AIAA JOURNAL

Vol. 51, No. 1, January 2013

Investigation of Flows Around a Rudimentary Landing Gear with

Advanced Detached-Eddy-Simulation Approaches
Zhixiang Xiao, Jian Liu, Kunyu Luo, Jingbo Huang, and Song Fu
Tsinghua University, 100084 Beijing, Peoples Republic of China

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

DOI: 10.2514/1.J051598
Unsteady and massively separated flows past the rudimentary landing gear are investigated using delayed
detached-eddy-simulation and improved delayed detached-eddy-simulation based on shear stress transport model.
To eliminate the unfavorable influence of large numerical dissipation, a high order symmetric total variation
diminishing scheme with adaptive dissipation approach is implemented. Three sets of grid, including the coarse,
medium and locally refined grids are applied. It is observed that the grid density effect is weak on the mean flows, but
significant on the instantaneous quantities. Both approaches present acceptable agreements with the available
experiments. Due to its wall-modeled large-eddy-simulation mode, improved delayed detached-eddy-simulation can
deliver slightly larger secondary separation and smaller horseshoe vortex on the aft wheels, predict the shear layer
instability a little more upstream and resolve smaller instantaneous structures. There are strong interactions between
vortices and sharp corners of the landing gear struts. Strongly periodical impingement of vortices also occurs from the
fore wheel onto the rear one as well as onto the downstream part of the front wheel itself. These flow behaviors cause
the high sound pressure level distribution and lead to surface noise.

Nomenclature
A~inv
CD
CDES

=
=
=

CL
CP
Cp;rms
CW

=
=
=
=

D
d
F
F2
fB

=
=
=
=
=

f~d

fdt

fe

FSST

k
LDDES

=
=

matrix of Roe-averaged
coefficient of drag
constant in detached eddy-simulation, delayed
detached-eddy-simulation, and improved delayed
detached-eddy-simulation
coefficient of lift
pressure coefficient
root mean square of pressure coefficient
constant in improved delayed detached-eddysimulation
diameter of rudimentary landing gears wheel
the distance to the nearest wall
inviscid flux of NavierStokes equations
blending function in shear stress transport model
function in improved delayed detached-eddysimulation
function in delayed detached-eddy-simulation or
improved delayed detached-eddy-simulation
function in improved delayed detached-eddysimulation
function in improved delayed detached-eddysimulation
shielding function in delayed detached-eddysimulation
turbulent kinetic energy
length scale of delayed detached-eddy-simulation

Presented as Paper 2012-2282 at the 18th AIAA/CEAS Aeroacoustics

Conference (33rd AIAA Aeroacoustics Conference), Colorado Springs, CO,
46 June 2012; received 4 September 2011; revision received 2 June 2012;
accepted for publication 3 June 2012; published online 27 November 2012.
Copyright 2012 by the American Institute of Aeronautics and Astronautics,
Inc. All rights reserved. Copies of this paper may be made for personal or
internal use, on condition that the copier pay the $10.00 per-copy fee to the
Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923;
include the code 1533-385X/12 and $10.00 in correspondence with the CCC.
*Associate Professor, School of Aerospace Engineering. Senior Member
AIAA.

M.Sc. Student, School of Aerospace Engineering.

Ph.D. Student, School of Aerospace Engineering. Student Member AIAA

Post-Doctor, School of Aerospace Engineering. Member AIAA.

Professor, School of Aerospace Engineering; [email protected].

Associate Fellow AIAA (Corresponding Author).

Lhybrid

LIDDES
LLES

=
=

LRANS

qL and
qR
Sij
t
T
U0 or
Uref
ui

ui0

xi

x; y; z
min
max
k

=
=
=
=
=
=
=

ij

=
=
=
=
=

=
=
=
=
=

=
Subscript:
i, j, k
i  12

=
=

length scale of detached-eddy-simulation, delayed

detached-eddy-simulation, or improved delayed
detached-eddy-simulation
length scale of delayed detached-eddy-simulation
turbulent length scale of large-eddy-simulation part
in delayed detached-eddy-simulation or improved
delayed detached-eddy-simulation
turbulent length scale of Reynolds averaged NavierStokes part in delayed detached-eddy-simulation or
improved delayed detached-eddy-simulation
left- and right-biased original variables in Navier
Stokes equations
mean strain rate tensor
time
total averaged time
freestream velocity
components of velocity in x, y, and z directions,
respectively
components of velocity fluctuation in x, y, and z
directions (u 0 , v 0 and w 0 )
components of coordinates (x, y, and z)
constant of turbulent kinetic energy equation
grid scale
grid scale in x, y, and z directions, respectively
minimum grid scale of x, y and z
maximum grid scale of x, y and z
turbulent Prandtl number in the turbulence kinetic
energy equation
molecular viscosity coefficient
turbulent eddy viscosity coefficient
density
Reynolds stress tensor
adaptive function of the numerical dissipation in
symmetric total variation diminishing scheme
specific dissipative rate
at grid center
at right interface of cell i in I-direction

I.

L
107

Introduction

ANDING gear is one of the major airframe noise sources when

aircraft approaches the airport. For some wide-body aircraft,

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

108

XIAO ET AL.

like Boeing 777 or Airbus 340, the airframe noise generated from the
landing gear can play a dominant role [1,2]. Many experiments [36],
numerical simulations [711], analytical tools [1215], and flight
tests [16] for high-fidelity model or simplified/real landing gears in
the past decade have been performed to investigate and explore the
mechanism of noise generation. A number of studies on low-noise
landing gears, such as SILENCER [17], Silent Aircraft [18], and
some other noise-reduction designs [19,20], have also been carried
out recently.
Noise prediction and noise-oriented design require mature and
unsteady numerical simulation methods at high Reynolds numbers.
Measurements and flight tests suggest that the frequency of landing
gear noise generally ranges from a few to thousands of Hertz. To
ensure the accuracy of noise simulation the time-step in the numerical
simulation should be small and the overall statistical-time should
be long enough. Therefore, the computational procedure is often
overwhelmingly time-consuming. Although there are numerous
references about the characteristic of near-fields and noise prediction,
the mechanism of noise generation from the landing gear still requires
further investigation due to the highly complex geometry as well as
the complicated flow phenomena involved, for instance, the massive
separations, shear layer instabilities, vortex interactions with the
landing gear components, and so on. Numerical simulation of the
massively separated flows past the landing gear is a great challenge
for both computational fluid dynamics (CFD) and computational
aero-acoustics (CAA). In fact, the unsteady CFD and CAA methods
should be carefully calibrated and verified by the comprehensive
flow characteristics of both the mean and instantaneous near-field
flow characteristics and far-field acoustic data.
Recently, a number of interesting experiments were focused on the
flow and acoustics of the landing gear models. Lazos [1] performed a
series of detailed wind-tunnel investigation of the flows past a
simplified landing gear (SLG) with some compulsory components
including four wheels and necessary struts. With the help of digital
particle image velocimetry (PIV), Lazos [1] provided the mean-flow
quantities on the surface and some spatial distributions. An oscillating vortex between the wheels on the groundside in the axial plane
occurs due to the geometric asymmetry caused by the presence of the
post. The impingement of the vortex on the aft wheel and the rotation
and translation between the wheels were thought as the main noisegenerating sources. Lazos [21] also examined the mean-flow surface
topological features on the fore and aft wheels of the SLG with oilflow visualization technique showing clearly the flow patterns on the
wheels. The turbulence quantities of the SLG with PIV were obtained
later [22], for instance, a region of strong turbulence activity in the
gap between the wheels on the groundside in a horizontal plane could
be seen. These studies provided important information in identifying
the potential significant noise sources and served for CFD validation.
Manoha et al. [23] also attempted, to build an extensive and accurate
experimental database for a simple landing-gear structure to enable
the validation of aeroacoustics numerical tools for landing gear
applications.
On the numerical side, Hedges et al. [24] computed the unsteady
flows past Lazoss SLG model using Spalart-Allmaras model (SA)
in Reynolds averaged Navier-Stokes equations (SA-RANS) [25]
and detached-eddy-simulation (SA-DES) [26]. The comparisons
between the two turbulence-model results for the mean surfacepressure distribution, instantaneous vorticity, resolved- and
modeled-Reynolds stresses, etc. were presented and SA-DES
performed much better than SA-URANS. However, the capability
of detached eddy simulation (DES) in simulating unsteady massive
separation was not well revealed because of the limited unsteady
experiment data.
SLG experiments by Lazos [1,21,22] are very helpful to validate
CFD codes and turbulence simulation models. It was chosen as one of
the test cases of European sixth framework project, detached eddy
simulation for industry aerodynamics (DESider, 2004-2007) [27,28].
However, SLG experiments did not supply sufficient unsteady data,
which is of vital importance for the understanding of the noise
generation. Also, SLG experiments were strongly Reynolds-number
dependent. For instance, laminar separation occurred in some regions

