Report of Chimera method in NSMB

Anthony Ponce, Thibaut Deloze, Yannick Hoarau, Jan Dusek

Laboratoire des sciences de lIngnieur, de lInformatique et de lImagerie,
Universite de Strasbourg, 2 rue Boussingault, 67000 Strasbourg, France
September 27, 2013
1 Introduction
One of major key in CFD is the spatial discretization. The discretization is
not problematic when the settings are simple and static. It becomes when the
conguration contains complex geometry and dynamic body, especially for the
structured grids. In this case, the grid generation is the limiting factor for the
simulations. To overcome this problem, some solutions are proposed like the Im-
mersed Boundary Method (IBM), Volume Of Fluid (VOF) or Chimera method.
The last one, chimera method (or called overset grid method) appears increas-
ingly for complex research applications and more and more for huge industrial
applications.
The Chimera method was initiated by Benek et al. [1] and the method is
subsequently been improved and it is adapted to many cases ([2, 3, 4, 5, 6, 7, 8, 9,
10, 11, 12, 13]). Its principle is to decompose the complex full domain into simple
sub-domains independently meshed by curvilinear grids. The only constraint
is a superposition of grids to allow the link of the ow description. Beyond
simplifying the mesh generation, this technique oers a powerful solution to
manage moving bodies.
The chimera processus is detailed for structured grid by Meakin [11]. The
process of chimera method is divided into four main steps like described Land-
mann [14] (g. 1):
1
st
step : Detection of overlapped cells. This process consists in nding
which cells are overlapped by other cells. The test is based only on the
coordinates of the cells.
2
nd
step : Determination of the overlapped cell status. Three kinds of
overlapped cells exist [14] :
a) Interpolated cells : the ow data are obtained by interpolation on
other overlapping grids.
b) Calculated cells : the values come from the solution of the equations
governing the ow.
c) Hole cells (or blanked/masked cells) : the values of this kind of cells
are never used in the discretisation schemes or overlapped boundary
conditions and they are not interpolated nor calculated.
1
3
rd
step : Calculation of the interpolations parameters. In this step, the
donor cells and the associated weights that compose the interpolation are
evaluated for each valid overlapped cell.
4
th
step : Calculation of the interpolation values.
grid 3 (flat)
grid 1 (airfoil)
grid 2 (flap)
a)
b)
A B C
W
NS
n
W
HC
n
W
NS
n
W
HC
n
interpolation
grid 2
(flap)
grid 1
(airfoil)
d)
D
grid 2
(flap)
grid 1
(airfoil)
overlapped cells
c)
A B C D E
e)
g
(a
Figure 1: Main steps of the chimera method.
In the present work, a three-dimensional structured multi-block chimera
method is implemented in order to simulate the free falling sphere falling in
a tube (Fig? 2). A basic chimera method is used and the main works focus on
the increasing of the speed process with a study on each step of chimera algo-
rithm. This work allowed to nd out a new criteria for a quickly and eciently
hole cutting process. The developped chimera method is fully automatic with
few, if not none, of user input and pre-processing.
Y
cylindri
spherical grid
XY-plane
YZ-plane
Figure 2: Representation of the cylindrical grid that discretizes the pipe space
and the spherical grid representing the sphere.
2
The paper is organized as follows: the section 2, the details of each step of
the implemented chimera method are discussed. In this section is described the
study of the bucket method for detection of overlapped cells, the denition of a
new criteria for hole cutting algorithm and the justication of the interpolation
scheme. In Section 3 we report the results of test cases to validate the accuracy
and the eciency. We nd here the results of the ow around a sphere, then
the ow around the sphere close to a plane wall and, to nish, the results of free
falling sphere in a tube.
2 The ow solver
We implemented the chimera method in the ow solver called Navier-Stokes
Multi-Block (NSMB) [15, 16]. The NSMB code solves the compressible Navier-
Stokes equations using a nite volume formulation on Multi-Block structured
grids and parallelized using the Message Passing Interface.
Among the many discretization schemes present in NSMB, in this study we
use a central scheme with 4th order articial dissipation [17] for spatial discreti-
sation and an implicit dual time stepping solved by Lower-Upper Symmetric
Gauss-Seidel method (LS-SGS) for time integration. The articial compress-
ibility method [18] is employed to full the incompressible ow.
The governing equations are the preconditioned unsteady Navier-Stokes equa-
tion written in conservative form with dual-time stepping. Using the Cartesian
coordinates (x,y,z), these equations can be expressed in a conservative form as
follows :
P
1

