You are on page 1of 24

AIAA 2017-3623

AIAA AVIATION Forum


5-9 June 2017, Denver, Colorado
23rd AIAA Computational Fluid Dynamics Conference

An Immersed Boundary Method for preliminary


design aerodynamic studies of complex configurations
∗ ∗ ∗ ∗
Stéphanie Péron Thomas Renaud Marc Terracol Christophe Benoit

Ivan Mary
ONERA - The French Aerospace Lab, F-92322, Châtillon, France

This paper presents a method to compute flows around arbitrary obstacles providing a
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

CFD solution within an acceptable time frame, by alleviating the tedious task of meshing
complex aerodynamics configurations. This method is based on an Immersed Boundary
Method (IBM), where the bodies are immersed within the mesh and relies on the automatic
generation of adaptive Cartesian grids. The workflow starts by generating the Cartesian
mesh around the different bodies involved into the study given the surface mesh of the
different geometrical components and the near-wall resolution around them. The present
IBM requires in a preprocessing step to compute some information that will be used during
the simulation in order to mimic the wall boundary condition near bodies. The present
approach is based on a Cartesian mesh, thus a wall model is used to mitigate the number
of points required to solve the RANS equations with an acceptable accuracy. A grid
convergence study is achieved on the turbulent transonic flow around a RAE2822 profile,
in order to quantify which near-wall resolution is necessary for a given accuracy of the
solution. Once these figures are determined, a simulation of the turbulent transonic flow
around the NASA Common Research Model is achieved. Finally a demonstration of the
capability of the method to perform CFD simulations of a complex geometry is achieved
on a helicopter rotor head.

I. Introduction
Thanks to the growth of the computational power and improvements of CFD solvers within the last
decades, the aerospace industry relies today mostly on CFD simulations during the design phase, since RANS
simulations can be run within a day. However, the geometrical complexity of the simulated configurations
has increased too, taking into account small details, such as track fairings on an aircraft or rotor head
components of a helicopter, making the mesh generation the major bottleneck of the CFD workflow.
Today, even preliminary design studies must be achieved within a short delay for competitiveness concerns.
This means that efficient tools are required to perform parametric studies and evaluate quickly the influence
of modifying a shape or adding details on the performances of an aircraft. High-fidelity CFD tools are
not necessary at that stage; lifting-line tools can be used to get trends quickly but models are not well-
adapted to complex configurations. Low-fidelity CFD (e.g. Euler solutions) could be appropriate but the
mesh generation is the barrier to override. The Immersed Bounday Methods (IBM) can be seen as a good
compromise between the quality of the solution and how quickly it can be obtained. This concept refers
initially to the work of Peskin1, 2 , which employed a novel approach many decades ago to simulate biological
flows onto Cartesian grids which did not conform to the geometry. The obstacles laying in the flow are taken
into account by introducing a forcing term into the momentum equations. Since then, many variants of this
approach have been developed, as quoted by Mittal and Iaccarino3 . A first approach consists in introducing
a continuous source term and is well suited for flows with immersed elastic boundaries1, 4 . In this context,
the source term represents the exchange of momentum between the fluid and solid through a law based on
the theory of elasticity. However, in the limit of rigid boundaries, this problem is not generally well-posed,
leading to a lack of stability and accuracy. Several discrete forcing methods have been developed for flow
∗ Research scientist, Department of Aerodynamics, Aeroelasticity & Acoustics (DAAA)

1 of 24

American Institute of Aeronautics and Astronautics


Copyright © 2017 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
simulations around solid bodies, among which the ghost-cell direct forcing approach, as developed by Mittal
et al.5 , Fadlun et al.6 and Tseng et al.7 . The IBM can be used on the whole geometry7, 8 or locally9, 10 to
capture the blocking effects of geometrical details. A similar approach consists in cutting cells that intersect
the geometry which has proven efficient and robust for inviscid flow simulations around complex geometries
(see Coirier & Powell11 and Berger & Aftosmis12 among others) . The use of Cartesian grids with local grid
refinement in combination with embedded obstacles (with either immersed boundary or cut-cell methods)
seems to be well-suited for a high-level of automation and computational efficiency8, 12, 13 . Although the use
of adaptive Cartesian grids around arbitrary immersed obstacles is conceptually attractive, the resolution of
high Reynolds number flows requires wall models14, 15 to restrict the number of points within the boundary
layer.
In this paper, we propose an approach that enables to get the aerodynamic trends quickly with an ac-
ceptable accuracy to address pre-design concerns. It consists in taking advantage of the Immersed Boundary
Method (IBM) onto a set of locally refined Cartesian grids to reduce significantly the mesh generation effort.
Input data are the flow conditions, the surface meshes describing the obstacles, the near-wall spacing around
the set of obstacles and the extent of the computational domain. Inviscid and viscous simulations can be
achieved. However, in order to prevent the number of points to be huge when treating three-dimensional
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

turbulent flows at high Reynolds number, a wall function is applied to counterbalance the inability of Carte-
sian cells to refine in the wall normal direction. The paper is organized as follows: first the present Immersed
Boundary Method is presented, based on a direct forcing formulation. The way the solution is forced at
some target cells is detailed and a focus will be made on the location of image points. In a second part, the
way this approach is integrated within a CFD workflow is presented. Then simulations of the turbulent flow
around a RAE2822 profile are achieved in order to validate the method. A comparison between a body-fitted
approach and the IBM Cartesian approach is achieved. A wall-resolved solution on the body-fitted mesh is
first compared to solutions on the same mesh using a wall function applied successively closer to the wall. A
grid convergence study on the IBM Cartesian mesh is then achieved and the influence of the wall function
and the IBM Cartesian mesh resolutions onto the solution accuracy will be discussed. Finally, a turbulent
flow simulation of the NASA Common Research Model configuration is achieved, showing that the method
is able to provide an acceptable solution while alleviating the mesh generation effort. A RANS simulation
of a static helicopter rotor head will demonstrate the capabilities of the approach to deal with geometrically
complex geometries will be done.

