You are on page 1of 23

AIAA 2011-3387

20th AIAA Computational Fluid Dynamics Conference


27 - 30 June 2011, Honolulu, Hawaii

A Cartesian cut-cell method for sharp moving


boundaries
C. G
unther, L. Schneiders, M. Meinke, and W. Schroder
Institute of Aerodynamics, RWTH Aachen University, 52062 Aachen, Germany

D. Hartmann
Division of Engineering and Applied Science, California Institute of Technology, Pasadena, CA 91125, USA
The extension of a three-dimensional, strictly conservative Cartesian cut-cell based
method to simulate flows bounded by complex geometries with moving parts like in-cylinder
flows is presented. The method handles complex geometric details of three-dimensional
technical objects such as sharp edges and acute-angled features, which may lead to cells
which are multiply cut by the geometry and possibly split into separate parts. Moving
boundaries are represented by the zero-contour of an advected level-set function. The
method is applied to compute full engine cycles. An analysis of the types of cut cells generated for several realistic engine geometries is given. Furthermore, the flow of an internal
combustion engine with a moving piston and opening/closing valves is discussed.

I.

Introduction

Three-dimensional in-cylinder flow simulations are an important tool in engine development to analyze
and understand the characteristic flow phenomena in internal combustion engines. This application puts
special demands on a simulation method due to the complexity of the geometry and the presence of moving
components such as piston and valves. A suitable method must allow simple and efficient mesh generation
and update with respect to the moving parts of the boundary while providing an exact representation of the
geometric details. Immersed boundary methods based on non-boundary-fitted Cartesian grids turn out to be
an excellent choice for this application, as they feature automatic mesh generation and moving boundaries
can be treated without remeshing. The accurate representation of the embedded boundaries in the Cartesian
background mesh, however, is challenging.
Following Mittal and Iaccarino,1 the various approaches to treat immersed boundaries reported in the
literature can be classified into continuous and discrete forcing approaches. Methods using the continuous
forcing approach smear out the interface at the immersed boundary over an area corresponding to a multiple
of the local cell size.2 Moreover, internal flows are very difficult to compute with this technique due to the
influence of the unphysical outside flow (see Peskin3 for a discussion) and appear impracticable for technical
applications such as the flow in internal combustion engines. On the other hand, the discrete forcing approach
is able to retain a sharp fluid-to-boundary interface and does not involve any unphysical outside flow. This
is an important feature when dealing with turbulent flows which makes an accurate description of the flow
close to the immersed boundary essential. Discrete-forcing finite-volume methods, usually termed as cutcell methods, are more difficult to implement but offer a variety of advantages over other discrete-forcing
approaches. They ensure strict global and local conservation of mass, momentum, and energy and provide
a high flexibility with regard to the grid structure, thus being amenable to local mesh refinement. Recently,
several sharp immersed interface approaches on Cartesian grids have been reported in the literature.48
However, there exist only few examples for compressible viscous flow,9, 10 which, however, are not strictly
Research

Scientist, Institute of Aerodynamics, W


ullnerstrae 5a, 52062 Aachen, c.guenther@aia.rwth-aachen.de
Scientist, Institute of Aerodynamics, W
ullnerstrae 5a, 52062 Aachen, l.schneiders@aia.rwth-aachen.de
Senior Scientist, Institute of Aerodynamics, W
ullnerstrae 5a, 52062 Aachen, m.meinke@aia.rwth-aachen.de
Head of Institute of Aerodynamics, W
ullnerstrae 5a, 52062 Aachen, office@aia.rwth-aachen.de
Postdoctoral Scholar, Division of Engineering and Applied Science, 1200 East California Boulevard, Pasadena, CA 91125,
hartmann@caltech.edu
Research

1 of 23
Copyright 2011 by Institute of Aerodynamics. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.

American Institute of Aeronautics and Astronautics

conservative. A novel approach which allows the computation of viscous compressible three-dimensional
flows in a strictly conservative way has been reported by Hartmann et al.11, 12
Due to the geometric intricacy of the grid-boundary intersection in three space dimensions, the number of
three-dimensional implementations of related cut-cell methods is limited and only few approaches proposed
in the literature are able to handle non-smooth, three-dimensional technical objects of arbitrary complexity,
such as internal combustion engines. The method of Hartmann et al.11, 13 as well is restricted to rather
smooth geometries as will be demonstrated in section II.B. Based on this method, the purpose of this
study is to develop a conservative Cartesian cut-cell method for the three-dimensional compressible NavierStokes equations, which allows multiple and complex intersections of the boundary with the cut cells. Being
coupled to a multi-purpose level-set solver12 used for the boundary representation, the presented approach
is extended to problems involving general moving boundaries described by a signed-distance function. A
continuous discretization in time near the moving interface prevents nonphysical oscillations in the solution.
The modeling of engine specifics such as the opening and closing events of the valves is done by taking
advantage of the level-set representation of the boundary.
The organization of this paper is as follows. Section II outlines the developed numerical methods for
complex cell-boundary intersections, boundary movement, and opening/closing of the valves. In section III,
the results for the validation of the method are summarized. Numerical findings of the flow in internal
combustion engines in steady flow cases as well as the flow in simplified engine geometries considering the
movement of the valves and the piston are shown in section IV. Finally, the findings are summarized in
section V.

II.

Numerical method

The method approximately solves the nondimensionalized Navier-Stokes equations for compressible fluids
in integral form
Z
Z
Q
dV +
H n dA = 0,
(1)
A
V t
where n is the outward unit normal vector on the surface dA on locally refined Cartesian grids. The quantity
Q = [%, %v, %E]T is the vector of the conservative variables and H is the flux vector containing an inviscid
i
v
part H and a viscous part H

%v
0
1
i
v

H=H H =
(2)
%vv
+

.
Re
%v(%E + p)
v + q
The discretization of these equations on the inner computational domain, i.e., not at the immersed
boundaries in space and time is not altered compared to the original approach for smooth boundaries.
Details can be found in.11
II.A.

Embedded boundary treatment

In Cartesian cut-cell methods, cells are usually classified in three categories. These are internal cells located
completely inside the fluid domain , external cells located completely outside the fluid domain , which
are discarded, and boundary or cut cells intersected by the geometry. The boundary cells are reshaped such
that the entire control volume V n of the reshaped cell lies inside the computational domain as illustrated
n . Finally, following Hartmann et al.,11
in Fig. 1. The cell center is shifted to the new volumetric cell center x
the application of boundary conditions on the boundary surfaces n is done by prescribing the primitive
variables on ghost cells.
II.B.

Non-smooth geometries

The treatment of the cut cells and ghost cells in the original method11 requires that the geometric intersection
of a boundary cell can be described by exactly one surface, see Fig. 1. However, in technical applications, the
embedded boundaries usually contain sharp edges. Their presence, often in conjunction with convex features
and acute angles, can lead to cells multiply cut by the geometry, where two or more boundary surfaces are

2 of 23
American Institute of Aeronautics and Astronautics

fluid face
cut faces
cut faces

outside domain
z

Figure 1. Reshaping of a Cartesian cut cell. The dark shaded surface is the resulting boundary surface, the light
shaded outside part of the cell is omitted.

needed to describe the boundary cell resulting from the geometric intersection. Cells which are split into
separate parts by the geometry may occur as well, as illustrated in Fig. 2(a). A possible remedy to this
problem is further refinement of these cells. Unfortunately, in various rather common cases refinement does
not directly eliminate all the multiple-cut cells, see Fig. 2(b). Since abrupt changes in the grid spacing are to
be avoided in sensitive flow regions, it is usually not sufficient just to refine the multiple-cut cells. This in turn
increases the probability that new multiple-cut cells on the new cell level occur. Consequently, computational
costs are raised without increasing the overall solution quality. However, any necessity for user intervention
must be avoided to keep the benefits of automated meshing, especially regarding the treatment of moving
boundaries. Hence, a more general and robust representation of geometric intersections is required.

