A simple approach for vortex core visualization 1

A simple approach for vortex core visualization

Jiajia Li and Pablo M. Carrica*
IIHR-Hydroscience and Engineering, The University of Iowa, Iowa City, IA 52242, USA
*Corresponding author, [email protected]

SUMMARY
We propose a method to visualize vortex cores based on manipulation of the pressure field
produced by isolated vortices in incompressible flow. Under ideal conditions the function
D  2 p  2 p yields an approximate distance to vortex centerlines. As opposed to local methods
to identify coherent structures, isosurfaces of D produce a field of vortex tubes equidistant to the
vortex core center which, ideally, are independent of vortex intensity or size. In contrast to other
line-vortex identification methods, which typically rely on algorithms to detect vortex core lines
and frequently need complex implementations, the proposed method can be computed from the
local Eulerian velocity and pressure fields as easily as vortex identification methods such as the Q
and λ2 criteria. D  2 p  2 p results in the exact distance to the core center for a Rankine vortex
and is in general valid for the region of a vortex where there is pure rotation, yielding an
approximation to the distance farther from the core in other simple one-dimensional vortex models.
The methodology performs well in all tests we attempted, though limitations are presented and
discussed. The method is demonstrated for a canonical Burgers vortex, a Bodewadt vortex,
homogeneous isotropic turbulent flow, the wake of a propeller, a heaving plate, a turning
containership and the airwake of a surface combatant. The proposed method helps to better
visualize vortical flow fields by displaying vortex cores, complementing methods like Q and λ2
which display vortical volumes.
KEY WORDS: Vortices, Computational Fluid Dynamics.
A vortex filament is generally recognized as
a region where rigid body rotation has a
1. Introduction
maximum and constitutes the center of the
While the definition of a vortex is still being vortex. The definition of a vortex core is more
debated, the need to visualize flow fields has complicated since it involves identifying the
led to the development of a considerable boundary between the rotational and the
number of vortex identification methods. irrotational flow, becoming subjective (see
Though none of these methods is perfect or different points of view on this in [1-3]). This
universal, due to limitations and a natural definition of the interface between rotational
consequence of the lack of a vortex definition, and irrotational flow is already arbitrary for
many of them have been in use for a number of simple one-dimensional vortex considering the
years, and are frequently tested by users to find diffusion of vorticity by viscosity, and
the method that best fits their particular becomes even more so for complex three
problem and needs. dimensional flows where vortices may not be
symmetric, and are subject to interaction with
  A simple approach for vortex core visualization 2

other vortices and flow features like pressure 2Q  Ω  S

2 2
(1)
gradients and strain.
In recent review papers, Epps [4] and where Ω   v  v T  / 2 and
Volkov et al. [5] broadly classify vortex S    v   v T  / 2 are the antisymmetric and
identification methods into region-type
symmetric components representing rates of
methods, which attempt to identify the volume
rotation and strain. A relation between Q and
of the vortices, and line-type methods, which
the Laplacian of the pressure can be obtained
seek to detect the lines joining the vortex
by taking the divergence of the dimensionless
centers. These last methods are typically
incompressible Navier-Stokes equation,
expensive and require specific algorithmic
implementations. Vortex identification v 1 2
  v   v  P  v (2)
methods are also classified as Eulerian, where t Re
Eulerian quantities (velocity and/or pressure
fields) are used to identify vortices, or where reference values of length and velocity
Lagrangian, typically based on particle are used to non-dimensionalize Eq. (2). The
trajectories. In addition, Eulerian methods can result is [9]
be local, depending only on local variables, or 1
global, which require larger regions in the Q  2 P (3)
2
domain. Eulerian, local methods are usually
the easiest to implement and are therefore most It is important to note that, in addition to the
widely used. We present in this paper an second invariant of the velocity gradient, Q can
Eulerian local method to identify vortex cores, be replaced by the Laplacian of the pressure,
which can be easily combined with Eulerian and that a positive Q can be interpreted as a
local vortex identification methods to provide negative source for the Poisson Eq. (3),
the location of vortex cores. 2 P  2Q  0 . A positive Q naturally results in
a local pressure drop due to a negative pressure
1.1. Eulerian vortex identification source. Thus the Q-criterion to define a vortex
methods can be also stated as regions where sources in
The Q-criterion vortex identification method the pressure Poisson equation are negative. In
was one of the first Eulerian local approaches, the case of a rigid body rotation this source is
and we dwell here in some detail as it is
2 P  22  0 (4)
relevant to the discussion in section §2. It was
proposed by Hunt et al. [6], defining vortex We note here that, while vortices always
regions where Q, the second invariant of cause negative sources for the pressure Poisson
velocity gradient v , is positive. The method Eq. (3), other sources can be present in the flow
enjoyed immediate popularity due to its that significantly affect the pressure field
simplicity and ability to efficiently display caused by vortices, thus the pressure minimum
vortical structures. While the Q-criterion has will not necessarily be located in the vortex
some shortcomings and alternative methods core.
have been proposed since (either Eulerian or A second criticism to the Q-criterion is
Lagrangian [1, 3, 7, 8]), it remains widely used related to the need to stablish a threshold, such
today. One of the criticisms to the Q-criterion
that any point with Q  Qt is identified as a
is that, though is Galilean invariant, is not an
objective method [2] and produces spurious vortex if a pressure minimum occurs within
structures when the system of reference is this region (see discussion in [9]). In complex
rotating. Q is defined as flows with a wide range of spatial scales a low
  A simple approach for vortex core visualization 3