II. Description of the Immersed Boundary Method


II.A. Discretization of the Navier-Stokes equations
The Navier-Stokes equations for a compressible flow can be expressed as follows:
 ∂ρ ∂
 + (ρuj ) = 0
∂t ∂xj




 ∂ρui ∂ ∂p ∂
+ (ρui uj ) = − + (σij ) i = 1, 2, 3 (1)

 ∂t ∂xj ∂xi ∂xj

 ∂ρE ∂ ∂ ∂

 + ((ρE + p) uj ) = − (Qj ) + (σij ui )
∂t ∂xj ∂xj ∂xj

where ρ denotes the fluid density, u the velocity vector, p the pressure, ρE the total energy per unit mass,
σ the viscous stress tensor and Q the heat flux vector. In our approach, the system (1) is solved for interior
cells using a cell-centered Finite-Volume method based on a directional five-cell stencil. W will denote the
state variables W = (ρ, ρu, ρv, ρw , ρE) in the following.

II.B. Principle of the Direct Forcing Formulation


To take into account embedded obstacles to which the mesh does not conform, a direct forcing Immersed
Boundary Method has been developed, similar to the one proposed by Fadlun6 and Nakahashi8 , which
consists in imposing a value of the flow state W at some cells in the vicinity of the obstacle in order to
recover implicitly the wall boundary condition. A discrete forcing is introduced at these particular cells into
the Navier-Stokes equations 1 for compressible flows in our case: First, cells that lie inside obstacles are
marked as solid points and those lying outside are marked as fluid points. The presence of walls is taken

2 of 24

American Institute of Aeronautics and Astronautics


into account by identifying some cells (called IB target cells in the following) located at the fringe of solid
and fluid points and modifying the state variables W such that a wall boundary condition can be implicitly
recovered. Figure 1-(a) displays an example of an arbitrary obstacle defined by a curve, immersed into a
2D Cartesian mesh. In that case, blanked points inside the obstacle (blue-colored region in figure 1-(b))
are surrounded by red-colored points defining IB target points. Fluid points are outside of the obstacle in
that case (green-colored region) and are computed by the CFD solver. The IB blanked (solid) cells are not
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

(a) (b)

Figure 1. (a) Cartesian mesh and immersed obstacle; (b) nature of points: fluid points in green, IB target points in
red, IB blanked points in blue.

updated by the solver during the simulation. Consequently, the IB target cells can be seen as ghost cells
on the other side of a wall boundary condition and are used in the stencil of some of their neighbouring
interior cells. These points must be updated in order to model the influence of the immersed obstacle using
information in the fluid in the vicinity of the obstacle. Depending of the physical boundary condition (e.g.
slip or no slip) or the governing equations (Euler, Navier-Stokes) and the flow regime (laminar, turbulent,
with a high Reynolds number), the IB target cells can lie inside or outside the obstacles as detailed in figure
2.
Values of W are forced at IB target points to satisfy the wall boundary condition at the wall using a set

(a) (b) (c)

Figure 2. Sketch of a 5-point directional stencil involving IB target points (a): for a slip boundary condition; (b): for
a no-slip condition; (c): using a wall function.

of forcing points within the fluid, that will be called IB image points. First, two methods to reconstruct the
flow field W at these IB target points are presented, assuming their corresponding image points are known.
Then, we will explain how we determine the location of these image points.

3 of 24

American Institute of Aeronautics and Astronautics


II.C. Reconstruction of the state variables at IB target points
The state variables W can be forced at IB target points either by a linear reconstruction of W at IB
image points or by applying a wall function, which is detailed hereafter. In both cases, we assume that the
pressure, density and temperature are equal for a given target point and its corresponding image point. Both
reconstructions are then applied on the velocity vector and also on the Spalart-Allmaras turbulent variable
ν̃ where applicable.

II.C.1. Linear reconstruction of the velocity


To illustrate how the state variables W are reconstructed at IB target points, consider the sketch displayed
in figure 3. The obstacle is represented by the curve S, splitting the domain into a solid region (on the
bottom) and a fluid region (on the top). Here, point A is defined as an IB target point, lying inside the
obstacle. Its corresponding wall point N is obtained by the projection onto S following the normal vector

−n . Its image point is B and lies within the fluid region but does not match with a point where the state
variables W are computed. Its surrounding points D1, D2, D3, D4 will be considered as donor points to
interpolate the state variables W at point B.
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

Then the pressure pA and density ρA at IB target points A are directly imposed as their corresponding

Figure 3. 2D sketch describing the present direct forcing IBM approach. Solution W at IB target point A is built up
using the corresponding interpolated value of W at its image point B.

values at IB image point B.


As it is shown on the sketch, the image points are not necessarily the mirror points of target points, thus
a linear reconstruction of the velocity must be applied. According to whether a no-slip or a slip boundary
condition has to be recovered at the wall, two cases are considered.

In the case of a no-slip boundary condition where k→



u k=0 at the wall, a one-dimensional linear interpo-


lation is applied given u (B) as follows:


− distA,N →

u (A) = u (B) ,
distB,N
where distA,N and distB,N are the signed distance of IB target point A and IB image point B to the wall
point N respectively.

Now, consider the case of a slip boundary condition. The velocity vector → −u can be decomposed in a
normal and a tangential vector as:

−u =→ −
ut+→ −un.
In order to recover k u n k = 0 at the wall, a linear reconstruction is applied on →

− −
u n and not →

u anymore:


− distA,N →−
u n (A) = u n (B) .
distB,N
The tangential velocity components are then obtained by → −
u t (A) = → −
u (B) − k→

u n (B)k→

n , where k→

u n (B)k is


the magnitude of the normal velocity at IB image point B and n is the normals to the wall defined at point

4 of 24

American Institute of Aeronautics and Astronautics


A. If the one-equation Spalart-Allmaras turbulence model is applied to solve the RANS equations, then the
pseudo-viscosity ν̃ is recovered by the same linear interpolation to recover ν̃=0 at the wall. However, in
practice, a wall model is applied.
Here, IB target points are defined by the m first layers of solid points, with m=2 is the number of ghost
cells required by the solver. It can be pointed out that if the point B is too close to the wall, then its donor
cell might contain an IB target point, which makes it impossible to explicitly interpolate the flow field at
image point B. If this point B is located too far from the wall, then the linear reconstruction will result in
a wrong estimation of the flow field at IB target point, which is especially true in boundary layers.