(a)

(b)

Figure 2. Cells multiply cut or split by the geometry in Cartesian grids. (a): Geometries for combustion engine
simulations. (b): Common generic configuration of two planes at an angle forming a sharp edge. A detail of the grid
at the highest refinement level is shown together with problematic cells at three different refinement levels.

II.B.1.

Multiple ghost cell formulation

To account for the specifics of flows bounded by complex geometries, we have to omit the rule used in
Hartmann et al.11 that each boundary cell contains exactly one boundary surface with one ghost cell
attached to it. Instead, multiple separate boundary surfaces, each equipped with a particular ghost cell, are
allowed to be attached to a cell. Contrary to the previous approach, the ghost cell center x
i for surface i is
located at:



x
i = xs i + x
xs i ni ni ,
(3)
as depicted in Fig. 3. Here, x
is the centroid of the reshaped cut cell, xs i is the centroid of the respective
boundary surface i, and the normal ni of the surface is assumed to point into the outside domain. For the
application of boundary conditions, an additional interpolation point xnIP is introduced which is computed
by mirroring the ghost cell along the boundary surface. To compute the variable values on the ghost cell,
the variables are first interpolated to the matching interpolation point xIP . Then, boundary conditions can
be applied by setting the ghost cell value n to
nN = nIP

or

nD = 2 nIP

(4)

for Neumann and Dirichlet type boundary conditions, respectively. This approach ensures that the prescribed value in the boundary surface centroid is recovered by a linear interpolation between ghost cell and
interpolation point. In addition, a meaningful continuation of the flow field is provided for the gradient reconstruction on the boundary cells. For the computation of the wall shear stresses, the approach of Hartmann
et al.11 for adiabatic no-slip walls can easily be adopted.
3 of 23
American Institute of Aeronautics and Astronautics

~
X1
n1
Xs1
2

~
X2

Xs2
n1

XIP
1
^ XIP
X

Figure 3. Representation of the boundaries in a cell with cell centroid x


which is multiply cut by the geometry using
one ghost cell located at x
i and the corresponding interpolation point xiIP per boundary surface i.

II.B.2.

Computation of cut cell information

The computation of the generalized reshaped cut cells is a nontrivial task. Therefore, we have implemented
a modification of the Marching Cubes algorithm.14, 15 Based on information about the location, i.e., inside/outside the fluid region, of the corner vertices of a Cartesian cell, an index can be calculated which
identifies the correct configuration for the examined cell out of a list of 256 possible polygon configurations
within the cell. Using the cut points on the cell edges, all necessary cut cell information can then be generated
by basic geometric computations. Care must be taken to choose the correct topology for some ambiguous
configurations by using additional connectivity information from the description of the embedded geometry
or by comparing the case information of neighboring cells. Special attention has to be paid to so-called split
cells which are divided by the geometry into two or more separate unconnected parts, as sketched in Fig. 4.
For such cells, the neighbor information has to be updated for the resulting cell parts and their respective
neighboring cells. The less restrictive treatment of boundary cut cells allows to preserve relevant features of
the geometry such as distinct edges or interfaces with different adjacent boundary conditions. To achieve
this, a cut cell representation with multiple boundary surfaces must be created taking into account additional
points where the edges or interfaces to be sustained cut through the cell faces. Further details on this can
be found in G
unther et al.16
II.B.3.

Treatment of small cells

To avoid problems with numerical instabilities generated by cells of very small volume emerging at the
embedded boundaries, a combined cell-merging/cell-linking approach is adopted. This approach merges the

^2
X

^1
X

Figure 4. Split cell which is divided by the geometry into two separate unconnected parts. The two parts are shaded
in light red and orange, and their cell centroids x
n are displayed in the respective color. Connections to neighboring
cells which have to be updated are indicated by dotted arrows.

4 of 23
American Institute of Aeronautics and Astronautics

volume of a small cell with a neighboring cell of sufficient volume, called the master cell. The volumetric
centers of both cells are shifted to the volumetric center of the merged cell volume. The surfaces between the
original master and slave cells are discarded, while the other surfaces are retained and linked to the combined
cell. The boundary surfaces of both master and small cells are not merged as proposed by Hartmann et al.11
but kept unchanged and are also linked to the combined cell to preserve the topology of the combined cell.
II.C.

Boundary motion

The basic extension of the presented method to general moving boundary problems is straightforward.
According to the new boundary position after each time step, the cut-cell approach reduces the problem of
adjusting the computational grid to the recomputation of the boundary cells cut by the moving parts of the
geometry. The representation of the moving boundaries, the general numerical discretization close to the
moving interface and the treatment of engine specifics such as the opening and closing events of the valves
and the piston movement will be discussed in the following subsections.
II.C.1.

Level-set representation of the boundary

A suitable approach to describe the moving boundary is the representation by a scalar level-set function
(x, t), which is specified as a signed distance function with respect to the moving boundary. This means
the body outline is given by the zero contour of the level set function = 0 termed 0 . Points outside
the flow field are indicated by a negative value < 0 and cells inside the fluid domain are indicated by a
positive value > 0. At prescribed boundary motion a given spatially constant extension velocity vector
f (t) is introduced such that the evolution of in time is defined by

+ f = 0.
t

(5)

The computation of uses a highly accurate level-set method13, 17, 18 coupled to the flow solver in a dualmesh approach.12 This methodology integrates well into the previously described modified Marching Cubes
approach used for the generation of the cut-cell information. The cut points on the edges of a boundary cell
can be obtained by interpolating the scalar level-set function while its sign directly carries the information
if a point is located inside or outside the fluid domain.
To unify the representation of fixed and moving geometries, all parts of the geometry are implicitly
described by a signed distance function instead of an explicit description by e.g., a surface triangulation.
Usually for technical applications such as combustion engines, the geometry can be divided into separate
parts which are either fixed or moving with a unique prescribed velocity function. As long as all these different
parts or groups are located far from each other, they can be represented using a single level-set function on
one grid without losses in accuracy. The identification and labeling of grid cells corresponding to the separate
parts is a necessary prerequisite for the assignment of the now spatially variable level-set extension velocity
f (x, t) in the complete level-set domain and for the application of correct velocity boundary conditions on
the cut cells in the flow grid. If all bodies are clearly disconnected at sufficient distance, this task is trivial.
A simple way to do this is by starting at a point xb known to be located inside the respective body with
(xb ) < 0, marking it with the respective body Id b and subsequently marking all unmarked neighboring
cells of already marked cells, until the zero contour of the level-set function is reached. This procedure is
repeated for all bodies. The level-set extension velocity f (x, t) is then prescribed inside the identified bodies
and on their boundary with the respective prescribed body velocity. A hyperbolic extension ensures a valid
and meaningful velocity prescription in the remainder of the level-set domain.
However, if two parts are very close to each other, the local resolution of the level-set grid may not
be sufficient to describe both bodies without losses in accuracy and correctness of the geometry, which is
illustrated in Fig. 5. In such cases or when different parts overlap, the identification of the different parts
is complex and error-prone. The incorrect specification of the extension velocity f (x, t) may lead to a rapid
deterioration of the level-set representation of the original rigid body geometry and the wrong imposition of
the velocity boundary condition on the flow grid leads to strong disturbances in the flow solution. Therefore,
in our approach the different bodies and parts are represented by different level-set functions i , i > 0. After
the separate level-set functions i have been updated in each time step, a combined level-set function 0
together with the correct body Id labeling b0 is assembled using the following algorithm for each level-set
grid cell:
5 of 23
American Institute of Aeronautics and Astronautics

Figure 5. Close-up of a valve approaching the valve seat. Here, the level set resolution is too low to capture both bodies
correctly. The white line indicates the interpolated zero contour of the level-set function which is strongly disturbed
between the two bodies. Colors show the level-set function . The original bodies are indicated by the gray shaded
surfaces.