Qt will produce false positives, and overly the center line crosses the element. This
busy structures. A high Qt , on the other hand, algorithm is simple and easy to parallelize,
though it tends to fail for vortex core lines with
will display less and tighter structures, clearing high curvature [12]. This problem was studied
the picture and allowing easier observation of and solutions proposed by Roth [13] in terms
the stronger vortices, but weaker vortices are of parallel vector operators, which can be used
lost when they may be relevant. to express most definitions of vortex cores. By
Other Eulerian local vortex identification choosing appropriate pairs of vectors and
methods have been proposed with various Boolean operators, Roth [13] was able to group
degrees of success. There is no consensus on the methods of Sujudi & Haimes [11], Levy et
what method is ultimately superior; instead all al. [14], Banks & Singer [15], Kida & Miura
have strengths and weaknesses. Jeong & [16] and Strawn et al. [17] into a single
Hussain [3] proposed the 2 -criterion, which algorithm implementation framework.
uses a negative second eigenvalue of the To our knowledge, all current vortex core
second order tensor Ω2  S2 to define a vortex. line detection methods fall in the category of
The 2 -criterion, as the Q-criterion, is widely algorithmic methods, with different degrees of
difficulty. Most of the proposed schemes can
used and relatively simple to evaluate, though
be very complex to implement and rarely used.
computation of the eigenvalues of Ω2  S2 and The method to visualize vortex core proposed
ordering is more involved than the in this paper, though it cannot find the exact
computation of Q, which only requires center of the vortex, can be used to
derivatives of the velocity. Chakraborty et al. approximate the core lines using isosurfaces of
[1] introduced the ci / cr enhanced swirling small D as shown in subsequent sections. In
strength criterion, an improvement to the addition, it is Eulerian and does not require
original ci swirling strength criterion method specific implementations, and can be used with
of Zhou et al. [10]. the capabilities to process velocity and
pressure fields available in most commercial or
1.2. Vortex core line identification open source software to postprocess flow
methods solutions.
Vortex core line identification methods are
mostly based on local or global algorithms that 2. Distance to a vortex core line
detect the center of rotation of vortices. One of
the most widely used methods was developed In this section we present a methodology to
by Sujudi & Haimes [11], and involves produce vortex tubes at a given distance
detection of a center of rotation in individual around vortex center lines as isosurfaces of a
cells. The velocity field is interpolated on each function D obtained from the Eulerian local
tetrahedra and v is computed. If the velocity and pressure.
eigenvalues of v has one real R and two 2.1. The distance function D
complex conjugate C eigenvalues, then the In absence of strong stretching by external
velocity field is projected into the plane normal strain, the inner portion of the core of a vortex
to the eigenvector corresponding to R . The can be approximated by rigid body rotation,
process to find the center of rotation involves thus the azimuthal velocity is a linear function
detecting if the line of rotation intersects the with the radial distance to the vortex center line
tetrahedron faces, if two points are found then and the angular rotation velocity in an inertial
  A simple approach for vortex core visualization 4

frame of reference, v  r . The pressure unacceptable noise. Recalling that the

gradient is in this case Laplacian of the pressure and Q are related by
Eq. (3), then a version of Eq. (8) that requires
dP only first derivatives of velocity and pressure
 2 r (5)
dr and results in smoother contours is
Equation (5) is exact near the core line for P
isolated axisymmetric vortices, and an D (9)
Q
approximation when the vortices are immersed
in pressure gradients generated by other flow Finally, since only regions where Q  0 are of
features like nearby vortices or geometrical interest, then the distance to the vortex core
boundaries. Since according to Eq. (4) the line can be written as
Laplacian of the pressure in this case is
P
2 P  22 , dividing by the Laplacian of the D (10)
pressure we obtain a function D that provides max  Q,  
the distance to the vortex center line
where  is a small positive number to prevent
dP  1 d  rdP / dr   dividing by zero and make D large when Q is
Dr /  (6) small. If strong pressure gradients not caused
dr  2r dr  by rotation are present, due for instance to local
In a general three dimensional flow D can be flow acceleration, then Eq. (7) can be
expressed as simplified to
2 P  n P  n
D (7) D (11)
   P  n  n max  Q ,  