(W) +

I

t
(W) +

x
(f f
v
) +

y
(g g
v
) +

z
(h h
v
) = 0 (1)
The state vector W and the the inviscid uxes f, g and h are given by :
W = (p, u, v, w)
T
f =
_
u, u
2
+p, uv, uw
_
T
g =
_
v, vu, v
2
+p, vw
_
T
h =
_
w, wu, wv, w
2
+p
_
T
where is the density, u v and w are the Cartesian components of velocity, p
is the pressure, dt is the physical time step and d the pseudo-time step. The
viscous uxes f
v
, g
v
and h
v
are given by :
f
v
= (0,
xx
,
xy
,
xz
)
T
g
v
= (0,
yx
,
yy
,
yz
)
T
h
v
= (0,
zx
,
zy
,
zz
)
T
with the shear stress tensor given by
3

xx
=
2
3

_
2
u
x

v
y

w
z
_

xy
=
yx
=
_
v
x
+
u
y
_

yy
=
2
3

u
x
+ 2
v
y

w
z
_

xz
=
zx
=
_
w
x
+
u
z
_

zz
=
2
3

u
x

v
y
+ 2
w
z
_

yz
=
zy
=
_
v
z
+
w
y
_
where is the dynamic viscosity (Stokes hypothesis).
The modied unit matrix

I and the pre-conditioning matrix Pfor the pseudo-

time equation system are given by

I =
_

_
0 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_

_
and P =
_

2
0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_

_
The coecient can be seen as a relaxation parameter for the pseudo-time
solution and is dened as