II.C.2. Reconstruction using wall functions


As the Reynolds number increases, this IBM approach on Cartesian grids cannot maintain wall spacings of
y + ≤ 1 at a reasonable cost. The use of a wall function is a key point to perform the present approach
for turbulent flow simulations since it allows the first grid point to be placed at y + ≈ 50 − 100. However,
wall functions on Cartesian grids do not decrease the number of points as strongly as they can do if applied
on curvilinear conformal grids due to the isotropic nature of cells. Figure 4 displays the IB target point A
and its image point B, for which the variables WB are interpolated from the computed cells. Instead of
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

a linear reconstruction, a wall function is applied between the image point B and the wall. Note that IB
target points are the m first layers of fluid points above the obstacles, since they must lie inside the fluid for
the wall function to be applied. As for the slip boundary condition, the velocity vector is decomposed into

Figure 4. Wall function IBM: IB target point is A and corresponding image point is B. In red dots are IB target points
around the obstacle; in blue dots, their image points.

a tangential and a normal vector, which are here treated in different ways. For the tangential velocity, the
following reconstruction is considered:

− UA →

u t (A) = u t (B) ,
UB
where U = k→
P
−u (P )k denotes the magnitude of the tangential velocity at any point P . The evaluation of
t
UA is achieved by applying a wall function, detailed in the next paragraph II.C.3.
The normal velocity at image point B is obtained by a simple projection:


u n (B) = (→

u (B) · →

n )→

n,

where →

n denotes the unit normal vector at the wall passing through points A and B.
The tangential velocity at IB image point B can be expressed by:

− →
− → −
u t (B) = (→

u (B) · t ) t ,

5 of 24

American Institute of Aeronautics and Astronautics


We could have imposed the flow to be locally parallel to the wall, that means → −
u n = 0, but this tends in
practice to delay the separation on massively separated flows. Thus we have kept a 1D linear interpolation
to compute the normal velocity:

− distA,N →

u n (A) = u n (B) .
distB,N
The three components of the velocity vector → −u at point A are then obtained by summing the correspond-
ing normal and tangential components. As previously said, the tangential components of the velocity are
obtained by applying a wall function.

II.C.3. Wall functions


The most common function to describe the evolution of the velocity within an equilibrium turbulent boundary
layer (zero-pressure gradient) is the log law of the wall defined as:
1
u+ = log(y + ) + β , (2)
κ
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

u ρw y u τ
where u+ = and y + = , with κ=0.41 is the Vón Kármán constant and β = 5.2; uτ denotes the
uτ µw
friction velocity, ρw and µw the values of density and viscosity at the wall, assumed equal to their values at
corresponding image points B.
However, the limitation of the log law is that it is not able to model the inner and buffer layers of the boundary
layer, which is critical in our approach since the dimensionless wall distance y + cannot be controlled at image
points since they can be more or less close to the wall, due to the Cartesian topology of the mesh. Several
analytical functions have been developed to fit the three sub-layers, such as the law developed by Musker16 ,
which results in:
 + " 9.6
#
(y + + 10.6)

+ 2y − 8.15
u = 5.424 arctan
16.7
+ log 2 − 3.52 . (3)
(y +2 − 8.15 y + + 86)

This expression is explicit for the velocity magnitude and can be applied in the same way as the log law. It
must be noted that expressions 2 and 3 involve the skin friction velocity uτ , which is unknown. The first
step of the process is therefore to estimate its value using a Newton iterative algorithm.

In the case of a RANS modeling using Spalart-Allmaras model17 , the pseudo-viscosity ν̃ must also be
estimated at IB target point A. Under the assumption of an equilibrium boundary layer, ν̃ can be defined
as
1
ν̃ = κ uτ y ,
fv1
where fv1 is the damping function of Spalart-Allmaras model, which is a nonlinear function of ν̃, thus ν̃ is
also obtained using a Newton iterative algorithm.

II.D. Location of image points


For both reconstructions (linear and with a wall function), the location of image points must be chosen care-
fully to minimize modeling errors. They must be as close as possible to the wall for the linear reconstruction
or the wall function. On the other hand, they must not lie within a donor stencil that contains an IB target
point since its state variables W is precisely unknown, preventing the interpolation to be explicit. Moreover,
in concave wall regions, the image points must not fall into the solid on the other side of the concavity. Thus
a front of the first computed cells is determined as detailed in Péron et al.18 . A sketch is displayed in figure
5 where red points are IB target points. If image points are the symmetrical of target points with respect to
the wall, then some may fall within the gap between the front (in green) and the wall (in blue) where they
are not interpolable. The image points are thus defined by a projection following normals of target points,
which has the advantage of maintaining a process as automated as possible.

6 of 24

American Institute of Aeronautics and Astronautics


Figure 5. Example of IB target points (in red), valid and invalid symmetrical points (in green and orange), front
(projected) point (in blue) and front mesh (thick green lines).
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

III. IBM workflow on Cartesian adaptive grids


In this section, we present the workflow that enables to perform an IBM simulation starting from a set of
discrete surfaces describing the obstacles and the resolution requirements near these obstacles, the reference
state and governing equations. Our objective is to be able to perform simulations within a short delay for
preliminary design purposes and thus to be able to break down the barrier of the mesh generation for complex
geometries while allowing a sligthly lower fidelity of the solution, combined with an efficient solver. The IBM
approach on adaptive Cartesian grids offer a fully automated method to achieve this goal. A workflow has
been developed, based on Cassiopee functions19 , to perform the mesh generation, the IBM pre-processing
and the update of IB target points at each time step. A dedicated Cartesian solver, called FastS, developed at
ONERA, computes the fluid points using a 2nd -order accurate method in order to maintain an affordable cost
of the simulation despite the number of Cartesian points is generally bigger than the number of curvilinear
points discretizing the same configuration.
The whole workflow is based on a set of Python functions, sharing the same data structure, namely the
CGNS/Python representation20 . Initially, a CGNS/Python tree defining the obstacles is defined. The final
mesh is stored in another tree, where its flow solution, boundary conditions and information for transfer
between overlapping Cartesian grids and IBM updates are stored.

III.A. Pre-processing
First, a set of Cartesian grids is automatically generated starting from the set of obstacles and resolution
requirements near each of them. This mesh is composed by a set of uniform Cartesian grids, based on a
skeleton defined by an octree21 . Different levels of refinements can be set near each obstacle and prescribed
in some fluid regions (e.g. in the wake). Cartesian grids overset with a minimum overlapping, such that
information between domains is achieved by general Chimera transfers. An example of a Cartesian mesh that
is automatically generated is presented in figure 6. For that case, the input data is a 1D-mesh discretizing
the profile and the cell size required near the profile, equal to 1/1000 of the chord length here. The quadtree
mesh is first built (figure 6-(a)), with the finest refinement level located near the profile and fills the whole
computational domain. Each element of the quadtree is then filled with a Cartesian grid of a user-specified
number of cells per direction, which are then extended at their borders to overlap such that m layers of cells
can be interpolated (where m=2 corresponds to the number of ghost cells required by the solver here). In
order to reduce the number of cells, grids that are entirely inside the solid are removed, as displayed in figure
6-(b). That example consists in a mesh of 195,000 cells with 104 grids, with an actual minimum spacing of
10−4 c, where c is the chord length. The IBM preprocessing is then achieved, based on several geometrical
algorithms initially developed for overset grids. Some of the steps are illustrated by the IBM preprocessing
of the previous NACA0012 configuration.
- a blanking is performed to mark solid points, using a X-Ray technique22 or by a line-of-sight algo-
rithm19 , as displayed in figure 7.