where  P  n  n is the pressure gradient which in some cases yields better results than
Eq. (10), at the cost of a more complex
projected in the direction of a unit vector n
computation. Alternatively, 2 can be used to
normal to the vortex axis. n can be obtained
from the eigenvector corresponding to 2 as compute D, since Q = −λ2/2 in planar flows
defined in the 2 criterion [3]. If the pressure P  n
D (12)
gradient is dominated by the pressure field max   2 / 2,  
produced by the vortex, which as will be
which essentially computes Q in the plane of
discussed in §3 is in most situations the case
rotation, removing effects of pressure
since D is relevant in locations inside vortex
variations in direction of the axis of rotation.
cores, we can then simplify Eq. (7) to
Eq. (12), however, is not immune to external
P pressure fields in the plane of rotation.
D (8)
1 2
 P 2.2. Simple vortex models
2
Gerz et al. [18] presents several one-
since in this case the pressure gradient and
dimensional vortex models in the context of a
Laplacian are dominated by the local vortex.
study on aircraft wing tip vortices. All models
Though Eq. (8) can be computed if the pressure
share two important properties. Since the
field is available from either CFD or
center of the core rotates as a rigid body, the
experiments, it involves second derivatives of
vorticity gradient normal to the vortex center
the pressure, and in some cases may result in
  A simple approach for vortex core visualization 5

line is zero at the vortex axis,  r  0 , [18], but that results in a deviation from the
implying that the azimuthal velocity increases rigid body rotational speed for r  0 .
linearly with r , v ~ r . The second is that the
vorticity vanishes far from the core, and thus
the azimuthal velocity decreases as v ~ 1/ r .
The simplest vortex model is the Rankine
vortex, where the vorticity is constant within
the vortex core, and zero outside,
 r r 1
v   (13)
1 / r r 1
where the velocity is nondimensionalized with
 /  2 a  , the azimuthal velocity at a
characteristic vortex core size a as defined in
Leweke et al. [19]. a is used as reference
length for the equations in dimensionless form.
The Rankine vortex suffers from a
discontinuity in the velocity gradient at the
vortex core edge r  1 . A more realistic model
is the Lamb-Oseen vortex model,


v  1/ r 1  er
2

 (14)

Using the one-dimensional version of D, Eq.

(6), we obtain in cylindrical coordinates
dP  1 d  rdP / dr   dv Fig. 1: Rankine and Lamb-Oseen vortices, v (top) and
D /   v / (15) D (bottom).
dr  2r dr  dr
For the Rankine vortex, D is then Since derived for the rigid body region of the
vortex core, D is only valid for small r, which
 r r 1 occurs when Q is positive. Notice that for the
D (16)
 r r 1 Lamb-Oseen vortex D is approximately 10%
larger than r for r  0.3 (30% of the vortex
and for the Lamb-Oseen vortex is core), providing an estimate of the error

D

r 1  e r
2

 (17)
incurred when the D function is displayed.
Adding straining to the Lamb-Oseen vortex
e  r  2r 2  1  1
2
results in the Burgers vortex model, a good
example of a vortex affected by an external
The azimuthal velocity v and the function D pressure gradient
for Rankine and Lamb-Oseen vortices are
vr   r , v  1/ r 1  e r  , vz  2 z (18)
2

shown in Fig. 1. Note that the core for the  

Lamb-Oseen vortex is not located exactly at
r  1 ; some authors apply a correction on the In Eq. (18)   4 /  is a dimensionless
exponent to displace the core back to r  1 parameter that controls the strength of the
  A simple approach for vortex core visualization 6