1. Store the level-set value 1 of the first separate level-set function in 0 and set the body Id b0 = b1 .
2. For all subsequent level-set functions i , i > 1 compare i and 0 :
If both values are positive or equal zero, set 0 = min(0 , i ) and b0 to the corresponding body
Id.
q
2
If both values are negative or equal zero, set 0 = 0 2 + i . Choose the body Id b0 corresponding to the maximum absolute value |0 | and |1 |.
If the values carry different signs, set 0 = min(0 , i ), i.e., choose the negative level-set value.
Set the body Id b0 correspondingly.
The flow solver uses only the combined level-set function 0 and the body Id information b0 corresponding
to this data set for the cut cell computation and the imposition of velocity boundary conditions, so the
interface between the solvers does not change if a single or multiple level-set functions are used to describe
the geometry.
For all level-set functions i , 0 i, the same grid is used which is is created, refined, and updated based on
the combined level-set function 0 . Due to the narrow band approach implemented in the level-set solver,12
each separate level-set function i , i > 0 is only computed on a small part of the level-set grid belonging to
the respective computing band, as sketched in Fig. 6. This means that the computational overhead remains
restricted to the computation of the combined level-set function 0 using the previously described algorithm
and the identification and maintenance of the different zero levels i0 and computing bands of the separate
level-set functions i . The memory requirements are increased though, since all properties which differ for
the level-set functions i need to be stored for each function separately.
While for technical applications such as internal combustion engines, the number of different parts nmax
remains relatively small, this is not necessarily the case for other applications where this approach could be
useful, e.g., the motion of a cluster of rigid particles in the flow field. To reduce the memory cost, a dynamic
split/merge procedure is applied. Using this procedure, most of the bodies are represented in a main level-set
function m . If the minimum distance of two bodies represented by this single level-set function m drops
below a critical level, the level-set function is split such that one body b and its corresponding signed distance
function is transferred to another level-set function i , i 6= m where the minimum distance to other bodies
is guaranteed. When no conflicts exist anymore, the body can be transferred back to the previous level-set
function i . Thanks to the implemented high-order constrained reinitialization equation,12, 13, 17 only the
level-set values on the cells intersected by the zero-contour 0 corresponding to the respective body b have
to be transferred, since the reinitialization process builds up a valid signed distance function in the complete
computing band. When the approach of different bodies can be assumed to occur only infrequently or if the
number of different bodies nmax is small, a rather simple algorithm is sufficient to decide in which level-set
function i the transferred body b should be placed. The most simple alternative is to create a new, separate
level-set function for such a body which is cleared after the body has been re-merged to the main level-set
function m . Otherwise, a more elaborate algorithm is needed which tries to minimize the total number of
different level-sets needed to hold all bodies, e.g., graph coloring algorithms. Here, a tradeoff must be made
between memory requirements and computational complexity of the split/merge procedure.

6 of 23
American Institute of Aeronautics and Astronautics

side walls

valve

piston

Figure 6. Level-set representation of a simplified two-dimensional engine geometry. The geometry contains four
different bodies each of which is described by its own level-set function. The computing grid valid for all separate
level-set functions i and the combined level-set function 0 is shown. The level-set functions i for each separate
component on the respective computing band together with the zero-level i0 are displayed with an offset normal to
the plane for better visualization. The zero-levels of the separate level-set functions i0 are also shown as yellow lines
in the grid.

II.C.2.

Time-continuous discretization near the moving boundary

If the treatment of small cells with the combined cell-merging/cell-linking approach as explained in section
II.B.3 is kept unchanged for cells cut by the moving boundary, the local grid structure and discrete operators
in the vicinity of the moving boundary change discontinuously related to the merging and separation of small
cells, as discussed by Hartmann et al.19 Such discontinuous changes in the spatial discretization operators
lead to strong numerical oscillations in the solution. To avoid this effect, small cells originating at the moving
boundary are not merged. Instead, on very small cells with a cell volume V below a certain threshold value,
e.g., V < 0.5hd , where h is the cell length in regular grid areas and d is the number of space dimensions,
a second variable update is computed by interpolation in addition to the variable update done by a usual
time integration step. Unlike the conservative variable update, interpolation preserves stability but the
formulation is no longer conservative. A smooth transition between both variable updates is ensured by
applying a continuously differentiable blending function depending on the cell volume. At the same time,
the blending function ensures that the influence of the interpolated variable update is as small as possible.
To correct the conservation error due to the interpolation, a flux redistribution technique20, 21 on a larger
stencil of conservation cells is introduced.
The gradient reconstruction stencil and thus the computed gradients on near-boundary cells are still
subject to discontinuous changes when cells are created or disappear due to the boundary motion and
when uncut fluid cells become cut cells and vice versa. A weighted least-squares approach with inverse
distance weighting is applied for the gradient reconstruction on the near-boundary cells. The inverse distance
weighting is improved such that additionally the weights of smaller cells in the stencil are reduced depending
smoothly on the distance between the cell center and the moving boundary. Thus, newly created cells are
introduced continuously in the reconstruction stencil of their surrounding cells and cells to be submerged
are continuously faded out, thus leading to a smooth change in the computed gradient values. Finally, the
ghost cells which are needed in the reconstruction stencil of the cut cells are introduced beforehand in the
stencil when the distance of an uncut fluid cell to the boundary drops below a certain value. Also, their
weights are smoothly increased as the boundary distance of the fluid cell decreases to avoid their sudden
appearance in the stencil when the fluid cell is cut. Together, these measures eliminate abrupt changes in the
spatial discretization in the presence of moving boundaries and retain a high quality of the solution without
boundary-movement-induced oscillations.

7 of 23
American Institute of Aeronautics and Astronautics

II.C.3.

Valve opening and closing

The simulation of in-cylinder flows not only requires a method for moving boundaries and a complex geometry, also topological changes of the domain must be handled when the valves approach the valve seat
and finally close, such that the computational domain is decomposed into different separate parts. When
boundary fitted grids are used,2224 the cells in the valve gap are usually compressed using an arbitrary
Lagrangian-Eulerian approach during valve closing until a certain minimum thickness is reached. Then, the
valve gap region is either remeshed or, in the case of structured meshes, one cell layer is removed and the
remainder of the cells is adjusted to fill the entire valve gap again. In either case, the variables must be
interpolated to the new grid. When a certain minimum valve lift corresponding to a minimum gap width
is reached, all remaining cells in the gap are removed and the interface to the neighboring cell layers is
treated as solid wall. The strong mesh deformations and the interpolation introduce additional numerical
errors to the solution. Immersed boundary methods, however, in the forcing approach or the cut-cell form
avoid remeshing and interpolation and do retain a good grid quality. Unfortunately, a general and robust
treatment of situations where different rigid bodies that are fixed or moving at different velocities approach
each other closely is not trivial for Cartesian cut-cell methods and a straightforward solution has not been
reported in the literature to the best of the authors knowledge. In the simulation of Verzicco et al.25, 26
which represents the first application of an immersed boundary method to in-cylinder flow, the valve is fixed
and the piston is modeled as a separate part with a finite distance to the cylinder walls. Senecal et al.27
report a cut-cell method for in-cylinder flow, which does not use a level-set representation of the boundary
but is based on surface triangulations of all engine parts instead to model complete engine cycles. It has
the possibility of performing a discrete shift of the valve triangulation to the closed position when a certain
minimum valve lift is reached. Further attempts to develop methods based on immersed boundary or cut-cell
approaches for in-cylinder flows including the motion of piston and valves are very rare.
The reason for the difficulties is that, analog to boundary fitted approaches, most methods require each
cell to have a minimum number of neighbors to guarantee a valid discretization stencil. This will not be the
case if the gap between the bodies gets small compared to the cell size. Even if boundary conditions can
be used instead of neighbor information or the stencil size can be decreased, the quality of the solution will
deteriorate rapidly when the gap resolution becomes poor. Mesh refinement can help to reduce the resolvable
gap width, but at some point a sudden transition to the closed state is necessary. An automated general
and robust approach for the discrete closure of non-resolvable gaps must
1. identify the presence of such gaps without user intervention and information about the system state,
2. identify all cells located in the gap which will be removed or become outside cells, and
3. provide a valid boundary description which represents the closed state.
While these conditions can easily be satisfied by boundary fitted, structured methods based on grid inherent
information and user input during mesh preprocessing as sketched above, i.e., condition 1 is weakened, an
approach fulfilling all these requirements being valid for Cartesian cut-cell and general immersed boundary
methods has not been reported in the literature, yet.
Identification of unresolvable gaps In the present method, the level-set representation of all boundaries is used to identify unresolvable gaps and to provide a valid boundary description in which the gap is
closed and thus eliminated. As discussed earlier in section II.C.1, the quality of a multi-body representation
using a single level-set function decreases if the distance
between two bodies b1 ,b2 is smaller than a few cell
lengths h. If the distance satisfies b1 ,b2 < min = 2h, the topology of the bodies can not be maintained
and originally unconnected surfaces receive artificial connections, see Fig. 5. Since this limit also marks the
minimum resolution of gaps the flow solver is able to work with reliably, it is chosen as limit distance min
to initiate gap closure. An approximation of the distance i,b1 ,b2 between the body b1 which is closest and
the body b2 which is second-closest to a cell i can be computed using
i,b1 ,b2 |bi 1 | + |bi 2 |,