7 of 24

American Institute of Aeronautics and Astronautics


(a) (b)

Figure 6. Example of a mesh around a NACA0012 profile for an Euler simulation: (a) quadtree skeleton mesh; (b)
resulting Cartesian mesh.
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

- the signed distance field is then computed;


- the IBM preprocessing is performed:
– marking of the m layers of IB target points at the fringe of blanked points (green dots in figure
8-(a));
– computation of the normal vectors at IB target points as the gradient of the signed distance;
– projection of the IB target points following these normal vectors onto the obstacles to get the
corresponding wall points (red dots in figure 8-(b));
– determination of the front of first valid donor cells (blue lines in figure 8-(c));
– projection of the IB target points following the normal direction onto that front (blue dots in
figure 8-(d));
– computation of the interpolation data of image points (donor cell indices and weights);
– computation of the interpolation data of interpolated points for Chimera transfers between over-
lapping Cartesian grids.

- initializing of the flow field (according to the reference state in the tree describing the obstacles);
- distribution of Cartesian grids on NP processors.
All the information is stored in the CGNS/Python tree (distance field, nature of cells, Chimera interpolation
data, IB reconstruction data,...), which will be used during the CFD simulation.

III.B. Implementation of IBM approach within the CFD simulation


III.B.1. Description of the FAST solver environment
A CFD demonstrator named FAST (Flexible Aerodynamic Solver Technology) has been developed at ON-
ERA for two years23 , whose objective is to provide a software architecture and numerical techniques allowing
for a high level of efficiency, flexibility and upgradeability. This demonstrator is made by a set of independent
modules, each of them defining a CFD solver dedicated to Cartesian, curvilinear and polyhedral grids, which
rely on the CGNS/Python data representation. These modules work in an interoperable way with other
modules, most of them being Cassiopee modules devoted to pre,co and post-processing. Each component
acts through Python functions on the same shared data (CGNS/Python tree in memory). Interior points of a
Cartesian/structured/polyhedral grid are updated by the corresponding dedicated solver. A function enables
to perform Chimera transfers between donor and receptor cells within Connector module of Cassiopee. This
function has been generalized to handle also the update of IB target points, since the first step consists in a
kind of Chimera interpolation, but for the IB image points.
Each CFD solver solves the compressible Euler and Navier-Stokes equations using a second-order accurate

8 of 24

American Institute of Aeronautics and Astronautics


Figure 7. IBM preprocessing of a NACA0012 profile: blanking of cells inside the obstacle.
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

(a) (b)

(c) (d)

Figure 8. Closeup view near the leading edge of a NACA0012 profile: (a) IB target points are represented by green
dots, the profile by the red curve; (b) red dots correspond to the IB wall points, resulting from the projection of IB
target points following normals onto the profile; (c) front of first computed cells in blue; (d) blue dots are the IB image
points obtained by projection of target points following normals onto the front.

Finite-Volume Method only for interior points of each grid. The turbulence modeling is handled by the
Spalart-Allmaras model17 in the RANS mode, whereas a MILES approach is retained for LES24 . For RANS
computations, the Roe-MUSCL scheme is used with a first-order accurate implicit time integration. Jacobian
approximations are those proposed by Jameson & Yoon25 and Coakley26 , wheras the linear system is solved
by the LU-SGS method25 . For LES computations, an hybrid centered/upwind scheme is retained to manage
a good compromise between robustness and accurate simulation of the turbulent small eddies27 , whereas
the temporal integration is achieved by a three-step Runge-Kutta explicit scheme, or by a 2nd -order implicit
Gear scheme with local Newton sub-iterations28 .
In this paper, the Cartesian solver of FAST is used to perform Euler and RANS simulations with Spalart-
Allmaras model.

9 of 24

American Institute of Aeronautics and Astronautics


III.B.2. Update of IBM points at each sub-iteration
In our framework, Cartesian grids overlap such that two layers of cells are interpolable from their neighbouring
grids. FAST Cartesian solver computes the interior points of each grid, physical boundary conditions are
then applied (e.g. farfield) and then IBM updates and generalized Chimera transfers between overlapping
grids must be achieved.
In order to know whether the IB target points must be updated before the Chimera transfers are applied or
not, we have considered the case of an immersed obstacle crossing a border between two overlapping grids
of same level of refinement. In that case, the update of the solution must be consistent with the one on
a single grid. The conclusion is that the IB target points must be updated before applying the transfers
between overlapping grids, which act as ghost cell updates. This has been taken into account within the
pre-processing: donor cells of IB image points cannot be Chimera interpolated cells, but Chimera donor cells
can be IB target points.
The update of IB target cells and Chimera-like interpolated cells are very similar, since the first step of the
IBM updates consists in an interpolation of the IB image points. The second step consists reconstructing
the velocity and the viscosity-like variable by either the linear formulation or the Musker’s wall function.
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

IV. Validation: turbulent flow simulation of a transonic RAE2822 airfoil


Our objective here is to quantify the accuracy of the IBM approach on Cartesian grids to simulate a
turbulent flow with an acceptable number of cells. For that purpose, RANS simulations of the transonic
viscous flow around a RAE2822 profile are performed using the IBM approach with a wall model and
compared to a reference solution computed on a conformal curvilinear mesh where the boundary layer is
resolved by the mesh. Thus, the influence of the wall model and the IBM method on Cartesian grids will be
considered separately. First, the wall model will be evaluated on the conformal mesh by applying it closer
and closer to the wall and comparing the solutions with the wall-resolved solution. Then, a grid convergence
study will be presented on the Cartesian mesh to highlight the impact of the mesh refinement near the wall
on the wall-modeled IBM solution.

IV.A. Description of the test-case


The flow conditions are defined as follows:
- the freestream Mach number is M∞ =0.73;
- the angle of attack is α = 2.79◦ ;
- the chord-based Reynolds number is 6.5 millions.