but probably not in others. To overcome these disadvantages of SLG,

a rudimentary landing gear (RLG) was investigated to assess and
evaluate hybrid Reynolds averaged Navier-Stokes and large-eddysimulation (RANS-LES) methods in simulations of unsteady flow
physics over RLG and to explore the modeling capabilities in
analysis of flow-generated acoustic noise source in relation to farfield sound propagation [29]. The RLG has larger axles with
rectangular cross sections and triggers the transition of the wheel
boundary layer with tripping. The wheels are the same as the SLG.
This model has many advantages such as a manageable level of
complex geometry, fully turbulent boundary layer, essentially
converged solution on affordable grids and low Reynolds-number
sensitivity. For these reasons, RLG model received concentrated
studies in a number of consorted investigations, like the European
seventh Framework project Advanced Turbulence Simulation for
Aerodynamic Application Challenges (ATAAC) [30], AIAA
Workshop on Benchmark problems for Airframe Noise Computations (BANC) [31].
In fact, Spalart et al. [29] were probably the first to make use of
the RLG data in the numerical investigation with SA-URANS and
SA-DDES (Delayed-DES). Instantaneous vorticity, pressure fluctuations, mean pressure coefficients, and streamlines on the surface
were demonstrated. Xiao et al. [32] studied the flow past RLG
with URANS. The computational pressure coefficients on the
wheels presented acceptable agreements with the measurements
and numerical results of reference [29]. However, only large-scale
structures were captured and the small-scale structures were almost
missing due to the diffusive URANS model. Krajnovic et al. [33]
simulated the flows using partially averaged NavierStokes (PANS)
and large-eddy simulation (LES) with the measurements and they
found some significant different performances between LES and
PANS. Furthermore, Spalart et al. [34] and Wang et al. [35] coupled
the DES-type methods with Ffowcs-Williams/Hawkings calculation to preliminarily predict the far-field aeroacoustics of the RLG
model.
Karthikeyan and Venkatakrishnan [36] successfully applied
photogrammetry method to reproduce the three-dimensional surface
flow visualization on the RLG model from two-dimensional images
using an off-the-shelf camera in a realistic wind-tunnel environment.
The surface-flow patterns are of great significance in the understanding of the flow physics with very complex flow topologies of
three-dimensional separation. Venkatakrishnan et al. [37] also presented the measurements on the RLG surface recently, including oilflow visualization, steady/unsteady pressure coefficients, and mean
forces. The flowfield in space and the unsteady flow features were
still absent but could be supplemented by dynamic pressure data and
possibly off-body flow measurements in subsequent phases [37]. In
summarizing the recent computational work with LES and/or RANSLES hybrid methods using the latest RLG data, Spalart and Mejia
[38] noted that the aerodynamic agreement is better than the acoustic
quantities. Significant improvement in the RANS-LES hybrid
methods is thus still needed.
In this article, delayed detached-eddy-simulation (DDES) [3941]
and improved delayed detached-eddy-simulation (IDDES) [42]
based on the two-equation k--shear-stress-transport (SST) [43]
turbulence model are applied to simulate the complex and unsteady
flow past RLG. Three sets of grids with 3.6, 11, and 13 million grid
points are applied to investigate the effects of grid density and the
most suitable grid is selected for the further investigations. The flow
features, including the surface pressure, pressure fluctuation, and
surface-flow patterns, by DDES and IDDES, are mainly investigated
in comparison with available measurements. The generation of high
sound pressure level (SPL) regions on the surface are discussed
through the analysis of the vortex motion.
This paper is arranged as follows. Section II presents a brief
description of turbulence simulation models and numerical methods.
Section III discusses the mean and instantaneous results of RLG
model based on three grids and DDES and IDDES and the possible
causes of noise generation. The last section is the conclusions.

109

XIAO ET AL.

II.

Turbulence Modeling and Numerical Methods

To accurately resolve the high Reynolds number turbulent flow

past RLG, two issues are very important. One is the turbulence
simulation model, and the other is the numerical discretization
scheme, especially the numerical dissipation level associated with the
scheme employed. Furthermore, the combination of the advanced
turbulence modeling methods and adaptive dissipation may be more
important than either alone.

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

A. Advanced Detached-Eddy-Simulation Approaches:

Delayed Detached-Eddy-Simulation and Improved
Delayed Detached-Eddy-Simulation

Even with limited computational resources the combination of

LES with RANS can achieve reasonably well in terms of both
efficiency and accuracy in computing the flows with massive
separation at high Reynolds number. This modeling strategy of
turbulent flows, often referred to as the RANS-LES hybrid method,
has recently become much favored in the study of the unsteady and
geometry-dependent separated flows. The original DES model [26],
proposed by Spalart et al. in 1997, achieved widespread acceptance in
industrial CFD communities. However, some inherent shortcomings
had also been identified in the original DES through further
investigation. These shortcomings included erroneous activities of
the near wall damping terms in LES mode, incursion of LES mode
inside boundary layer, gray area and log-layer mismatch. Many of
these had been successfully addressed in later revisions (DDES [41]
and IDDES [42]), while some still remained.
Differing from the original DDES (proposed by Spalart et al. [41]
in 2006, called as DDES-2006) and IDDES [42], the background
turbulence model in this paper is not taken as the one-equation S-A
model, but the two-equation SST model. In this article, the earlier
version of DDES (proposed by Menter et al. in 2003 [39,40]) and
IDDES [42] are applied. The detailed formulations are omitted here
and can be easily found in the original references.
These two advanced DES methods can be constructed through
modifying the dissipation term of the turbulent kinetic energy (TKE)
equation. After introducing a length scale, Lhybrid , the TKE equation
can be given in tensor form as
k uj k



t
xj
xj


 
1
k
k31
(1)
 t
 ij Sij
k
xj
Lhybrid

where the length scale Lhybrid differentiates one hybrid RANS-LES

approach from the other.
1. Length-Scale Definition of Delayed Detached-Eddy-Simulation

To formulate a DDES hybrid method the length scales can be

defined as
Lhybrid  LDDES  minLLES 1 FSST ; LRANS 

(2)

This length scale LDDES is opting between the turbulence length scale
LRANS , defined as k12 
, for RANS calculation and the
filter length scale LLES , (CDES ), for LES calculation with
 maxx; y; z. FSST is a damping parameter taking the form
of F2 in the SST model. It equals one near the wall and zero outside
the boundary layer. This parameter prevents LLES from becoming less
than LRANS in the near-wall region with locally refined grids and can
thus delay the switch from original DES model to LES in the
boundary layer where RANS is needed. It is called as SST-DDES. If
FSST takes zero, it reverts to SST-DES [44].
Equation (2) can also be approximately written as
Lhybrid  LDDES  FSST LRANS  1 FSST  LLES

(3)

It has the similar formulation of DDES-2006 [41], shown later in

Eq. (5).

2. Length Scale Definition of Improved Delayed

Detached-Eddy-Simulation

For the IDDES, the length scale of Lhybrid can be written as

Lhybrid  LIDDES  f~d 1  fe  LRANS  1 f~d  LLES (4)
Here, the length scales of LRANS and LLES are defined the same as
those in DDES. The grid scale is, however, redefined as
 minmaxCw max ; Cw d; min ; max , where Cw is a constant,
d is the distance to the nearest wall, min is taken as minx; y; z,
and max is equal to maxx; y; z. Function f~d is defined as
max1 fdt ; fB , which is determined by both the geometry part fB
and the flow part 1 fdt .
When fe is equal to zero, LIDDES in Eq. (4) can be written as
LIDDES  LDDES  f~d LRANS  1 f~d  LLES

(5)

and IDDES reverts to DDES-2006 [41]. When fe is larger than zero

and f~d is equal to fB , LIDDES in Eq. (4) can be written as
LIDDES  LWMLES  fB 1  fe  LRANS  1 fB  LLES
(6)
and IDDES acts in wall-modeled LES (WMLES) mode near the
wall.The detailed formulations of functions fB , fe , fdt , etc., can be
found in the original reference [42].
From Eqs. (3) and (5), the shielding function is the only difference
between DDES and DDES-2006. From Eqs. (4) and (5), it can be
found that IDDES has the WMLES mode, but DDES does not. In the
authors previous study [45], the above three versions of advanced
DES-type methods based on SST model were analyzed in detail and
their performances for tandem cylinder flows were well demonstrated.
B. Numerical Methods

Also, it has been shown [46] the numerical dissipation of spatial

schemes played a key role on both the mean and instantaneous flow
features. It was illustrated that excessive numerical dissipation would
prevent the generation of small-scale structures, suppress the flow
separation, and reduce the range of the resolved frequencies in the
computation of flow past tandem cylinders.
1. Spatial Scheme with Adaptive Dissipation

When LES is applied to simulate turbulent flows the numerical

dissipation should be low enough with very fine grids to resolve the
appropriate turbulence scales. However, they often suffer from the
spurious oscillation due to the coarse grid. The adaptive dissipation
scheme, whose dissipation becomes large near the wall and in
the irrotational region while very small in the separation region
[44,46,47], is an appropriate choice.
In this article, the high-order symmetric total variation diminishing
(STVD) scheme based on the upwind Roe scheme, which combines
sixth order symmetric scheme with fifth order weighted essential
non-oscillating interpolation (S6WENO5) [48], with adaptive
dissipation, can successfully resolve the massively separated flows.
The reason for choosing STVD schemes is that this algorithm allows
one to independently control the dispersion and dissipation errors in
the solution. The invicid flux of NavierStokes equations is given as
Fi12  Fsymmetric;i12
|{z}
6th order center scheme

1 ~
jAinv jqR qL i12
2
|{z}

5th order WENO based on Roe scheme

(7)
where the adaptive function varies from zero to one.
In this subsection, the detailed formulations of the adaptive
dissipation function are not reillustrated here as they can be found
in the references [44,46,47]. This approach, combining DDES or
IDDES with adaptive dissipation scheme, is also applied directly to
the RLG computation. Figure 1 shows the distribution of the modeled
eddy viscosity and the adaptive dissipation coefficient on the x-z

110

XIAO ET AL.