2
= max(
2
min
U
2

, C

U
2
)
with
min
= 5 and C

= 0.05.
3 The chimera method
Details and improvements to meet our constraints (free fall body in a conned
space) are described following the same steps of the description (detection of
overlapped cells, determination of the type of overlapped cells, ...)
3.1 Detection of overlapped cells
The search of overlapped cells is based on a test of inclusion on the coordinates.
In order to accelerate the search, the mapping of coordinates in a virtual uniform
Cartesian grid is performed. This method is well known as the bucket method
[19, 20, 21]. An inverse mapping algorithm creates an index array that link
virtual grid to the real coordinates. The search begins on the virtual grid and
it continues on the physical associated grid points. The creation of link array is
time consuming but the gain in the search time is signicant (at least a factor
20).
This algorithm is well known but the number of the virtual cells (N
v
) is a
parameter not clearly dened. A simple test is performed in order to determi-
nate the optimum value of N
v
in function of the number of real coordinates N.
We consider two uniform three-dimensional Cartesian grids (N
x
= N
y
= N
z
=
N = 128) overlapping entire the cube 1x1x1 domain. Identical results were
obtained on a coarser and a ner grid. The cpu-time is determined for range
1 N
v
256. In addition to the compromise between cpu-time decrease for
virtual grid creation and cpu-time increase for searching the overset, the size
required by the link array must be taken into account. The results, represented
4
on gure 3, show that the optimum value for the virtual cell number N
v
is the
half of the number of physical coordinates N.The same test is performed for a
grid with tangent hyperbolic distribution and, in this case, the optimum value
correspond to N.
N
c
p
u
-
t
i
m
e
0 50 100 150 200 250
10
0
10
1
10
2
10
10
10
cpu-time of creation of link array
cpu-time for search
sum of cpu-time (creation+search)
array size of virtual grid link
v
64=N/2 128=N
Figure 3: Representation of the cpu-time used for the indexation of real coor-
dinates in virtual grid, cpu-time for searching overset and the size of the index
between both grids versus the number of virtual box N
v
in one direction.
Another issue concerns the points that are inside a solid body. In this case,
the cells are not detected as overlapped cells. To remove these cells, two meth-
ods are implemented. For simple cases (cylinder, sphere) we use an analytical
function and for general cases. We compute the dot product of the vector from
the nearest wall cell center to cell center and the vector of the associated wall
normal vector. If the dot product is negative, the cell is in the solid region,
otherwise the cell is out of the solid region. This second method is 30 times
more expensive than the rst one, for the conguration of a sphere in a tube,
with 6 millions of cells and 65 000 wall cells.
3.2 Determination of the type of overlapped cells
The second step of the method is to dene the type of the overlapped cells
(calculated, interpolated, hole). In this step, among several overlapped cell,
the best candidate to predict the ow state (calculated cell) and to give the
ow solution to the other cells (interpolated cells) is chosen. The criterion
that justies the choice is based on the best resolution of the physical problem.
We distinguish four techniques based on the criteria associated to the follow
quantities : user-dened grid ranking, user-dened cell quality, cell size, cell
size in the wall-normal direction.
For simple overset, the denition on an overlapped hierarchy of each block
is enough for the cell type selection. All the cells of the highest ranking of
the grid are dened as calculated cells. This solution is simple and fast but
not suitable for complex overlapped layers where the denition of chimera grid
ranking by the user is then needed. Based on the fact that the cell size is
a criterion indicating the best discretization, Siikonen et al. [21] or Liao et al.
[22] justify this choice by assigning the calculated cells type to the smallest ones.
5
This criterion is locally adapted, automatic and requires no user input. A third
criterion is added to the previous one and based on a user-dened cell quality.
This way the user can force a specic kind of cell (calculated or interpolated) by
associating to this cell a high quality value. This technique can force cell type by
protecting or immunizing cells but the user input is binding and does not follow
the grid movement. The cell size in the wall-normal direction was introduced
by Landmann [14]. The selection use the intersection segment between normal
nearest wall and cell boundary. This technique is fully automatic and it is based
on an important physical quality : the boundary layer. This test requires to
search for the nearest wall and to calculate the intersection for each cell. In
order to nd a lighter, accurate and automatic criterion, we developed a similar
criterion based on the distance of a cell to the nearest local wall. The nearest