straining velocity field  vr , vz  respect to the (see Fig. 2, top). For   0.05 the difference
strength of the vortical structure represented by between the total pressure gradient and the
v . The resulting total dimensionless pressure pressure gradient due to the azimuthal velocity
is 0.03 at r  0 , and quickly decreases to less
gradient is shown in Fig. 2 for different  and than 0.01 for larger r . A stronger straining
z values. For weak straining,   0.05 , the field causes a considerable pressure field,
external pressure gradient is negligible which causes Eq. (10) to predict a deformed
compared to the pressure gradient induced by distance function D, while Eq. (7) still
the flow rotation, resulting in no deformation produces good results (see Fig. 2, bottom). In
of the predicted tubes using the approximate most practical cases use of Eq. (7) produces
Eq. (10) respect to predictions with the more results that are very similar to those produced
accurate Eq. (7) (see Fig. 3). A stronger by Eq. (10). While more accurate and immune
straining flow with   0.2 causes a high to effects of external pressure fields on the
pressure gradient in the axial direction of the ability to display vortex tubes, Eq. (7) will in
vortex  z  , resulting in higher total pressure most cases result in noisier images caused by
the computation of the second derivatives of
gradient as z increases. The result is a pressure.
distortion of the isosurface of D further out
from the straining plane at higher z , deviating
from the desired vortex tube. Notice that using
only the pressure gradient normal to the axis,
Eq. (7), the vortex tube is recovered with little
effect on the distance to the center, as shown in
Fig. 3. This is explained by the little effect that
the straining field has in the radial pressure
gradient, depicted in Fig. 2.
P vr2 v2 P P vz2
  ,  0,  (19)
r r r  z z
In the Burgers vortex the radial inflow
produces a pressure gradient caused by the
straining velocity field, independent of the
vortical structure represented by v , resulting
in a simple example where a pressure field,
external to the vortex, affects the behaviour of
different methodologies to compute D. It must
be noted that Q is not affected by the straining
velocity field, with Q  0 isosurfaces yielding
a cylinder of radius R  1.1a . In the case of
low to moderate strain, the pressure gradient is
dominated by the azimuthal component of the
velocity and isosurfaces of D  0.4a from Eq.
Fig. 2: Burgers vortex: total pressure gradient (top) and
(10) produce a tube with actual distance to the radial pressure gradient and velocity (bottom) for
vortex axis of approximately R  0.355a , various values of  and z .
indistinguishable from results using Eq. (7)
  A simple approach for vortex core visualization 7

tubes of constant D. We show in this section

that for a wide range of flows the proposed
method reveals vortex cores well, but presence
of walls may cause strong pressure gradients
that mask the pressure gradients resulting from
weaker vortices. External pressure gradients
on a vortex caused by nearby vortices are in
general small and do not cause significant
effects on D. This is because the pressure
gradient induced by a vortex in the irrotational
external region decays as dP / dr ~ 1/ r 3 . For a
pair of Rankine vortices 1 and 2 at a distance
b larger than the characteristic radius of either
vortex, direct superposition of individual
pressure gradients in the core of vortex 1 with
Fig. 3: Isosurfaces of D = 0.5 for   0.05 (a, c) and characteristic radius a1 results in
  0.2 (b, d) with D computed using Eq. (10) (a, b)
dP / dr 2 1
and Eq. (7) (c, d).  1  22 (20)
dP1 / dr 1  b / a1  13

3. Discussion and examples

where 2 is a nearby vortex. Equation. (20)
does not use the complex Navier-Stokes
In this and following sections D is computed solutions to the interaction between the two
using Eq. (10) unless otherwise stated, vortices [19], but provides a first-order
selecting D with values of a few grid points to estimation of the pressure effects that nearby
show the vortex cores. Smaller values of D vortices cause on a local vortex. If a smaller
make the isosurfaces look like vortex core
vortex 2 is located inside the core of a co-
lines, but values smaller than the grid spacing
cannot display isosurfaces. Note that, since rotating local larger vortex 1 , in the core of
small values of D maintain the displayed tubes the smaller vortex D can be estimated as
inside the rigid rotation vortex cores, coloring
the isosurfaces of D with absolute vorticity 12 r   22  b  r 
D (21)
provides an approximation to the vorticity in 12   22
the core.
since Q  12  22 . Since smaller vortices tend
3.1. Limitations to spin faster than larger vortices, Eq. (21)
The method proposed relies in detecting the implies that D favors smaller vortices, just as
rigid rotating core of a vortex based on the Q and 2 do. The preceding discussion
assumption that the local pressure gradient is highlights two limitations of the proposed
caused by isolated vortices. This assumption of method. The first is that the computation of D
course fails when significant pressure is affected by external pressure gradients (not
gradients caused by other sources are present. caused by the local vortex), resulting in
One example is the Burgers vortex shown in deformation of the tubes displayed by
§2.2, where strain causes an external pressure isosurfaces of D. Results show that strong
gradient that results in a deformation of the external pressure gradients can mask weaker
  A simple approach for vortex core visualization 8