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

Fig. 1 Modeled eddy viscosity ratio and the distribution of the adaptive dissipation coefficient- by DDES.

plane cross the post, the x-z plane past the center of the wheels
(yD  0), the symmetric plane of the RLG model (zD  0), and
the central plane of wheels (zD  0.44), respectively. As shown in
Fig. 1, is low enough in the separation region where the flow is
dominated by turbulence. In the irrotational region and near the wall,
is close to unity, and the scheme reverts to the original S6WENO5
scheme.
2. Other Numerical Methods

The computations are all based on an in-house compressible

solver, unsteady NavierStokes equation solver (UNITs), in a cellcentered finite-volume formulation. The TKE and specific
dissipation rate transport equations are solved and decoupled with
the mean flow equations. The approach is parallelized using domaindecomposition and message-passing-interface strategies for the
platform on computer clusters.
A modified fully implicit low-upper symmetric Gauss-Seidel
method with Newton-like sub-iteration in pseudo time is taken as the
time marching method when solving the mean flow and the
turbulence model equations. Global nondimensional time stepping is
implemented to capture the unsteady properties of the massive
separation flows.
All the computations of DDES and IDDES start from the
flowfields solved by URANS.

III.

Results and Discussion

In the present work, research attention is mainly focused on the

mean surface patterns, pressure fluctuation on the surface, surface

Fig. 2

Computational domain and dimensions of RLG model [30].

flow patterns, and instantaneous flow structures of the RLG. Firstly,

the grid density effects are investigated using a coarse, a medium, and
then with a locally refined meshes. Secondly, DDES and IDDES are
both applied to present their performances based on the most
reasonable results among the above three meshes. Lastly, the
generation mechanism of high SPL regions on the RLG surface can
be well understood through analyzing the near-field flows in space.
The velocity of the freestream is 40 ms; the Reynolds number
based on the diameter of the wheel is 1 106 and the angle of attack is
zero. Normalized uniform time step is taken as 0.005. In a physical
time step, 80 subiterations are applied to achieve second order
temporal accuracy due to the compressible solver for low-speed
flows.
The computational domain and the dimensions of RLG model are
presented in Fig. 2. The computational domain is the same as the test
cross section in the wind tunnel, recommended by BANC-I [31,38].
In-flow conditions are specified at the computational domain flow
inlet and the outlet conditions are applied at the domain exit. No-slip
wall condition is applied on the surface of RLG model. The tunnel
boundary-layer thickness is not provided, so invisicid wall
conditions are applied to the top-, bottom- and, side-tunnel wall
boundaries [38].
It is noted that the main flow is in the x-direction, the wingside
means near the wing along the y-direction, and the groundside means
near the ground in the adverse direction of y-direction.
A. Effects of Grid Density

In Fig. 3, the medium grid, Grid-M, corresponds to the mandatory

grid in project ATAAC, generated by ANSYS. It has about 11 million

Fig. 3 Comparisons of the three grids around RLG model (grids in x-z
plane with yD
0).

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

XIAO ET AL.

Fig. 4

Grids on the specific planes and on the surface.

grid points. The streamwise grid scales between the wheels and
struts are almost isotropic, about 0.01D. After the aft wheels, the
streamwise grid scales become larger and larger gradually. The
streamwise grid scales at 1D after the rear wheel are about 0.022D.
Unfortunately, in the further downstream of RLG, the streamwise
grid scales become too coarse in a short distance (about 0.13D at 2D
after the aft wheels) and the filter length scales () of the LES part are
generally taken as the streamwise grid scale, leading to large subgrid
scale eddy viscosity. At the same time the resolution of small
structures can be greatly suppressed by the coarse streamwise grids.
To resolve more small-scale structures in the wake of the RLG
model, and capture more reasonable near-fields flow, the grids are
then locally refined only in the wake to allow more streamwise grid
clustering. The locally refined grid (Grid-LF) has two million cells
more than Grid-M. The streamwise grid scale is about 0.035D at 2D
after the aft wheels.
Furthermore, to explore the grid-density effects, a coarse grid
(Grid-C) is also generated, which has about 70% grid points in all
three x-, y- and z-directions compared with the medium grid,
respectively. Grid-C has about 3.6 million grid points. The local grid
scales of Grid-C are about one and a half times of those of Grid-M and
Grid-LF, especially around the RLG model. The main differences
among the three grids are shown in Fig. 3.
Due to a shared topology of Grid-C, M, and LF, several specific
slices of Grid-C, such as the symmetric plane (x-y plane with
zD  0), the central plane of wheels (x-y plane with zD  0.44),
the cross plane through the aft wheel center (y-z plane with
xD  0:58), and the cross plane through the post (y-z plane at
xD  0), can be used to demonstrate the grid features, shown in
Fig. 4. Although the configuration of RLG is very complex,
multiblock point-to-point structured grids are generated with good
orthogonality near the wall.
Table 1 presents the comparisons of the aerodynamic force
coefficients of the whole model using the three grids between the
Table 1 Comparisons of force coefficients using
different grid density for the whole RLG

Exp. [38]
Grid-LF
Grid-M
Grid-C

Grid points (million)

13
11
3.6

T U0 D
41.0
55.0
42.0

111

CD
1.60
1.77
1.78
1.82

CL
0.65
0.30
0.30
0.31

measurements and computations. The experimental data, labeled as

Exp., are taken from Spalart and Mejia [38]. Though different, the
averaged time for each set of grids is enough to obtain the mean force,
pressure, pressure fluctuations, and so on.
The experimental drag coefficient (CD ) is 1.60, whereas all the
computational results are larger than it. The drag coefficient of GridC (1.83) is not only larger than the experiment (1.60), but it is also
larger than those of both Grid-M (1.78) and GridLF (1.77). The
relative errors with the experiments of Grid-C, M and LF are 13.8,
11.3 and 10.6%, respectively.
The absolute value of the experimental lift coefficient (CL ) is very
large (0.65), whereas all the absolute values of computational CL
using the three grids are much smaller (about 0.30). Spalart and
Mejia [38] summarized six groups computational lift coefficients,
ranging from 0.18 to 0.28, and commented that almost all the
CFD results are higher than the experiment; the agreement on lift is
very poor, for reasons unknown.
The comparisons of Grid-C, M, and LF on the force, minimum and
maximum pressure coefficients (Cp ), and sound pressure level on the
fore and aft wheels are presented in Table 2. In project ATAAC
Boeing provided the original dynamic pressure data measured by
kulites on the wheels, post, beam, and struts surface, which includes
200,000 samples for each kulite at 10 K Hz. It means that the overall
test time is 20 s. After averaging and analyzing these data, we have
reconstructed the pressure and SPL distributions on the wheels
surface. The experimental data from the dynamic pressure are named
as Exp.-revB**. From the comparisons of Exp. and Exp.-revB on the
pressure coefficients and SPL, some minor differences are found.
One possible reason is that the pressure coefficients of Exp. are
obtained from the statistic pressure ports but not the unsteady
pressure from kulites. Because the mean pressure and SPL on the
wheels surface of Exp.-revB can be conveniently plotted and
compared with computations in three-dimensional visualization they
are taken as the measurements in the following subsections.
1) The computational drag coefficients for wheels are quite good
and the difference between measurements (Exp.) and computations is
much smaller than those of the whole model. The difference from the
measurements and computations on lift is acceptable (0.016 for fore
wheel and 0.003 for aft wheel using Grid-LF), whereas the lift
coefficients (CL ) have a difference of 0.35 for the whole model,
shown in Table 1. The possible reason for lift disagreement is the
**Mejia K., Rudimentary Landing Gear- Description of Dynamic Data &
Files (rev. B), ATAAC Project Internal Communication, 8 February 2012.

112

XIAO ET AL.