Steady RANS simulations are performed using FAST structured and Cartesian solvers and using Roe-MUSCL
scheme for the spatial discretization.

IV.B. Meshes
The same body-fitted mesh is used for both the wall-resolved and wall-modeled simulations (figure 9). Five
Cartesian meshes of increased resolutions in the vicinity of the profile are automatically generated and used
within the wall-modeled IBM approach; they are displayed in figures 10. Table 1 provides the main features
of the six meshes involved in this study. The number of cells of finest levels within the boundary layer of
height δ ≈ 0.007 c on the suction side of the profile at station x/c=0.3 are compared. It can be pointed out
that the size of the Cartesian mesh increases significantly with the near-wall resolution, in comparison with
the body-fitted mesh, which is due to the isotropic and uniform spacing along the profile.

IV.C. Comparison of the wall-resolved and wall-modeled solutions on the body-fitted struc-
tured mesh
Here, the RANS wall-resolved and the wall-modeled approaches are compared on the same body-fitted
structured mesh. In order to mimic the IBM approach that will be presented afterwards, the forcing by the
wall function is applied on two layers of cells on the wall-normal grid line j0 and j0 +1. Cells below these layers

10 of 24

American Institute of Aeronautics and Astronautics


Figure 9. Body-fitted structured mesh.
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

(a) (b)

(c) (d)

Figure 10. IBM Cartesian meshes: (a) IBM #1 (coarsest); (b) IBM #2; (c) IBM #3; (d) IBM #4 (finest).

are not considered by the CFD solver. Decreasing j0 towards the physical wall boundary enables to show
separately the influence of the wall model onto the solution such that the influence of the Cartesian IBM itself
will be more easily highlighted later on. The cells where the forcing is applied are distant of 1.5e−3c (case
#1) to 1.8e−4c (case #4) to the wall and case # 5 consists of a forcing at the first layers to compare with
the RANS wall-resolved solution. The wall-modeled simulations compare well with the reference simulation
considering the density contours displayed in figures 11 and 12. Velocity profiles are extracted at two different
stations upstream (x/c=0.3) and downstream (x/c=0.65) of the shock on the suction side. As it is shown
in figure 13, the velocity converges to the reference solution as the forcing is applied closer to the wall.
Note that the velocity is no longer significant below the first point where the wall model is applied, which
is due to our post-processing here, which interpolates directly the solution between the wall and the first

11 of 24

American Institute of Aeronautics and Astronautics


Mesh # zones # cells near-wall mesh resolution # boundary layer cells at x/c = 0.3
body-fitted 1 65,532 2.0e-6 54
IBM #1 7 347,000 1.5e-3 4
IBM #2 97 623,000 7.5e-4 9
IBM #3 130 2.37 M 3.5e-4 19
IBM #4 176 2.6 M 1.8e-4 34
Table 1. Conformal and Cartesian mesh features.

wall-modeled point. Figures 14-(a) and (b) display the eddy viscosity profiles (non-dimensionalized by the
molecular viscosity) through the boundary layer at the same streamwise stations. Points below a distance
corresponding to that of the first forced cells are not relevant and are not represented here. It can be
noticed that the wall-modeled solution converges towards the boundary-layer resolved solution as the forced
point gets closer to the wall, and faster downstream of the shock since the boundary layer is thicker than
upstream. Figures 15 display the skin pressure and friction distributions. It can be noticed that the skin
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

pressure distribution is not sensitive to the distance of forced cells to the wall. It appears clearly shifted
along the profile for the first four wall-modeled simulations but converges towards the reference solution.
As displayed in table 2, the order of magnitude of y + for forced cells is 50 to 400 (minimum and maximum
values for all the simulations are details in table 2). For the first four simulations (WL #1 to #4), the cells
where the wall function is forced are not within the boundary layer at the leading edge, which could explain
the discrepancies on the skin friction coefficient.

Mesh Min Max Mean


BF WL1 4.57 648 409
BF WL2 2.78 346 210
BF WL3 2.43 181 106
BF WL4 1.10 91 51
BF WL5 0.01 5.9 2.7
Table 2. RAE2822 configuration: ranges of y + with respect to the distance of the cells to the wall where the wall
function is applied.

IV.D. Comparison of the IBM wall-modeled solution on Cartesian grids


Here, the influence of the mesh resolution near the wall for wall-modeled IBM approach on a set of Cartesian
grids is studied. Solutions are compared with the body-fitted wall-resolved solution presented previously,
which will be our reference solution. Figure 16 displays the density contours for the five IBM simulations on
Cartesian grids, showing a good agreement with the reference solution in figure 11. As done in the previous
section, the velocity profile is extracted at the same streamwise stations on the suction side of the profile.
Figure 17 shows that the velocity converges towards the reference solution as the mesh resolution increases
(by dividing the cell spacing by a factor of 2).
Figures 14-(a) and (b) display the eddy viscosity profiles (non-dimensionalized by the molecular viscosity)
through the boundary layer at the same streamwise stations. Similarly, points below a distance corresponding
to that of the first forced cells are not relevant and are not represented here. The wall-modeled IBM solution
converges also towards the boundary-layer resolved solution. The same trends as for the wall-modeled
solution on body-fitted grids can be noticed upstream of the shock. However, downstream of the shock, the
eddy viscosity was under-estimated before reaching the convergence, whereas here it tends to decrease.
Similarly to the previous study, the skin pressure coefficient for the four IBM simulations on Cartesian
grids compare well with the reference solution (figure 19-(a)). The shock position is shifted upstream using
the IBM approaches, which cannot be explained yet. Unsurprisingly, the skin friction is also shifted for the
Cartesian IBM solutions but seems to converge towards the reference solution, except at the leading edge
and at x/c=0.55, which correspond to the shock region (figure 19-(b)). Similar trends were observed for the
wall-modeled solution on the conformal mesh, so this is probably due to the wall function itself rather than

12 of 24

American Institute of Aeronautics and Astronautics


Figure 11. RAE2822 profile, M∞ =0.73. Wall-resolved RANS solution on body-fitted mesh: density contours.
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

(a) (b)

(c) (d)

Figure 12. RAE2822 profile, M∞ =0.73. Wall-modeled RANS solution on body-fitted mesh: density contours (a) case
#1; (b) case #2; (c) case # 3: (d) case #4.

to the Cartesian approach.