(6)

where bi 1 and bi 2 denote the signed distances held in the level-set functions corresponding to b1 and b2 .
Assuming exact values for the signed distance functions, this relation is exact for parallel walls and switches
over to an inequality if the angle between the walls increases. Note that it is indispensable that both
8 of 23
American Institute of Aeronautics and Astronautics

bodies are represented by different level-set functions, since otherwise the distance to the second body bi 2
would not be available. Strictly speaking, relation (6) is only valid on cells where both bi 1 and bi 2 are
positive, so a reasonable extension is provided by using the absolute values. This makes sure that cells with
a centroid located in the outside domain possesses a higher value i,b1 ,b2 than neighboring cells located in
the
fluid domain. The increment for such cells intersected by the corresponding body contour is at most

2
h.
Assuming the second closest body for each cell is known, the presence of a non-resolvable gap and the
2
cells located inside the gap are thus identified using the following procedure:
1. Compute i,b1 ,b2 for all cells i.

2. If i,b1 ,b2 < 2h on one or more cells i, a gap is present. Mark the corresponding cells as gap cells.

3. In this case, mark all the unmarked neighbor cells i of the gap cells if i,b1 ,b2 < 23 2 h holds for these
cells. Repeat until no cell is newly marked.
This ensures that once a resolvable gap is detected, all the cells located in this gap are reliably identified,
including cells the centroid of which is located in the outside domain.
Boundary representation for the closed state To provide a valid description of the boundary
where the gap is eliminated, the level-set function 0 is reinitialized. If a gap is detected, the combined
level-set function 0 is incremented by half of the minimum resolvable gap width min /2
0 0 + min /2.

(7)

00 = {(x) : 0 (x) = 0}

(8)

Then, the zero level


of the resulting level-set function is determined. The zero level is shifted exactly such that it describes
a closed contour where the gap is closed and does not follow the separate body contours with a gap in
between. Subsequently, reinitialization is performed based on this new, shifted zero level. This constructs a
signed distance function whose level contours correspond to the closed contour inside the former gap and its
neighboring region, but represents the original situation in the remainder of the domain. It is a must to use
a reinitialization procedure which conserves the zero level of the level-set function such as the constrained
reinitialization developed by Hartmann et al.13, 17 When the reinitialization is finished, the level-set function
is shifted back to the original levels by applying the reverse shift
0 0 min /2.

(9)

The result of this procedure is a valid collected level-set function 0 in the complete level-set domain which
is initialized as signed distance function with respect to a boundary description where the gap is closed.
The corrected function 0 is known to the flow solver together with the cells marked as gap cells which
were formerly located inside the gap and must be removed during the current time step. To reduce the
computational costs for the reinitialization, this procedure should not be applied to the complete level-set
domain but only to the cells inside a region close to the gap. Note that the underlying separate level-set
functions i are not modified by this routine, that is, no information gets lost and no disturbances are
introduced in these functions. The result of the closing procedure is illustrated in Fig. 7.
As long as the gap should remain closed, i.e., the gap width is unresolvable, this procedure should be
applied, even if the gap width tends to zero and the gap would effectively be closed without applying the
procedure, too. This ensures a smooth boundary description at the interfaces between the bodies and no
jumps of the boundary occur in time due to boundary movement. In addition, overlaps between two bodies
can be treated consistently which renders the method extremely robust in case the boundary description is
not perfect or the prescribed body motion is not exact. This makes this approach very attractive for fluidstructure interaction problems where contact between rigid bodies may occur. However, the sharp interface
between two bodies which are in contact is not preserved by this approach but smoothed over a few cell
lengths. In cases where the preservation of this sharp interface is important, the integration of the previously
developed multiple ghost cell formulation in the moving boundary context is required. Alternatively, further
refinement in the interface region reduces the smoothing effect as min is reduced.

9 of 23
American Institute of Aeronautics and Astronautics

(a)

(b)

Figure 7. Gap between a simplified valve and the valve seat before (a) and after (b) the closing procedure is applied.
The original bodies are displayed by the gray shaded surfaces. The red lines indicate the body outlines, i.e., the zero
levels, as represented in the separate level-set functions i . The thick white line indicates the zero level of the combined
level-set function 0 . The colors correspond to the value of the combined level-set function 0 . Thin white lines are
contours of the combined level-set function 0 plotted in intervals of 0 = 0.005 in the range 0 = {0.03, . . . , 0.03}

Reopening of a gap To reopen a gap in the combined level-set representation 0 when the gap width is
large enough, it is sufficient not to call the reinitialization procedure after the time update of the separate
level-set functions i . Since the flow cells located in the previously closed gap have no valid time history,
smooth initial values for the velocities, pressure, and density are provided by solving a boundary value
problem on these cells using the Laplace equation. The necessary boundary conditions are provided at the
cut cells intersected by the body surfaces forming the gap and at the cells on the border of the gap which
also existed in the previous time step and therefore have a valid time history. Dirichlet boundary conditions
are used for the velocity using the cell velocity of the previous time step, if available, and the respective body
velocity otherwise. For the pressure and the density, Dirichlet boundary conditions are used at cells with a
valid time history and homogeneous Neumann conditions are applied on the remaining cut cells enclosing
the gap. The resulting initialization of the cells in the gap is shown in Fig. 8.

(a)

(b)

(c)

Figure 8. Smooth initialization by solving the Laplace equation on the cells inside a gap after reopening: (a): velocity
in the x-direction, (b): pressure, (c): density. Cells which have a valid time history which are not initialized by solving
the Laplace equation are shaded dark.

