5 views

Uploaded by ashoku2

g

- 12_piatkowski_chosen_aspects_4_2015.pdf
- v52-9
- 2003 28 Maintenance
- Engineering_Calculations Web Sight
- FM Lecture (Week13)
- E-Tech piston rings
- Assign 0708 Topics
- Contherm® Core Scraped-Surface Heat Exchanger
- Synopsis Ssss
- EUROPLEXU
- Freed(1975)_Huggins coefficient for the viscosity of polymer solutions.pdf
- York Compressors
- 73 Thermal (9)
- sdpreport-151109082415-lva1-app6892
- AmsoilBreak-In Oil (SAE 30) g2881
- Paper Water Hammer
- Chapter 5
- 01 Resonance-Free SWATH
- Computer Application in Hydraulic Engineering-7th Edition
- ttt

You are on page 1of 23

27 - 30 June 2011, Honolulu, Hawaii

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

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.

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.

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.

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

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.

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.

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.

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.

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.

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)

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, chamfered edge

fine, sharp edge

fine, chamfered edge

0.03

0.02

0.01

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

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

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

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.

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

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

- 12_piatkowski_chosen_aspects_4_2015.pdfUploaded byrenatoeliege
- v52-9Uploaded byMeBituman
- 2003 28 MaintenanceUploaded byvishal.nitham
- Engineering_Calculations Web SightUploaded byatharva_surfotech
- FM Lecture (Week13)Uploaded byMahtab Ilyas
- E-Tech piston ringsUploaded byCopyMaster2
- Assign 0708 TopicsUploaded bygetsweet
- Contherm® Core Scraped-Surface Heat ExchangerUploaded byKhánh Ngân
- Synopsis SsssUploaded byKonul Alizada
- EUROPLEXUUploaded byMark Guo
- Freed(1975)_Huggins coefficient for the viscosity of polymer solutions.pdfUploaded byhume975
- York CompressorsUploaded byFernando Huaroc
- 73 Thermal (9)Uploaded byShashank Dubey
- sdpreport-151109082415-lva1-app6892Uploaded byTechnos_Inc
- AmsoilBreak-In Oil (SAE 30) g2881Uploaded byamsoilromania
- Paper Water HammerUploaded byPatel Amit
- Chapter 5Uploaded bysptbala
- 01 Resonance-Free SWATHUploaded byIvan Nilsen
- Computer Application in Hydraulic Engineering-7th EditionUploaded bykawser zaman
- tttUploaded byrpraj3135
- AerodynamicsUploaded byGaurav Chaturvedi
- 58103-mt----advanced transport phenomenaUploaded bySRINIVASA RAO GANTA
- TutorswfUploaded byAnghelo Alcalde
- ME200-Fluid Mechanics and Machinery (1)Uploaded byvishnu
- BernoulliUploaded bymohdyashfi mpp
- Catalog CourseUploaded bymsyarifudin
- 2_HydrodynamicTheoryUploaded bytwinkytom
- AssignementUploaded byPool King
- fluentUploaded byJeremy Hernandez
- Prerequisites QuizUploaded byedmundac

- Thermal Conductivity of Hot Plate MethodUploaded byashoku2
- Machine Dynamics Question Bank1Uploaded byashoku2
- Cse Ext ResultUploaded byashoku2
- Natural Convection HT07.docUploaded byashoku2
- Natural Convection HT07.docUploaded byashoku2
- 566343049494905Uploaded byashoku2
- Common Rail Direct InjectionUploaded byashoku2
- Gen Terms - Std. TEC2016.docUploaded byashoku2
- Gen Terms - Std. TEC2016.docUploaded byashoku2
- Ht-10 Heat Transfer in Natural ConvectionUploaded byashoku2
- SolidsUploaded byashoku2
- SolidsUploaded byashoku2
- Manufacturing Technologyone MarksUploaded byashoku2
- Manufacturing Processess Assinment -II October -2016Uploaded byashoku2
- Manufacturing Technologyone Marks AaUploaded byashoku2
- Factors for Calculating DepreciationUploaded byashoku2
- Engineering Mechanics_A. K. Tayal (1)Uploaded byashoku2
- Nova Educational Society - Contact UsUploaded byashoku2
- ME1001(1).pdfUploaded byashoku2
- Pricelist.xlsxUploaded byashoku2
- Artigo_13_05_05Uploaded byashoku2
- CFX12_03_Physics1Uploaded byashoku2
- KotaUploaded byPradeep Mainrai
- COHEN-SUTHERLAND Algorithm - Learn With ShamimUploaded byashoku2
- Renewable Energy Harvesting by Vortex-Induced Motions Review and Benchmarking of ThechnologiesUploaded byashoku2
- 109ME0417Uploaded byashoku2
- CFD Analysis of Different Bluff Bodies-456 (1).pdfUploaded byArup Das
- Sponsorship Form1Uploaded byashoku2
- Electrical EngineeringUploaded byRahul Agarwal
- A1S1 2017Uploaded byashoku2

- PSE4 Lecture Ch32Uploaded byRuthwik Kuruturi
- Concrete Flow Using Rheological parameters.pdfUploaded byPanagiotis Stamatis
- Newton vs LeibnizUploaded byMax Mejía Dextre
- Class03 ChemistryG12 Notes and HomeworkUploaded byAndy Rei Kou
- Thermodynamics (Ch. 16) AP ProblemsUploaded byHasantha Perera
- Solutions for Electic ChargeUploaded byBreanna Venable
- ECE_105_FNL_Fall_2014_final.pdfUploaded byYuChenQian
- ASVAB Mechanical Comprehension Practice Test 2Uploaded byASVABTestBank
- TATA Blue BookUploaded bymrswcecivil
- 201805 FHSC1014 Mechanics Tutorial 6 S (1)Uploaded bySteve Thaw
- James Prescott JouleUploaded byAndruşa Ioana
- DESIGN AND STRESS ANALYSIS OF AIRCRAFT STRUCTUREUploaded byKugan Kugan Madiwanan
- r5310105-structural analysis - iiUploaded bysivabharathamurthy
- TEMA 6bUploaded byEmilio Sordo Zabay
- M C DeoUploaded byumar_contact
- SG2215 Exam Questions.pdfUploaded byStephanie Edwards
- PRESENTATION SHAFT AND AXLES.pptxUploaded byAfiq Bobby
- Tan Thesis Defense 0814 FinalUploaded byRadhika Rebellious
- n Body ProblemUploaded byTeka Kam
- On Site Maintenance of Strategic Rotating Machines - May 2009.pdfUploaded byHernan Giraut
- Laboratory 5 - EntropyUploaded byzaqbasalmau
- assign3 (1)Uploaded byEXAMENQUIMESTRAL
- Center of MassUploaded bytimun
- VectorsUploaded byTanu Sharma
- Determination of Metracentric HeightUploaded byArbel Acuril
- may 17r.pdfUploaded byTayyaba Mumtaz Khan
- Corbel ProgramUploaded byShem Barro
- Naval Architecture Equation Summary SheetUploaded byAnonymous JSHUTp
- L. Sinclair Instructional Design Key AssessmentUploaded byllsinclair
- INTRO.pdfUploaded byMelmar Jangerald Sabio