Table 2

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

Exp.-revB
Exp. [38]
Grid-LF
Grid-M
Grid-C

Comparisons of mean force, pressure, and pressure fluctuation parameters for the fore and aft wheels with different grid density

CD;p fore CL;p fore CD;p aft CL;p aft Cp;min fore Cp;max fore Cp;min aft Cp;max aft SPLmin fore SPLmax fore SPLmin aft SPLmax aft

2.53
1.07
2.32
1.03
117
144
121
149
0.24
0.05
0.17
0.03
2.11
1.01
2.00
0.90
112
147
121
151
0.233
0.066
0.170 0.027
2.10
1.02
2.29
0.97
113
145
119
149
0.241
0.066
0.169 0.029
2.09
1.04
2.15
0.98
112
144
121
149
0.237
0.071
0.175 0.031
2.08
1.02
2.39
1.01
98
143
115
150

different pressure distribution on the beam, struts, and the invicid

wind-tunnel wall hypothesis.
2) It seems that the computational minimums of pressure
coefficients on the fore wheels agree with the experimental data from
the pressure ports (Exp.) better than those from the kulites (Exp.revB). On the aft wheel Cp;min coefficients of Grid-C and LF show
better agreement with Exp.-revB than Exp. However, Grid-M agrees
with Exp. better. All the computational Cp;max coefficients on the fore
and aft wheels by three grids agree well with the Exp. and Exp.-revB.
3) When comparing the SPL on the wheels, both Grid-M and LF
agree better than Grid-C, although there is 45 dB difference
(between the computations and Exp.-revB data) of the SPLmin on the
fore wheel. Generally, Grid-C performs the worst.
Figure 5a presents the magnitude of vorticity at four planes,
including the symmetric plane of the whole model (zD  0), the
central plane of the wheel (zD  0.44), the plane cross the post
(yD  0.4), and the plane cross the wheel centers (yD  0). From
the comparisons, the effects of the grid density and distribution have a
significant influence on the instantaneous vorticity distribution. GridC presents relatively large-scale structures. Grid-M, due to the coarse
grid in the wake, the vorticity diminishes quickly behind the RLG
model. Grid-LF presents plenty of small-scale structures in the wake
due to its reasonable grid scale distribution in streamwise direction.
The same tendency is shown in Fig. 5b. The iso-surfaces of
instantaneous Q-criterion using the three grids are presented. GridLF exhibits more and smaller flow structures in the wake than those
by Grid-C, or even by Grid-M.
Figure 5c presents the comparisons of root mean square of pressure
coefficients (Cp;rms ) at three sections in z-direction (zD  0.292,
0.440, and 0.588, respectively) on the fore and aft wheels. From this
figure, the Grid-C performs worst due to its coarse grid scale in x-, yand z-direction. Grid-M and LF performs similarly and presents
acceptable results except for some regions such as the separation of
the fore wheel and the reattachment on the aft wheel.
Comparisons of the numerical results with different grid density
suggest that Grid-C is too coarse for the RLG flow simulation. GridM and LF perform similarly, whereas Grid-LF presents a little more
reasonable instantaneous flow structures in the wake and a little better
surface-pressure fluctuation. In the rest of this article, all the results
will be based on Grid-LF, whose streamwise grids in the wake are
locally refined.
B. Effects of Delayed Detached-Eddy-Simulation and Improved
Delayed Detached-Eddy-Simulation

In this subsection, the performances of DDES and IDDES will be

presented through comparing the mean and instantaneous flows past
RLG with the experiments as detailed as possible, mainly focusing on
the surface flows, such as pressure, pressure fluctuation, and surface
flow patterns, and so on.
1. Mean Cp on the Surface

First, the computational Cp coefficients by DDES and IDDES on

the surface, such as the beam, the struts, the post, the fore wheel, and
the aft wheels, are compared with the measurements of Exp.-revB.
Figure 6 shows the comparisons of Cp at 17 points on the beam,
struts, and post. The computational results of DDES and IDDES are
almost the same, except at point 3. Points 10 and 13 are in the
stagnation region on the windward surface, which corresponds to Cp
of 1. Maximum of disagreements about 0.2 occur at points 7, 9, and
16. Points 7 and 16 are on the side of the post; the difference is

possibly caused by the shear-layer instability from the sharp edge.

The feature of the shear layer from the upstream strut can also lead to
disagreement of Cp at point 9 on the beam of the groundside. At point
3, IDDES predicts a little larger Cp (about 0.2) than DDES.
The comparisons of Cp on the fore wheels are shown in Fig. 7. The
Cp contours on the forward, rearward, inboard, and outboard surfaces
are demonstrated respectively. At the same time, the comparisons of
Cp coefficients at three sections in the z-direction (zD  0:292,
0.44, and 0.588, respectively) are also presented.
1) Little difference between DDES and IDDES is distinguished on
the forward surface. Both of them can well predict the pressure at the
stagnation region (Region A, where Cp coefficients range from 0.8 to
1.0) and attached regions on the windward surface. The relatively
obvious difference is in the low-pressure region (Region I, where Cp
coefficients are between 2.4 and 1.6) on the outboard surface and
the experimental Region I looks a little larger than that of DDES and
IDDES.
2) From the rearward surface of the experiments, there is an
obvious low-pressure region (Region B, where Cp coefficients range
from 1.4 and 1.1). Both DDES and IDDES can capture this
region, but they all underpredict its range and IDDES presents the
smaller one. At the same time, the measurements present another
relatively low-pressure region (Region C, where Cp coefficients are
between 0.8 and 0.75) on the wingside and both DDES and
IDDES underpredict it again.
3) Due to the junction flow of strut and inboard surface of the
wheel, high-pressure region (Region D, where Cp coefficients are
between 0.6 and 0.7) on the inboard surface is observed in the
computations and measurements and both DDES and IDDES present
a little larger Region D than the measurements. Four other specific
regions, called as Region E (where Cp coefficients range from 1.35
to 1.25), F (Cp : 1.35, 1.1), G (Cp : 0.3, 0.05) and H (Cp : 0.8,
0.75), are also observed on the inboard surface. Both DDES and
IDDES can predict regions D, E, G and H, but they all look a little
smaller than those of the measurements. Unfortunately, both DDES
and IDDES cannot well predict Region F.
4) Due to the attached flow on the outboard surface there is little
difference between DDES and IDDES, and they match the
measurements well, although the experimental low-pressure Region I
looks a little larger than that of the computations.
5) After analyzing the surface pressure coefficients contours from
the four views, DDES differs only slightly from IDDES. Some gaps
in quantity between computations and measurements of Cp are
presented in three spanwise sections. At the section of zD  0:292
near the inboard surface, the disagreement of the computations and
measurements (from 100 to 255 deg) corresponds to the different
range of Region E, F and H on the inboard surface. At the middle
plane (zD  0:440) of the wheels, both DDES and IDDES can well
predict the measurements, and the difference is really small. At the
section of zD  0:588 near the outboard surface, both DDES and
IDDES underpredict Region I, which leads to the disagreement (from
0 to 60 and from 325 to 360 deg) between the computations and
measurements.
The comparisons of mean Cp coefficients on the aft wheels are
shown in Fig. 8.
1) A very small difference is observed between the computations
and measurements on the windward surface where the flow is almost
attached. DDES and IDDES differ from each other very little. The
high-pressure region (Region J, where Cp coefficients range from 0.6
to 1.0) is mainly caused by the impingements and reattachments from
the fore wheels. The low-pressure region (Region L, where Cp

113

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

XIAO ET AL.

Fig. 5 Effects of grid density on the computations, comparisons are made for coarse, medium, and locally refined meshes. a) Slices of instantaneous
vorticities; b) Instantaneous Q-criterion; c) Cp;rms on the wheels.

coefficients range from 2.2 to 1.6) of the measurements looks a

little smaller than that of both DDES and IDDES. Between Region J
and L, Region K (where Cp coefficients range from 0.15 to 0.05)
has a little different shape and range between the computations and
measurements.

2) A relatively large difference between the computations and

measurements is observed from the rearward surface. In the
measurements no low-pressure region (Region M, where Cp
coefficients range from 1.35 to 1.1) appears, whereas it can be
observed from DDES and IDDES, especially from DDES.

114

XIAO ET AL.

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

Fig. 6

Comparisons of mean Cp on the beam, struts, and post.

3) Due to the junction flow of strut and the inboard surface of the aft
wheel, high-pressure Region N can be observed from computations
and measurements on the inboard surface. DDES presents an obvious
low-pressure Region O, whereas there is no such region in
measurements.

4) From the outboard surface, both DDES and IDDES present a

little larger low-pressure region (Region P) than measurements and
Region P by IDDES is a bit larger than that of DDES.
5) Then, after analyzing the pressure contours from the four
views, it is found that DDES and IDDES perform almost the
same. At the section of zD  0:292, both DDES and IDDES
overpredict the low-pressure region (from 30 to 90 deg), which
corresponds to Region L; at the middle plane of zD  0.440, the
disagreement (from 0 to 55 deg) is on the windward surface,
corresponding to Region K, and at the section of zD  0.588,
the computational results look a little smaller than the
measurements from 118 to 180 deg, corresponding to the lowpressure Region M.
2. Pressure Fluctuations on the Surface

Due to the unsteady and massive separation flow past the RLG
model the pressure fluctuations on the RLG model surface are
investigated in this subsection. The comparisons of the SPL

Fig. 7 Comparisons of the mean Cp distribution on the fore wheel.

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

XIAO ET AL.