vortices, but pressure gradients caused by with boundary conditions

nearby vortices are typically innocuous. A
F  0   0, G  0   0, H  0   0
second limitation is that larger vortices tend to (24)
be masked by smaller vortices, a trend also F     0, G     1
observed in most other Eulerian vortex
identification methods [20]. A third limitation The Bodewadt vortex develops a boundary
is that the proposed method is restricted to layer that reduces the rotational velocity of the
incompressible flows as shown in its vortex near the wall, inducing an inflow
derivation. Performance of D when applied to toward the vortex axis, as shown in Fig. 4. The
non-symmetric vortices has not been studied, flow is similar to the “tea cup” flow, and vortex
but the example in §3.4 seems to show that identification methods should display a vortex
elliptic vortices can be displayed by reaching all the way to the wall. Jeong &
isosurfaces of D. Hussain [3] showed that the Q-criterion fails to
identify the whole volume as a vortex. Figure
3.2. Bodewadt vortex 4 shows that 2 also fails to identify the near-
Bodewadt [21] studied a vortex of rotational wall region as a vortex, as local strain
rate  normal to a solid wall and extending to dominates at the boundary layer. Notice that
infinity in the opposite direction. Schwiderski Jeong & Hussain [3] claim that 2 remains
& Lugt [22] presented solutiuons of the negative all the way to the wall, but
Bodewadt vortex and the similar problem of observation of the solution, also shown in their
the von Karman vortex, questioning the Fig. 11, reveals a small positive zone near the
validity of the original Bodewadt solution as it wall. Isosurfaces of Q or 2 , however, are
is unable to predict boundary layer separation unable to provide any useful information, since
at the axis of rotation. We use the original the vortex is present everywhere and thus the
solution of Bodewadt [21], as done by Jeong & volume should be unbounded. Isosurfaces of D
Hussain [3], because while simpler still reveals = 0.1 computed from Eq. (7) can display the
limitations of vortex identification methods. vortex axis almost to the wall, while coloring
Defining reference length  /  and velocity with vorticity shows the effect of the wall on
 , the solution can be expressed in a the vortex strength. When D is computed with
cylindrical coordinate system as Eq. (12), it fails to display the core close to the
wall as 2 is positive, but still displays the
ur  rF  z 
location of the core further out. This example
u  rG  z  nicely shows how isosurfaces of D can
uz  H  z  (22) complement isosurfaces of Q or 2 to display
vortex structures and cores.
1 2
P r  P0  z 
2
where F  z  , G  z  , H  z  are determined by
the following expressions
F 2  G 2  HF   1  F 
2 FG  HG  G
(23)
H   HH   P0
2F  H   0
  A simple approach for vortex core visualization 9

isosurfaces of Q is more dramatic for larger

vortices, where low values of D locate the
isosurfaces deep into the core where the
pressure is lower, while the Q-criterion
produces larger structures naturally located
farther from the vortex axis where pressure is
higher. While weaker vortices in regions of
higher pressure are mostly absent for Q  500
, they are displayed well by D . This is clear
near the corner ( xmax , ymax , zmin ) and around
the center of the ymax face. On the other hand,
Fig. 4: Bodewadt vortex. Isosurface of D = 0.1
computed with Eq. (7) colored with axial vorticity (left). Q shows the size of the vortices and, by not
Isosurfaces of D = 0.1 computed with Eq. (12) colored showing weaker vortices, complements D as
with pressure (right). The right panel also shows
isosurfaces of Q = 0 and 2  0 , with streamlines over
a vortex visualization tool. The turbulent field
shown in Fig. 5 contains too much information
those surfaces.
and can hardly be used for visualization.
3.3. Homogeneous isotropic turbulence Filtering the velocity and pressure fields
reveals larger scales, not discernible when
This example uses an instantaneous solution using the original flow field. For example, the
of the homogeneous isotropic turbulence DNS pressure field can be filtered by
simulation by Li et al. [23], which was

obtained in a 10243 uniform grid in a domain P  x    P  x  G  x  r  dr (25)

extending  0, 2  in all three directions, for
where G is the Gaussian filter operator. P is
Re  418 . Results are displayed for a fraction
then a low-pass filtered pressure. The smaller
of the original solution at time step t  10 , pressure scales can be then obtained as the
covering half of the grid in each direction (size deviation P  P  P . A similar procedure can
5123 ). Figure 5 shows isosurfaces of produce a velocity deviation u ' , and u ' and
D  0.012 (top), approximately two cells in P  can be used to compute D from Eq. (10).
size, and Q  500 (bottom), colored with We employ a Gaussian operator with standard
pressure. Notice the presence of large-scale deviation of 10 grids cells to isolate larger
pressure fluctuations, of scale much larger than structures. Figure 6 shows isosurfaces of
the resolved vortices shown by either Q or D Q  20 for the filtered solution along with
. These large-scale fluctuations do not seem to isosurfaces of D  0.0075 for the deviation,
affect the display of the vortex tubes of size removing structures with pressures higher than
D  0.012 , as the large-scale structure P  0.5 for the smaller scales to clear the
pressure gradients are mild. In regions of picture. The resulting larger vortical structures
weaker vortices where Q  500 does not displayed by isosurfaces of Q = 20 exhibit
display structures, as the corner near ( xmax , lower rotation rates than smaller structures,
and higher pressures. Smaller scale structures
ymin , zmin ), D produces vortex tubes of
displayed by D tend to rotate around the larger
approximately equal size as desired. Stronger structures, but are also present inside the
vortices, which are characterized by low isosurfaces of Q  20 , displaying locally the
pressure in the core, are displayed clearly as lowest pressures in the field.
blue tubes in Fig. 5 (top). The difference with
  A simple approach for vortex core visualization 10