As explained above, in the present approach the cells disappearing in an artificially closed gap are removed
together with their remaining mass, momentum, and energy content without performing a conservative
redistribution on the cells in the vicinity of the gap. This is common in many approaches for structured, body
fitted solvers. Likewise, the cells appearing in a reopened gap are initialized without enforcing conservation
with the surrounding cells. It is evident that the lost or gained mass, momentum, and energy during both
opening and closing is proportional to the minimum resolvable volume in the gap. Therefore, refinement
will definitely reduce this effect. The consequences of this simplification will be discussed in more detail in
section IV.C.
Prescription of velocities in the vicinity of a closed gap The prescription of correct boundary
conditions on moving boundary surfaces is crucial for accuracy and stability of the flow solver. If the
prescribed body velocity does not match the rate of change in the cell volume on a particular boundary
cell, i.e., if the so-called geometric conservation law28 is not fulfilled, strong unphysical oscillations in the

10 of 23
American Institute of Aeronautics and Astronautics

solution occur.29 The strict compliance with this geometric conservation law is a general problem for cutcell methods.30 We found that for rigid bodies the prescription of the body velocity on the body surface in
general produces good results. However, in the vicinity of an artificially closed gap resulting from the closing
procedure, where at least one of the bodies is not at its final closed position and still moves with a nonzero velocity, the body velocity on the cut cells between the bodies is not well defined anymore. Therefore
care must be taken to specify on each cell a velocity which matches the actual displacement velocity of the
boundary in this cell as closely as possible and prevents any jumps once the gap has been closed. In our
method, interpolation of the velocities corresponding to the two bodies forming the gap in a small region
close to the gap has proven to yield reliable results. Refinement reduces the extent of the smoothed region in
the vicinity of the closed gap and thus, the effect of an inappropriate prescription of the velocity boundary
conditions is reduced as well. A further discussion of this issue is provided in the following section II.C.4.
Refinement provides a solution to generally increase the accuracy during valve closing and opening and
to reduce the uncertainties related to a non-conservative formulation induced by the discrete grid changes.
Therefore, it is certainly advantageous to include an adaptive refinement procedure in the process of closing
and opening the valves, which refines the cells in the valve gap region repeatedly when the resolution in
the valve gap falls below a certain limit. The closing procedure is finally initialized when the maximum
refinement stage is attained and the corresponding minimum gap width min is reached. When the valve
reopens again, the cells may be gradually coarsened until the initial refinement level is regained.
The capability provided by this method to automatically close small gaps arising between different rigid
bodies when the resolution is not sufficient anymore certainly renders the Cartesian cut-cell method more
robust and opens up various other applications where different bodies or structures can contact each other,
such as fluid-structure interaction in biological applications, e.g., artificial heart valve simulations or the
simulation of phonation.
II.C.4.

Piston movement

The description of the piston movement is rather simple in boundary fitted methods, as this can be done
by an arbitrary-Lagrangian-Eulerian-type mesh compression. In immersed boundary methods, the piston
can either be modeled as an independent part of the geometry with a small distance to the cylinder liner or
the size of the piston can exactly match the engine bore, thus maintaining constant contact to the cylinder
walls. Verzicco et al.25, 26 chose the first possibility in their simulations to avoid the necessity to treat the
sliding interface between the piston and the cylinder walls. However, in this case the gap between the piston
and the cylinder walls has to be resolved properly and the computational domain has to be extended to
include the fluid on the lower side of the piston as well, which may considerably increase the problem size
and introduces additional sources of error due to the complex behavior of the flow close to the piston gap.
Using the framework developed for the treatment of valve closing, the present method is able to model the
piston according to the second possibility.
If the piston is prescribed such that it fits exactly in the cylinder liner or even penetrates the walls, the
problem is similar to that of modeling the movement of a valve up to its final position, when the valve gap
has been artificially closed. If no special treatment is applied, the boundary description on cells being cut
by the piston and cylinder wall jumps discontinuously due to the first-order approximation of the boundary
in the fluid cells. This introduces large oscillations in the flow solution. At the same time, the problem to
specify the correct velocity boundary condition on these cells remains. In contrast, when the interface is
treated analogously to real gaps as illustrated in Fig. 9 with the procedure described in the previous section
II.C.3 without a special treatment, the occurrence of jumps is prevented and the approximate specification
of velocity boundary conditions on the affected cells is simplified. Also in this case, the prescription of
interpolated velocities based on the velocities of the two bodies in contact is a good choice, since this
provides a smooth transition of the velocities between the two bodies and is able to approximately mimic the
actual boundary displacement. Nevertheless, it has to be stated that we encountered a very strong influence
of the interpolation function.
Hence, the representation of the interfacial cells with two separate boundary surfaces in the original setup
using the multiple ghost cell formulation as described in sections II.B.1 and II.B.2 without the smoothing
treatment is favorable. The introduction of a higher-order representation of the boundary on such selected
cells allows to preserve the original well-defined sharp edge between the piston and the cylinder walls. This
is indicated by the red lines in Fig. 9. Consequently, this interface moves continuously and discrete jumps
of the geometry on the interfacial cells are prohibited. Likewise, the boundary conditions can be prescribed
11 of 23
American Institute of Aeronautics and Astronautics

Figure 9. Piston penetrating the walls in a 2D engine setup with a close-up in the intersection region (cf. Fig. 6). The
red lines indicate the contours of the piston and the cylinder walls, as represented in the separate level-set functions
i . The colors correspond to the value of the combined level-set function 0 resulting from the gap treatment of the
intersection region. The white line indicates the corresponding smoothed combined contour.

without further complications corresponding to the respective body velocity, substituting the crucial velocity
interpolation function on these cells. The application of the multiple ghost cell formulation to such cells cut
by two different bodies at least one of which is moving will be part of our future work.

III.

Validation

The basic numerical method was validated for various generic configurations with curved embedded
bodies where no cells multiply cut by the geometry occur.11, 31 In addition, validation results for reacting
flows are available.12 The deviation in the results when the extended formulation for complex geometries is
applied to these test setups is negligibly small.
Uniform flow past a cube To demonstrate the correct modification of the flow field when multiple ghost
cells are attached to one cell, we consider the laminar three-dimensional flow past a cube. As illustrated in
Fig. 10, the extended method allows to preserve the sharp leading edges important for the flow separation
in their real position, while the original method produces chamfered edges if the geometry is not perfectly
aligned with the mesh, thus altering the flow field. The cube centroid is located in the center of the
coordinate system and the x-direction is the streamwise direction. Two different meshes are used. With L
being the edge length of the cube, a relatively coarse mesh width h 0.055L in the close vicinity of the
cube, corresponding to a resolution of 18 grid points on each side of the cube, is used to increase the effect
of the surface modification. The finer mesh uses a minimum mesh width h 0.03125L which corresponds
to a resolution of 32 grid points on each side of the cube. The grid resolution in the vicinity of the cube
is roughly equivalent to the coarsest and medium resolution Saha32 used in in his study of the flow around
a cube. The extended method is expected to have a bigger influence on the solution on the coarse grid
than on the fine grid, since grid refinement decreases the discrepancy between the original geometry and the
representation generated by the original approach. We compute the flow at a Mach number M = 0.1 and a
Reynolds number ReL = 100 based on the freestream values, where the flow field is steady and symmetric.
Fig. 11 shows a comparison of the streamlines in the z-symmetry plane for both grids when the original
method assuming a single boundary surface per cut cell or the modified method for complex geometries
is applied. All setups are able to reproduce the basic flow field features reported by Saha.32 Deviations

Figure 10. Comparison of the representation of a cube with sharp edges using the original method (left) and using the
present method (right). The edges are chamfered using the original approach due to the first-order approximation of
the boundary in the cut cells.