local wall is the wall in the same block where the cell is dened (gure 4). With
this criterion, the overlapped cell with the smallest local near wall distance
is calculated and the others are interpolated. This approach guarantees the
resolution of Navier-Stokes equations in the nearest wall region and consequently
each boundary layer is accurately computed. Moreover, this technique is based
on the local wall distance and this quantity does not change with moving grid
(an update is needed for deformed meshes) whereas the criterion based on the
wall normal intersection need an update after each grid movement.
Mesh 1 Mesh 2
C
1
Global nearest wall distance Local nearest wall distance
Figure 4: Denition of local nearest wall distance and global nearest wall dis-
tance for the cell C
1
of the mesh 1 in the conguration of two overlapped polar
meshes.
In the case of the sphere in a circular tube, the criterion based on the cell size
complicates the generation of the mesh. The same problem may be encountered
in the conguration of a sphere near a plane wall. The cell aspect ratio of
the near wall mesh is not the same as that of the spherical wall mesh and the
cell volume is very dicult to control. In our case of a sphere in a tube with a
cylindrical grid overlapped by a spherical grid, the criterion based on the volume
does not give a proper solution. The new criterion based on the local near wall
distance give the best answer. The gure 5 represents the visualization of the
calculated cells in the 3 planes crossing the sphere center for the both cases.
6
Figure 5: Visualization of the calculated cells for the chimera cell selection
before cell selection (1
st
row), using cell volume criterion (2
nd
row) and using
the nearest local wall distance criterion (3
rd
row) for the cross section of the
tube (1
st
line), the streamwise plane in the symmetric plane (2
nd
line) and
streamwise plane perpendicular to the symmetric plane (3
rd
line)
3.3 Buer and hole cells
The next step is the addition of a buer layer. The buer cells are cell between
interpolated and calculated cells. Their role is to move away the interpolation
communications and thereby to avoid implicit interpolation. An implicit inter-
polation is an interpolation that uses an interpolated cell as a donor cell. The
more the interpolation needs a large number of donor cells, the more the buer
layer needs to be wide. The importance of wide overset is admitted and de-
scribed in several papers [23, 24]. The set-up of the buer layer is implemented
with an user-input. The user give a xed number of layer and the algorithm
changes the type of the cell from interpolated to calculated over this layer.
The last type of possible cells is the hole cell. This kind of cell is a non-
necessary cell and it concerns interpolated cells. This cell does not step in any
discretization schemes. In order to save computation time, the interpolation
process is avoided for this cell. After the buer layers, a user-input imposes a
number of interpolated cells in the hole region. These new cells are considered
as ghost-cells. The minimum number of ghost-cells is two because the numerical
schemes used in NSMB require two neighbor cells (we have used fourth order
central scheme with fourth order articial dissipation).
7
3.4 The interpolation
The interpolation is a key element of the chimera method. It allows the pas-
sage of ow information between overlapped grids and it creates the link. The
inuence of interpolation has already been investigated in several studies (Delfs
[25], Sherer & Scott [26] and Sengupta et al. [27]). The simplest treatment of
an interface consists in the interpolation of all state variables [1, 28, 6, 29, 30].
However, this approach does not guarantee the global mass conservation as ex-
plained Wang et al. [31]. Some studies found a solution to ensure the global
conservation for compressible ows [32] or for incompressible ows (Wright and
Shyy [33] or Tang et al. [34]) with a correction of interpolated values. The ac-
curacy of interpolation is important and dierent schemes of interpolation are
used : interpolation based on Lagrange polynomials [35, 36, 24, 27, 26], trilin-
ear interpolation [37, 34, 38, 39], quadratic interpolation [40] or tetravolumic
interpolation [41, 21].
The interpolation scheme depends on the physical problem and the overset
width. A highly-accurate interpolation needs many donor cells and it requires
a wide overset. In our case of the sphere in a circular pipe, the gap between the
sphere wall and tube wall does not provide a wide overset and a large stencil of
donor cells. For this reason, we chose some interpolation that use a low number
of donor cells. The choice focus itself on the trilinear, tetravolumic and weighted
inverse distance interpolations.
The inverse distance weighted interpolation is based on the distance between
the points. The implementation is simple and its advantage is the exibility of
the stencil of points. The formula used is given by the following relation :
f
M
=