Fig. 6: Homogeneous isotropic turbulence in a periodic

box. Isosurfaces of Q  20 of the filtered velocity field
and isorufaces of D  0.0075 of the deviation velocity
and pressure fields, coloured with pressure. The vortex
cores displayed with D are subject to a cutoff showing
only pressures lower than P  0.5 .

Fig. 5: Homogeneous isotropic turbulence in a periodic

box. Isosurfaces of D  0.012 (top) and Q  500
(bottom) coloured with pressure.

Fig. 7: Propeller E1619 operating at J  0.3 .

Isosurfaces of D  0.0125D0 colored with local rotation
rate  / 0 .
  A simple approach for vortex core visualization 11

3.4. Propeller wake instabilities 3.5. Heaving plate

E1619 is a 7-bladed generic submarine A heaving plate produces a rich vortical
propeller designed at INSEAN (now INM) in system, with periodic tip vortex separation and
Rome, Italy. The propeller has been used as a trailing vortices that form a reverse von
benchmark to test CFD capabilities to predict Karman streets [27]. In this case a simulation
propeller performance and wakes [24]. At low was performed with a zero-thickness square
advance coefficients the tip vortices exhibit plate of side L with free stream velocity V , at
elliptic instabilities, merging and eventual a Reynolds number Re  VL /   2000 . The
breakdown. Some of these effects have been plate was subject to a sinusoidal heave
discussed by Di Mascio, A. et al. [25]. In this oscillation with amplitude a / L  0.25 and a
section we show example results for an period TV / L  1 . A fully structured grid with
advance coefficient J  V0 /   0 D0   0.3 , 18 M grid points was used. Figure 8 shows
where V0 is the advance speed, and  0 and D0 instantaneous vortical structures displayed
with isosurfaces of Q  60 and D  0.015 .
are the propeller rotational speed and diameter,
Isosurfaces of D show fairly constant diameter,
respectively. Figure 7 shows isosurfaces of
indicating that D is showing an approximate
D  0.0125D0 colored with local rotation rate constant distance to the vortex centerlines
 Ω nondimensionalized with the while no artificial vortex cores are observed.
propeller rotational speed, and pressure at the This example, however, displays well regions
centerplane. In a heavily loaded propeller with where D, and also Q, fail to display vortices.
large number of blades the pitch between tip Vortex interaction does not affect D
vortices is small compared to the propeller significantly for vortices not too close to each
diameter and vortex radius, resulting in strong other. In regions of weaker and vanishing
vortex interaction. The rotation rates of the tip vorticity, as displayed by isosurfaces of D
vortices when shedding exceeds 100 , colored with vorticity magnitude, isosurfaces
of D tend to deviate from the distance to the
decaying as they evolve in the propeller’s core center. This is a natural consequence of
wake. The blade tip vortices are co-rotating low pressure gradients caused by the weaker
helical vortices that develop elliptic vortices, becoming comparable to pressure
instabilities [19], as discussed by Leweke et al. gradients caused by the primary flow of the
[26] for a horizontal wind turbine rotor. These moving plate or by nearby stronger vortices.
vortex instabilities are characterized by short- The vortex cores predicted by D can be
wavelength perturbation on the vortex core compared with the vortex cores predicted by
shape, and can be clearly seen in the tip the algorithm of Sujudi and Haimes [11] in the
vortices in Fig. 7 as they develop after the bottom of Fig. 8, where it can be noticed that
vortices are shed from the blade tip. Merging this particular vortex core line identification
of two continuous tip vortices can also be method based on velocity tensor eigenvalues
observed in Fig. 7 (fifth and sixth vortices from fails to predict weaker vortices. Notice that
the left), as the elliptic instability continuously while Sujudi and Haimes method predicts a
growing into the propeller wake, generating a vortex line, in Fig. 8 is shown as a tube with
large number of streamwise vortices as the tip volume to facilitate visualization.
vortices breakdown. The very strong hub
vortex and the weaker blade root vortices are
also clearly exposed by isosurfaces of D.
  A simple approach for vortex core visualization 12