12 of 23
American Institute of Aeronautics and Astronautics

(a)

(b)

(c)

(d)

Figure 11. Uniform flow past a cube at ReL = 100. Streamlines on z-symmetry plane: (a) coarse grid, chamfered edge
(original method), (b) coarse grid, sharp edge (present method), (c) fine grid, chamfered edge, (d) fine grid, sharp
edge.

in the prediction of the thickness of the separation region and the length of the recirculation zone can be
observed. To illustrate this fact more clearly, Fig. 12 compares the velocity magnitude and streamlines and
the pressure obtained on the fine grid in the z-symmetry plane. The upper part of the respective picture
shows the solution using the modified method for complex geometries, while the solution assuming a single
boundary surface per cut cell is depicted below. The velocity and the pressure field show visible differences,
especially in the length of the recirculation region and the corresponding pressure distribution behind the
cube. Also, the displacement of the separation point on the leading edge produced by the basic approach
and the corresponding difference in the shape of the low pressure region are evident.
The x-velocities on the x-axis are displayed in Fig. 13 for all four setups and both grids. While the
recirculation length predicted by the basic method on the coarse mesh is evidently too short, it is found that
the predicted recirculation length lc using the extended approach resulting in a sharp edge is close to the
value of lc = 2.00 found by Saha32 both for the coarse and the fine mesh. As expected, the difference in the
solutions on the coarse mesh is bigger than on the fine mesh. However, the recirculation length predicted by
the original method on the fine mesh is still smaller than that predicted on the coarse mesh by the extended

(a)

(b)

Figure 12. Uniform flow past a cube at ReL = 100. Upper part of the pictures: solution with the extended method
(sharp edge); lower part of the pictures: solution with the basic method (chamfered edge). (a) Velocity magnitude and
streamlines (z-symmetry plane), (b) pressure.

13 of 23
American Institute of Aeronautics and Astronautics

coarse, sharp edge


coarse, chamfered edge
fine, sharp edge
fine, chamfered edge

0.03

0.02
0.01

Table 1. Comparison of the recirculation length lc


for the uniform flow past a cube at ReL = 100 obtained with the extended method (sharp edge) and
the original method (chamfered edge).

lc

-0.01
-0.02

0.5

1.5

2.5

x/L

sharp edge

chamfered edge

1.9752
1.9831

1.8942
1.9529

coarse grid
fine grid
Saha32

2.0

Figure 13. x-velocity component on the x-axis downstream of


the cube. Thick red lines indicate the solution with the extended
method (sharp edge), thin blue lines indicate the solution with
the basic method (chamfered edge). Solutions on the coarse grid
are shown with a dashed line, solutions on the fine grid are shown
with a continuous line.

approach. This shows that the treatment of a sharp feature which is of major importance for the flow field
using the extended approach as proposed in section II.B.2 may considerably reduce errors due to low mesh
resolution in the vicinity of the feature.
Vortex induced vibrations of a circular cylinder To validate the numerical method for the treatment of moving boundaries, Hartmann et al.33 conducted simulations of a generic fluid-structure interaction
setup, the vortex induced vibrations of a circular cylinder. The flow past an elastically mounted circular
cylinder with diameter D restrained to move in the cross-flow direction in response to the fluid and elastic forces is investigated, ignoring the influence of the spring mounting on the flow. The set-up is shown
schematically in Fig. 14(a). The Mach number is M = 0.1 and the Reynolds number based on the freestream
values and the cylinder diameter is ReD = 150. For this setup, the cylinder is known to couple with the
unsteady vortex shedding process when the vortex formation frequency comes close to the natural vibration
frequency of the cylinder-spring system. This coupling results in a vibration of the cylinder in response to
the unsteady lift force due to the vortex shedding. The initial position of the cylinder center is slightly offset
in the y-direction such that a spring force is generated at t = 0, which in turn initiates the oscillatory motion
of the cylinder. In Fig. 14(b) the results for the maximum amplitude of the cylinder obtained for different
values of the parameter Ured are compared with results given in the literature by Ahn & Kallinderis34 and
Borazjani et al.35 The present results are in convincing agreement with the reference data. For more details
the reader is referred to Hartmann et al.33

(a)

(b)

Figure 14. Vortex induced vibrations of a circular cylinder at ReD = 150. (a) Sketch of the test case set-up. (b)
Maximum amplitude Ymax of the cylinder as a function of the spring stiffness.33

14 of 23
American Institute of Aeronautics and Astronautics

IV.
IV.A.

Results

Cut cell generation for different internal combustion engine geometries

We first analyze the types of cut cells generated for Cartesian grids refined at the boundary up to a certain
refinement level r for two realistic engine geometries. The mesh width at refinement level r can be computed
by hr = h2r0 , where h0 defines the maximum spatial extent of the computational domain in x-,y- or z-direction.
The results are displayed in Table 2. For each boundary refinement level, the total number of cut cells Nc
generated is compared with the number of special cut cells with multiple cut surfaces Nmc . The number of
split cells Nspl , i.e., the cells which are divided by the geometry into two or more separate unconnected parts,
related to the Nmc cells is indicated separately. For engine A, no split cells occur and the number of special
cut cells is relatively low. For engine B, however, which has a more detailed and less smooth geometry,
the number of special cut cells is higher and split cells are generated. While for engine A, the number
of multiple-cut-surface cells slightly decreases with increasing refinement level, it increases considerably for
engine B. This result confirms the necessity of an elaborate method to handle arbitrary cut cells.
Table 2. Number of cut cells Nc , absolute number of cut cells with multiple boundary surfaces Nmc , and number of
split cells Nspl belonging to Nmc at different levels of refinement L for two engine geometries.

Engine A

IV.B.

Nc

Nmc

Nspl

7
8
9
10

64814
250513
1002414
3789452

12
8
8
6

0
0
0
0

Engine B

Nc

Nmc

Nspl

7
8
9
10
11

24304
96753
357842
1531610
5229609

20
19
20
38
110

6
6
8
11
44

Flow in internal combustion engines - steady flow cases

Computations of the flow in the two single cylinder engines introduced in the previous section IV.A are
performed successfully for two steady flow cases. Setup I assumes an infinitely long intake stroke, thus
inflow is prescribed at the end of the intake ports and outflow is prescribed in the piston cross section. In
setup II the fluid is sucked in through the combustion chamber via the inlet and outlet ports, thus inflow is
prescribed at the end of the intake ports and outflow is prescribed at the end of the outlet ports. To limit
the problem size, these computations are performed at very low engine speeds for which the flow field in the
combustion chamber should remain laminar. The resulting velocity field for setup I of engine A is shown

(a)

(b)

Figure 15. In-cylinder velocity field for a steady flow case in engine A, setup I. Velocity magnitude, arrows indicate
the flow direction: (a) Re = 660, (b) Re = 2200.

15 of 23
American Institute of Aeronautics and Astronautics

in Fig. 15 for two Reynolds numbers Re = 660 and Re = 2200 based on the mean intake velocity and the
intake port diameter. The formation of two strong ring vortices beneath the inlet valves is obvious, which
Dannemann et al.36 likewise found to be the dominant large-scale structures in this engine operated at a
relevant engine speed of 1500rpm. Since no outlet geometry is available for engine A, the setup II is not
considered for this engine.
The flow fields of engine B at Reynolds numbers Re = 1400 for setup I and Re = 700 for setup II are
depicted in Fig. 16. The intake port of this engine is designed to generate a strong tumble vortex. The
formation of ring vortices beneath the inlet valves is not as pronounced as for engine A. In contrast, the jet
emerging from the upper part of the valve gap is clearly predominant. It leads to the formation of the strong
tumble vortex in a motored setup, when the moving piston is present.

(a)

(b)
Figure 16. In-cylinder velocity field for steady flow cases in engine B. Velocity magnitude, arrows indicate the flow
direction: (a) setup I, Re = 1400, (b) setup II, Re = 700.

16 of 23
American Institute of Aeronautics and Astronautics

IV.C.