Fig. 8

115

Comparisons of the mean Cp distribution on the aft wheel.

distribution on the wheels and pressure fluctuations at 12 points on

the surface are presented from Figs. 913.
The comparisons of SPL on the fore wheels are presented in Fig. 9.
1) On the forward surface, DDES and IDDES perform almost the
same, and they present much smaller SPL (about 5 dB) on the
windward surface (Region A), where the flow is stagnated, attached
and turbulent, than the measurements. In the attached region, both
DDES and IDDES act in RANS mode. Thus, the disagreement with
the experiment is foreseeable.
2) On the rearward surface, the SPL is much larger than that on the
windward surface and the difference between computations and
measurements is relatively small. The main reason is that both DDES
and IDDES act in LES mode and the flow is really unsteady and
turbulent. The relatively high-SPL regions (B, C, Q, R and S)
often correspond to the characteristic flow features, such as the
unsteady separations or reattachments.
The comparisons of SPL on the aft wheels are presented in Fig. 10.
1) It is easily found that the SPL level on the windward surface is
much higher than that of the fore wheels; the main reason is that the
aft wheel is down-washed by the turbulent wake of the fore one.

Avery high-SPL region (Region T, about 148 dB) on the groundside

and a relatively high-SPL region (Region U, about 147 dB) on the
wingside can be clearly observed. Both Region T and U are near
the inboard side of the wheels, and they are possibly caused by the
unsteady impingements of the vortices from the front wheels. Both
DDES and IDDES can well predict these two regions. IDDES
presents a little more reasonable shape and distribution than those of
DDES. In addition, both DDES and IDDES present another high SPL
region (Region V, about 140 dB) near the outboard side of the wheel,
while no such region can be observed in the measurements.
2) On the rearward surface, both DDES and IDDES overpredict the
SPL level by about 2 dB, and they present a larger Region W than the
measurements. In this region, both DDES and IDDES even present
another higher SPL region (Region X, about 146 dB). On the
wingside, DDES and IDDES overpredict the Region Y (about
140 dB). Region W, X and Y are mainly caused by the unsteady
separations.
The comparisons of SPL on the whole RLG model are presented in
Fig. 11. Due to the insufficient kulites on the beam, struts, and post,
the measurements only on the wheels are reconstructed and plotted to

116

XIAO ET AL.

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

Fig. 9

Comparisons of the SPL distribution on the fore wheel.

compare from four views (down-front, down-rear, up-front, and uprear). Generally, DDES and IDDES perform similarly.
1) As mentioned before, the computational SPL distributions
match the measurements well except the attached-flow region
including the windward surface of the front wheel and outboard
surfaces of front and rear wheels. Both DDES and IDDES are in
RANS mode and underpredict the SPL level naturally.
2) It is very interesting that the maximum SPL regions, especially
Region T, are found on the groundside. In addition, a very high SPL
region (Region Z, from 146 to 150 dB) can be observed on the beam
surface. It is possibly caused by the unsteady separations and
reattachments between the vortices and the sharp corner of the beam
and struts.
The comparisons of the Cp;rms coefficients on the wheels at three
spanwise sections are presented in Fig. 12.
On the fore wheels, both DDES and IDDES generally underpredict
the Cp;rms coefficients, especially on the windward surface. On the
leeward surface, obvious differences can be easily observed.
1) At the section near the inboard surface (zD  0:292), the
disagreements (from 105 to 210 deg) between the computations and
measurements, corresponding to the high-SPL distribution of Region
B and Q, are shown in Fig. 9.
2) At the middle plane (zD  0.440), the difference (from 135 to
225 deg) is relatively small and the computational peak looks a little
smaller than that of measurements.
3) At the section near the outboard surface (zD  0.588), DDES
performs a little better than IDDES from 160 to 200 deg,
corresponding to the high-SPL Region S shown in Fig. 9.
On the aft wheels, DDES and IDDES perform better than those on
the fore wheels, and both of them can well predict the pressure
fluctuation features.
1) At the section near the inboard surface (zD  0:292), IDDES
performs better than DDES from 0 to 65 deg, although a little
disagreement occurs between IDDES and measurements, corresponding to the high-SPL Region T on the groundside shown in
Fig. 10. From 65 to 250 deg, both DDES and IDDES perform well.
From 250 to 275 deg, both DDES and IDDES cannot well predict
the peak. From 275 to 360 deg, IDDES well matches with the
measurements and it performs better than DDES, corresponding to
the high-SPL Region U on the wingside shown in Fig. 10.
2) At the middle plane of the wheel (zD  0.440) both DDES and
IDDES perform well. The peaks at about 45, 155, 235, and 320 deg
correspond to high-SPL Region T, X, Y and U shown in Fig. 10.

3) At the section near the outboard surface (zD  0.588) both

DDES and IDDES underpredict the Cp;rms coefficients because this
section is near the outboard surface from 0 to 139 deg. However, both
of them give higher Cp;rms values than the measurements from 140 to
170 deg, corresponding to Region V shown in Fig. 10.
The comparisons of power spectral density (PSD) at 12 points are
presented in Fig. 13. Point 1, 2, and 3 are on the rearward surface of
the aft wheel, the post, and the downstream strut, respectively. Point
4, 5, and 6 are on the windward surface of the fore wheel, the post, and
the upstream strut, respectively. Point 7, 8, and 9 are on the rearward
surface of the fore wheels. Point 10 and 11 are on the forward surface
of the aft wheels, and Point 12 is on the rearward surface of the aft
wheels.
1) DDES and IDDES perform similarly at Point 2, 3, 4, and 6. At
Point 1, IDDES presents a little larger PSD than DDES, whereas
IDDES presents a little smaller PSD than DDES at Point 5. Generally,
both DDES and IDDES can well predict the pressure fluctuation at
low frequency, where Strouhal number is less than 10. Distinct
disagreement at high frequency is observed when Strouhal number is
greater than 10.
2) At Point 1 DDES and IDDES slightly overpredict the PSD. No
primary frequency can be observed from numerical results and
measurements. At point 2 and 3 almost no primary frequency is
observed in the experiment.
3) At Point 4, 5, and 6, an obvious primary frequency (around
St  1.0) can be observed. Both DDES and IDDES show good
agreement with the measurements. From Figs. 9 and 10, some highSPL regions are found, such as B, Q, C, T, U and W, which
correspond to Point 7, 9, 8, 10, 11, and 12, respectively. The PSD and
frequency features by IDDES at these six points (from Point 7 to 12)
in these regions are also presented. Generally, IDDES well predicts
the PSD when St number is less than 10. IDDES underpredicts the
PSD in the high-frequency part.
4) At Point 7 and 12, an obvious primary frequency, where St is
about 1.0, can be observed. It means that the flow separates
periodically.
5) At Point 10 and 11 the flow is reattached periodically with St
about 1.0.
3. Surface Flow Patterns

In this subsection the surface flow patterns of the RLG model are
investigated. The measurements are taken from the photos of the
wind-tunnel model [36,37].

Fig. 10 Comparisons of the SPL distribution on the aft wheel.

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

XIAO ET AL.

Fig. 11 Comparisons of the computed SPL contours from IDDES and DDES with the experiment for the whole model.

Fig. 12 Comparisons of root mean square of Cp on the fore and aft wheels at sections zD
0.292, 0.440, and 0.588.

117

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

118

XIAO ET AL.

Fig. 13 Comparisons of the power spectral density at twelve typical locations on the wheels.

According to the analysis of surface topologies [21,49,50], we

apply the similar legends for the characteristic flow features in the
current work, shown in Fig. 14. Typical flow patterns presented
include Focus of Attachment (FoA), Focus of Separation (FoS),
Node of Attachment (NoA), Node of Separation (NoS), Line of
Attachment (LoA), Line of Separation (LoS), Saddle of Attachment
(SoA) and Saddle of Separation (SoS).
In Fig. 15 the surface flow patterns of the fore wheels are analyzed.
As shown in Fig. 15a the flow is almost attached on the windward and
outboard surface. Stagnation point, which is a typical node of
attachment (NoA0), can be easily observed. The region around NoA0
corresponds to the high-pressure Region A in Fig. 7 and low-SPL
Region A in Fig. 9. DDES performs similarly with IDDES and both
of them well match the experiment with transition tripping.
Figure 15b presents the comparisons of the surface flow patterns
on the inboard surface. Because the rectangular strut is mounted on
the inboard surface the flow past the strut and wheel has typical
junction flow features. Generally, both DDES and IDDES present
reasonable agreement with the measurements and the horseshoe

vortex can be easily observed. Due to the large size of strut, the range
of horseshoe vortex, which is labeled by the dashed line of separation
(LoS0), is also very large.
1) If we observe the horseshoe vortex carefully, we find that the
horseshoe vortex before the strut in streamwise direction by DDES is
a little smaller than that of the experiment, and the one by IDDES is
the smallest. It means that the saddle of separation (SoS0) by DDES
or IDDES is more downstream than that of the experiment. Table 3
presents the coordinates of SoS0. The main difference is the
streamwise position. The SoS0 by IDDES and DDES is about
0.079D and 0.019D more downstream than that of experiment. In
addition, the local surface flow patterns around SoS0 are zoomed in,
and very complex surface-flow structures are demonstrated by
DDES, such as two saddles of separation (SoS-z0 and SoS-z1) and a
node of attachment (NoA-z0). From the same similar figure by
IDDES only SoS0 is observed.
2) Two nodes of attachment (NoA1 before the strut and NoA2
behind the strut), which correspond to two high-pressure Regions D
and G in Fig. 7, respectively, can be easily observed.