Thanks to these comparisons, it can be assessed that the IBM # 3 mesh, corresponding to a near-wall
spacing of 3.5 × 10−4 c seems to be a good compromise between the quality of the solution and the mesh
size. Given the ranges of y + (cf. table 3), this means that the wall-modeled IBM approach is valid for
y + ≈ 100 - 200 for that type of flow regime. Extrapolating this mesh resolution in 3D, the resulting number
of points should be about one billion for a configuration similar to the NASA Common Research Model29 .
Berger and Aftosmis15 have recently proposed a model that relaxes the near-wall resolution requirements by
improving the wall model, such that it is also valid in the wake region of the boundary layer, which seems
to be a way to restrain the number of cells.

13 of 24

American Institute of Aeronautics and Astronautics


(a) (b)
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

(c) (d)
Figure 13. RAE2822 profile, M∞ =0.73: velocity profile with respect to the distance to the wall starting at two stations
on the upper side of the profile: (a,b) x/c=0.3, (c,d) x/c=0.65. Wall-modeled solutions on IBM Cartesian meshes
(dashed lines), wall-resolved solution on body-fitted mesh (solid line).

(a) (b)
Figure 14. RAE2822 profile, M∞ =0.73: eddy viscosity profile with respect to the distance to the wall starting at two
stations on the upper side of the profile (a) x x
c =0.3 and (b) c =0.65. Wall-model solution (dashed lines), wall-resolved
solution (solid line), body-fitted mesh.

V. Applications
Two applications are presented here. The first application is a simulation of the baseline configuration of
the NASA Common Research Model, considered in the fifth Drag Prediction Workshop29 . The objective is
to quantify how accurate is the present IBM approach for transonic flow simulations of aircrafts, regarding
the time spent to build up the mesh. The second application is rather a demonstration of the capabilities
of the present method, highlighting the ease to build up a geometrically complex configuration and the

14 of 24

American Institute of Aeronautics and Astronautics


Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

Figure 15. RAE2822 configuration: (a) pressure coefficient and (b) skin friction for wall-model simulations compared
with wall-resolved simulation (black dashed line).

robustness of the IBM preprocessing.

V.A. RANS turbulent flow simulation of the Common Research Model configuration
The baseline configuration of the NASA Common Research Model (see figure 20-(a)) is considered here. The
freestream Mach number is M∞ = 0.85, the angle of attack is α = 2.175◦ and the Reynolds number based
on is Rec =5 millions.
Our objective here is to demonstrate how the present IBM approach on Cartesian grids is accurate to deal
with aircraft configurations. One challenge we have to tackle is the limitation of the number of points.
Indeed, in order to capture the leading edge of the wing, a fine resolution must be set all along the span.
In our approach, the near-wall resolution varies on the surface: figure 20-(b) displays the three regions of
different resolutions. It is set to 0.4%c around the fuselage, 0.2%c around the wing, except at the leading

15 of 24

American Institute of Aeronautics and Astronautics


(a) (b)
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

(c) (d)

Figure 16. RAE2822 configuration, M∞ =0.73. Wall-model IBM approach on Cartesian grids: density contours for (a)
case IBM #1; (b) case IBM #2; (c) case IBM #3; (d) case # 4.

and trailing edges, at the wing tip and wing/fuselage junction where the required resolution is 0.1%c, where
c is the wing mean chord length. The resulting Cartesian mesh is made by 691 million points, extended of 43
chords in the three directions: currently, the mesh of the whole computational domain is built, but defining
a symmetry plane should be easily performed later. A view at mid-span is represented in figure 21-(a).
The preprocessing is automatically achieved for the case of wall-modeled IBM with IB image points as close
as possible from the IB target cells. A RANS simulation using Musker’s wall model to force the flow state
at IB target cells is achieved. RANS equations are solved using Roe’s spatial scheme and Spalart-Allmaras
turbulence model. Figure 21-(b) displays the convergence of the solution after 10 000 iterations. A cut in
the volume mesh at y=33% of the wing span is presented in figure 23, highlighting a transonic flow over the
wing. The pseudo-viscosity ν̃ of the Spalart-Allmaras model is more intense in the wake as expected and
provides a good information about the boundary layer thickness. Table 4 displays the mean, minimum and
maximum values of y + for the different mesh components. As expected, the boundary layer, its values are
too high on the fuselage and in the mid-chord region of the wing since the mesh resolution in these regions
were on purpose much coarser than at wing tip and near the leading and trailing edges.

Mesh Min Max Mean


IBM#1 0.18 812 312
IBM#2 1.58 413 167
IBM#3 0.03 217 81
IBM#4 0.05 109 42
Table 3. RAE2822 configuration: ranges of y + with respect to the size of the IB target cells.

16 of 24

American Institute of Aeronautics and Astronautics


(a) (b)
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

(c) (d)
Figure 17. RAE2822 profile, M∞ =0.73: velocity profile with respect to the distance to the wall starting at two stations
on the upper side of the profile: (a,b) x/c=0.3, (c,d) x/c=0.65. Wall-modeled solutions on IBM Cartesian meshes
(dashed lines), wall-resolved solution on body-fitted mesh (solid line).

(a) (b)
Figure 18. RAE2822 configuration, M∞ =0.73: eddy viscosity profile with respect to the distance to the wall starting
at two stations on the upper side of the profile (a) x x
c =0.3 and (b) c =0.65. Wall-modeled solutions on IBM Cartesian
meshes (dashed lines), wall-resolved solution on body-fitted mesh (solid line).

Results are compared with a solution obtained by a body-fitted multiblock structured approach, where
the boundary layer is resolved by the mesh. The solver parameters are the same for the Cartesian and
structured grids. The Cartesian solver is actually an optimized version of the structured solver for an
improved efficiency. Figure 24 displays the skin pressure coefficients at different spanwise sections of the
wing, from the fuselage/wing junction to the wing tip. Solutions are in good agreement despite a quite
coarse near-wall resolution for the IBM approach in comparison with the previous study on the RAE2822
case. However, discrepancies can be noted near the wing tip: the shock is too sharp with the IBM approach.

17 of 24

American Institute of Aeronautics and Astronautics


Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

Figure 19. RAE2822 configuration, M∞ =0.73: wall-modeled IBM approach on Cartesian grids: (a) skin pressure
coefficient distribution ; (b) skin friction distribution.