Flow through a valve gap during valve closing and opening

To analyze the impact of the discrete closing procedure described in section II.C.3, a simple generic twodimensional valve setup consisting of a valve-like moving object and two bodies forming the valve seat are
considered. The freestream conditions are based on a uniform flow field resulting in a Mach number M = 0.1
and a Reynolds number ReL = 200 based on the valve disk diameter Dvd . The topology of the successively
refined Cartesian grids used for the simulations is shown in Fig. 17. Three different refinement levels in the
vicinity of the valve gap are used. The relevant parameters of the grids are summarized in table 3. The
valve moves with a prescribed sinusoidal oscillation in the x-direction. The location of its centroid is given
by following function:
x = A cos(t) + x0 .
(10)
The quantity A is the amplitude given as 0.1 times the valve disk diameter Dvd = 1, x0 is the initial position
of the valve, and is the angular frequency which is prescribed according to a Strouhal number Sr = 5
based on the freestream velocity u .

Figure 17. Topology of the refined Cartesian grid used for the simulation of the flow through a valve during closing
and opening. Three refinement levels in the vicinity of the valve are used.

Since the resolvable gap width scales proportional to the cell length, the volume enclosed in the gap before
closing should also be reduced proportionally. Fig. 18 shows details of the grids just before the discrete closing
procedure is initiated. The reduction of the minimum resolvable gap width using successively refined grids
is clearly visible. While for grid 1, the valve lift at closing is still a considerable fraction of the maximum
valve lift, the valve lift at which discrete closure is initiated for grid 3 is negligible compared to the maximum
valve lift.
In Fig. 19, the remaining volume and nondimensionalized mass in the valve gap before closing is shown
for all grids. As expected, the gap volume corresponding to the minimum resolvable gap width decreases
almost linearly when the refinement level is increased. The decrease in mass is slightly lower. This can be

(a)

(b)

(c)

Figure 18. Details of the grids in the vicinity of the valve gap before the discrete closing procedure is initiated. (a)
grid 1, (b) grid 2, (c) grid 3.

17 of 23
American Institute of Aeronautics and Astronautics

0.006

volume
mass

0.005
Table 3. Parameters of the grids with increasing maximum refinement
level r. The cell length hmin at the highest refinement level, the minimum resolvable gap width min as well as the number of cells resolving
the valve disk Nvd and the number of cells resolving the length of the
valve gap Ng are given.

grid

hmin

min

Nvd

Ng

1
2
3

12
13
14

0.014648
0.007324
0.003662

0.020716
0.010358
0.005179

68
137
273

10
20
40

0.004
0.003
0.002
0.001
0

grid
Figure 19. Remaining volume and nondimensionalized mass in the valve gap before the
discrete closing procedure is initiated for the
three grids given in Table 3.

explained by the fact that the viscous forces dominate the flow in such narrow gaps. Thus, the fluid density
increases if the reduction of the gap volume due to the valve motion is stronger than the mass flow of the
fluid being squeezed out of the gap.
The flow field at the valve setup during one complete cycle of valve closing and opening is depicted in Fig.
20. In the pictures in the first row, starting from the fully opened position the valve moves in the negative
x-direction until the fully closed state is reached in Fig. 20(d). Thereafter, the valve reopens and moves in
the positive x-direction. The flow between the valve and the valve seat is strongly accelerated when the valve
is closing, thus producing strong jets which form vortical structures downstream of the valve. The vortical
structures persist when the valve has fully closed and no jets exist. During reopening, the fluid is sucked in
the valve gap from both sides and the flow is separated around the pointed corner of the valve disk. When
the valve lift is high enough, the jets in the valve gap are reformed.

(a)

(b)

(c)

(d)

(e)

(f)

Figure 20. Flow through a generic valve: Instantaneous velocity fields for six instants of time during closing and
reopening. Colors show the nondimensionalized velocity magnitude, arrows indicate the flow direction.

18 of 23
American Institute of Aeronautics and Astronautics

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

Figure 21. Flow through a valve: Instantaneous pressure fields for eight instants of time just before and after closing.
Colors and contour lines show the nondimensionalized pressure.

Fig. 21 illustrates the pressure field at the discontinuous closure of the valve. Fig. 21(a) shows the pressure
field just before the closure of the valve is initiated. In the subsequent figures the formation, propagation,
and interaction of the pressure waves initiated by the discrete closure on both sides of the gap are shown.
IV.D.

Flow in a simplified engine geometry - transient flow case including piston and valve
movement

Finally, the flow in a two-dimensional simplified engine geometry in a motored setup is simulated. That
is, the movement of the piston and the valves is included. The engine model consists of four parts and is
equivalent to that shown in Fig. 6. The parts are the valve, the piston, and two side walls which also form
the valve seat and the intake port region. The piston is modeled such that it overlaps the engine walls. The
engine model is embedded in a bigger flow domain to avoid the explicit prescription of an inflow boundary at
the end of the intake port. Initially, the flow in the complete domain is at rest. The movement of the piston
is specified according to equation (10) at an amplitude A = 0.65. The angular frequency is prescribed
such that the Reynolds number based on the maximum piston velocity and the piston diameter is ReP = 90.
This value is chosen relatively low to avoid high refinement levels in the valve gap region in the early intake
and exhaust phases when very high velocities occur in the still narrow valve gaps. The movement of the
valve during the intake and exhaust strokes is prescribed by
x = A sin2 (t) + x0 ,

(11)

with A = 0.2 and the same angular frequency as used for the motion of the piston. This means that the
valve starts to open at 0 crank angle CA, reaches its maximum lift at 90 CA, and the closed position again
at 180 CA.
Fig. 23 shows a time series of instantaneous flow fields in the model engine. Each row corresponds
approximately to one stroke. Starting with the intake stroke, the piston moves downwards and the valve
opens. The flow is sucked in through the valve gap and strong vortices are created downstream of the
valve. The vortices are stretched as the piston moves to the bottom dead center (Figs. 23(e), (f)). During
the compression phase, the vortices become smaller and decay. The expansion phase (Figs. 23(k)-(o)) is
characterized by hardly any vortical motion. When the valve finally reopens again at the beginning of the
exhaust stroke, high velocities occur in the valve gaps. Strong jets are formed and the flow is clearly detached
at the sharp edge of the valve seat.

19 of 23
American Institute of Aeronautics and Astronautics

Figure 22. Topology of the refined Cartesian grid and nondimensionalized lengths used for the simulation of the flow
in a simplified two-dimensional engine model. The engine model is embedded in a bigger flow domain and consists of
four different parts: the valve, the piston, and two side walls also forming the valve seat and intake port.

V.

Summary

An extension of a validated conservative Cartesian cut-cell method for general three-dimensional problems
of compressible viscous flow bounded by complex technical geometries was presented. Validation results
showed that the flow in the vicinity of complex, sharp edges can be more accurately described. The successful
application of the improved method to different steady flow cases in real engine geometries at low engine
speeds yielded promising results. A further extension of this method to problems including moving boundaries
thus enabling the movement of piston and valves was discussed. A validation test case showed convincing
agreement with reference data given in the literature. The method was applied to the flow through a
closing and reopening valve and a simplified two-dimensional engine model under motored conditions, i.e.,
the movement of piston and valves was simulated. In the near future, the present method will be applied
to three-dimensional engine geometries under engine-relevant conditions and engine speeds using large-eddy
simulation.

Acknowledgments
This research was performed as part of the Cluster of Excellence Taylor-Made Fuels from Biomass,
which is funded by the German Excellence Initiative. The support is gratefully acknowledged.

20 of 23
American Institute of Aeronautics and Astronautics

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

(i)

(j)

(k)

(l)

(m)

(n)

(o)

(p)

(q)

(r)

(s)

(t)

Figure 23. Flow in a simplified two-dimensional engine model. A complete motored cycle consisting of intake, compression, expansion, and exhaust stroke. Colors correspond to the nondimensionalized velocity magnitude, arrows indicate
the flow direction.