Fig. 14 Some typical surface flow patterns [21,49, 50] (from left to right: FoA, FoS, NoA, NoS, LoA, LoS, SoA, and SoS).

119

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

XIAO ET AL.

Fig. 15 Comparisons of the surface flow patterns on the fore wheel surface. a) On the windward and outboard surface; b) On the inboard surface; c) On
the leeward and outboard surface.

3) Two focuses of separation are observed near the strut. One is on

the wingside (FoS0) and another is on the groundside (FoS1).
4) In addition, three other saddles of separation can also be
observed, including SoS1, SoS2 and SoS3, by both DDES and
IDDES.
The flow patterns on the rearward surface are shown in Fig. 15c.
1) A focus of separation (FoS2) on the groundside can be
obviously found by DDES, IDDES, and experiments. After
comparing the pressure and SPL contours, FoS2 corresponds to lowpressure Region B in Fig. 7 and high-SPL Region B in Fig. 9. It
means that this kind of separation can lead to strong unsteady
pressure fluctuation.
2) Near the outboard surface another focus of separation (FoS3)
can be identified by DDES and IDDES. Because FoS3 is relatively
weak its effects on the pressure and SPL are not directly presented.
3) Between FoS2 and FoS3 two saddles of separation (SoS4 and
SoS5) can be observed.
4) On the wingside both DDES and IDDES present a node of
separation (NoS0), which correspond to high-SPL region C, shown
in Fig. 9, and a saddle of separation (SoS7).
5) It is interesting that a saddle of separation (SoS6) and a node of
attachment (NoA3) by DDES, corresponding to high-SPL Regions
Q and R, shown in Fig. 9, are observed, whereas they cannot be
explicitly presented by IDDES.
6) The secondary separation near the outboard surface corresponds
to the high-SPL Region S shown in Fig. 9. It indicates that the
separation on the shoulder of the wheel is strong and unsteady.

The comparisons of the surface flow patterns on the aft wheels

between the computations and measurements are presented in
Fig. 16.
Figure 16a presents the surface flow patterns on the windward and
outboard surface. Both DDES and IDDES well match the experiment
and the difference is really small. The surface flow patterns on the
windward surface of the aft wheels are a bit similar with those of the
fore wheel, but some differences still exist. The flow on the aft wheel
is attached as a line of attachment (LoA0) or impingement line, but
not as a node of attachment (NoA0) on the fore wheels, shown in
Fig. 15a. The LoA0 corresponds to high-pressure Region J in Fig. 8
and high-SPL Region V in Fig. 10. It is mainly caused by the wake of
the fore wheels. In addition, a saddle of separation (SoS8) and a node
of attachment (NoA4) are observed on the wingside, which are
caused mainly by the wake of fore wheels, the strut, and beam. Its
very interesting that the surface flow around the extremely large-SPL
Region T is very smooth without any singularities.

Table 3

the coordinates of SoS0

Coordinates of SoS0
xD
Exp.
0.982
DDES
0.963
IDDES
0.903

yD
zD
0.0156

0.0233 0.256
0.0202 0.256

XIAO ET AL.

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

120

Fig. 16 Comparisons of surface flow patterns. a) On the windward and outboard surface; b) On the leeward and outboard surface; c) On the inboard
surface.

On the rearward surface, shown in Fig. 16b, the flow patterns

seem relatively simpler than those on the fore wheels. Its very
easy to observe a focus of separation (FoS4), which just looks like
FoS2 shown in Fig. 15c on the rearward surface of the fore wheel,
with a saddle of separation (SoS9). FoS4 corresponds to lowpressure Region M in Fig. 8 and high-SPL Regions W and X in
Fig. 10. The significant difference between DDES and IDDES is

that the secondary line of separation (LoS1) by DDES is much

shorter (about 0.155 D) than that of IDDES. Then, IDDES shows
a better agreement with the experiments. Table 4 presents the
comparison of the end coordinates of the line of separation
(LoS1).
On the inboard surface, shown in Fig. 16c, the horseshoe vortex
can also be easily observed due to the junction flow past the strut and

XIAO ET AL.

Table 4

The difference of LoS1

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

End coordinates of LoS1 xD

yD
zD
Exp.

DDES
1.057
0.0944 0.564
IDDES
1.068 0.0607 0.551

the aft wheel. The range of the horseshoe vortex is labeled as line of
separation (LoS2).
1) Generally, DDES and IDDES can present acceptable results
with the experiments, especially on the groundside. Both of them
present the reasonable horseshoe vortex, the node of attachment on
the strut (NoA5), and the saddle of separation (SoS10) on the
wingside, and so on. NoA5 corresponds to the high-pressure Region
N in Fig. 8. The range of the horseshoe vortex by IDDES is a little
smaller than that of DDES, and it presents better agreements with the
experiment.

121

2) DDES presents similar surface flow patterns with those of

IDDES. Both of them present two saddles near the strut, where one is
on the groundside (SoS11) and the other is on the wingside (SoS14).
Also, the node of attachment (NoA6) near the strut, the node of
separation (NoS1) behind the strut, the downstream saddle of
separation (SoS12), the downstream node of attachment (NoA7), and
the saddle of separation (SoS13) are observed by DDES and IDDES.
The difference between DDES and IDDES is that SoS12, NoA7, and
SoS13 by IDDES are more downstream than those by DDES.
Furthermore, the locations of NoA7 and SoS13 are just reversed by
these two methods.
3) The local surface flow patterns around SoS10 also are magnified
and presented in the same figure. The flow topologies by DDES look
much more complex than those by IDDES. In this region, two saddles
of separation (SoS-z2 and SoS-z3) and a node of attachment (NoAz1) by DDES are demonstrated, whereas only one saddle of
separation (SoS10) is presented by IDDES.
4) The local surface flow patterns on the groundside are also
presented and labeled by the dashed square. Again, DDES presents

Fig. 17 Surface flow patterns on the several components. a) On the beam, struts, and post; b) On the wheels, beam, and struts; c) On the whole model.

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

122

XIAO ET AL.

Fig. 18 The comparisons of instantaneous flows by DDES and IDDES. a) Q-isosurface (Q
5.0); b) Vorticity across the post and the center of the
wheels; c) Vorticity at the middle plane of wheels and on the symmetry plane of the RLG model.

more complex surface flow patterns than those of IDDES. A saddle of

separation (SoS-z4) and a node of attachment (NoA-z2) by DDES are
presented, whereas no specific flow features are presented by IDDES.
Figure 17 presents the surface flow patterns on the several
components of the RLG model.
1) From the comparisons of surface flow patterns on the struts,
beam, and post with sharp corners, shown in Fig. 17a, the surface
flow is very complex on the groundside surface of the beam. Both
DDES and IDDES perform well and the difference between them is
very small. From this figure, some specific flow features, such as
node of attachment, saddle of separation, focus of separation, and so
on, are presented.
2) From the top view of the surface flow on the wheels, beam, and
struts, shown in Fig. 17b, DDES and IDDES perform similarly and
the difference between the computations and experiments is really
small.
3) Figure 17c presents the surface flows on the rearward surface of
the overall RLG model. In other words, this figure can be considered
as the combination of all the components of the model. Both DDES
and IDDES present acceptable agreements with experiments. The
main differences are the secondary separation (LoS1) on the leeward
surface shown in Fig. 16b, and the range of horseshoe vortex (LoS2)
on the inboard surface of the aft wheels shown in Fig. 16c. As
mentioned before IDDES performs better than DDES in terms of the
secondary separation and the range of the horseshoe vortex on the aft
wheels.

RLG model. With the same grid IDDES seems to resolve smaller
scales than those by DDES.
Figure 18b presents the vorticities distribution in x-z plane at two
normal sections. On the x-z plane through the center of the wheels
(yD  0), the scales of the small structures by IDDES look a little
smaller than those by DDES. On the x-z plane through the post, the
shear layer instability by IDDES from the post occurs more upstream
and develops wider than those by DDES.
Figure 18c presents the vorticities distribution in x-z plane at two
spanwise sections. On the middle plane of the wheel, where
zD  0.44, the shear layers detached from both the groundside and
wingside of the fore wheels instabilize and breakdown into small
scale of structures. These structures periodically impinge the aft
wheels and lead to high-SPL region on the aft wheels, like Regions T
and U shown in Fig. 10. On the symmetric plane of the RLG model
(zD  0), a very strong shear layer is generated from the sharp
corner of the beam, and the shear layer instability by IDDES occurs a

4. Instantaneous Flows in Space

In the above subsections the mean pressure coefficients, pressure

fluctuation, and surface flow patterns are investigated through
comparisons with the available measurements. In this subsection, we
mainly investigate the instantaneous flow features by DDES and
IDDES.
Figure 18a presents the isosurface of Q criterion (Q  5.0),
contoured by the relative streamwise velocity. Both DDES and
IDDES can well predict the small-scale motions, especially in the
wake of the post, between the wheels, and in the wake of the whole

Fig. 19 High-SPL regions and interactions between the vortices and

sharp corners. Letters A, B and T denote the high-SPL regions.

123

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

XIAO ET AL.