ship is self-propelled at Froude number

Fr  V0 / L0 g  0.095 , where L0 and V0 are
the ship length and speed and used to non-
dimensionalize all variables. The speed is
controlled by an autopilot that varies the
rotational velocity of the propeller. In a 20/5
maneuver test the rudder is turned in the
opposite direction at the maximum rudder rate
every time the ship heading checks +/− 5
degrees, resulting in a zigzag trajectory. In this
case the propeller produces thrust mostly
through pressure lift on the blades, resulting in
high pressure gradients external to the vortex-
induced pressure gradients.
As in the example in §3.4, the propeller
produces strong hub and blade tip vortices,
which mostly dominate the vortical structure
of the flow. The rudder also produces vortices,
and affects the propeller vortices through
complex interaction. Figure 9 shows a view of
the stern with the propeller/rudder system from
below, with the propeller rotating clockwise
looking from stern to bow. Isosurfaces of
Q  15000 can only display propeller/rudder
vortices, since these are much stronger than
other vortices generated by the flow, and have
Fig. 8: Vortical structures for a heaving square plate.
Isosurface of Q  60 are shown in transparent grey,
high dimensionless vorticity magnitude (
while isosurfaces of D  0.015 show the vortex cores ω  200 ). Isosurfaces of lower Q produce
colored with vorticity magnitude (top). The bottom plate large numbers of structures that obscure the
shows vortex core lines (thickened for clarity using lines picture. Isosurfaces of D  0.001 show well
with spheres instead of points) obtained with the method
of Sujudi & Haimes [11]. the tip vortices far from the propeller where Q
fails, and also display more structures present
3.6. Containership propeller-rudder in the flow, in particular the bilge vortex
interaction generated by the drift of the ship as it turns, and
In this case the propeller/rudder interaction the weakened blade tip vortices downstream of
flow of the KCS containership during a zigzag the rudder. The bilge vortices, visible on the
maneuver in shallow water is visualized [28]. port side of the propeller shaft hub, are of
Computations were performed with a hybrid special interest since they are ingested by the
RANS/LES turbulence modeling approach, propeller, and they change strength and axial
and thus only the largest vortices are resolved. vorticity sign as the zigzag maneuver causes
The instantaneous solution shown in Fig. 9 is drift to starboard or port.
taken from a simulation of KCS performing a
20/5 zigzag maneuver in shallow water with
water depth to ship draft ratio h / T  1.2 . The
  A simple approach for vortex core visualization 13

pressure gradients poses limitations to the

method as discussed in the paper, but in the
many examples shown these limitations appear
important only when weak vortices are
affected by strong pressure gradients caused by
the primary flow around a blunt body.
Examples show that D can effectively display
vortex tubes around the core in most flows, and
that can complement well other methods to
visualize turbulent structures.
Depending on the flow configurations,
properly simplified formulations of D can be
used for better visualization. However, in most
the examples shown in this paper the simplest
expression D  P max  Q,   works well.
Suggested isosurfaces of D of a few cell
lengths at locations of interests with  a couple
of order of magnitude smaller than the lowest
expected Q produces tubes of approximate
diameter 2D .

ACKNOWLEDGEMENTS
This work was partially supported by the US
Office of Naval Research grants N00014-17-1-
2082 and N00014-17-1-2293, Drs. Thomas Fu
and Ki-Han Kim program officers.

REFERENCES
Fig. 9: Propeller-rudder interaction for containership in
zigzag maneuver. Isosurfaces of D  0.001 (top) and
Q  15000 (bottom) coloured with dimensionless
[1] Chakraborty, P., Balachandar, S., and
vorticity magnitude.
Adrian, R. J., 2005, "On the relationships
between local vortex identification schemes,"
J. Fluid Mech., 535, pp. 189-214.
4. Conclusion [2] Haller, G., 2005, "An objective definition
of a vortex," J. Fluid Mech., 525, pp. 1-26.
We discussed a method to compute a
function D that provides the approximate [3] Jeong, J., and Hussain, F., 1995, "On the
distance to a vortex centerline for Identification of a Vortex," J. Fluid Mech.,
incompressible flows. The method assumes 285, pp. 69-94.
that the pressure gradient is mainly caused by [4] Epps, B. P., 2017, "Review of Vortex
the local vortex, as opposed to pressure Identification Methods," AIAA SciTec Forum.
gradients caused by other nearby vortices or
flow boundaries. This effect of external
  A simple approach for vortex core visualization 14