21 of 23
American Institute of Aeronautics and Astronautics

References
1 Mittal,

R. and Iaccarino, G., Immersed boundary methods, Ann. Rev. Fluid Mech., Vol. 37, 2005, pp. 239261.
C., The immersed boundary method, Acta Numer. 2002 , 2002, pp. 139.
3 Peskin, C., Numerical Analysis of Blood Flow in the Heart, J. Comput. Phys., Vol. 25, 1977, pp. 220252.
4 Barad, M., Colella, P., and Schladow, S., An adaptive cut-cell method for environmental fluid mechanics, Int. J.
Numer. Meth. Fluids, Vol. 60, 2009, pp. 473 514.
5 Cieslak, S., Khelil, S. B., Choquet, I., and Merlen, A., Cut cell strategy for 3-D blast waves numerical simulations,
Shock Waves, Vol. 10, 2001, pp. 421429.
6 Kang, S., Iaccarino, G., and Moin, P., Accurate immersed-boundary reconstructions for viscous flow simulations, AIAA
J., Vol. 47 (7), 2009, pp. 17501760.
7 Meyer, M., Devesa, A., Hickel, S., Hu, X., and Adams, N., A conservative immersed interface method for Large-Eddy
Simulation of incompressible flows, J. Comput. Phys., Vol. 229, 2010, pp. 63006317.
8 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, J. Comput. Phys., Vol. 227, 2008, pp. 48254852.
9 Ghias, R., Mittal, R., and Dong, H., A sharp interface immersed boundary method for compressible viscous flows, J.
Comput. Phys., Vol. 225, 2007, pp. 528553.
10 de Tullio, M., de Palma, P., Iaccarino, G., Pascazio, G., and Napolitano, M., An immersed boundary method for
compressible flows using local grid refinement, J. Comput. Phys., Vol. 225, 2007, pp. 20982117.
11 Hartmann, D., Meinke, M., and Schr
oder, W., A strictly conservative Cartesian cut-cell method for compressible viscous
flows on adaptive grids, Comput. Meth. Appl. Mech. Eng., Vol. 200, 2011, pp. 10381052.
12 Hartmann, D., Meinke, M., and Schr
oder, W., A level-set based adaptive-grid method for premixed combustion,
Combust. Flame, Vol. 158, 2010, pp. 13181339.
13 Hartmann, D., Meinke, M., and Schr
oder, W., The constrained reinitialization equation for level set methods, J.
Comput. Phys., Vol. 229, 2010, pp. 15141535.
14 Lorensen, W. and Cline, H., Marching Cubes: A high resolution 3D surface construction algorithm, Computer Graphics,
Vol. 21, 1987, pp. 163169.
15 Lewiner, T., Lopes, H., Vieira, A. W., and Tavares, G., Efficient implementation of Marching Cubes cases with topological guarantees, J. Graphics Tools, Vol. 8, 2003, pp. 115.
16 G
unther, C., Hartmann, D., Meinke, M., and Schr
oder, W., A level-set based cut-cell method for flows with complex
moving boundaries, Proceedings of the V European Conference on Computational Fluid Dynamics ECCOMAS CFD 2010 ,
edited by J. C. F. Pereira and A. Sequeira, ECCOMAS, Paris, 2010.
17 Hartmann, D., Meinke, M., and Schr
oder, W., Differential Equation Based Constrained Reinitialization for Level Set
Methods, J. Comput. Phys., Vol. 227, 2008, pp. 68216845.
18 Hartmann, D., Meinke, M., and Schr
oder, W., On accuracy and efficiency of constrained reinitialization, Int. J. Numer.
Meth. Fluids, Vol. 63, 2010, pp. 13471358.
19 Hartmann, D., Schneiders, L., Meinke, M., and Schr
oder, W., A numerical method for moving-boundary problems of
compressible viscous flow, Proceedings of the V European Conference on Computational Fluid Dynamics ECCOMAS CFD
2010 , edited by J. C. F. Pereira and A. Sequeira, ECCOMAS, Paris, 2010.
20 Colella, P., Graves, D., Keen, B., and Modiano, D., A Cartesian grid embedded boundary method for hyperbolic
conservation laws, J. Comput. Phys., Vol. 211, 2006, pp. 347366.
21 Pember, R., Bell, J., Colella, P., Crutchfield, W., and Welcome, M., An adaptive Cartesian grid method for unsteady
compressible flow in irregular regions, J. Comput. Phys., Vol. 120, 1995, pp. 278304.
22 Amsden, A. A., KIVA-3V: A Block-Structured KIVA Program for Engines with Vertical or Canted Valves, Tech. Rep.
LA13313-MS, Los Alamos National Lab., NM (United States), 1997.
23 Jasak, J., g. Weller, H., and Nordin, N., In-Cylinder CFD Simulation Using a C++ Object-Oriented Toolkit, SAE
Technical Paper , Vol. 2004-01-0110, 2004.
24 Lucchini, T., DErrico, G., Jasak, H., and Tukovic, Z., Automatic Mesh Motion with Topological Changes for Engine
Simulation, SAE Technical Paper , Vol. 2007-01-0170, 2007.
25 Verzicco, R., Mohd-Yusof, J., Orlandi, P., and Haworth, D., Large Eddy Simulation in Complex Geometric Configurations Using Boundary Body Forces, AIAA Journal, Vol. 38, 2000, pp. 427433.
26 Fadlun, E. A., Verzicco, R., Orlandi, P., and Mohd-Yusof, J., Combined Immersed-Boundary Finite-Difference Methods
for Three-Dimensional Complex Flow Simulations, J. Comput. Phys., Vol. 161, 2000, pp. 3560.
27 Senecal, P. K., Richards, K. J., Pomraning, E., Yang, T., Dai, M. Z., McDavid, R. M., Patterson, M. A., Hou, S., and
Shethaji, T., A New Parallel Cut-Cell Cartesian CFD Code for Rapid Grid Generation Applied to In-Cylinder Diesel Engine
Simulations, SAE Technical Paper , Vol. 2007-01-0159, 2007.
28 Thomas, P. D. and Lombardd, C. K., Geometric Conservation Law and Its Application to Flow Computations on
Moving Grids, AIAA Journal, Vol. 17, 1978, pp. 10301037.
29 Farhat, C., Geuzaine, P., and Grandmont, C., The discrete geometric conservation law and the nonlinear stability of
ALE schemes for the solution of flow problems on moving grids, J. Comput. Phys., Vol. 174, 2001, pp. 669694.
30 Bayyuk, S. A., Computation of inviscid compressible flows about arbitrary geometries and moving boundaries, Ph.D.
thesis, University of Michigan, 2008.
31 Hartmann, D., Meinke, M., and Schr
oder, W., An adaptive multilevel multigrid formulation for Cartesian hierarchical
grid methods, Comput. Fluids, Vol. 37, 2008, pp. 11031125.
32 Saha, A., Three-dimensional numerical simulations of the transition of flow past a cube, Phys. Fluids, Vol. 16, 2004,
pp. 16301646.
2 Peskin,

22 of 23
American Institute of Aeronautics and Astronautics

33 Hartmann, D., Schneiders, L., Schr


oder, W., and Shashikanth, B. N., On the Interaction of a Vortex Pair with a Freely
Moving Cylinder, AIAA Paper , Vol. 2010-4749, 2010.
34 Ahn, H. T. and Kallinderis, Y., Strongly coupled flow/structure interactions with a geometrically conservative ALE
scheme on general hybrid meshes, J. Comput. Phys., Vol. 219, 2006, pp. 671696.
35 Borazjani, I., Ge, L., and Sotiropoulos, F., Curvilinear immersed boundary method for simulating fluid structure
interaction with complex 3d rigid bodies, J. Comput. Phys., Vol. 227, 2008, pp. 75877620.
36 Dannemann, J., Pielhop, K., Klaas, M., and Schr
oder, W., Cycle resolved multi-planar flow measurements in a four-valve
combustion engine, Experiments in Fluids, Vol. 50, 2011, pp. 961976.

23 of 23
American Institute of Aeronautics and Astronautics

You might also like