N
i=1
f
i
d

N
i=1
1
d

i
(2)
where N is the number of the stencil points, f
i
the value at the point i, is
the weighting exponent (in our case equal to 2) and d
i
is the distance from the
stencil point i to the interpolation point M.
The second interpolation is the tetravolumic interpolation based on the tetra-
hedral volumes composed between the four nearest cells. The interpolation
weights are given by the opposite tetrahedral volume :
f
M
=

i=A,B,C
W
i
f
i
with W
A
=
V
BCM
V
ABC
, W
B
=
V
ACM
V
ABC
and W
C
=
V
ABM
V
ABC
with W the weight of the interpolation and V
ACM
the volume of the triangle
(ABC). In three dimensions, the triangle volume is replaced by the tetrahedral
volume.
The third interpolation is the trilinear one. We use a trilinear interpolation
based on a linear interpolation on each direction. In two-dimensional example
of the gure ??, the rst linear interpolation yields the values f
Q
and f
P
then
a second linear interpolation provides f
M
. In three-dimensions, three steps of
linear interpolation are required.
8
3.5 Grid-to-grid interpolation test case
The rst results concern the interpolation of data generated by analytic func-
tions from one grid onto another. These tests allow us to isolate the interpo-
lation properties from other aspects of the solver. The analysis validates the
interpolation and quanties the accuracy of each interpolation scheme.
The grids are composed of two uniform square Cartesian grids, the back-
ground grid and the foreground one which is rotated by 45(Fig.6 (a)). An
analytical function is superimposed on the background grid and then interpo-
lated onto the foreground one. The dierence between the interpolated values
and the analytical solution on the foreground mesh gives the interpolation error
on each cell. Five meshes with dierent renement are considered. The space
discretization (X) is characterized by the number of points (N
x
) and ranges
from N
x
= 34 for the coarse grid to N
x
= 514 for the ne grid.
Three two-dimensional analytic functions are considered :
F
1
(x, y) = cos
_
x
3
_
sin
_
y
3
_
(3)
F
2
(x, y) = 5
2
xexp
_

_
x
2
+y
2
2
_
(4)
F
3
(x, y) =
28x
x
2
+ 9(y
2
+ 9)
cos
_
20x
z + 3
_
cos
_
x
3
_
sin
_
z
3
_
(5)
The L
2
-norm error is calculated using the following equation :
E
2
=
__

|E|
2
dxdy
_
1/2
(6)
with the error E = F
analytic
F
interpolated
.
The gure 6 represents the L
2
-norm error for the function F
3
versus the
spatial discretization step X in logarithmic scale. The behavior of all functions
is similar and the order-of-accuracy of each interpolation is calculated. For the
tetravolumic interpolation the orders are 2.00, 2.04 and 2.04 for the function
F
1
, F
2
and F
3
, for the inverse distance weighted 1.33, 1.25 and 1.31 and for
trilinear interpolation 2.03, 2.06 and 2.06. The order-of-accuracy of the trilinear
interpolation (equal to 2.04) is similar to that of the tetravolumic interpolation
(equal to 2.00). However, the value of L
2
-norm error for trilinear interpolation
is smaller than that of the tetravolumic interpolation. The inverse distance
weighted interpolation is less accurate with an order equal to 1.31 at the best.
The advantage of the inverse distance weighted interpolation is the simple
and exible algorithm but the accuracy is poor. The accuracy of the tetravolu-
mic interpolation is better but the stencil of points is more rigid. The trilinear
interpolation is a good compromise with a good accuracy and a exible stencil
of points with the possibility of extrapolation but its implementation is heavy.
The trilinear interpolation is used for the following simulations.
One test in this way has been performed. The test is the same kind of grid-to-
grid interpolation test presented in the paper, section 4.1. The interpolated grid
is a cartesian grid with a xed distribution, 128x128 cells. The cell size of the
donor cells varies. The L
2
-norm error of the function F
3
(5) is represented versus
9
x
y
0 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Block 1
Block 2
(a)
x
L
2
-
n
o
r
m
e
r
r
o
r
0.01 0.02 0.03 0.040.050.06 0.07
10
-5
10
-4
10
-3
10
-2
tetravolumic
inverse distance weighted
trilinear

(b)
Figure 6: Coarse meshes used for the grid-to-grid interpolation test case (a) and
L
2
-norm error versus the x for the function F
3
(Eq. 5) (b).
the ratio aspect of the interpolated cell over donor cell (gure 7). The results
show that the cell aspect ration have an small inuence on interpolated values.
The main inuence of the big aspect ratio is on the diusion in the Navier-
Stokes equations. When the ow crosses the overset border with a dramatic
changement of cell size, the ow prediction is aected. We didnt had this result
to the paper. To complete our answer, we can had that the chimera connection
can be compared to a block-block connection (except for the interpolation) :
if the cell aspect ration is too big this will induce numercial dissipation and
we have veried this with turbulent chimera simulations. So one of the ajor
rule when building chimera grids is that at the interface we have to respect an
equivalent aspect ratio.
/
L
2
-
n
o
r
m
e
r
r
o
r
1 2 3
0
0.0002
0.0004
0.0006
0.0008
0.001
trilinear
inverse distance weighted
tetravolumic
x x 1 2
x / x
|
|