Fig. 20 Generation of some high SPL regions on the RLG model surface.

little more upstream than that of DDES. In addition, the scales of

structures by IDDES look smaller than that by DDES.
C. Generation of High Sound Pressure Level Regions on the Surface

The SPL distribution on the groundside is presented in Fig. 19

together with the mean vorticity contours around the RLG model using
DDES. Three high-SPL regions are marked as Region T, A and B.

Regions A and B are the sharp corner regions of the central

longitudinal strut and downstream horizontal strut. Very strong
interactions between vortices and the downstream sharp corners lead
to these two high-SPL regions where the maximum of SPL is over
150 dB.
Region T, which is mentioned several times before, can be
understood as the consequence of vortex, detached from the fore

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

124

XIAO ET AL.

wheel, periodically impinging on the rear wheel. Nine snapshots of

the instantaneous vorticities at zD  0.4 are presented in Fig. 20a.
Here, the flow is from the right to the left. It is seen that a vortex
(Vortex i, marked by a dashed rectangle) near the groundside
detaches from the fore wheel and has a significant influence on
Region T. It grows up (1), develops (2), evolves and splits (34), and
combines with another vortex near the fore wheel (5). It develops
again (6), detaches from the fore wheel (7), approaches to the aft
wheel (8), and eventually impinges on the windward aft wheel
surface (9). The strong periodic impingement of the vortex leads to
the very high-SPL of Region T. In fact, another vortex (Vortex ii),
detached from the fore wheel near the minimum of y-coordinates, just
passes by the aft wheel without impingement on the rear wheel.
From the front view of the aft wheel, we can find another high-SPL
Region U on the wingside. As presented in Fig. 20b, Region U is
mainly caused by the impingement of the Vortex iii that directly
detaches from the fore wheel on the wingside. This vortex movement
is marked in a dashed-dot square.
From the rear view of the fore wheel, high-SPL Regions B can be
observed, shown in Fig. 20c. It is found that a vortex (Vortex iv)
detaches from the fore wheel, splitting into two vortices. One vortex
goes downstream while the other impinges on the fore wheel itself.
Then, Region B is mainly caused by the adverse impingement and
unsteady separation. On the wingside, Vortex iii, detached from the
fore wheel, encounters shear-layer instability. It breaks down and
impinges onto the aft wheel surface. At the same time, some smallscale structures go reversely to the leeward surface of fore wheel
(marked as the dashed array). The relatively high-SPL regions C and
R on the leeward surface of the fore wheel are mainly caused by the
interactions or impingements between the unsteady adverse vortices
and the fore wheel surface.

IV.

Conclusions

Delayed detached-eddy-simulation (DDES) and improved

delayed detached-eddy-simulation (IDDES) based on the k-shear-stress-transport (SST) model with the adaptive dissipation
symmetric total variation diminishing scheme are applied, in the
present work, to simulate the unsteady and massively separated flows
around rudimentary landing gear (RLG) model.
To accurately simulate this flow three sets of grids with 3.6 million,
11 million, and locally refined 13 million grid points are used to
explore the effects of grid density by SST-DDES. The computational
grid density has a weak effect on the force and mean pressure
coefficients on the wheels, but has a more important influence on the
instantaneous flow structures and pressure fluctuations. To capture
and resolve the small turbulence structures, the grid scales should be
locally refined and enlarged in the wake smoothly. The reason for the
significant difference of lift coefficients is not clear.
Based on the locally refined grids in the wake, the computational
results by DDES and IDDES, such as mean pressure coefficients,
pressure fluctuation, surface flow patterns, and instantaneous flow
structures are investigated and compared with measurements as
detailed as possible.
Generally, both DDES and IDDES with the adaptive dissipation
scheme can present reasonable agreements with the experiments,
such as the horseshoe vortex on the inboard surface of both fore and
aft wheels, secondary separation on the rearward surface of both
wheels, and focus of separation on the rearward surface of the both
wheels.
1) Both DDES and IDDES underpredict the pressure fluctuation
on the windward and outboard surface where the flow is attached as
both of them act in Reynolds averaged Navier-Stokes (RANS) mode.
2) IDDES performs on Cp;rms a little better than DDES on the aft
wheels due to its wall-modeled large-eddy-simulation (WMLES)
mode in large-eddy-simulation.
3) DDES often presents more complex surface flow patterns than
those by IDDES. IDDES presents more reasonable secondary
separation and range of horseshoe vortex than those by DDES on the
rearward and inboard surface of both fore and aft wheels due to its
WMLES mode near the wall.

4) IDDES can resolve smaller scale structures than those by

DDES.
The cause of the high sound pressure level (SPL) regions on the
surface of the overall RLG model is highlighted in this work. The
interactions between the vortices and the sharp corner of the
longitudinal strut (Region B) and the downstream horizontal strut
(Region A) lead to high SPL in these two regions. The strongly
periodic impingement of the detached vortex on the back surface of
the fore wheel also causes high SPL though the vortex impingement,
as shown in Region R and C, is actually in the adverse direction.

Acknowledgments
This work was partly supported by European Union (EU) project
Advanced Turbulence Simulation for Aerodynamic Application
Challenges (ATAAC, Contract No. ACP8-GA-2009-233710)
and partly by National Science Foundation of China (Grant
Nos. 10932005 and 11072129). The authors thank ANSYS for the
mandatory grid in project ATAAC. The authors also thank Shanghai
Supercomputer Center and Tsinghua National Laboratory for
Information Science and Technology for the computational
resources.

References
[1] Lazos, B. S., Mean Flow Features Around the Inline Wheels of FourWheel Landing Gear, AIAA Journal, Vol. 40, No. 2, 2002, pp. 193198.
doi:10.2514/2.1642
[2] Dobrzynski, W., Almost 40 Years of Airframe Noise Research:
What Did We Achieve?, Journal of Aircraft, Vol. 47, No. 2, 2010,
pp. 353367.
doi:10.2514/1.44457
[3] Guo, Y. P., Yamamoto, K. J., and Stoker, R. W., Experimental Study on
Aircraft Landing Gear Noise, Journal of Aircraft, Vol. 43, No. 2, 2006,
pp. 306317.
doi:10.2514/1.11085
[4] Remillieux, M. C., Camargo, H. E., Ravetta, P. A., Burdisso, R. A., and
Ng, W. F., Novel Kevlar-Walled Wind Tunnel for Aeroacoustic Testing
of a Landing Gear, AIAA Journal, Vol. 46, No. 7, 2008, pp. 16311639.
doi:10.2514/1.33082
[5] Li, Y., Smith, M. G., and Zhang, X., Identification and Attenuation of a
Tonal-Noise Source on an Aircrafts Landing Gear, Journal of Aircraft,
Vol. 47, No. 3, 2010, pp. 796804.
doi:10.2514/1.43183
[6] Yokokawa, Y., Imamura, T., Ura, H., Kobayashi, H., Uchida, H., and
Yamamoto, K., Experimental Study on Noise Generation of a TwoWheel Main Landing Gear, AIAA Paper 2010-3973, 2010.
[7] Souliez, F. J., Long, L. N., Morris, P. J., and Sharma, A., Landing Gear
Aerodynamic Noise Prediction Using Unstructured Grids, International Journal of Aeroacoustics, Vol. 1, No. 2, 2002, pp. 115135.
doi:10.1260/147547202760236932
[8] Lockard, D. P., Khorrami, M. R., and Li, F., High Resolution
Calculation of a Simplified Landing Gear, AIAA Paper 2004-2887,
2004.
[9] Drage, P., Wiesler, B., van Beek, P., van Lier, L., Parchen, R., and Tibaut,
P., Prediction of Noise Radiation from Basic Configurations of
Landing Gears by Means of Computational Aeroacoustics, Aerospace
Science and Technology, Vol. 11, No. 6, 2007, pp. 451458.
doi:10.1016/j.ast.2007.03.002
[10] Van Mierlo, K. J., Takeday, K., and Peersz, E., Computational Analysis
of the Effect of Bogie Inclination Angle on Landing Gear Noise, AIAA
Paper 2010-3971, 2010.
[11] Noelting, S., Brs, G. A., Dethioux, P., Van de Ven, T., and Vieito, R., A
Hybrid Lattice-Boltzmann/FW-H Method to Predict Sources and
Propagation of Landing Gear Noise, AIAA Paper 2010-3976, 2010.
[12] Guo, Y. P., A Statistical Model for Landing Gear Noise Prediction,
Journal of Sound and Vibration, Vol. 282, Nos. 12, 2005, pp. 6187.
doi:10.1016/j.jsv.2004.02.021
[13] Lopes, L. V., Brentner, K. S., and Morris, P. J., Framework for a
Landing-Gear Model and Acoustic Prediction, Journal of Aircraft,
Vol. 47, No. 3, 2010, pp. 763774.
doi:10.2514/1.36925
[14] Guo, Y. P., Effects of Local Flow Variations on Landing Gear Noise
Prediction and Analysis, Journal of Aircraft, Vol. 47, No. 2, 2010,
pp. 383391.
doi:10.2514/1.43615

Downloaded by QUEEN MARY UNIVERSITY OF LONDON on November 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J051598

XIAO ET AL.