[5] Volkov, K. N., Emel'yanov, V. N., Identification, Representation,

Teterina, I. V., and Yakovchuk, M. S., 2017, Reconstruction," In Conference on
"Visualization of vortical flows in Visualization, pp. 132-139.
computational fluid dynamics," Comput.
[16] Kida, S., and Miura, H., 1998,
Math. Math. Phys., 57(8), pp. 1360-1375.
"Identification and analysis of vortical
[6] Hunt, J., Wray, A., and Moin, P., 1988, structures," European J. Mech. B/Fluid, 17(4),
"Eddies, streams, and convergence zones in pp. 471-488.
turbulent flows," Tech. Rep. CTR-S88. Center
[17] Strawn, R. C., Kenwright, D. N., and
for Turbulence Research.
Ahmad, J., 1999, "Computer Visualization of
[7] Green, M. A., Rowley, C. W., and Haller, Vortex Wake Systems," AIAA J., 37(4), pp.
G., 2007, "Detection of Lagrangian coherent 511-512.
structures in three-dimensional turbulence," J.
[18] Gerz, T., Holzapfel, F., and Darracq, D.,
Fluid Mech., 572, pp. 111-120.
2002, "Commercial aircraft wake vortices,"
[8] Haller, G., 2015, "Lagrangian Coherent Progr. Aerospace Sci., 38(3), pp. 181-208.
Structures," Annu. Rev. Fluid Mech., 47, pp.
[19] Leweke, T., Le Dizes, S., and Williamson,
137-162.
C. H. K., 2016, "Dynamics and Instabilities of
[9] Dubief, Y., and Delcayre, F., 2000, "On Vortex Pairs," Annu. Rev. Fluid Mech., 48, pp.
coherent-vortex identification in turbulence," 507-541.
J. Turbulence, 1, pp. 1-22.
[20] Bürger, K., Treib, M., Westermann, R.,
[10] Zhou, J., Adrian, R. J., Balachandar, S., Werner, S., Lalescu, C., Szalay, A., Meneveau,
and Kendall, T. M., 1999, "Mechanisms for C., and Eyink, G., 2012, "Vortices within
generating coherent packets of hairpin vortices vortices: hierarchical nature of vortex tubes in
in channel flow," J. Fluid Mech., 387, pp. 353- turbulence." arXiv:1210.3325v2 [physics.flu-
396. dyn]
[11] Sujudi, D., and Haimes, R., 1995, [21] Boedewadt, U. T., 1940, "Die
"Identification of swirling flow in 3-D vector Drehstromung uber festem Grund," Z. Angew.
fields," AIAA Paper No. AIAA 95-1715. Math. Mech. , 20, pp. 241-253.
[12] Garth, C., Tricoche, X., Wiebel, A., and [22] Schwiderski, E. W., and Lugt, H. J., 1964,
Joy, K., 2008, "On the role of domain-specific "Rotating Flows of von Kármán and
knowledge in the visualization of technical Bödewadt," Phys. Fluids, 7(6), pp. 867-875.
flows," Comp. Graphics Forum 27, 1007-
[23] Li, Y., Perlman, E., Wan, M. P., Yang, Y.
1014.
K., Meneveau, C., Burns, R., Chen, S. Y.,
[13] Roth, M., 2000, "Automatic extraction of Szalay, A., and Eyink, G., 2008, "A public
vortex core lines and other line-type features turbulence database cluster and applications to
for scientific visualization," PhD thesis, Swiss study Lagrangian evolution of velocity
Federal Institute of Technology Zurich. increments in turbulence," J. Turbulence,
[14] Levy, Y., Degani, D., and Seginer, A., 9(31), pp. 1-29.
1990, "Graphical Visualization of Vortical [24] Chase, N., and Carrica, P., 2013,
Flows by Means of Helicity," AIAA J., 28(8), "Submarine propeller computations and
pp. 1347-1352. application to self-propulsion of DARPA
[15] Banks, D. C., and Singer, B. A., 1994, Suboff," Ocean Eng., 60, pp. 68–80.
"Vortex Tubes in Turbulent Flows:
  A simple approach for vortex core visualization 15

[25] Di Mascio, A., Muscari, R., and

Dubbioso, G., 2014, "On the wake dynamics of
a propeller operating in drift," J. Fluid Mech.,
754, pp. 263-307.
[26] Leweke, T., Quaranta, H. U., Bolnot, H.,
Blanco-Rodriguez, F. J., and Le Dizes, S.,
2014, "Long- and short-wave instabilities in
helical vortices," Journal of Physics:
Conference Series, 524, 012154
[27] Triantafyllou, M. S., Techet, A. H., and
Hover, F. S., 2004, "Review of experimental
work in biomimetic foils," IEEE J. Oceanic
Eng., 29(3), pp. 585-594.
[28] Carrica, P., Mofidi, A., Eloot, K., and
Delefortrie, G., 2016, "Direct simulation and
experimental study of zigzag maneuver of
KCS in shallow water," Ocean Eng., 112, pp.
117-133.