E

|
|
2
,

/


|
|

E

|
|
2
,

=
1
1 2 3
1.00
1.05
1.10
1.15
1.20
ttravolumique
inverse la distan
trilinaire
1 2
=
Figure 7: L
2
-norm error of the function F
3
(5) versus the ratio aspect of the
interpolated cell over donor cell.
10
4 Validation of the method
The method described in the previous paragraph will now be validated. Three
test cases are : the simulation of ow around a sphere in an innty domain, a
ow around a sphere translating uniformly along a at wall, and nally a sphere
falling freely in a tube lled with a Newtonian uid at rest. Each test case
complexity rises, so one part of a single ow, then an interaction between two
walls dened by two independent chimera meshes, then the case of interaction
uid / solid of a free fall body represented by on overset grid. These test cases
are also selected for the presence of numerical or/and experimental results that
making the validation deeper.
4.1 Flow past a sphere
The ow around a sphere is the topic of large experimental studies or numerical
studies [42, 43, 44, 45, 46, 47, 48]. The ow state is well known and clearly
dened by a single parameter, the Reynolds number (Re = Ud/, with U the
freestream velocity, d the diameter of the sphere and the kinetic viscosity).
For low Reynolds number (Re < 212) the ow is steady and axisymmetric, then
after Re > 212 the axisymmetry is broken into a steady non-axisymmetric ow
and the Hopf bifurcation appears at Re = 273 where the ow beacom unsteady
with horse-shoe vortex shedding.
This case test is the rst step of validation and it can demonstrate the
ability and the accuracy of the chimera method in a simple three dimensional
computation hugly referenced. The simulations are performed by two set of
grids : an chiemra grid composed by a cartesian background grid overlapped by
a spherical grid and a classical body-tted mesh.
The representation of the isolines of streamwise velocity in the streamwise
plane (gure 8) shows the wake behind the sphere with a recirculation. The
shape of the vortex is a torus like described in the literature (Johnson et al. [46]
for example). The dashed line on the gure 8 represents the external border of
the spherical grid. The ow information crosses correctly this overset border.
0.2 0.9
0
0.1
0
.2
0
.3
0.9
0
0.2
0.7
0.7
0.8
0.4 0.5
0.6
0.7
0.9
0.9
0
0.1
0.2
0.4
0.6
0.7
0.8 0.9
(a)
(b)
Figure 8: Representation, on the streamwise plane, the isovalues of the stream-
wise velocity (a) and the streamlines (b)
The comparison of the results is based on three physical quantities : the
drag coecient (C
D
), the recirculation length (L) and the separation angle ().
For this range of Reynolds number, the ow is axisymmetric and the lift force
is zero. The comparison of the results on the drag coecient of Bouchet et al.
[45] and the present study do not exceed 0.52%. The very closer results show
11
the good resolution of overlapped grids. The shape descriptions (L and ) are
in good agreement with the previous works reported on gure 9. The results are
compared with datas obtained by experimental studies (Taneda [42], Nakamura
[49] and Roos & Willmarth [50]) and by numerical studies (Tomboulides et al.
[48], Mittal [51], Magnaudet et al. [52], Bagchi & Balachandar [53] and Bouchet
et al. [45]).
Reynolds number
D
r
a
g
c
o
e
f
f
i
c
i
e
n
t
50 100 150 200
0.8
1
1.2
1.4
1.6 Bouchet et al.
Magnaudet et al.
Roos & Wilmarth
Present study with chimera mesh
Present study with no-chimera mesh
Reynolds number
R
e
c
i
r
c
u
l
a
t
i
o
n
l
e
n
g
t
h
(
L
/
d
)
50 100 150 20
0.4
0.6
0.8
1
1.2
1.4
1.6
Bouchet et al.
Mittal
Taneda
Tomboulides et al.
Present study with no-chimera mesh
Present study with chimera mesh
Reynolds number
S
e
p
a
r
a
t
i
o
n
a
n
g
l
e
(
d
e
g
r
e
e
)
50 100 150 200
115
120
125
130
135
140
Bouchet et al.
Magnaudet et al.
Mittal
Taneda
Present study with no-chimera mesh
Present study with chimera mesh
Figure 9: Drag coecient C
D
, recirculation length (L/d) and separated angle
versus Reynolds number for the ow past a sphere and 50 Re 200.
4.2 Flow around a sphere translating along a plane wall
To complete the study of the three-dimensional chimera method, we consider a
conguration with an interaction between two walls yielding a border of over-
lapping (Fig. 10). We simulate the ow past a sphere in a uniform translation
parallel to a plane wall which has been previously reported by Zeng et al. [54]
and Takemura & Magnaudet [55]. The parameters for this conguration are
the distance between the sphere center and the wall (L/d) and the Reynolds
number Re = U.d/, with U the velocity of the sphere translation and the
kinematic viscosity of the uid. The geometry is meshed with a rst Cartesian
grid, which is rened along the plane wall, and a second spherical one, which
is rened near the sphere wall. The criterion for chimera cell selection is based
12
on the denition of a boundary layer for the background grid where the cells
are always calculated. Elsewhere the cells of spherical mesh are calculated. The
motion of the sphere is implemented with the wall condition of the plate and
with an inlet boundary condition. The mesh for parallel simulations is decom-
posed into 32 sub-blocks. The range of the Reynolds number is 10 Re 250
and the studied distance is L/d = 1 and L/d = 0.75.
The presence of the plane wall breaks the axisymmetric geometry, and at
all Reynolds numbers a lift force exists. Two mechanisms are the source of this
force (Takemura & Magnaudet [55]). The rst is linked to the strong interaction
between the wall and the wake of the sphere. The gure 11 shows the isovalues
of the streamwise velocity in the wall normal plane for the case Re = 200 and
L/d = 1.00. The distribution of the ow looses the up/top asymmetric and it
results into a lift force directed away from the wall. The second mechanism is
linked to the high velocity in the gap associated to a low pressure. This behavior
gives a force with opposite direction (attractive force) to the rst mechanism.
The resultant force tends to push the sphere away from the wall. The streamlines
plotted along the streamwise wall normal plane (gure 12) represents the vortex
structure for Re = 200 and L/d = 0.75. The same case performed by Zeng et al.
[54] is plotted sideline our results for comparison (gure 12). The overset border
is not visible and proves one more time that the chimera communication is
sucient. Zeng et al. [54] found the same representation of the ow reproduced
in gure 12.
Figure 10: The chimera mesh in the plane perpendicular of the plane wall across
the sphere center for L/D = 1.
Figure 11: Isovalues of the streamwise velocity in the plane normal to the wall
(-0.3 to 1.1 with a increment of 0.2)
The results on the aerodynamic coecients are in agreement with those of
Zeng et al. [54] both for the drag and lift coecients (gure 12). The dierences
13
between the results of Zeng et al. [54] and ours remain below 1%.
Reynolds number
D
r
a
g

