. This is
suggestive that the simple average minimizes the
random uncertainty in the transformation from velocity
field to vortex properties.
References
Baker GR, Barker SJ, Bofah KK, Saffman PG (1974) Laser
anemometer measurements of trailing vortices in water. J Fluid
Mech 65:325–336
Bhagwat MJ, Leishman JG (2000) Measurements of bound and wake
circulation on a helicopter rotor. J Aircr 37(2):227–234
Cierpka C, Weier T, Gerbeth G (2008) Evolution of vortex structures
in an electromagnetically excited separated flow. Exp Fluids
45:943–953
Cucitore R, Quadrio M, Baron A (1999) On the effectivenss and
limitations of local criteria for the identification of a vortex. Eur
J Mech B Fluids 18(2):261–282
Devenport WJ, Rife MC, Liapis SI, Follin GJ (1996) The structure
and development of a wing-tip vortex. J Fluid Mech 312:67–106
Heineck JT, Yamauchi GK, Wadcock AJ, Lourenco L, Abrego AI
(2000) Application of three-component PIV to a hovering rotor
wake. In: Proceedings of the American Helicopter Society 56th
annual national forum, Virginia Beach
Kindler K, Mulleners K, Richard H, van der Wall BG, Raffel M
(2011) Aperiodicity in the near field of full-scale rotor blade tip
vortices. Exp Fluids 50:1601–1610
Leishman JG (1996) Seed particle dynamics in tip vortex flows.
J Aircr 33(4):823–825
Leishman JG (1998) Measurements of the aperiodic wake of a
hovering rotor. Exp Fluids 25(4):352–361
Lim J, van der Wall BG (2010) An assessment of 3-C PIV analysis
methodology for HART II measured data. Tech Rep RDMR-AF-
10-03, NASA/TM-2010-216024, US Army AFDD (RDECOM)
McAlister KW (2004) Rotor wake development during the first
revolution. J Am Helicopter Soc 49(4):371–390
Raffel M, Seelhorstt U, Willert C (1998) Vortical flow structures at a
helicopter rotor model measured by LDV and PIV. Aeronaut J,
102(1014):221–227
Fig. 9 Scatter in vortex center location for individual samples for the
HART II position 17. The results using the planar fit are compared
with those using different methods summarized by Lim and van der
Wall (2010)
Exp Fluids (2012) 53:1191–1202 1201
123
Raffel M, Willert W, Wereley S, Kompenhans J (2007) Particle image
velocimetry: a practical guide, 2nd edn. Springer, Berlin
Ramasamy M, Johnson B, Huismann T, Leishman JG (2007a) A new
method for estimating turbulent vortex flow properties from
stereoscopic DPIV measurements. In: Proceedings of the Amer-
ican Helicopter Society 63rd annual national forum, Virginia
Beach
Ramasamy M, Leishman JG, Lee TE (2007b) Flow field of a rotating
wing MAV. J Aircr 44(4):1236–1244
Ramasamy M, Johnson B, Leishman JG (2009) Turbulent tip vortex
measurements using dual-plane stereoscopic particle image
velocimetry. AIAA J 47(8):1826–1840
Ramasamy M, Paetzel R, Bhagwat MJ (2010) Aperiodicity correction
for rotor tip vortex measurements. In: Proceedings of the
American Helicopter Society 67th annual national forum,
Virginia Beach
Schram C, Rambaud P, Riethmuller ML (2004) Wavelet based eddy
structure eduction from a backward facing step flow investigated
using particle image velocimetry. Exp Fluids 36:233–245
Takahashi RK, McAlister KW (1987) Preliminary study of a wing-tip
vortex using laser velocimetry. Tech Rep TM 88343, NASA
van der Wall BG, Richard H (2006) Analysis methodology for 3C-
PIV data of rotary wing vortices. Exp Fluids 40(5):798–812
van der Wall BG, Schneider O (2007) Conditional averaging
methodology for periodic data with time jitter and spatial
scatter. In: Proceedings of the 33rd European rotorcraft forum,
ERF, Kazan
Vogt A, Baumann P, Kompenhans J, Gharib M (1996) Investigations
of a wing tip vortex in air by means of DPIV. In: Proceedings of
the 19th AIAA advanced measurement and ground testing
technology conference, New Orleans, AIAA-96-2254
Wadcock A, Yamauchi GK, Solis E, Pete AE (2011) PIV measure-
ments in the wake of a full-scale rotor in forward flight. In: 29th
AIAA applied aerodynamics conference, Honolulu
Yamauchi GK, Burley CL, Mercker E, Pengel K, Janakiram R (1999)
Flow measurements of an isolated model tilt rotor. In: Proceed-
ings of the American Helicopter Society 55th annual national
forum, Montreal
1202 Exp Fluids (2012) 53:1191–1202
123