On the pressure side, results are very similar. The skin friction distribution at the same sections are compared
on figure 25 with the multiblock structured solution. It can be noticed that the IBM solution follows the
same trends as what was observed previously on the RAE2822 test-case. The most significant discrepancies
can be highlighted at the leading edge of the wing and in the vicinity of the shock but it can be recalled that
the near-wall resolution corresponds to the coarsest ones of the RAE2822 grid convergence study. These are
good results regarding the meshing effort and the solution accuracy.

V.B. Demonstration of a RANS simulation of a helicopter rotor head


Here, a RANS simulation of a static helicopter rotor head is presented in order to show that the method
enables to deal with complex geometries. The freestream Mach number is M∞ = 0.2057 and the angle of
attack is −2◦ , the rotorhead is tilted by −5◦ of shaft angle. The mesh is automatically generated given the

18 of 24

American Institute of Aeronautics and Astronautics


Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

(a) (b)

Figure 20. NASA CRM baseline configuration: (a) configuration; (b) different colors define different resolutions near
the corresponding surfaces.

Component Mean Min Max


Fuselage 540 0.005 3477
Pressure side 263 0.006 1006
Suction side 246 0.015 1552
Wing tip 220 1.46 551
Leading edge 187 0.28 439
Trailing edge 88 0.13 394
Table 4. NASA CRM configuration: ranges of y + with respect to the size of the IB target cells.

surface mesh (displayed in figure 26-(a)) and the near-wall resolution is 1/1000 D, where D is the diameter
of the rotor head. The resulting mesh involves 211 millions points on 2348 Cartesian grids. A steady RANS
simulation is achieved on the static rotor head, using FAST Cartesian solver using Roe-MUSCL spatial
scheme. The IB target points are updated thanks to Musker’s wall model. A slice in the fluid of the Mach
number contours and the pressure coefficient on the rotor head are displayed in figure 27, showing that the
geometrical IBM preprocessing is robust to perform such geometrically complex configurations. Preliminary
comparisons with a previous body-fitted approach using overset grids30 show promising perspectives to deal
with that kind of configurations.

19 of 24

American Institute of Aeronautics and Astronautics


Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

(a) (b)

Figure 21. NASA CRM configuration: (a) view of the mesh at 33% of the wing span; (b) convergence of density
residuals.

(a) (b)

Figure 22. (a) NASA CRM configuration: isocontours of the Mach number at midspan; (b) Spalart-Allmaras pseudo-
viscosity.

20 of 24

American Institute of Aeronautics and Astronautics


Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

(a) (b)

Figure 23. NASA CRM configuration: pressure coefficient distribution on the surface (a) pressure side; (b) suction
side of the wing.

(a) (b) (c)


Figure 24. NASA CRM configuration: comparison of skin pressure coefficient at different spanwise sections on the
wing with a body-fitted multiblock structured approach: (a) near the fuselage/wing junction; (b) at mid-span; (c) at
wing tip.

21 of 24

American Institute of Aeronautics and Astronautics


(a) (b) (c)
Figure 25. NASA CRM configuration: comparison of skin friction coefficient at different spanwise sections on the wing
with a body-fitted multiblock structured approach: (a) near the fuselage/wing junction; (b) at mid-span; (c) at wing
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

tip.

Figure 26. Helicopter rotor head configuration: (a) surface mesh; (b) slice of the IBM Cartesian mesh.

Figure 27. Helicopter rotor head configuration: (a) Mach contours; (b) Pressure coefficient on the rotor head.

22 of 24

American Institute of Aeronautics and Astronautics


VI. Discussions and perspectives
This paper presented an Immersed Boundary Method based on a direct forcing approach, whose objec-
tive is to be able to perform CFD simulations around arbitrary bodies without the daunting task of mesh
generation with an acceptable flow accuracy. This method relies on the automatic generation of adaptive
Cartesian grids. Since the IBM approach requires to locate some forcing points, namely the IB image points,
a method is proposed to determine them automatically in a preprocessing phase, which ensures that they are
located within the flow whatever the geometry is and as close as possible to the wall for a relevant forcing
term, especially when using a wall function. A workflow has been developed, based on Cassiopee modules
for the automatic adaptive Cartesian mesh generation, the IBM preprocessing and co/post-processing. An
efficient dedicated Cartesian solver of the ONERA FAST suite is used to update the fluid points, whereas
a transfer function of Cassiopee’s Connector module, originally developed for Chimera transfers, has been
generalized to all transfers, including the IB target point updates.
This enables to build up quickly the aerodynamic trends of a body by a CFD RANS solution within one
night, with only the surface meshes as input data, which should be promising for preliminary design studies.
As a wall function must be applied to avoid a wall-resolved mesh resolution leading to prohibitive com-
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

putational resources, the influence of the IBM Cartesian mesh approach and the wall function must be
determined. Consequently, a grid convergence study has been achieved on a RAE2822 profile to determine
the maximum cell size that provides a reasonably accurate solution in terms of skin pressure and friction
coefficients. For this case typical of a transonic turbulent flow regime, an acceptable mesh size near the wall
is roughly of 5.10−4 c, which leads to roughly a billion points on a 3D transonic configuration.
This method has been applied on the RANS simulation of the NASA Common Research Model. Despite a
high number of points in comparison with a body-fitted approach, it has been proved to provide a solution
with a good accuracy regarding the fact that the mesh was automatically generated. Mesh requirements
won’t be significantly increased if additional components are applied (e.g. the vertical and horizontal planes
or the pylon/nacelle). A preliminary RANS simulation of the flow around a helicopter rotor head has also
been achieved and demonstrates the ease and robustness of the preprocessing. Future works will consist in
going further into the validation of this approach on several flow regimes and on more complex configurations.
A non-uniform spacing along an obstacle could be achieved to reduce the number of cells, but this would
be done at the expense of the automation of the mesh generation. In order to control the mesh resolution
automatically, an adaption to the boundary layer thickness could be performed and an improvement of the
wall model is another way to reduce the mesh requirements, following the idea of Berger and Aftosmis.
Another perspective will be to deal with bodies in relative motion. For instance, it will be useful to combine
a body-fitted approach for helicopter blades and the IBM approach for the rotor head and the fuselage.

VII. Acknowledgments
The authors would like to thank Alexandre Limare and Robin Nez for their fruitful contributions to the
development of this method.

References
1 Peskin, C. S., “Flow patterns around heart valves: a numerical method,” Journal of Computational Physics, Vol. 10,

No. 2, 1972, pp. 252–271.