c
o
e
f
f
i
c
i
e
n
t

(
C
D
)
50 100 150 200 250
0
1
2
3
4
5
Present study
Zeng et al.
C
D
C
L
a)
Figure 12: The drag and lift coecient for L/d = 1.00 and versus Reynolds
numbers and b,c) streamlines in the plane normal to the wall for the case Re =
200, L/d = 0.75 for the present study (left) and extract to Zeng et al. [54]
(right)
5 Conclusion
We have developed an overlapped grid methodology in NSMB allowing the sim-
ulation of steady/unsteady, three-dimensional, incompressible ows in complex
and conned geometries. Arbitrary overlapped sub-domain meshes can be re-
composed automatically.
We carried out a series of numerical test cases to validate the algorithm and
to assess its feasibility in complex geometries. The computed results demon-
strated that our method gives good results both for simple two-dimensional
conguration and for complex conned three-dimensional ones. For the three-
dimensional test cases of the unconned and conned sphere, the dierences
with the literature are less than 1%. This proves that the communications in
overlapping interfaces do not induce any distortion.
The automatic detection of overlapped cell allows the simulation of moving
bodies and the next step is to simulate particles transported in conned ows.
References
[1] J. A. Benek, J. L. Steger, and F.C. Dougherty. A exible grid embedding
technique with application to the euler equations. AIAA Paper, 83-1944,
1983.
[2] W.D. Henshaw and D.W. Schwendeman. An adaptive numerical scheme
for high-speed reactive ow on overlapping grids. Journal of Computational
Physics, 191(2):420447, 2003.
14
[3] W.D. Henshaw and D.W. Schwendeman. Moving overlapping grids with
adaptive mesh renement for high-speed reactive and non-reactive ow.
Journal of Computational Physics, 216(2):744779, 2006.
[4] AK Kapila, DW Schwendeman, JB Bdzil, and WD Henshaw. A study
of detonation diraction in the ignition-and-growth model. Combustion
Theory and Modelling, 11(5):781822, 2007.
[5] PG Buning, IT Chiu, S. Obayashi, YM Rizk, and JL Steger. Numerical
simulation of the integrated space shuttle vehicle in ascent. In AIAA At-
mospheric Flight Mechanics Conference, Minneapolis, MN, pages 265283,
1988.
[6] M. Hinatsu and JH Ferziger. Numerical computation of unsteady incom-
pressible ow in complex geometry using a composite multigrid technique.
International Journal for Numerical Methods in Fluids, 13(8):971997,
1991.
[7] R.L. Meakin. Moving body overset grid methods for complete aircraft
tiltrotor simulations. In AIAA Computational Fluid Dynamics Conference,
11 th, Orlando, FL, Technical Papers. Pt. 2, volume 6(9), 1993.
[8] D.G. Pearce, S.A. Stanley, F.W. Martin Jr, R.J. Gomez, G.J. Le Beau, P.G.
Buning, W.M. Chan, I.T. Chiu, A. Wulf, and V. Akdag. Development of
a large scale Chimera grid system for the space shuttle launch vehicle. In
AIAA, 31st Aerospace Sciences Meeting, Reno, NV, volume 0533, 1993.
[9] R. Maple and D. Belk. A new approach to domain decomposition, the
beggar code. Numerical Grid Generation in Computational Fluid Dynamics
and Related Fields, Pineridge Press Limited, pages 305314, 1994.
[10] D. Jespersen, TH Pulliam, and P. Buning. Recent enhancements to OVER-
FLOW. AIAA paper, 970644, 1997.
[11] R. L. Meakin. Chapter 11 : Composite overset structured grids, in : Hand-
book of grid generation, J.F. Thompson, B.K. Soni and N.P. Weatherill.
CRC, 1999.
[12] R.L. Meakin and N.E. Suhs. Unsteady aerodynamic simulation of multiple
bodies in relative motion. In AIAA Computational Fluid Dynamics Con-
ference, 9 th, Bualo, NY, volume AIAA-89-1996, pages 643657, 1989.
[13] W.D. Henshaw and D.W. Schwendeman. Parallel computation of three-
dimensional ows using overlapping grids with adaptive mesh renement.
Journal of Computational Physics, 227:74697502, 2008.
[14] B. Landmann and M. Montagnac. A highly automated parallel Chimera
method for overset grids based on the implicit hole cutting technique. In-
ternational Journal for Numerical Methods in Fluids, 2010.
[15] J. Vos, A. Rizzi, D. Darracq, and E. Hirschel. Navier-Stokes solvers in
European aircraft design. Progress in Aerospace Sciences, 38:601697, 2002.
15
[16] J. Vos, A. Rizzi, A. Corjon, E. Chaput, and E. Soinne. Recent Advances
in aerodynamics inside the NSMB (Navier-Stokes Multiblock) Consortium.
AIAA paper, 98-0225, 1998.
[17] A. Jameson. Analysis and design of numerical scheme for gas dynamics, 1 :
articial diusion, upwind biased, limiters and their eect on accuracy and
multigrid convergence. Computational Fluid Dynamics, 4:171218, 1995.
[18] A.J. Chorin. Numerical solution of the navier-stokes equations. J. Math.
Computation, 22:745, 1968.
[19] R.L. Meakin. A new method for establishing intergrid communication
among systems of overset grids. AIAA Paper, 91-1586, 1991.
[20] H. Samet. The design and analysis of spatial data structures. Addison-
Wesley, 1990.
[21] Timo L. Siikonen, Patrick P. Rautaheimo, and Esa J. Salminen. Numerical
techniques for complex aeronautical ows. European Congress on Compu-
tational Methods in Applied Sciences and Engineering ECCOMAS 2000,
2000.
[22] Wei Liao, Jinsheng Cai, and Her Mann Tsai. A multigrid overset grid ow
solver with implicit hole cutting method. Computer Methods in Applied
Mechanics and Engineering, 196:1701 1715, 2007.