[15] Lopes, L. V., Prediction of Landing Gear Noise Reduction and

Comparison to Measurements, AIAA Paper 2010-3970, 2010.
[16] Piet, J. F., Davy, R., Elias, G., Siller, H., Chow, L. C., Seror, C., and
Laporte, F., Flight Test Investigation of Add-On Treatments to Reduce
Aircraft Airframe Noise, AIAA Paper 2005-3007, 2005.
[17] Dobrzynski, W. M., Schning, B., Chow, L. C., Wood, C., Smith, M.,
and Seror, C., Design and Testing of Low Noise Landing Gears,
International Journal of Aeroacoustics, Vol. 5, No. 3, 2006,
pp. 233262.
doi:10.1260/1475-472X.5.3.233
[18] Quayle, A. R., Dowling, A. P., Babinsky, H., Graham, W. R., Shin, H.C., and Sijtsma, P., Landing Gear for a Silent Aircraft, AIAA Paper
2007-231, 2007.
[19] Oerlemans, S., Sandu, C., Molin, N., and Piet, J.-F., Reduction of
Landing Gear Noise Using Meshes, AIAA Paper 2010-3972, 2010.
[20] Smith, M. G., Chow, L. C., and Molin, N., Control of Landing Gear
Noise Using Meshes, Proceeding of the 16th AIAA/CEAS
Aeroacoustics Conference, AIAA Paper 2010-3974, 2010.
[21] Lazos, B. S., Surface Topology on the Wheels of a Generic Four-Wheel
Landing Gear, AIAA Journal, Vol. 40, No. 12, 2002, pp. 24022411.
doi:10.2514/2.1608
[22] Lazos, B. S., Reynolds Stresses Around the Wheels of Simplified FourWheel Landing Gear, AIAA Journal, Vol. 42, No. 1, 2004, pp. 196198.
doi:10.2514/1.597
[23] Manoha, E., Bult, J., Ciobaca, V., and Caruelle, B., LAGOON:
Further Analysis of Aerodynamic Experiments and Early Aeroacoustics
Results, AIAA Paper 2009-3277, 2009.
[24] Hedges, L. S., Travin, A. K., and Spalart, P. R., Detached-Eddy
Simulations Over Simplified Landing Gear, Journal of Fluids
Engineering, Vol. 124, 2002, pp. 413423.
doi:10.1115/1.1471532
[25] Spalart, P. R., and Allmaras, S. R., A One-Equation Turbulence Model
for Aerodynamic Flows, AIAA Paper 92-0439, Jan. 1992.
[26] Spalart, P. R., Jou, W. H., Strelets, M., and Allmaras, S. R., Comments
on the Feasibility of LES for Wings, and on a Hybrid RANS/LES
Approach, First AFOSR International Coference on DNS/LES,
Advanced in DNS/LES, edited by Liu, C., and Liu, Z., Greyden Press,
Ruston, LA, Columbus, OH, Aug. 1997.
[27] DESider Project, http://cfd.mace.manchester.ac.uk/desider/index2
.htlm.
[28] Haase, W., Braza, M., and Revell, A., DESiderA European Effort on
Hybrid RANS-LES Modelling, Notes on Numerical Fluid Mechanics
and Multidisciplanry Design, Vol. 103, SpringerVerlag, 2009.
[29] Spalart, P., Shur, M., Strelets, M., and Travin, A., Initial RANS and
DDES of a Rudimentary Landing Gear, Notes on Numerical Fluid
Mechanics and Multidisciplinary Design, Vol. 111, Progress in Hybrid
RANS-LES Modelling, Springer-Verlag, Berlin, Heidelberg, 2010,
pp. 101110.
[30] Billard, F., ATAAC Project, http://cfd.mace.manchester.ac.uk/
ATAAC/WebHome [retrieved 27 Sept. 2012].
[31] Choudhari, M., Workshop on Benchmark problems for Airframe Noise
Computations-I, https://info.aiaa.org/tac/ASG/FDTC/DG/BECAN_
files_/Workshop_June_2010_Final_problem_Statements/Contact_
Information.pdf [retrieved 14 Oct. 2009].
[32] Xiao, Z. X., Liu, J., and Fu, S., Calculations of Massive Separation
around Landing-Gear-like Geometries, Journal of Hydrodynamics,
Ser. B, Vol. 22, No. 5, 2010, pp. 926931.
doi:10.1016/S1001-6058(10)60054-6
[33] Krajnovic, S., Larusson, R., Helgason, E., and Basara, B., PANS of
Rudimentary Landing Gear, AIAA Paper 2011-3109, 2011.
[34] Spalart, P., Shur, M., Strelets, M., and Travin, A., Towards Noise
Prediction for Rudimentary Landing Gear, IUTAM Symposium on
Computational Aero-Acoustics for Aircraft Noise Prediction, Vol. 6,
Procedia Engineering, Elsevier Science BV, Amsterdam, Netherlands,
2010, pp. 283292.

125

[35] Wang, L., Mockett, C., Knacke, T., and Thiele, F., Noise prediction of a
rudimentary landing gear using Detached-Eddy Simulation, Notes on
Numerical Fluid Mechanics and Multidisciplinary Design, Vol. 117,
Progress in Hybrid RANS-LES Modelling, Springer-Verlag, Berlin,
Heidelberg, 2012, pp. 279289.
[36] Karthikeyan, N., and Venkatakrishnan, L., Application of Photogrammetry to Surface Flow Visualization, Experiments in Fluids,
Vol. 50, No. 3, 2011, pp. 689700.
doi:10.1007/s00348-010-0978-x
[37] Venkatakrishnan, L., Karthikeyan, N., and Mejia, K. M., Experiment
Studies on Rudimentary Four Wheel Landing Gear, AIAA Paper 2011354, 2011.
[38] Spalart, P. R., and Mejia, K. M., Analysis of Experimental and
Numerical Studies of the Rudimentary Landing Gear, AIAA Paper
2011-355, 2011.
[39] Menter, F. R., and Kuntz, M., Adaptation of EddyViscosity Turbulence
Models to Unsteady Separated Flow Behind Vehicles, The
Aerodynamics of Heavy Vehicles: Trucks, Buses and Trains,
Springer-Verlag, Berlin, Heidelberg, 2004, pp. 339352.
[40] Fu, S., Xiao, Z. X., Chen, H. X., Zhang, Y. F., and Huang, J. B.,
Simulation of Wing-Body Junction Flows with Hybrid RANS/LES
Methods, International Journal of Heat and Fluid Flow, Vol. 28, No. 6,
2007, pp. 13791390.
doi:10.1016/j.ijheatfluidflow.2007.05.007
[41] Spalart, P. R., Deck, S., Shur, M., Squares, K. D., Strelets, M., and
Travin, A., A New Version of Detached-Eddy Simulation, Resistant to
Ambiguous Grid Densities, Theoretical and Computational Fluid
Dynamics, Vol. 20, No. 3, 2006, pp. 181195.
doi:10.1007/s00162-006-0015-0
[42] Shur, M. L., Spalart, P. R., Strelets, M., and Travin, A., A Hybrid
RANS-LES Approach with Delayed-DES and Wall-Modelled LES
Capabilities, International Journal of Heat and Fluid Flow, Vol. 29,
No. 6, 2008, pp. 16381649.
doi:10.1016/j.ijheatfluidflow.2008.07.001
[43] Menter, F. R., Two-Equation Eddy-Viscosity Turbulence Models for
Engineering Applications, AIAA Journal, Vol. 32, No. 8, 1994,
pp. 15981605.
doi:10.2514/3.12149
[44] Strelets, M., Detached Eddy Simulation of Massively Separated
Flows, AIAA Paper 2001-0879, 2001.
[45] Xiao, Z. X., Liu, J., Huang, J. B., and Fu, S., Comparisons of Three
Improved DES Methods on Unsteady Flows Past Tandem Cylinders,
Notes on Numerical Fluid Mechanics and Multidisciplinary Design,
Vol. 117, Progress in Hybrid RANS-LES Modelling, Springer-Verlag,
Berlin, Heidelberg, 2012, pp. 231243.
[46] Xiao, Z. X., Liu, J., Huang, J. B., and Fu, S., Numerical Dissipation
Effect on the Massive Separation around Tandem Cylinders, AIAA
Journal, Vol. 55, No. 5, 2012, pp. 11191136.
doi:10.2514/1.J051299
[47] Mockett, C., A Comprehensive Study of Detached-Eddy Simulation,
Doctoral Thesis, Technical Univ. Berlin, 2009.
[48] Usta, E., Application of a Symmetric Total Variation Diminishing
Scheme to Aerodynamic of Rotors, Ph.D. Dissertation, Georgia Inst. of
Technology, School of Aerospace Engineering, Atlanta, GA, 2002.
[49] Tobak, M., and Peake, D. J., Topology of Three-Dimensional
Separated Flows, Annual Review of Fluid Mechanics, Vol. 14,
Jan. 1982, pp. 6185.
doi:10.1146/annurev.fl.14.010182.000425
[50] Chapman, G. T., and Yates, L. A., Topology of Flow Separation on
Three-Dimensional Bodies, Applied Mechanics Review, Vol. 44, No. 7,
1991, pp. 329345.
doi:10.1115/1.3119507

P. Tucker
Associate Editor