2 Peskin, C. S., “The immersed boundary method,” Acta numerica, Vol. 11, 2002, pp. 479–517.
3 Mittal, R. and Iaccarino, G., “Immersed Boundary Methods,” Annu. Rev. Fluid Mech., Vol. 37, 2005, pp. 239–261.
4 R. P. Beyer and R. J. LeVeque, “Analysis of a One-Dimensional Model for the Immersed Boundary Method,” SIAM

Journal on Numerical Analysis, Vol. 29, No. 2, 1992, pp. 332–364.


5 Mittal, R., Dong, H., Bozkurttas, M., Najjar, F., Vargas, A., and von Loebbecke, A., “A versatile sharp interface

immersed boundary method for incompressible flows with complex boundaries,” Journal of Computational Physics, Vol. 227,
No. 10, 2008, pp. 4825–4852.
6 Fadlun, E. A., Verzicco, R., Orlandi, P., and Mohd-Yusof, J., “Combined immersed-boundary finite-difference methods

for three-dimensional complex flow simulations,” Journal of Computational Physics, Vol. 161, No. 1, 2000, pp. 35–60.
7 Tseng, Y.-H. and Ferziger, J. H., “A ghost-cell immersed boundary method for flow in complex geometry,” Journal of

Computational Physics, Vol. 192, No. 2, 2003, pp. 593–623.


8 Nakahashi, K., “Immersed boundary method for compressible Euler equations in the Building-Cube Method,” AIAA

paper 2011-3386, 2011.

23 of 24

American Institute of Aeronautics and Astronautics


9 Zhu, W. J., Behrens, T., Shen, W. Z., and Sørensen, J. N., “Hybrid immersed boundary method for airfoils with a

trailing-edge flap,” AIAA Journal, Vol. 51, No. 1, 2012, pp. 30–41.
10 Mochel, L., Weiss, P.-É., and Deck, S., “Zonal Immersed Boundary Conditions: Application to a High-Reynolds-Number

Afterbody Flow,” AIAA Journal, Vol. 52, No. 12, 2014, pp. 2782–2794.
11 Coirier, W. J. and Powell, K. G., “Solution-adaptive Cartesian cell approach for viscous and inviscid flows,” AIAA

Journal, Vol. 34, No. 5, 1996, pp. 938–945.


12 Berger, M. and Aftosmis, M. J., “Progress towards a Cartesian cut-cell method for viscous compressible flow,” AIAA

paper 2012-1301, 2012.


13 Brehm, Christoph and Barad, Michael F. and Kiris, Cetin C., “Open Rotor Computational Aeroacoustic Analysis with

an Immersed Boundary Method,” 54th AIAA Aerospace Sciences Meeting, 2016, p. 0815.
14 Capizzano, F., “Turbulent wall model for immersed boundary methods,” AIAA Journal, Vol. 49, No. 11, 2011, pp. 2367–

2381.
15 Berger, M. J. and Aftosmis, M. J., “An ODE-based Wall Model for Turbulent Flow Simulations,” AIAA paper 2017-0528,

2017.
16 Musker, A., “Explicit expression for the smooth wall velocity distribution in a turbulent boundary layer,” AIAA Journal,

Vol. 17, No. 6, 1979, pp. 655–657.


17 Spalart, P. R. and Allmaras, S. R., “A one-equation turbulence model for aerodynamic flows,” AIAA Journal, Vol. 94,

1992.
18 Péron, S., Benoit, C., Gleize, V., Mary, I., and Terracol, M., “A mixed overset grid/immersed boundary approach for
Downloaded by PURDUE UNIVERSITY on June 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/6.2017-3623

CFD simulations of complex geometries,” AIAA paper 2016-2055, 2016.


19 Benoit, C., Péron, S., and Landier, S., “Cassiopee: A CFD pre- and post-processing tool,” Aerospace Science and

Technology, Vol. 45, 2015, pp. 272 – 283.


20 Rumsey, C. L., Wedan, B., Hauser, T., and Poinot, M., “Recent updates to the CFD general notation system (CGNS),”

50th AIAA Aerospace Sciences Meeting, Vol. 10, 2012, pp. 6–2012.
21 Péron, S. and Benoit, C., “Automatic off-body overset adaptive Cartesian mesh method based on an octree approach,”

Journal of Computational Physics, Vol. 232, No. 1, 2013, pp. 153 – 173.
22 Meakin, R. L., “Object X-Rays for Cutting Holes in Composite Overset Structured Grids,” AIAA paper 2001-2537, 2001.
23 Mary, I., “Flexible Aerodynamic Solver Technology in an HPC environment,” Maison de la Simulation Seminars, 2016,

http://www.maisondelasimulation.fr/seminar/data/201611 slides 1.ppt.


24 Boris, J., Grinstein, F., Oran, E., and Kolbe, R., “New Insights into Large Eddy Simulation,” Fluid Dynamics Research,

Vol. 10, No. 4-6, 1992, pp. 199–228.


25 Jameson, A. and Yoon, S., “Lower-upper implicit schemes with multiple grids for the Euler equations,” AIAA Journal,

Vol. 25, No. 7, 1987, pp. 929–935.


26 Coakley, T., “Implicit upwind methods for the compressible Navier-Stokes equations,” AIAA Journal, Vol. 23, No. 3,

1985, pp. 374–380.


27 Laurent, C., Mary, I., Gleize, V., Lerat, A., and Arnal, D., “DNS database of a transitional separation bubble on a flat

plate and application to RANS modeling validation,” Computers & Fluids, Vol. 61, 2012, pp. 21–30.
28 Daude, F., Mary, I., and Comte, P., “Self-Adaptive Newton-based iteration strategy for the LES of turbulent multi-scale

flows,” Computers & Fluids, Vol. 100, 2014, pp. 278–290.


29 Levy, D. W., Laflin, K. R., Tinoco, E. N., Vassberg, J. C., Mani, M., Rider, B., Rumsey, C. L., Wahls, R. A., Morrison,

J. H., Brodersen, O. P., Crippa, S., Mavriplis, D. J., and Murayama, M., “Summary of Data from the Fifth Computational
Fluid Dynamics Drag Prediction Workshop,” Journal of Aircraft, Vol. 51, No. 4, 2014, pp. 1194–1213.
30 Renaud, T., Le Pape, A., and Péron, S., “Numerical Analysis of hub and fuselage drag breakdown of a helicopter

configuration,” CEAS Aeronautical Journal, Vol. 4, No. 4, 2013, pp. 409–419.

24 of 24

American Institute of Aeronautics and Astronautics

You might also like