[23] Jubaraj Sahu and Joseph L. Steger. Numerical simulation of three-
dimensional transonic ows. International Journal for Numerical Methods
in Fluids, 10(8):855873, 1990.
[24] G. Desquesnes, M. Terracol, E. Manoha, and P. Sagaut. On the use of a
high order overlapping grid method for coupling in CFD/CAA. Journal of
Computational Physics, 220:355382, 2006.
[25] J.W. Delfs. An overlapped grid technique for high resolution CAA schemes
for complex geometries. AIAA Paper, Seventh AIAA/CEAS Aerocoustics
Conference, Reno, May 2001, 2001-2199, 2001.
[26] S.E. Sherer and J.N. Scott. High-order compact nite-dierence methods
on general overset grids. Journal of Computational Physics, 210:459496,
2005.
[27] T.K. Sengupta, VK Suman, and N. Singh. Solving NavierStokes equa-
tion for ow past cylinders using single-block structured and overset grids.
Journal of Computational Physics, 229(1):178199, 2010.
[28] G. Chesshire and WD Henshaw. Composite overlapping meshes for the
solution of partial dierential equations. Journal of Computational Physics,
90:164, 1990.
[29] HS Tang and T. Zhou. On nonconservative algorithms for grid interfaces.
SIAM Journal on Numerical Analysis, 37:173193, 1999.
[30] CJ Freitas and SR Runnels. Simulation of uidstructure interaction using
patched-overset grids. Journal of uids and Structures, 13:191207, 1999.
16
[31] Z J Wang. A fully conservative interface algorithm for overlapped grids.
Journal of Computational Physics, 122:96106, 1995.
[32] Man Mohan Rai. A relaxation approach to patched-grid calculations with
the euler equations. Journal of Computational Physics, 66:99 131, 1986.
[33] JA Wright and W. Shyy. A pressure-based composite grid method for the
Navier-Stokes equations. Journal of computational physics, 107:225238,
1993.
[34] HS Tang, S. Casey Jones, and F. Sotiropoulos. An overset-grid method
for 3D unsteady incompressible ows. Journal of Computational Physics,
191:567600, 2003.
[35] W.D. Henshaw. A fourth-order accurate method for the incompressible
Navier-Stokes equations on overlapping grids. Journal of Computational
Physics, 113:1325, 1994.
[36] Y. Zang and RL Street. A composite multigrid method for calculating
unsteady incompressible ows in geometrically complex domains. Interna-
tional Journal for Numerical Methods in Fluids, 20:341362, 1995.
[37] B. Hubbard and H. C. Chen. A chimera scheme for incompressible viscous
ows with application to submarine hydrodynamics. AIAA Paper, 25th
AIAA Fluid Dynamics Conference, 2210, 1990.
[38] Jean-Jacques Chattot and Y. Wang. Improved treatment of intersecting
bodies with the chimera method and validation with a simple and fast ow
solver. Computers & uids, 27:721740, 1998.
[39] W. Liao, J. Cai, and H.M. Tsai. A multigrid overset grid ow solver with
implicit hole cutting method. Computer Methods in Applied Mechanics and
Engineering, 196:17011715, 2007.
[40] T.M. Burton and J.K. Eaton. Analysis of a fractional-step method on
overset grids. Journal of Computational Physics, 177:336364, 2002.
[41] CH Tai, Y. Zhao, and KM Liew. Parallel computation of unsteady in-
compressible viscous ows around moving rigid bodies using an immersed
object method with overlapping grids. Journal of Computational Physics,
207:151172, 2005.
[42] S. Taneda. Experimental investigation of the wake behind a sphere at low
Reynolds numbers. J. Phys. Soc. Japan, 11(10):11041108, 1956.
[43] RH Magarvey and R.L. Bishop. Transition ranges for three-dimensional
wakes. Canadian Journal of Physics, 39(10):14181422, 1961.
[44] M. Provansal and D. Ormieres. Transition to turbulence in the wakes of
axisymmetrical objects. In Symposium on Turbulent Shear Flows, 11 th,
Grenoble, France, pages 177, 1997.
[45] G. Bouchet, M. Mebarek, and J. Dusek. Hydrodynamics forces acting on
a rigid xed sphere in early transitional regimes. European Journal of
Mechanics B/Fluids, 25:321336, 2006.
17
[46] TA Johnson and VC Patel. Flow past a sphere up to a Reynolds number
of 300. Journal of Fluid Mechanics, 378:1970, 1999.
[47] B. Ghidersa and Jzn Dusek. Breaking of axisymmetry and onset of un-
steadiness in the wake of a sphere. Journal of Fluid Mechanics, 423:3369,
2000.
[48] A.G. Tomboulides and S.A. Orszag. Numerical investigation of transitional
and weak turbulent ow past a sphere. Journal of Fluid Mechanics, 416:45
73, 2000.
[49] I. Nakamura. Steady wake behind a sphere. Physics of Fluids, 19:5, 1976.
[50] FW Roos and WW Willmarth. Some experimental results on sphere and
disk drag. AIAA Journal, 9:285291, 1971.
[51] R. Mittal. Planar symmetry in the unsteady wake of a sphere. AIAA
journal, 37(3):388390, 1999.
[52] J. Magnaudet, M. Rivero, and J. Fabre. Accelerated ows past a rigid
sphere or a spherical bubble. Part 1. Steady straining ow. Journal of
Fluid Mechanics, 284:97135, 1995.
[53] P. Bagchi and S. Balachandar. Steady planar straining ow past a
rigid sphere at moderate Reynolds number. Journal of Fluid Mechanics,
466:365407, 2002.
[54] Lanying Zeng, S. Balachandar, and Paul Fischer. Wall-induced forces on a
rigid sphere at nite Reynolds number. J. Fluid Mech., 536:125, 2005.
[55] Fumio Takumura and Jacques Magnaudet. The transverse force on clean
and contaminated bubbles rising near a vertical wall at moderate Reynolds
number. J. Fluid Mech., 495:235253, 2003.
18